net: ena: Add a driver for Amazon Elastic Network Adapters (ENA)
authorNetanel Belgazal <netanel@annapurnalabs.com>
Wed, 10 Aug 2016 11:03:22 +0000 (14:03 +0300)
committerDavid S. Miller <davem@davemloft.net>
Sat, 13 Aug 2016 00:12:08 +0000 (17:12 -0700)
This is a driver for the ENA family of networking devices.

Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 files changed:
Documentation/networking/00-INDEX
Documentation/networking/ena.txt [new file with mode: 0644]
MAINTAINERS
drivers/net/ethernet/Kconfig
drivers/net/ethernet/Makefile
drivers/net/ethernet/amazon/Kconfig [new file with mode: 0644]
drivers/net/ethernet/amazon/Makefile [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/Makefile [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_admin_defs.h [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_com.c [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_com.h [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_common_defs.h [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_eth_com.c [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_eth_com.h [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_eth_io_defs.h [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_ethtool.c [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_netdev.c [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_netdev.h [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_pci_id_tbl.h [new file with mode: 0644]
drivers/net/ethernet/amazon/ena/ena_regs_defs.h [new file with mode: 0644]

index 415154a..a769778 100644 (file)
@@ -74,6 +74,8 @@ dns_resolver.txt
        - The DNS resolver module allows kernel servies to make DNS queries.
 driver.txt
        - Softnet driver issues.
+ena.txt
+       - info on Amazon's Elastic Network Adapter (ENA)
 e100.txt
        - info on Intel's EtherExpress PRO/100 line of 10/100 boards
 e1000.txt
diff --git a/Documentation/networking/ena.txt b/Documentation/networking/ena.txt
new file mode 100644 (file)
index 0000000..2b4b6f5
--- /dev/null
@@ -0,0 +1,305 @@
+Linux kernel driver for Elastic Network Adapter (ENA) family:
+=============================================================
+
+Overview:
+=========
+ENA is a networking interface designed to make good use of modern CPU
+features and system architectures.
+
+The ENA device exposes a lightweight management interface with a
+minimal set of memory mapped registers and extendable command set
+through an Admin Queue.
+
+The driver supports a range of ENA devices, is link-speed independent
+(i.e., the same driver is used for 10GbE, 25GbE, 40GbE, etc.), and has
+a negotiated and extendable feature set.
+
+Some ENA devices support SR-IOV. This driver is used for both the
+SR-IOV Physical Function (PF) and Virtual Function (VF) devices.
+
+ENA devices enable high speed and low overhead network traffic
+processing by providing multiple Tx/Rx queue pairs (the maximum number
+is advertised by the device via the Admin Queue), a dedicated MSI-X
+interrupt vector per Tx/Rx queue pair, adaptive interrupt moderation,
+and CPU cacheline optimized data placement.
+
+The ENA driver supports industry standard TCP/IP offload features such
+as checksum offload and TCP transmit segmentation offload (TSO).
+Receive-side scaling (RSS) is supported for multi-core scaling.
+
+The ENA driver and its corresponding devices implement health
+monitoring mechanisms such as watchdog, enabling the device and driver
+to recover in a manner transparent to the application, as well as
+debug logs.
+
+Some of the ENA devices support a working mode called Low-latency
+Queue (LLQ), which saves several more microseconds.
+
+Supported PCI vendor ID/device IDs:
+===================================
+1d0f:0ec2 - ENA PF
+1d0f:1ec2 - ENA PF with LLQ support
+1d0f:ec20 - ENA VF
+1d0f:ec21 - ENA VF with LLQ support
+
+ENA Source Code Directory Structure:
+====================================
+ena_com.[ch]      - Management communication layer. This layer is
+                    responsible for the handling all the management
+                    (admin) communication between the device and the
+                    driver.
+ena_eth_com.[ch]  - Tx/Rx data path.
+ena_admin_defs.h  - Definition of ENA management interface.
+ena_eth_io_defs.h - Definition of ENA data path interface.
+ena_common_defs.h - Common definitions for ena_com layer.
+ena_regs_defs.h   - Definition of ENA PCI memory-mapped (MMIO) registers.
+ena_netdev.[ch]   - Main Linux kernel driver.
+ena_syfsfs.[ch]   - Sysfs files.
+ena_ethtool.c     - ethtool callbacks.
+ena_pci_id_tbl.h  - Supported device IDs.
+
+Management Interface:
+=====================
+ENA management interface is exposed by means of:
+- PCIe Configuration Space
+- Device Registers
+- Admin Queue (AQ) and Admin Completion Queue (ACQ)
+- Asynchronous Event Notification Queue (AENQ)
+
+ENA device MMIO Registers are accessed only during driver
+initialization and are not involved in further normal device
+operation.
+
+AQ is used for submitting management commands, and the
+results/responses are reported asynchronously through ACQ.
+
+ENA introduces a very small set of management commands with room for
+vendor-specific extensions. Most of the management operations are
+framed in a generic Get/Set feature command.
+
+The following admin queue commands are supported:
+- Create I/O submission queue
+- Create I/O completion queue
+- Destroy I/O submission queue
+- Destroy I/O completion queue
+- Get feature
+- Set feature
+- Configure AENQ
+- Get statistics
+
+Refer to ena_admin_defs.h for the list of supported Get/Set Feature
+properties.
+
+The Asynchronous Event Notification Queue (AENQ) is a uni-directional
+queue used by the ENA device to send to the driver events that cannot
+be reported using ACQ. AENQ events are subdivided into groups. Each
+group may have multiple syndromes, as shown below
+
+The events are:
+       Group                   Syndrome
+       Link state change       - X -
+       Fatal error             - X -
+       Notification            Suspend traffic
+       Notification            Resume traffic
+       Keep-Alive              - X -
+
+ACQ and AENQ share the same MSI-X vector.
+
+Keep-Alive is a special mechanism that allows monitoring of the
+device's health. The driver maintains a watchdog (WD) handler which,
+if fired, logs the current state and statistics then resets and
+restarts the ENA device and driver. A Keep-Alive event is delivered by
+the device every second. The driver re-arms the WD upon reception of a
+Keep-Alive event. A missed Keep-Alive event causes the WD handler to
+fire.
+
+Data Path Interface:
+====================
+I/O operations are based on Tx and Rx Submission Queues (Tx SQ and Rx
+SQ correspondingly). Each SQ has a completion queue (CQ) associated
+with it.
+
+The SQs and CQs are implemented as descriptor rings in contiguous
+physical memory.
+
+The ENA driver supports two Queue Operation modes for Tx SQs:
+- Regular mode
+  * In this mode the Tx SQs reside in the host's memory. The ENA
+    device fetches the ENA Tx descriptors and packet data from host
+    memory.
+- Low Latency Queue (LLQ) mode or "push-mode".
+  * In this mode the driver pushes the transmit descriptors and the
+    first 128 bytes of the packet directly to the ENA device memory
+    space. The rest of the packet payload is fetched by the
+    device. For this operation mode, the driver uses a dedicated PCI
+    device memory BAR, which is mapped with write-combine capability.
+
+The Rx SQs support only the regular mode.
+
+Note: Not all ENA devices support LLQ, and this feature is negotiated
+      with the device upon initialization. If the ENA device does not
+      support LLQ mode, the driver falls back to the regular mode.
+
+The driver supports multi-queue for both Tx and Rx. This has various
+benefits:
+- Reduced CPU/thread/process contention on a given Ethernet interface.
+- Cache miss rate on completion is reduced, particularly for data
+  cache lines that hold the sk_buff structures.
+- Increased process-level parallelism when handling received packets.
+- Increased data cache hit rate, by steering kernel processing of
+  packets to the CPU, where the application thread consuming the
+  packet is running.
+- In hardware interrupt re-direction.
+
+Interrupt Modes:
+================
+The driver assigns a single MSI-X vector per queue pair (for both Tx
+and Rx directions). The driver assigns an additional dedicated MSI-X vector
+for management (for ACQ and AENQ).
+
+Management interrupt registration is performed when the Linux kernel
+probes the adapter, and it is de-registered when the adapter is
+removed. I/O queue interrupt registration is performed when the Linux
+interface of the adapter is opened, and it is de-registered when the
+interface is closed.
+
+The management interrupt is named:
+   ena-mgmnt@pci:<PCI domain:bus:slot.function>
+and for each queue pair, an interrupt is named:
+   <interface name>-Tx-Rx-<queue index>
+
+The ENA device operates in auto-mask and auto-clear interrupt
+modes. That is, once MSI-X is delivered to the host, its Cause bit is
+automatically cleared and the interrupt is masked. The interrupt is
+unmasked by the driver after NAPI processing is complete.
+
+Interrupt Moderation:
+=====================
+ENA driver and device can operate in conventional or adaptive interrupt
+moderation mode.
+
+In conventional mode the driver instructs device to postpone interrupt
+posting according to static interrupt delay value. The interrupt delay
+value can be configured through ethtool(8). The following ethtool
+parameters are supported by the driver: tx-usecs, rx-usecs
+
+In adaptive interrupt moderation mode the interrupt delay value is
+updated by the driver dynamically and adjusted every NAPI cycle
+according to the traffic nature.
+
+By default ENA driver applies adaptive coalescing on Rx traffic and
+conventional coalescing on Tx traffic.
+
+Adaptive coalescing can be switched on/off through ethtool(8)
+adaptive_rx on|off parameter.
+
+The driver chooses interrupt delay value according to the number of
+bytes and packets received between interrupt unmasking and interrupt
+posting. The driver uses interrupt delay table that subdivides the
+range of received bytes/packets into 5 levels and assigns interrupt
+delay value to each level.
+
+The user can enable/disable adaptive moderation, modify the interrupt
+delay table and restore its default values through sysfs.
+
+The rx_copybreak is initialized by default to ENA_DEFAULT_RX_COPYBREAK
+and can be configured by the ETHTOOL_STUNABLE command of the
+SIOCETHTOOL ioctl.
+
+SKB:
+The driver-allocated SKB for frames received from Rx handling using
+NAPI context. The allocation method depends on the size of the packet.
+If the frame length is larger than rx_copybreak, napi_get_frags()
+is used, otherwise netdev_alloc_skb_ip_align() is used, the buffer
+content is copied (by CPU) to the SKB, and the buffer is recycled.
+
+Statistics:
+===========
+The user can obtain ENA device and driver statistics using ethtool.
+The driver can collect regular or extended statistics (including
+per-queue stats) from the device.
+
+In addition the driver logs the stats to syslog upon device reset.
+
+MTU:
+====
+The driver supports an arbitrarily large MTU with a maximum that is
+negotiated with the device. The driver configures MTU using the
+SetFeature command (ENA_ADMIN_MTU property). The user can change MTU
+via ip(8) and similar legacy tools.
+
+Stateless Offloads:
+===================
+The ENA driver supports:
+- TSO over IPv4/IPv6
+- TSO with ECN
+- IPv4 header checksum offload
+- TCP/UDP over IPv4/IPv6 checksum offloads
+
+RSS:
+====
+- The ENA device supports RSS that allows flexible Rx traffic
+  steering.
+- Toeplitz and CRC32 hash functions are supported.
+- Different combinations of L2/L3/L4 fields can be configured as
+  inputs for hash functions.
+- The driver configures RSS settings using the AQ SetFeature command
+  (ENA_ADMIN_RSS_HASH_FUNCTION, ENA_ADMIN_RSS_HASH_INPUT and
+  ENA_ADMIN_RSS_REDIRECTION_TABLE_CONFIG properties).
+- If the NETIF_F_RXHASH flag is set, the 32-bit result of the hash
+  function delivered in the Rx CQ descriptor is set in the received
+  SKB.
+- The user can provide a hash key, hash function, and configure the
+  indirection table through ethtool(8).
+
+DATA PATH:
+==========
+Tx:
+---
+end_start_xmit() is called by the stack. This function does the following:
+- Maps data buffers (skb->data and frags).
+- Populates ena_buf for the push buffer (if the driver and device are
+  in push mode.)
+- Prepares ENA bufs for the remaining frags.
+- Allocates a new request ID from the empty req_id ring. The request
+  ID is the index of the packet in the Tx info. This is used for
+  out-of-order TX completions.
+- Adds the packet to the proper place in the Tx ring.
+- Calls ena_com_prepare_tx(), an ENA communication layer that converts
+  the ena_bufs to ENA descriptors (and adds meta ENA descriptors as
+  needed.)
+  * This function also copies the ENA descriptors and the push buffer
+    to the Device memory space (if in push mode.)
+- Writes doorbell to the ENA device.
+- When the ENA device finishes sending the packet, a completion
+  interrupt is raised.
+- The interrupt handler schedules NAPI.
+- The ena_clean_tx_irq() function is called. This function handles the
+  completion descriptors generated by the ENA, with a single
+  completion descriptor per completed packet.
+  * req_id is retrieved from the completion descriptor. The tx_info of
+    the packet is retrieved via the req_id. The data buffers are
+    unmapped and req_id is returned to the empty req_id ring.
+  * The function stops when the completion descriptors are completed or
+    the budget is reached.
+
+Rx:
+---
+- When a packet is received from the ENA device.
+- The interrupt handler schedules NAPI.
+- The ena_clean_rx_irq() function is called. This function calls
+  ena_rx_pkt(), an ENA communication layer function, which returns the
+  number of descriptors used for a new unhandled packet, and zero if
+  no new packet is found.
+- Then it calls the ena_clean_rx_irq() function.
+- ena_eth_rx_skb() checks packet length:
+  * If the packet is small (len < rx_copybreak), the driver allocates
+    a SKB for the new packet, and copies the packet payload into the
+    SKB data buffer.
+    - In this way the original data buffer is not passed to the stack
+      and is reused for future Rx packets.
+  * Otherwise the function unmaps the Rx buffer, then allocates the
+    new SKB structure and hooks the Rx buffer to the SKB frags.
+- The new SKB is updated with the necessary information (protocol,
+  checksum hw verify result, etc.), and then passed to the network
+  stack, using the NAPI interface function napi_gro_receive().
index 429fc61..50ce33d 100644 (file)
@@ -636,6 +636,15 @@ F: drivers/tty/serial/altera_jtaguart.c
 F:     include/linux/altera_uart.h
 F:     include/linux/altera_jtaguart.h
 
+AMAZON ETHERNET DRIVERS
+M:     Netanel Belgazal <netanel@annapurnalabs.com>
+R:     Saeed Bishara <saeed@annapurnalabs.com>
+R:     Zorik Machulsky <zorik@annapurnalabs.com>
+L:     netdev@vger.kernel.org
+S:     Supported
+F:     Documentation/networking/ena.txt
+F:     drivers/net/ethernet/amazon/
+
 AMD CRYPTOGRAPHIC COPROCESSOR (CCP) DRIVER
 M:     Tom Lendacky <thomas.lendacky@amd.com>
 M:     Gary Hook <gary.hook@amd.com>
index 2ffd634..8cc7467 100644 (file)
@@ -24,6 +24,7 @@ source "drivers/net/ethernet/agere/Kconfig"
 source "drivers/net/ethernet/allwinner/Kconfig"
 source "drivers/net/ethernet/alteon/Kconfig"
 source "drivers/net/ethernet/altera/Kconfig"
+source "drivers/net/ethernet/amazon/Kconfig"
 source "drivers/net/ethernet/amd/Kconfig"
 source "drivers/net/ethernet/apm/Kconfig"
 source "drivers/net/ethernet/apple/Kconfig"
index 1d349e9..a09423d 100644 (file)
@@ -10,6 +10,7 @@ obj-$(CONFIG_NET_VENDOR_AGERE) += agere/
 obj-$(CONFIG_NET_VENDOR_ALLWINNER) += allwinner/
 obj-$(CONFIG_NET_VENDOR_ALTEON) += alteon/
 obj-$(CONFIG_ALTERA_TSE) += altera/
+obj-$(CONFIG_NET_VENDOR_AMAZON) += amazon/
 obj-$(CONFIG_NET_VENDOR_AMD) += amd/
 obj-$(CONFIG_NET_XGENE) += apm/
 obj-$(CONFIG_NET_VENDOR_APPLE) += apple/
diff --git a/drivers/net/ethernet/amazon/Kconfig b/drivers/net/ethernet/amazon/Kconfig
new file mode 100644 (file)
index 0000000..99b3035
--- /dev/null
@@ -0,0 +1,27 @@
+#
+# Amazon network device configuration
+#
+
+config NET_VENDOR_AMAZON
+       bool "Amazon Devices"
+       default y
+       ---help---
+         If you have a network (Ethernet) device belonging to this class, say Y.
+
+         Note that the answer to this question doesn't directly affect the
+         kernel: saying N will just cause the configurator to skip all
+         the questions about Amazon devices. If you say Y, you will be asked
+         for your specific device in the following questions.
+
+if NET_VENDOR_AMAZON
+
+config ENA_ETHERNET
+       tristate "Elastic Network Adapter (ENA) support"
+       depends on (PCI_MSI && X86)
+       ---help---
+         This driver supports Elastic Network Adapter (ENA)"
+
+         To compile this driver as a module, choose M here.
+         The module will be called ena.
+
+endif #NET_VENDOR_AMAZON
diff --git a/drivers/net/ethernet/amazon/Makefile b/drivers/net/ethernet/amazon/Makefile
new file mode 100644 (file)
index 0000000..8e0b73f
--- /dev/null
@@ -0,0 +1,5 @@
+#
+# Makefile for the Amazon network device drivers.
+#
+
+obj-$(CONFIG_ENA_ETHERNET) += ena/
diff --git a/drivers/net/ethernet/amazon/ena/Makefile b/drivers/net/ethernet/amazon/ena/Makefile
new file mode 100644 (file)
index 0000000..eaeeae0
--- /dev/null
@@ -0,0 +1,7 @@
+#
+# Makefile for the Elastic Network Adapter (ENA) device drivers.
+#
+
+obj-$(CONFIG_ENA_ETHERNET) += ena.o
+
+ena-y := ena_netdev.o ena_com.o ena_eth_com.o ena_ethtool.o
diff --git a/drivers/net/ethernet/amazon/ena/ena_admin_defs.h b/drivers/net/ethernet/amazon/ena/ena_admin_defs.h
new file mode 100644 (file)
index 0000000..a46e749
--- /dev/null
@@ -0,0 +1,973 @@
+/*
+ * Copyright 2015 - 2016 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _ENA_ADMIN_H_
+#define _ENA_ADMIN_H_
+
+enum ena_admin_aq_opcode {
+       ENA_ADMIN_CREATE_SQ     = 1,
+
+       ENA_ADMIN_DESTROY_SQ    = 2,
+
+       ENA_ADMIN_CREATE_CQ     = 3,
+
+       ENA_ADMIN_DESTROY_CQ    = 4,
+
+       ENA_ADMIN_GET_FEATURE   = 8,
+
+       ENA_ADMIN_SET_FEATURE   = 9,
+
+       ENA_ADMIN_GET_STATS     = 11,
+};
+
+enum ena_admin_aq_completion_status {
+       ENA_ADMIN_SUCCESS                       = 0,
+
+       ENA_ADMIN_RESOURCE_ALLOCATION_FAILURE   = 1,
+
+       ENA_ADMIN_BAD_OPCODE                    = 2,
+
+       ENA_ADMIN_UNSUPPORTED_OPCODE            = 3,
+
+       ENA_ADMIN_MALFORMED_REQUEST             = 4,
+
+       /* Additional status is provided in ACQ entry extended_status */
+       ENA_ADMIN_ILLEGAL_PARAMETER             = 5,
+
+       ENA_ADMIN_UNKNOWN_ERROR                 = 6,
+};
+
+enum ena_admin_aq_feature_id {
+       ENA_ADMIN_DEVICE_ATTRIBUTES             = 1,
+
+       ENA_ADMIN_MAX_QUEUES_NUM                = 2,
+
+       ENA_ADMIN_RSS_HASH_FUNCTION             = 10,
+
+       ENA_ADMIN_STATELESS_OFFLOAD_CONFIG      = 11,
+
+       ENA_ADMIN_RSS_REDIRECTION_TABLE_CONFIG  = 12,
+
+       ENA_ADMIN_MTU                           = 14,
+
+       ENA_ADMIN_RSS_HASH_INPUT                = 18,
+
+       ENA_ADMIN_INTERRUPT_MODERATION          = 20,
+
+       ENA_ADMIN_AENQ_CONFIG                   = 26,
+
+       ENA_ADMIN_LINK_CONFIG                   = 27,
+
+       ENA_ADMIN_HOST_ATTR_CONFIG              = 28,
+
+       ENA_ADMIN_FEATURES_OPCODE_NUM           = 32,
+};
+
+enum ena_admin_placement_policy_type {
+       /* descriptors and headers are in host memory */
+       ENA_ADMIN_PLACEMENT_POLICY_HOST = 1,
+
+       /* descriptors and headers are in device memory (a.k.a Low Latency
+        * Queue)
+        */
+       ENA_ADMIN_PLACEMENT_POLICY_DEV  = 3,
+};
+
+enum ena_admin_link_types {
+       ENA_ADMIN_LINK_SPEED_1G         = 0x1,
+
+       ENA_ADMIN_LINK_SPEED_2_HALF_G   = 0x2,
+
+       ENA_ADMIN_LINK_SPEED_5G         = 0x4,
+
+       ENA_ADMIN_LINK_SPEED_10G        = 0x8,
+
+       ENA_ADMIN_LINK_SPEED_25G        = 0x10,
+
+       ENA_ADMIN_LINK_SPEED_40G        = 0x20,
+
+       ENA_ADMIN_LINK_SPEED_50G        = 0x40,
+
+       ENA_ADMIN_LINK_SPEED_100G       = 0x80,
+
+       ENA_ADMIN_LINK_SPEED_200G       = 0x100,
+
+       ENA_ADMIN_LINK_SPEED_400G       = 0x200,
+};
+
+enum ena_admin_completion_policy_type {
+       /* completion queue entry for each sq descriptor */
+       ENA_ADMIN_COMPLETION_POLICY_DESC                = 0,
+
+       /* completion queue entry upon request in sq descriptor */
+       ENA_ADMIN_COMPLETION_POLICY_DESC_ON_DEMAND      = 1,
+
+       /* current queue head pointer is updated in OS memory upon sq
+        * descriptor request
+        */
+       ENA_ADMIN_COMPLETION_POLICY_HEAD_ON_DEMAND      = 2,
+
+       /* current queue head pointer is updated in OS memory for each sq
+        * descriptor
+        */
+       ENA_ADMIN_COMPLETION_POLICY_HEAD                = 3,
+};
+
+/* basic stats return ena_admin_basic_stats while extanded stats return a
+ * buffer (string format) with additional statistics per queue and per
+ * device id
+ */
+enum ena_admin_get_stats_type {
+       ENA_ADMIN_GET_STATS_TYPE_BASIC          = 0,
+
+       ENA_ADMIN_GET_STATS_TYPE_EXTENDED       = 1,
+};
+
+enum ena_admin_get_stats_scope {
+       ENA_ADMIN_SPECIFIC_QUEUE        = 0,
+
+       ENA_ADMIN_ETH_TRAFFIC           = 1,
+};
+
+struct ena_admin_aq_common_desc {
+       /* 11:0 : command_id
+        * 15:12 : reserved12
+        */
+       u16 command_id;
+
+       /* as appears in ena_admin_aq_opcode */
+       u8 opcode;
+
+       /* 0 : phase
+        * 1 : ctrl_data - control buffer address valid
+        * 2 : ctrl_data_indirect - control buffer address
+        *    points to list of pages with addresses of control
+        *    buffers
+        * 7:3 : reserved3
+        */
+       u8 flags;
+};
+
+/* used in ena_admin_aq_entry. Can point directly to control data, or to a
+ * page list chunk. Used also at the end of indirect mode page list chunks,
+ * for chaining.
+ */
+struct ena_admin_ctrl_buff_info {
+       u32 length;
+
+       struct ena_common_mem_addr address;
+};
+
+struct ena_admin_sq {
+       u16 sq_idx;
+
+       /* 4:0 : reserved
+        * 7:5 : sq_direction - 0x1 - Tx; 0x2 - Rx
+        */
+       u8 sq_identity;
+
+       u8 reserved1;
+};
+
+struct ena_admin_aq_entry {
+       struct ena_admin_aq_common_desc aq_common_descriptor;
+
+       union {
+               u32 inline_data_w1[3];
+
+               struct ena_admin_ctrl_buff_info control_buffer;
+       } u;
+
+       u32 inline_data_w4[12];
+};
+
+struct ena_admin_acq_common_desc {
+       /* command identifier to associate it with the aq descriptor
+        * 11:0 : command_id
+        * 15:12 : reserved12
+        */
+       u16 command;
+
+       u8 status;
+
+       /* 0 : phase
+        * 7:1 : reserved1
+        */
+       u8 flags;
+
+       u16 extended_status;
+
+       /* serves as a hint what AQ entries can be revoked */
+       u16 sq_head_indx;
+};
+
+struct ena_admin_acq_entry {
+       struct ena_admin_acq_common_desc acq_common_descriptor;
+
+       u32 response_specific_data[14];
+};
+
+struct ena_admin_aq_create_sq_cmd {
+       struct ena_admin_aq_common_desc aq_common_descriptor;
+
+       /* 4:0 : reserved0_w1
+        * 7:5 : sq_direction - 0x1 - Tx, 0x2 - Rx
+        */
+       u8 sq_identity;
+
+       u8 reserved8_w1;
+
+       /* 3:0 : placement_policy - Describing where the SQ
+        *    descriptor ring and the SQ packet headers reside:
+        *    0x1 - descriptors and headers are in OS memory,
+        *    0x3 - descriptors and headers in device memory
+        *    (a.k.a Low Latency Queue)
+        * 6:4 : completion_policy - Describing what policy
+        *    to use for generation completion entry (cqe) in
+        *    the CQ associated with this SQ: 0x0 - cqe for each
+        *    sq descriptor, 0x1 - cqe upon request in sq
+        *    descriptor, 0x2 - current queue head pointer is
+        *    updated in OS memory upon sq descriptor request
+        *    0x3 - current queue head pointer is updated in OS
+        *    memory for each sq descriptor
+        * 7 : reserved15_w1
+        */
+       u8 sq_caps_2;
+
+       /* 0 : is_physically_contiguous - Described if the
+        *    queue ring memory is allocated in physical
+        *    contiguous pages or split.
+        * 7:1 : reserved17_w1
+        */
+       u8 sq_caps_3;
+
+       /* associated completion queue id. This CQ must be created prior to
+        *    SQ creation
+        */
+       u16 cq_idx;
+
+       /* submission queue depth in entries */
+       u16 sq_depth;
+
+       /* SQ physical base address in OS memory. This field should not be
+        * used for Low Latency queues. Has to be page aligned.
+        */
+       struct ena_common_mem_addr sq_ba;
+
+       /* specifies queue head writeback location in OS memory. Valid if
+        * completion_policy is set to completion_policy_head_on_demand or
+        * completion_policy_head. Has to be cache aligned
+        */
+       struct ena_common_mem_addr sq_head_writeback;
+
+       u32 reserved0_w7;
+
+       u32 reserved0_w8;
+};
+
+enum ena_admin_sq_direction {
+       ENA_ADMIN_SQ_DIRECTION_TX       = 1,
+
+       ENA_ADMIN_SQ_DIRECTION_RX       = 2,
+};
+
+struct ena_admin_acq_create_sq_resp_desc {
+       struct ena_admin_acq_common_desc acq_common_desc;
+
+       u16 sq_idx;
+
+       u16 reserved;
+
+       /* queue doorbell address as an offset to PCIe MMIO REG BAR */
+       u32 sq_doorbell_offset;
+
+       /* low latency queue ring base address as an offset to PCIe MMIO
+        * LLQ_MEM BAR
+        */
+       u32 llq_descriptors_offset;
+
+       /* low latency queue headers' memory as an offset to PCIe MMIO
+        * LLQ_MEM BAR
+        */
+       u32 llq_headers_offset;
+};
+
+struct ena_admin_aq_destroy_sq_cmd {
+       struct ena_admin_aq_common_desc aq_common_descriptor;
+
+       struct ena_admin_sq sq;
+};
+
+struct ena_admin_acq_destroy_sq_resp_desc {
+       struct ena_admin_acq_common_desc acq_common_desc;
+};
+
+struct ena_admin_aq_create_cq_cmd {
+       struct ena_admin_aq_common_desc aq_common_descriptor;
+
+       /* 4:0 : reserved5
+        * 5 : interrupt_mode_enabled - if set, cq operates
+        *    in interrupt mode, otherwise - polling
+        * 7:6 : reserved6
+        */
+       u8 cq_caps_1;
+
+       /* 4:0 : cq_entry_size_words - size of CQ entry in
+        *    32-bit words, valid values: 4, 8.
+        * 7:5 : reserved7
+        */
+       u8 cq_caps_2;
+
+       /* completion queue depth in # of entries. must be power of 2 */
+       u16 cq_depth;
+
+       /* msix vector assigned to this cq */
+       u32 msix_vector;
+
+       /* cq physical base address in OS memory. CQ must be physically
+        * contiguous
+        */
+       struct ena_common_mem_addr cq_ba;
+};
+
+struct ena_admin_acq_create_cq_resp_desc {
+       struct ena_admin_acq_common_desc acq_common_desc;
+
+       u16 cq_idx;
+
+       /* actual cq depth in number of entries */
+       u16 cq_actual_depth;
+
+       u32 numa_node_register_offset;
+
+       u32 cq_head_db_register_offset;
+
+       u32 cq_interrupt_unmask_register_offset;
+};
+
+struct ena_admin_aq_destroy_cq_cmd {
+       struct ena_admin_aq_common_desc aq_common_descriptor;
+
+       u16 cq_idx;
+
+       u16 reserved1;
+};
+
+struct ena_admin_acq_destroy_cq_resp_desc {
+       struct ena_admin_acq_common_desc acq_common_desc;
+};
+
+/* ENA AQ Get Statistics command. Extended statistics are placed in control
+ * buffer pointed by AQ entry
+ */
+struct ena_admin_aq_get_stats_cmd {
+       struct ena_admin_aq_common_desc aq_common_descriptor;
+
+       union {
+               /* command specific inline data */
+               u32 inline_data_w1[3];
+
+               struct ena_admin_ctrl_buff_info control_buffer;
+       } u;
+
+       /* stats type as defined in enum ena_admin_get_stats_type */
+       u8 type;
+
+       /* stats scope defined in enum ena_admin_get_stats_scope */
+       u8 scope;
+
+       u16 reserved3;
+
+       /* queue id. used when scope is specific_queue */
+       u16 queue_idx;
+
+       /* device id, value 0xFFFF means mine. only privileged device can get
+        *    stats of other device
+        */
+       u16 device_id;
+};
+
+/* Basic Statistics Command. */
+struct ena_admin_basic_stats {
+       u32 tx_bytes_low;
+
+       u32 tx_bytes_high;
+
+       u32 tx_pkts_low;
+
+       u32 tx_pkts_high;
+
+       u32 rx_bytes_low;
+
+       u32 rx_bytes_high;
+
+       u32 rx_pkts_low;
+
+       u32 rx_pkts_high;
+
+       u32 rx_drops_low;
+
+       u32 rx_drops_high;
+};
+
+struct ena_admin_acq_get_stats_resp {
+       struct ena_admin_acq_common_desc acq_common_desc;
+
+       struct ena_admin_basic_stats basic_stats;
+};
+
+struct ena_admin_get_set_feature_common_desc {
+       /* 1:0 : select - 0x1 - current value; 0x3 - default
+        *    value
+        * 7:3 : reserved3
+        */
+       u8 flags;
+
+       /* as appears in ena_admin_aq_feature_id */
+       u8 feature_id;
+
+       u16 reserved16;
+};
+
+struct ena_admin_device_attr_feature_desc {
+       u32 impl_id;
+
+       u32 device_version;
+
+       /* bitmap of ena_admin_aq_feature_id */
+       u32 supported_features;
+
+       u32 reserved3;
+
+       /* Indicates how many bits are used physical address access. */
+       u32 phys_addr_width;
+
+       /* Indicates how many bits are used virtual address access. */
+       u32 virt_addr_width;
+
+       /* unicast MAC address (in Network byte order) */
+       u8 mac_addr[6];
+
+       u8 reserved7[2];
+
+       u32 max_mtu;
+};
+
+struct ena_admin_queue_feature_desc {
+       /* including LLQs */
+       u32 max_sq_num;
+
+       u32 max_sq_depth;
+
+       u32 max_cq_num;
+
+       u32 max_cq_depth;
+
+       u32 max_llq_num;
+
+       u32 max_llq_depth;
+
+       u32 max_header_size;
+
+       /* Maximum Descriptors number, including meta descriptor, allowed for
+        *    a single Tx packet
+        */
+       u16 max_packet_tx_descs;
+
+       /* Maximum Descriptors number allowed for a single Rx packet */
+       u16 max_packet_rx_descs;
+};
+
+struct ena_admin_set_feature_mtu_desc {
+       /* exclude L2 */
+       u32 mtu;
+};
+
+struct ena_admin_set_feature_host_attr_desc {
+       /* host OS info base address in OS memory. host info is 4KB of
+        * physically contiguous
+        */
+       struct ena_common_mem_addr os_info_ba;
+
+       /* host debug area base address in OS memory. debug area must be
+        * physically contiguous
+        */
+       struct ena_common_mem_addr debug_ba;
+
+       /* debug area size */
+       u32 debug_area_size;
+};
+
+struct ena_admin_feature_intr_moder_desc {
+       /* interrupt delay granularity in usec */
+       u16 intr_delay_resolution;
+
+       u16 reserved;
+};
+
+struct ena_admin_get_feature_link_desc {
+       /* Link speed in Mb */
+       u32 speed;
+
+       /* bit field of enum ena_admin_link types */
+       u32 supported;
+
+       /* 0 : autoneg
+        * 1 : duplex - Full Duplex
+        * 31:2 : reserved2
+        */
+       u32 flags;
+};
+
+struct ena_admin_feature_aenq_desc {
+       /* bitmask for AENQ groups the device can report */
+       u32 supported_groups;
+
+       /* bitmask for AENQ groups to report */
+       u32 enabled_groups;
+};
+
+struct ena_admin_feature_offload_desc {
+       /* 0 : TX_L3_csum_ipv4
+        * 1 : TX_L4_ipv4_csum_part - The checksum field
+        *    should be initialized with pseudo header checksum
+        * 2 : TX_L4_ipv4_csum_full
+        * 3 : TX_L4_ipv6_csum_part - The checksum field
+        *    should be initialized with pseudo header checksum
+        * 4 : TX_L4_ipv6_csum_full
+        * 5 : tso_ipv4
+        * 6 : tso_ipv6
+        * 7 : tso_ecn
+        */
+       u32 tx;
+
+       /* Receive side supported stateless offload
+        * 0 : RX_L3_csum_ipv4 - IPv4 checksum
+        * 1 : RX_L4_ipv4_csum - TCP/UDP/IPv4 checksum
+        * 2 : RX_L4_ipv6_csum - TCP/UDP/IPv6 checksum
+        * 3 : RX_hash - Hash calculation
+        */
+       u32 rx_supported;
+
+       u32 rx_enabled;
+};
+
+enum ena_admin_hash_functions {
+       ENA_ADMIN_TOEPLITZ      = 1,
+
+       ENA_ADMIN_CRC32         = 2,
+};
+
+struct ena_admin_feature_rss_flow_hash_control {
+       u32 keys_num;
+
+       u32 reserved;
+
+       u32 key[10];
+};
+
+struct ena_admin_feature_rss_flow_hash_function {
+       /* 7:0 : funcs - bitmask of ena_admin_hash_functions */
+       u32 supported_func;
+
+       /* 7:0 : selected_func - bitmask of
+        *    ena_admin_hash_functions
+        */
+       u32 selected_func;
+
+       /* initial value */
+       u32 init_val;
+};
+
+/* RSS flow hash protocols */
+enum ena_admin_flow_hash_proto {
+       ENA_ADMIN_RSS_TCP4      = 0,
+
+       ENA_ADMIN_RSS_UDP4      = 1,
+
+       ENA_ADMIN_RSS_TCP6      = 2,
+
+       ENA_ADMIN_RSS_UDP6      = 3,
+
+       ENA_ADMIN_RSS_IP4       = 4,
+
+       ENA_ADMIN_RSS_IP6       = 5,
+
+       ENA_ADMIN_RSS_IP4_FRAG  = 6,
+
+       ENA_ADMIN_RSS_NOT_IP    = 7,
+
+       ENA_ADMIN_RSS_PROTO_NUM = 16,
+};
+
+/* RSS flow hash fields */
+enum ena_admin_flow_hash_fields {
+       /* Ethernet Dest Addr */
+       ENA_ADMIN_RSS_L2_DA     = 0,
+
+       /* Ethernet Src Addr */
+       ENA_ADMIN_RSS_L2_SA     = 1,
+
+       /* ipv4/6 Dest Addr */
+       ENA_ADMIN_RSS_L3_DA     = 2,
+
+       /* ipv4/6 Src Addr */
+       ENA_ADMIN_RSS_L3_SA     = 5,
+
+       /* tcp/udp Dest Port */
+       ENA_ADMIN_RSS_L4_DP     = 6,
+
+       /* tcp/udp Src Port */
+       ENA_ADMIN_RSS_L4_SP     = 7,
+};
+
+struct ena_admin_proto_input {
+       /* flow hash fields (bitwise according to ena_admin_flow_hash_fields) */
+       u16 fields;
+
+       u16 reserved2;
+};
+
+struct ena_admin_feature_rss_hash_control {
+       struct ena_admin_proto_input supported_fields[ENA_ADMIN_RSS_PROTO_NUM];
+
+       struct ena_admin_proto_input selected_fields[ENA_ADMIN_RSS_PROTO_NUM];
+
+       struct ena_admin_proto_input reserved2[ENA_ADMIN_RSS_PROTO_NUM];
+
+       struct ena_admin_proto_input reserved3[ENA_ADMIN_RSS_PROTO_NUM];
+};
+
+struct ena_admin_feature_rss_flow_hash_input {
+       /* supported hash input sorting
+        * 1 : L3_sort - support swap L3 addresses if DA is
+        *    smaller than SA
+        * 2 : L4_sort - support swap L4 ports if DP smaller
+        *    SP
+        */
+       u16 supported_input_sort;
+
+       /* enabled hash input sorting
+        * 1 : enable_L3_sort - enable swap L3 addresses if
+        *    DA smaller than SA
+        * 2 : enable_L4_sort - enable swap L4 ports if DP
+        *    smaller than SP
+        */
+       u16 enabled_input_sort;
+};
+
+enum ena_admin_os_type {
+       ENA_ADMIN_OS_LINUX      = 1,
+
+       ENA_ADMIN_OS_WIN        = 2,
+
+       ENA_ADMIN_OS_DPDK       = 3,
+
+       ENA_ADMIN_OS_FREEBSD    = 4,
+
+       ENA_ADMIN_OS_IPXE       = 5,
+};
+
+struct ena_admin_host_info {
+       /* defined in enum ena_admin_os_type */
+       u32 os_type;
+
+       /* os distribution string format */
+       u8 os_dist_str[128];
+
+       /* OS distribution numeric format */
+       u32 os_dist;
+
+       /* kernel version string format */
+       u8 kernel_ver_str[32];
+
+       /* Kernel version numeric format */
+       u32 kernel_ver;
+
+       /* 7:0 : major
+        * 15:8 : minor
+        * 23:16 : sub_minor
+        */
+       u32 driver_version;
+
+       /* features bitmap */
+       u32 supported_network_features[4];
+};
+
+struct ena_admin_rss_ind_table_entry {
+       u16 cq_idx;
+
+       u16 reserved;
+};
+
+struct ena_admin_feature_rss_ind_table {
+       /* min supported table size (2^min_size) */
+       u16 min_size;
+
+       /* max supported table size (2^max_size) */
+       u16 max_size;
+
+       /* table size (2^size) */
+       u16 size;
+
+       u16 reserved;
+
+       /* index of the inline entry. 0xFFFFFFFF means invalid */
+       u32 inline_index;
+
+       /* used for updating single entry, ignored when setting the entire
+        * table through the control buffer.
+        */
+       struct ena_admin_rss_ind_table_entry inline_entry;
+};
+
+struct ena_admin_get_feat_cmd {
+       struct ena_admin_aq_common_desc aq_common_descriptor;
+
+       struct ena_admin_ctrl_buff_info control_buffer;
+
+       struct ena_admin_get_set_feature_common_desc feat_common;
+
+       u32 raw[11];
+};
+
+struct ena_admin_get_feat_resp {
+       struct ena_admin_acq_common_desc acq_common_desc;
+
+       union {
+               u32 raw[14];
+
+               struct ena_admin_device_attr_feature_desc dev_attr;
+
+               struct ena_admin_queue_feature_desc max_queue;
+
+               struct ena_admin_feature_aenq_desc aenq;
+
+               struct ena_admin_get_feature_link_desc link;
+
+               struct ena_admin_feature_offload_desc offload;
+
+               struct ena_admin_feature_rss_flow_hash_function flow_hash_func;
+
+               struct ena_admin_feature_rss_flow_hash_input flow_hash_input;
+
+               struct ena_admin_feature_rss_ind_table ind_table;
+
+               struct ena_admin_feature_intr_moder_desc intr_moderation;
+       } u;
+};
+
+struct ena_admin_set_feat_cmd {
+       struct ena_admin_aq_common_desc aq_common_descriptor;
+
+       struct ena_admin_ctrl_buff_info control_buffer;
+
+       struct ena_admin_get_set_feature_common_desc feat_common;
+
+       union {
+               u32 raw[11];
+
+               /* mtu size */
+               struct ena_admin_set_feature_mtu_desc mtu;
+
+               /* host attributes */
+               struct ena_admin_set_feature_host_attr_desc host_attr;
+
+               /* AENQ configuration */
+               struct ena_admin_feature_aenq_desc aenq;
+
+               /* rss flow hash function */
+               struct ena_admin_feature_rss_flow_hash_function flow_hash_func;
+
+               /* rss flow hash input */
+               struct ena_admin_feature_rss_flow_hash_input flow_hash_input;
+
+               /* rss indirection table */
+               struct ena_admin_feature_rss_ind_table ind_table;
+       } u;
+};
+
+struct ena_admin_set_feat_resp {
+       struct ena_admin_acq_common_desc acq_common_desc;
+
+       union {
+               u32 raw[14];
+       } u;
+};
+
+struct ena_admin_aenq_common_desc {
+       u16 group;
+
+       u16 syndrom;
+
+       /* 0 : phase */
+       u8 flags;
+
+       u8 reserved1[3];
+
+       u32 timestamp_low;
+
+       u32 timestamp_high;
+};
+
+/* asynchronous event notification groups */
+enum ena_admin_aenq_group {
+       ENA_ADMIN_LINK_CHANGE           = 0,
+
+       ENA_ADMIN_FATAL_ERROR           = 1,
+
+       ENA_ADMIN_WARNING               = 2,
+
+       ENA_ADMIN_NOTIFICATION          = 3,
+
+       ENA_ADMIN_KEEP_ALIVE            = 4,
+
+       ENA_ADMIN_AENQ_GROUPS_NUM       = 5,
+};
+
+enum ena_admin_aenq_notification_syndrom {
+       ENA_ADMIN_SUSPEND       = 0,
+
+       ENA_ADMIN_RESUME        = 1,
+};
+
+struct ena_admin_aenq_entry {
+       struct ena_admin_aenq_common_desc aenq_common_desc;
+
+       /* command specific inline data */
+       u32 inline_data_w4[12];
+};
+
+struct ena_admin_aenq_link_change_desc {
+       struct ena_admin_aenq_common_desc aenq_common_desc;
+
+       /* 0 : link_status */
+       u32 flags;
+};
+
+struct ena_admin_ena_mmio_req_read_less_resp {
+       u16 req_id;
+
+       u16 reg_off;
+
+       /* value is valid when poll is cleared */
+       u32 reg_val;
+};
+
+/* aq_common_desc */
+#define ENA_ADMIN_AQ_COMMON_DESC_COMMAND_ID_MASK GENMASK(11, 0)
+#define ENA_ADMIN_AQ_COMMON_DESC_PHASE_MASK BIT(0)
+#define ENA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_SHIFT 1
+#define ENA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_MASK BIT(1)
+#define ENA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_SHIFT 2
+#define ENA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK BIT(2)
+
+/* sq */
+#define ENA_ADMIN_SQ_SQ_DIRECTION_SHIFT 5
+#define ENA_ADMIN_SQ_SQ_DIRECTION_MASK GENMASK(7, 5)
+
+/* acq_common_desc */
+#define ENA_ADMIN_ACQ_COMMON_DESC_COMMAND_ID_MASK GENMASK(11, 0)
+#define ENA_ADMIN_ACQ_COMMON_DESC_PHASE_MASK BIT(0)
+
+/* aq_create_sq_cmd */
+#define ENA_ADMIN_AQ_CREATE_SQ_CMD_SQ_DIRECTION_SHIFT 5
+#define ENA_ADMIN_AQ_CREATE_SQ_CMD_SQ_DIRECTION_MASK GENMASK(7, 5)
+#define ENA_ADMIN_AQ_CREATE_SQ_CMD_PLACEMENT_POLICY_MASK GENMASK(3, 0)
+#define ENA_ADMIN_AQ_CREATE_SQ_CMD_COMPLETION_POLICY_SHIFT 4
+#define ENA_ADMIN_AQ_CREATE_SQ_CMD_COMPLETION_POLICY_MASK GENMASK(6, 4)
+#define ENA_ADMIN_AQ_CREATE_SQ_CMD_IS_PHYSICALLY_CONTIGUOUS_MASK BIT(0)
+
+/* aq_create_cq_cmd */
+#define ENA_ADMIN_AQ_CREATE_CQ_CMD_INTERRUPT_MODE_ENABLED_SHIFT 5
+#define ENA_ADMIN_AQ_CREATE_CQ_CMD_INTERRUPT_MODE_ENABLED_MASK BIT(5)
+#define ENA_ADMIN_AQ_CREATE_CQ_CMD_CQ_ENTRY_SIZE_WORDS_MASK GENMASK(4, 0)
+
+/* get_set_feature_common_desc */
+#define ENA_ADMIN_GET_SET_FEATURE_COMMON_DESC_SELECT_MASK GENMASK(1, 0)
+
+/* get_feature_link_desc */
+#define ENA_ADMIN_GET_FEATURE_LINK_DESC_AUTONEG_MASK BIT(0)
+#define ENA_ADMIN_GET_FEATURE_LINK_DESC_DUPLEX_SHIFT 1
+#define ENA_ADMIN_GET_FEATURE_LINK_DESC_DUPLEX_MASK BIT(1)
+
+/* feature_offload_desc */
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK BIT(0)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_SHIFT 1
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK BIT(1)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_SHIFT 2
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK BIT(2)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_SHIFT 3
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK BIT(3)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_SHIFT 4
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK BIT(4)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_SHIFT 5
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK BIT(5)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_SHIFT 6
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK BIT(6)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_SHIFT 7
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK BIT(7)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK BIT(0)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_SHIFT 1
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK BIT(1)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_SHIFT 2
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK BIT(2)
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_HASH_SHIFT 3
+#define ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_HASH_MASK BIT(3)
+
+/* feature_rss_flow_hash_function */
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_FUNCTION_FUNCS_MASK GENMASK(7, 0)
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_FUNCTION_SELECTED_FUNC_MASK GENMASK(7, 0)
+
+/* feature_rss_flow_hash_input */
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_L3_SORT_SHIFT 1
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_L3_SORT_MASK BIT(1)
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_L4_SORT_SHIFT 2
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_L4_SORT_MASK BIT(2)
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_ENABLE_L3_SORT_SHIFT 1
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_ENABLE_L3_SORT_MASK BIT(1)
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_ENABLE_L4_SORT_SHIFT 2
+#define ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_ENABLE_L4_SORT_MASK BIT(2)
+
+/* host_info */
+#define ENA_ADMIN_HOST_INFO_MAJOR_MASK GENMASK(7, 0)
+#define ENA_ADMIN_HOST_INFO_MINOR_SHIFT 8
+#define ENA_ADMIN_HOST_INFO_MINOR_MASK GENMASK(15, 8)
+#define ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT 16
+#define ENA_ADMIN_HOST_INFO_SUB_MINOR_MASK GENMASK(23, 16)
+
+/* aenq_common_desc */
+#define ENA_ADMIN_AENQ_COMMON_DESC_PHASE_MASK BIT(0)
+
+/* aenq_link_change_desc */
+#define ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK BIT(0)
+
+#endif /*_ENA_ADMIN_H_ */
diff --git a/drivers/net/ethernet/amazon/ena/ena_com.c b/drivers/net/ethernet/amazon/ena/ena_com.c
new file mode 100644 (file)
index 0000000..3066d9c
--- /dev/null
@@ -0,0 +1,2666 @@
+/*
+ * Copyright 2015 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "ena_com.h"
+
+/*****************************************************************************/
+/*****************************************************************************/
+
+/* Timeout in micro-sec */
+#define ADMIN_CMD_TIMEOUT_US (1000000)
+
+#define ENA_ASYNC_QUEUE_DEPTH 4
+#define ENA_ADMIN_QUEUE_DEPTH 32
+
+#define MIN_ENA_VER (((ENA_COMMON_SPEC_VERSION_MAJOR) << \
+               ENA_REGS_VERSION_MAJOR_VERSION_SHIFT) \
+               | (ENA_COMMON_SPEC_VERSION_MINOR))
+
+#define ENA_CTRL_MAJOR         0
+#define ENA_CTRL_MINOR         0
+#define ENA_CTRL_SUB_MINOR     1
+
+#define MIN_ENA_CTRL_VER \
+       (((ENA_CTRL_MAJOR) << \
+       (ENA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_SHIFT)) | \
+       ((ENA_CTRL_MINOR) << \
+       (ENA_REGS_CONTROLLER_VERSION_MINOR_VERSION_SHIFT)) | \
+       (ENA_CTRL_SUB_MINOR))
+
+#define ENA_DMA_ADDR_TO_UINT32_LOW(x)  ((u32)((u64)(x)))
+#define ENA_DMA_ADDR_TO_UINT32_HIGH(x) ((u32)(((u64)(x)) >> 32))
+
+#define ENA_MMIO_READ_TIMEOUT 0xFFFFFFFF
+
+/*****************************************************************************/
+/*****************************************************************************/
+/*****************************************************************************/
+
+enum ena_cmd_status {
+       ENA_CMD_SUBMITTED,
+       ENA_CMD_COMPLETED,
+       /* Abort - canceled by the driver */
+       ENA_CMD_ABORTED,
+};
+
+struct ena_comp_ctx {
+       struct completion wait_event;
+       struct ena_admin_acq_entry *user_cqe;
+       u32 comp_size;
+       enum ena_cmd_status status;
+       /* status from the device */
+       u8 comp_status;
+       u8 cmd_opcode;
+       bool occupied;
+};
+
+struct ena_com_stats_ctx {
+       struct ena_admin_aq_get_stats_cmd get_cmd;
+       struct ena_admin_acq_get_stats_resp get_resp;
+};
+
+static inline int ena_com_mem_addr_set(struct ena_com_dev *ena_dev,
+                                      struct ena_common_mem_addr *ena_addr,
+                                      dma_addr_t addr)
+{
+       if ((addr & GENMASK_ULL(ena_dev->dma_addr_bits - 1, 0)) != addr) {
+               pr_err("dma address has more bits that the device supports\n");
+               return -EINVAL;
+       }
+
+       ena_addr->mem_addr_low = (u32)addr;
+       ena_addr->mem_addr_high = (u64)addr >> 32;
+
+       return 0;
+}
+
+static int ena_com_admin_init_sq(struct ena_com_admin_queue *queue)
+{
+       struct ena_com_admin_sq *sq = &queue->sq;
+       u16 size = ADMIN_SQ_SIZE(queue->q_depth);
+
+       sq->entries = dma_zalloc_coherent(queue->q_dmadev, size, &sq->dma_addr,
+                                         GFP_KERNEL);
+
+       if (!sq->entries) {
+               pr_err("memory allocation failed");
+               return -ENOMEM;
+       }
+
+       sq->head = 0;
+       sq->tail = 0;
+       sq->phase = 1;
+
+       sq->db_addr = NULL;
+
+       return 0;
+}
+
+static int ena_com_admin_init_cq(struct ena_com_admin_queue *queue)
+{
+       struct ena_com_admin_cq *cq = &queue->cq;
+       u16 size = ADMIN_CQ_SIZE(queue->q_depth);
+
+       cq->entries = dma_zalloc_coherent(queue->q_dmadev, size, &cq->dma_addr,
+                                         GFP_KERNEL);
+
+       if (!cq->entries) {
+               pr_err("memory allocation failed");
+               return -ENOMEM;
+       }
+
+       cq->head = 0;
+       cq->phase = 1;
+
+       return 0;
+}
+
+static int ena_com_admin_init_aenq(struct ena_com_dev *dev,
+                                  struct ena_aenq_handlers *aenq_handlers)
+{
+       struct ena_com_aenq *aenq = &dev->aenq;
+       u32 addr_low, addr_high, aenq_caps;
+       u16 size;
+
+       dev->aenq.q_depth = ENA_ASYNC_QUEUE_DEPTH;
+       size = ADMIN_AENQ_SIZE(ENA_ASYNC_QUEUE_DEPTH);
+       aenq->entries = dma_zalloc_coherent(dev->dmadev, size, &aenq->dma_addr,
+                                           GFP_KERNEL);
+
+       if (!aenq->entries) {
+               pr_err("memory allocation failed");
+               return -ENOMEM;
+       }
+
+       aenq->head = aenq->q_depth;
+       aenq->phase = 1;
+
+       addr_low = ENA_DMA_ADDR_TO_UINT32_LOW(aenq->dma_addr);
+       addr_high = ENA_DMA_ADDR_TO_UINT32_HIGH(aenq->dma_addr);
+
+       writel(addr_low, dev->reg_bar + ENA_REGS_AENQ_BASE_LO_OFF);
+       writel(addr_high, dev->reg_bar + ENA_REGS_AENQ_BASE_HI_OFF);
+
+       aenq_caps = 0;
+       aenq_caps |= dev->aenq.q_depth & ENA_REGS_AENQ_CAPS_AENQ_DEPTH_MASK;
+       aenq_caps |= (sizeof(struct ena_admin_aenq_entry)
+                     << ENA_REGS_AENQ_CAPS_AENQ_ENTRY_SIZE_SHIFT) &
+                    ENA_REGS_AENQ_CAPS_AENQ_ENTRY_SIZE_MASK;
+       writel(aenq_caps, dev->reg_bar + ENA_REGS_AENQ_CAPS_OFF);
+
+       if (unlikely(!aenq_handlers)) {
+               pr_err("aenq handlers pointer is NULL\n");
+               return -EINVAL;
+       }
+
+       aenq->aenq_handlers = aenq_handlers;
+
+       return 0;
+}
+
+static inline void comp_ctxt_release(struct ena_com_admin_queue *queue,
+                                    struct ena_comp_ctx *comp_ctx)
+{
+       comp_ctx->occupied = false;
+       atomic_dec(&queue->outstanding_cmds);
+}
+
+static struct ena_comp_ctx *get_comp_ctxt(struct ena_com_admin_queue *queue,
+                                         u16 command_id, bool capture)
+{
+       if (unlikely(command_id >= queue->q_depth)) {
+               pr_err("command id is larger than the queue size. cmd_id: %u queue size %d\n",
+                      command_id, queue->q_depth);
+               return NULL;
+       }
+
+       if (unlikely(queue->comp_ctx[command_id].occupied && capture)) {
+               pr_err("Completion context is occupied\n");
+               return NULL;
+       }
+
+       if (capture) {
+               atomic_inc(&queue->outstanding_cmds);
+               queue->comp_ctx[command_id].occupied = true;
+       }
+
+       return &queue->comp_ctx[command_id];
+}
+
+static struct ena_comp_ctx *__ena_com_submit_admin_cmd(struct ena_com_admin_queue *admin_queue,
+                                                      struct ena_admin_aq_entry *cmd,
+                                                      size_t cmd_size_in_bytes,
+                                                      struct ena_admin_acq_entry *comp,
+                                                      size_t comp_size_in_bytes)
+{
+       struct ena_comp_ctx *comp_ctx;
+       u16 tail_masked, cmd_id;
+       u16 queue_size_mask;
+       u16 cnt;
+
+       queue_size_mask = admin_queue->q_depth - 1;
+
+       tail_masked = admin_queue->sq.tail & queue_size_mask;
+
+       /* In case of queue FULL */
+       cnt = admin_queue->sq.tail - admin_queue->sq.head;
+       if (cnt >= admin_queue->q_depth) {
+               pr_debug("admin queue is FULL (tail %d head %d depth: %d)\n",
+                        admin_queue->sq.tail, admin_queue->sq.head,
+                        admin_queue->q_depth);
+               admin_queue->stats.out_of_space++;
+               return ERR_PTR(-ENOSPC);
+       }
+
+       cmd_id = admin_queue->curr_cmd_id;
+
+       cmd->aq_common_descriptor.flags |= admin_queue->sq.phase &
+               ENA_ADMIN_AQ_COMMON_DESC_PHASE_MASK;
+
+       cmd->aq_common_descriptor.command_id |= cmd_id &
+               ENA_ADMIN_AQ_COMMON_DESC_COMMAND_ID_MASK;
+
+       comp_ctx = get_comp_ctxt(admin_queue, cmd_id, true);
+       if (unlikely(!comp_ctx))
+               return ERR_PTR(-EINVAL);
+
+       comp_ctx->status = ENA_CMD_SUBMITTED;
+       comp_ctx->comp_size = (u32)comp_size_in_bytes;
+       comp_ctx->user_cqe = comp;
+       comp_ctx->cmd_opcode = cmd->aq_common_descriptor.opcode;
+
+       reinit_completion(&comp_ctx->wait_event);
+
+       memcpy(&admin_queue->sq.entries[tail_masked], cmd, cmd_size_in_bytes);
+
+       admin_queue->curr_cmd_id = (admin_queue->curr_cmd_id + 1) &
+               queue_size_mask;
+
+       admin_queue->sq.tail++;
+       admin_queue->stats.submitted_cmd++;
+
+       if (unlikely((admin_queue->sq.tail & queue_size_mask) == 0))
+               admin_queue->sq.phase = !admin_queue->sq.phase;
+
+       writel(admin_queue->sq.tail, admin_queue->sq.db_addr);
+
+       return comp_ctx;
+}
+
+static inline int ena_com_init_comp_ctxt(struct ena_com_admin_queue *queue)
+{
+       size_t size = queue->q_depth * sizeof(struct ena_comp_ctx);
+       struct ena_comp_ctx *comp_ctx;
+       u16 i;
+
+       queue->comp_ctx = devm_kzalloc(queue->q_dmadev, size, GFP_KERNEL);
+       if (unlikely(!queue->comp_ctx)) {
+               pr_err("memory allocation failed");
+               return -ENOMEM;
+       }
+
+       for (i = 0; i < queue->q_depth; i++) {
+               comp_ctx = get_comp_ctxt(queue, i, false);
+               if (comp_ctx)
+                       init_completion(&comp_ctx->wait_event);
+       }
+
+       return 0;
+}
+
+static struct ena_comp_ctx *ena_com_submit_admin_cmd(struct ena_com_admin_queue *admin_queue,
+                                                    struct ena_admin_aq_entry *cmd,
+                                                    size_t cmd_size_in_bytes,
+                                                    struct ena_admin_acq_entry *comp,
+                                                    size_t comp_size_in_bytes)
+{
+       unsigned long flags;
+       struct ena_comp_ctx *comp_ctx;
+
+       spin_lock_irqsave(&admin_queue->q_lock, flags);
+       if (unlikely(!admin_queue->running_state)) {
+               spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+               return ERR_PTR(-ENODEV);
+       }
+       comp_ctx = __ena_com_submit_admin_cmd(admin_queue, cmd,
+                                             cmd_size_in_bytes,
+                                             comp,
+                                             comp_size_in_bytes);
+       if (unlikely(IS_ERR(comp_ctx)))
+               admin_queue->running_state = false;
+       spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+
+       return comp_ctx;
+}
+
+static int ena_com_init_io_sq(struct ena_com_dev *ena_dev,
+                             struct ena_com_create_io_ctx *ctx,
+                             struct ena_com_io_sq *io_sq)
+{
+       size_t size;
+       int dev_node = 0;
+
+       memset(&io_sq->desc_addr, 0x0, sizeof(struct ena_com_io_desc_addr));
+
+       io_sq->desc_entry_size =
+               (io_sq->direction == ENA_COM_IO_QUEUE_DIRECTION_TX) ?
+               sizeof(struct ena_eth_io_tx_desc) :
+               sizeof(struct ena_eth_io_rx_desc);
+
+       size = io_sq->desc_entry_size * io_sq->q_depth;
+
+       if (io_sq->mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST) {
+               dev_node = dev_to_node(ena_dev->dmadev);
+               set_dev_node(ena_dev->dmadev, ctx->numa_node);
+               io_sq->desc_addr.virt_addr =
+                       dma_zalloc_coherent(ena_dev->dmadev, size,
+                                           &io_sq->desc_addr.phys_addr,
+                                           GFP_KERNEL);
+               set_dev_node(ena_dev->dmadev, dev_node);
+               if (!io_sq->desc_addr.virt_addr) {
+                       io_sq->desc_addr.virt_addr =
+                               dma_zalloc_coherent(ena_dev->dmadev, size,
+                                                   &io_sq->desc_addr.phys_addr,
+                                                   GFP_KERNEL);
+               }
+       } else {
+               dev_node = dev_to_node(ena_dev->dmadev);
+               set_dev_node(ena_dev->dmadev, ctx->numa_node);
+               io_sq->desc_addr.virt_addr =
+                       devm_kzalloc(ena_dev->dmadev, size, GFP_KERNEL);
+               set_dev_node(ena_dev->dmadev, dev_node);
+               if (!io_sq->desc_addr.virt_addr) {
+                       io_sq->desc_addr.virt_addr =
+                               devm_kzalloc(ena_dev->dmadev, size, GFP_KERNEL);
+               }
+       }
+
+       if (!io_sq->desc_addr.virt_addr) {
+               pr_err("memory allocation failed");
+               return -ENOMEM;
+       }
+
+       io_sq->tail = 0;
+       io_sq->next_to_comp = 0;
+       io_sq->phase = 1;
+
+       return 0;
+}
+
+static int ena_com_init_io_cq(struct ena_com_dev *ena_dev,
+                             struct ena_com_create_io_ctx *ctx,
+                             struct ena_com_io_cq *io_cq)
+{
+       size_t size;
+       int prev_node = 0;
+
+       memset(&io_cq->cdesc_addr, 0x0, sizeof(struct ena_com_io_desc_addr));
+
+       /* Use the basic completion descriptor for Rx */
+       io_cq->cdesc_entry_size_in_bytes =
+               (io_cq->direction == ENA_COM_IO_QUEUE_DIRECTION_TX) ?
+               sizeof(struct ena_eth_io_tx_cdesc) :
+               sizeof(struct ena_eth_io_rx_cdesc_base);
+
+       size = io_cq->cdesc_entry_size_in_bytes * io_cq->q_depth;
+
+       prev_node = dev_to_node(ena_dev->dmadev);
+       set_dev_node(ena_dev->dmadev, ctx->numa_node);
+       io_cq->cdesc_addr.virt_addr =
+               dma_zalloc_coherent(ena_dev->dmadev, size,
+                                   &io_cq->cdesc_addr.phys_addr, GFP_KERNEL);
+       set_dev_node(ena_dev->dmadev, prev_node);
+       if (!io_cq->cdesc_addr.virt_addr) {
+               io_cq->cdesc_addr.virt_addr =
+                       dma_zalloc_coherent(ena_dev->dmadev, size,
+                                           &io_cq->cdesc_addr.phys_addr,
+                                           GFP_KERNEL);
+       }
+
+       if (!io_cq->cdesc_addr.virt_addr) {
+               pr_err("memory allocation failed");
+               return -ENOMEM;
+       }
+
+       io_cq->phase = 1;
+       io_cq->head = 0;
+
+       return 0;
+}
+
+static void ena_com_handle_single_admin_completion(struct ena_com_admin_queue *admin_queue,
+                                                  struct ena_admin_acq_entry *cqe)
+{
+       struct ena_comp_ctx *comp_ctx;
+       u16 cmd_id;
+
+       cmd_id = cqe->acq_common_descriptor.command &
+               ENA_ADMIN_ACQ_COMMON_DESC_COMMAND_ID_MASK;
+
+       comp_ctx = get_comp_ctxt(admin_queue, cmd_id, false);
+       if (unlikely(!comp_ctx)) {
+               pr_err("comp_ctx is NULL. Changing the admin queue running state\n");
+               admin_queue->running_state = false;
+               return;
+       }
+
+       comp_ctx->status = ENA_CMD_COMPLETED;
+       comp_ctx->comp_status = cqe->acq_common_descriptor.status;
+
+       if (comp_ctx->user_cqe)
+               memcpy(comp_ctx->user_cqe, (void *)cqe, comp_ctx->comp_size);
+
+       if (!admin_queue->polling)
+               complete(&comp_ctx->wait_event);
+}
+
+static void ena_com_handle_admin_completion(struct ena_com_admin_queue *admin_queue)
+{
+       struct ena_admin_acq_entry *cqe = NULL;
+       u16 comp_num = 0;
+       u16 head_masked;
+       u8 phase;
+
+       head_masked = admin_queue->cq.head & (admin_queue->q_depth - 1);
+       phase = admin_queue->cq.phase;
+
+       cqe = &admin_queue->cq.entries[head_masked];
+
+       /* Go over all the completions */
+       while ((cqe->acq_common_descriptor.flags &
+                       ENA_ADMIN_ACQ_COMMON_DESC_PHASE_MASK) == phase) {
+               /* Do not read the rest of the completion entry before the
+                * phase bit was validated
+                */
+               rmb();
+               ena_com_handle_single_admin_completion(admin_queue, cqe);
+
+               head_masked++;
+               comp_num++;
+               if (unlikely(head_masked == admin_queue->q_depth)) {
+                       head_masked = 0;
+                       phase = !phase;
+               }
+
+               cqe = &admin_queue->cq.entries[head_masked];
+       }
+
+       admin_queue->cq.head += comp_num;
+       admin_queue->cq.phase = phase;
+       admin_queue->sq.head += comp_num;
+       admin_queue->stats.completed_cmd += comp_num;
+}
+
+static int ena_com_comp_status_to_errno(u8 comp_status)
+{
+       if (unlikely(comp_status != 0))
+               pr_err("admin command failed[%u]\n", comp_status);
+
+       if (unlikely(comp_status > ENA_ADMIN_UNKNOWN_ERROR))
+               return -EINVAL;
+
+       switch (comp_status) {
+       case ENA_ADMIN_SUCCESS:
+               return 0;
+       case ENA_ADMIN_RESOURCE_ALLOCATION_FAILURE:
+               return -ENOMEM;
+       case ENA_ADMIN_UNSUPPORTED_OPCODE:
+               return -EPERM;
+       case ENA_ADMIN_BAD_OPCODE:
+       case ENA_ADMIN_MALFORMED_REQUEST:
+       case ENA_ADMIN_ILLEGAL_PARAMETER:
+       case ENA_ADMIN_UNKNOWN_ERROR:
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static int ena_com_wait_and_process_admin_cq_polling(struct ena_comp_ctx *comp_ctx,
+                                                    struct ena_com_admin_queue *admin_queue)
+{
+       unsigned long flags;
+       u32 start_time;
+       int ret;
+
+       start_time = ((u32)jiffies_to_usecs(jiffies));
+
+       while (comp_ctx->status == ENA_CMD_SUBMITTED) {
+               if ((((u32)jiffies_to_usecs(jiffies)) - start_time) >
+                   ADMIN_CMD_TIMEOUT_US) {
+                       pr_err("Wait for completion (polling) timeout\n");
+                       /* ENA didn't have any completion */
+                       spin_lock_irqsave(&admin_queue->q_lock, flags);
+                       admin_queue->stats.no_completion++;
+                       admin_queue->running_state = false;
+                       spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+
+                       ret = -ETIME;
+                       goto err;
+               }
+
+               spin_lock_irqsave(&admin_queue->q_lock, flags);
+               ena_com_handle_admin_completion(admin_queue);
+               spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+
+               msleep(100);
+       }
+
+       if (unlikely(comp_ctx->status == ENA_CMD_ABORTED)) {
+               pr_err("Command was aborted\n");
+               spin_lock_irqsave(&admin_queue->q_lock, flags);
+               admin_queue->stats.aborted_cmd++;
+               spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+               ret = -ENODEV;
+               goto err;
+       }
+
+       WARN(comp_ctx->status != ENA_CMD_COMPLETED, "Invalid comp status %d\n",
+            comp_ctx->status);
+
+       ret = ena_com_comp_status_to_errno(comp_ctx->comp_status);
+err:
+       comp_ctxt_release(admin_queue, comp_ctx);
+       return ret;
+}
+
+static int ena_com_wait_and_process_admin_cq_interrupts(struct ena_comp_ctx *comp_ctx,
+                                                       struct ena_com_admin_queue *admin_queue)
+{
+       unsigned long flags;
+       int ret;
+
+       wait_for_completion_timeout(&comp_ctx->wait_event,
+                                   usecs_to_jiffies(ADMIN_CMD_TIMEOUT_US));
+
+       /* In case the command wasn't completed find out the root cause.
+        * There might be 2 kinds of errors
+        * 1) No completion (timeout reached)
+        * 2) There is completion but the device didn't get any msi-x interrupt.
+        */
+       if (unlikely(comp_ctx->status == ENA_CMD_SUBMITTED)) {
+               spin_lock_irqsave(&admin_queue->q_lock, flags);
+               ena_com_handle_admin_completion(admin_queue);
+               admin_queue->stats.no_completion++;
+               spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+
+               if (comp_ctx->status == ENA_CMD_COMPLETED)
+                       pr_err("The ena device have completion but the driver didn't receive any MSI-X interrupt (cmd %d)\n",
+                              comp_ctx->cmd_opcode);
+               else
+                       pr_err("The ena device doesn't send any completion for the admin cmd %d status %d\n",
+                              comp_ctx->cmd_opcode, comp_ctx->status);
+
+               admin_queue->running_state = false;
+               ret = -ETIME;
+               goto err;
+       }
+
+       ret = ena_com_comp_status_to_errno(comp_ctx->comp_status);
+err:
+       comp_ctxt_release(admin_queue, comp_ctx);
+       return ret;
+}
+
+/* This method read the hardware device register through posting writes
+ * and waiting for response
+ * On timeout the function will return ENA_MMIO_READ_TIMEOUT
+ */
+static u32 ena_com_reg_bar_read32(struct ena_com_dev *ena_dev, u16 offset)
+{
+       struct ena_com_mmio_read *mmio_read = &ena_dev->mmio_read;
+       volatile struct ena_admin_ena_mmio_req_read_less_resp *read_resp =
+               mmio_read->read_resp;
+       u32 mmio_read_reg, ret;
+       unsigned long flags;
+       int i;
+
+       might_sleep();
+
+       /* If readless is disabled, perform regular read */
+       if (!mmio_read->readless_supported)
+               return readl(ena_dev->reg_bar + offset);
+
+       spin_lock_irqsave(&mmio_read->lock, flags);
+       mmio_read->seq_num++;
+
+       read_resp->req_id = mmio_read->seq_num + 0xDEAD;
+       mmio_read_reg = (offset << ENA_REGS_MMIO_REG_READ_REG_OFF_SHIFT) &
+                       ENA_REGS_MMIO_REG_READ_REG_OFF_MASK;
+       mmio_read_reg |= mmio_read->seq_num &
+                       ENA_REGS_MMIO_REG_READ_REQ_ID_MASK;
+
+       /* make sure read_resp->req_id get updated before the hw can write
+        * there
+        */
+       wmb();
+
+       writel(mmio_read_reg, ena_dev->reg_bar + ENA_REGS_MMIO_REG_READ_OFF);
+
+       for (i = 0; i < ENA_REG_READ_TIMEOUT; i++) {
+               if (read_resp->req_id == mmio_read->seq_num)
+                       break;
+
+               udelay(1);
+       }
+
+       if (unlikely(i == ENA_REG_READ_TIMEOUT)) {
+               pr_err("reading reg failed for timeout. expected: req id[%hu] offset[%hu] actual: req id[%hu] offset[%hu]\n",
+                      mmio_read->seq_num, offset, read_resp->req_id,
+                      read_resp->reg_off);
+               ret = ENA_MMIO_READ_TIMEOUT;
+               goto err;
+       }
+
+       if (read_resp->reg_off != offset) {
+               pr_err("Read failure: wrong offset provided");
+               ret = ENA_MMIO_READ_TIMEOUT;
+       } else {
+               ret = read_resp->reg_val;
+       }
+err:
+       spin_unlock_irqrestore(&mmio_read->lock, flags);
+
+       return ret;
+}
+
+/* There are two types to wait for completion.
+ * Polling mode - wait until the completion is available.
+ * Async mode - wait on wait queue until the completion is ready
+ * (or the timeout expired).
+ * It is expected that the IRQ called ena_com_handle_admin_completion
+ * to mark the completions.
+ */
+static int ena_com_wait_and_process_admin_cq(struct ena_comp_ctx *comp_ctx,
+                                            struct ena_com_admin_queue *admin_queue)
+{
+       if (admin_queue->polling)
+               return ena_com_wait_and_process_admin_cq_polling(comp_ctx,
+                                                                admin_queue);
+
+       return ena_com_wait_and_process_admin_cq_interrupts(comp_ctx,
+                                                           admin_queue);
+}
+
+static int ena_com_destroy_io_sq(struct ena_com_dev *ena_dev,
+                                struct ena_com_io_sq *io_sq)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_admin_aq_destroy_sq_cmd destroy_cmd;
+       struct ena_admin_acq_destroy_sq_resp_desc destroy_resp;
+       u8 direction;
+       int ret;
+
+       memset(&destroy_cmd, 0x0, sizeof(struct ena_admin_aq_destroy_sq_cmd));
+
+       if (io_sq->direction == ENA_COM_IO_QUEUE_DIRECTION_TX)
+               direction = ENA_ADMIN_SQ_DIRECTION_TX;
+       else
+               direction = ENA_ADMIN_SQ_DIRECTION_RX;
+
+       destroy_cmd.sq.sq_identity |= (direction <<
+               ENA_ADMIN_SQ_SQ_DIRECTION_SHIFT) &
+               ENA_ADMIN_SQ_SQ_DIRECTION_MASK;
+
+       destroy_cmd.sq.sq_idx = io_sq->idx;
+       destroy_cmd.aq_common_descriptor.opcode = ENA_ADMIN_DESTROY_SQ;
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&destroy_cmd,
+                                           sizeof(destroy_cmd),
+                                           (struct ena_admin_acq_entry *)&destroy_resp,
+                                           sizeof(destroy_resp));
+
+       if (unlikely(ret && (ret != -ENODEV)))
+               pr_err("failed to destroy io sq error: %d\n", ret);
+
+       return ret;
+}
+
+static void ena_com_io_queue_free(struct ena_com_dev *ena_dev,
+                                 struct ena_com_io_sq *io_sq,
+                                 struct ena_com_io_cq *io_cq)
+{
+       size_t size;
+
+       if (io_cq->cdesc_addr.virt_addr) {
+               size = io_cq->cdesc_entry_size_in_bytes * io_cq->q_depth;
+
+               dma_free_coherent(ena_dev->dmadev, size,
+                                 io_cq->cdesc_addr.virt_addr,
+                                 io_cq->cdesc_addr.phys_addr);
+
+               io_cq->cdesc_addr.virt_addr = NULL;
+       }
+
+       if (io_sq->desc_addr.virt_addr) {
+               size = io_sq->desc_entry_size * io_sq->q_depth;
+
+               if (io_sq->mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
+                       dma_free_coherent(ena_dev->dmadev, size,
+                                         io_sq->desc_addr.virt_addr,
+                                         io_sq->desc_addr.phys_addr);
+               else
+                       devm_kfree(ena_dev->dmadev, io_sq->desc_addr.virt_addr);
+
+               io_sq->desc_addr.virt_addr = NULL;
+       }
+}
+
+static int wait_for_reset_state(struct ena_com_dev *ena_dev, u32 timeout,
+                               u16 exp_state)
+{
+       u32 val, i;
+
+       for (i = 0; i < timeout; i++) {
+               val = ena_com_reg_bar_read32(ena_dev, ENA_REGS_DEV_STS_OFF);
+
+               if (unlikely(val == ENA_MMIO_READ_TIMEOUT)) {
+                       pr_err("Reg read timeout occurred\n");
+                       return -ETIME;
+               }
+
+               if ((val & ENA_REGS_DEV_STS_RESET_IN_PROGRESS_MASK) ==
+                       exp_state)
+                       return 0;
+
+               /* The resolution of the timeout is 100ms */
+               msleep(100);
+       }
+
+       return -ETIME;
+}
+
+static bool ena_com_check_supported_feature_id(struct ena_com_dev *ena_dev,
+                                              enum ena_admin_aq_feature_id feature_id)
+{
+       u32 feature_mask = 1 << feature_id;
+
+       /* Device attributes is always supported */
+       if ((feature_id != ENA_ADMIN_DEVICE_ATTRIBUTES) &&
+           !(ena_dev->supported_features & feature_mask))
+               return false;
+
+       return true;
+}
+
+static int ena_com_get_feature_ex(struct ena_com_dev *ena_dev,
+                                 struct ena_admin_get_feat_resp *get_resp,
+                                 enum ena_admin_aq_feature_id feature_id,
+                                 dma_addr_t control_buf_dma_addr,
+                                 u32 control_buff_size)
+{
+       struct ena_com_admin_queue *admin_queue;
+       struct ena_admin_get_feat_cmd get_cmd;
+       int ret;
+
+       if (!ena_com_check_supported_feature_id(ena_dev, feature_id)) {
+               pr_info("Feature %d isn't supported\n", feature_id);
+               return -EPERM;
+       }
+
+       memset(&get_cmd, 0x0, sizeof(get_cmd));
+       admin_queue = &ena_dev->admin_queue;
+
+       get_cmd.aq_common_descriptor.opcode = ENA_ADMIN_GET_FEATURE;
+
+       if (control_buff_size)
+               get_cmd.aq_common_descriptor.flags =
+                       ENA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK;
+       else
+               get_cmd.aq_common_descriptor.flags = 0;
+
+       ret = ena_com_mem_addr_set(ena_dev,
+                                  &get_cmd.control_buffer.address,
+                                  control_buf_dma_addr);
+       if (unlikely(ret)) {
+               pr_err("memory address set failed\n");
+               return ret;
+       }
+
+       get_cmd.control_buffer.length = control_buff_size;
+
+       get_cmd.feat_common.feature_id = feature_id;
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)
+                                           &get_cmd,
+                                           sizeof(get_cmd),
+                                           (struct ena_admin_acq_entry *)
+                                           get_resp,
+                                           sizeof(*get_resp));
+
+       if (unlikely(ret))
+               pr_err("Failed to submit get_feature command %d error: %d\n",
+                      feature_id, ret);
+
+       return ret;
+}
+
+static int ena_com_get_feature(struct ena_com_dev *ena_dev,
+                              struct ena_admin_get_feat_resp *get_resp,
+                              enum ena_admin_aq_feature_id feature_id)
+{
+       return ena_com_get_feature_ex(ena_dev,
+                                     get_resp,
+                                     feature_id,
+                                     0,
+                                     0);
+}
+
+static int ena_com_hash_key_allocate(struct ena_com_dev *ena_dev)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+
+       rss->hash_key =
+               dma_zalloc_coherent(ena_dev->dmadev, sizeof(*rss->hash_key),
+                                   &rss->hash_key_dma_addr, GFP_KERNEL);
+
+       if (unlikely(!rss->hash_key))
+               return -ENOMEM;
+
+       return 0;
+}
+
+static void ena_com_hash_key_destroy(struct ena_com_dev *ena_dev)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+
+       if (rss->hash_key)
+               dma_free_coherent(ena_dev->dmadev, sizeof(*rss->hash_key),
+                                 rss->hash_key, rss->hash_key_dma_addr);
+       rss->hash_key = NULL;
+}
+
+static int ena_com_hash_ctrl_init(struct ena_com_dev *ena_dev)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+
+       rss->hash_ctrl =
+               dma_zalloc_coherent(ena_dev->dmadev, sizeof(*rss->hash_ctrl),
+                                   &rss->hash_ctrl_dma_addr, GFP_KERNEL);
+
+       if (unlikely(!rss->hash_ctrl))
+               return -ENOMEM;
+
+       return 0;
+}
+
+static void ena_com_hash_ctrl_destroy(struct ena_com_dev *ena_dev)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+
+       if (rss->hash_ctrl)
+               dma_free_coherent(ena_dev->dmadev, sizeof(*rss->hash_ctrl),
+                                 rss->hash_ctrl, rss->hash_ctrl_dma_addr);
+       rss->hash_ctrl = NULL;
+}
+
+static int ena_com_indirect_table_allocate(struct ena_com_dev *ena_dev,
+                                          u16 log_size)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_get_feat_resp get_resp;
+       size_t tbl_size;
+       int ret;
+
+       ret = ena_com_get_feature(ena_dev, &get_resp,
+                                 ENA_ADMIN_RSS_REDIRECTION_TABLE_CONFIG);
+       if (unlikely(ret))
+               return ret;
+
+       if ((get_resp.u.ind_table.min_size > log_size) ||
+           (get_resp.u.ind_table.max_size < log_size)) {
+               pr_err("indirect table size doesn't fit. requested size: %d while min is:%d and max %d\n",
+                      1 << log_size, 1 << get_resp.u.ind_table.min_size,
+                      1 << get_resp.u.ind_table.max_size);
+               return -EINVAL;
+       }
+
+       tbl_size = (1ULL << log_size) *
+               sizeof(struct ena_admin_rss_ind_table_entry);
+
+       rss->rss_ind_tbl =
+               dma_zalloc_coherent(ena_dev->dmadev, tbl_size,
+                                   &rss->rss_ind_tbl_dma_addr, GFP_KERNEL);
+       if (unlikely(!rss->rss_ind_tbl))
+               goto mem_err1;
+
+       tbl_size = (1ULL << log_size) * sizeof(u16);
+       rss->host_rss_ind_tbl =
+               devm_kzalloc(ena_dev->dmadev, tbl_size, GFP_KERNEL);
+       if (unlikely(!rss->host_rss_ind_tbl))
+               goto mem_err2;
+
+       rss->tbl_log_size = log_size;
+
+       return 0;
+
+mem_err2:
+       tbl_size = (1ULL << log_size) *
+               sizeof(struct ena_admin_rss_ind_table_entry);
+
+       dma_free_coherent(ena_dev->dmadev, tbl_size, rss->rss_ind_tbl,
+                         rss->rss_ind_tbl_dma_addr);
+       rss->rss_ind_tbl = NULL;
+mem_err1:
+       rss->tbl_log_size = 0;
+       return -ENOMEM;
+}
+
+static void ena_com_indirect_table_destroy(struct ena_com_dev *ena_dev)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       size_t tbl_size = (1ULL << rss->tbl_log_size) *
+               sizeof(struct ena_admin_rss_ind_table_entry);
+
+       if (rss->rss_ind_tbl)
+               dma_free_coherent(ena_dev->dmadev, tbl_size, rss->rss_ind_tbl,
+                                 rss->rss_ind_tbl_dma_addr);
+       rss->rss_ind_tbl = NULL;
+
+       if (rss->host_rss_ind_tbl)
+               devm_kfree(ena_dev->dmadev, rss->host_rss_ind_tbl);
+       rss->host_rss_ind_tbl = NULL;
+}
+
+static int ena_com_create_io_sq(struct ena_com_dev *ena_dev,
+                               struct ena_com_io_sq *io_sq, u16 cq_idx)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_admin_aq_create_sq_cmd create_cmd;
+       struct ena_admin_acq_create_sq_resp_desc cmd_completion;
+       u8 direction;
+       int ret;
+
+       memset(&create_cmd, 0x0, sizeof(struct ena_admin_aq_create_sq_cmd));
+
+       create_cmd.aq_common_descriptor.opcode = ENA_ADMIN_CREATE_SQ;
+
+       if (io_sq->direction == ENA_COM_IO_QUEUE_DIRECTION_TX)
+               direction = ENA_ADMIN_SQ_DIRECTION_TX;
+       else
+               direction = ENA_ADMIN_SQ_DIRECTION_RX;
+
+       create_cmd.sq_identity |= (direction <<
+               ENA_ADMIN_AQ_CREATE_SQ_CMD_SQ_DIRECTION_SHIFT) &
+               ENA_ADMIN_AQ_CREATE_SQ_CMD_SQ_DIRECTION_MASK;
+
+       create_cmd.sq_caps_2 |= io_sq->mem_queue_type &
+               ENA_ADMIN_AQ_CREATE_SQ_CMD_PLACEMENT_POLICY_MASK;
+
+       create_cmd.sq_caps_2 |= (ENA_ADMIN_COMPLETION_POLICY_DESC <<
+               ENA_ADMIN_AQ_CREATE_SQ_CMD_COMPLETION_POLICY_SHIFT) &
+               ENA_ADMIN_AQ_CREATE_SQ_CMD_COMPLETION_POLICY_MASK;
+
+       create_cmd.sq_caps_3 |=
+               ENA_ADMIN_AQ_CREATE_SQ_CMD_IS_PHYSICALLY_CONTIGUOUS_MASK;
+
+       create_cmd.cq_idx = cq_idx;
+       create_cmd.sq_depth = io_sq->q_depth;
+
+       if (io_sq->mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST) {
+               ret = ena_com_mem_addr_set(ena_dev,
+                                          &create_cmd.sq_ba,
+                                          io_sq->desc_addr.phys_addr);
+               if (unlikely(ret)) {
+                       pr_err("memory address set failed\n");
+                       return ret;
+               }
+       }
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&create_cmd,
+                                           sizeof(create_cmd),
+                                           (struct ena_admin_acq_entry *)&cmd_completion,
+                                           sizeof(cmd_completion));
+       if (unlikely(ret)) {
+               pr_err("Failed to create IO SQ. error: %d\n", ret);
+               return ret;
+       }
+
+       io_sq->idx = cmd_completion.sq_idx;
+
+       io_sq->db_addr = (u32 __iomem *)((uintptr_t)ena_dev->reg_bar +
+               (uintptr_t)cmd_completion.sq_doorbell_offset);
+
+       if (io_sq->mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
+               io_sq->header_addr = (u8 __iomem *)((uintptr_t)ena_dev->mem_bar
+                               + cmd_completion.llq_headers_offset);
+
+               io_sq->desc_addr.pbuf_dev_addr =
+                       (u8 __iomem *)((uintptr_t)ena_dev->mem_bar +
+                       cmd_completion.llq_descriptors_offset);
+       }
+
+       pr_debug("created sq[%u], depth[%u]\n", io_sq->idx, io_sq->q_depth);
+
+       return ret;
+}
+
+static int ena_com_ind_tbl_convert_to_device(struct ena_com_dev *ena_dev)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_com_io_sq *io_sq;
+       u16 qid;
+       int i;
+
+       for (i = 0; i < 1 << rss->tbl_log_size; i++) {
+               qid = rss->host_rss_ind_tbl[i];
+               if (qid >= ENA_TOTAL_NUM_QUEUES)
+                       return -EINVAL;
+
+               io_sq = &ena_dev->io_sq_queues[qid];
+
+               if (io_sq->direction != ENA_COM_IO_QUEUE_DIRECTION_RX)
+                       return -EINVAL;
+
+               rss->rss_ind_tbl[i].cq_idx = io_sq->idx;
+       }
+
+       return 0;
+}
+
+static int ena_com_ind_tbl_convert_from_device(struct ena_com_dev *ena_dev)
+{
+       u16 dev_idx_to_host_tbl[ENA_TOTAL_NUM_QUEUES] = { (u16)-1 };
+       struct ena_rss *rss = &ena_dev->rss;
+       u8 idx;
+       u16 i;
+
+       for (i = 0; i < ENA_TOTAL_NUM_QUEUES; i++)
+               dev_idx_to_host_tbl[ena_dev->io_sq_queues[i].idx] = i;
+
+       for (i = 0; i < 1 << rss->tbl_log_size; i++) {
+               if (rss->rss_ind_tbl[i].cq_idx > ENA_TOTAL_NUM_QUEUES)
+                       return -EINVAL;
+               idx = (u8)rss->rss_ind_tbl[i].cq_idx;
+
+               if (dev_idx_to_host_tbl[idx] > ENA_TOTAL_NUM_QUEUES)
+                       return -EINVAL;
+
+               rss->host_rss_ind_tbl[i] = dev_idx_to_host_tbl[idx];
+       }
+
+       return 0;
+}
+
+static int ena_com_init_interrupt_moderation_table(struct ena_com_dev *ena_dev)
+{
+       size_t size;
+
+       size = sizeof(struct ena_intr_moder_entry) * ENA_INTR_MAX_NUM_OF_LEVELS;
+
+       ena_dev->intr_moder_tbl =
+               devm_kzalloc(ena_dev->dmadev, size, GFP_KERNEL);
+       if (!ena_dev->intr_moder_tbl)
+               return -ENOMEM;
+
+       ena_com_config_default_interrupt_moderation_table(ena_dev);
+
+       return 0;
+}
+
+static void ena_com_update_intr_delay_resolution(struct ena_com_dev *ena_dev,
+                                                u16 intr_delay_resolution)
+{
+       struct ena_intr_moder_entry *intr_moder_tbl = ena_dev->intr_moder_tbl;
+       unsigned int i;
+
+       if (!intr_delay_resolution) {
+               pr_err("Illegal intr_delay_resolution provided. Going to use default 1 usec resolution\n");
+               intr_delay_resolution = 1;
+       }
+       ena_dev->intr_delay_resolution = intr_delay_resolution;
+
+       /* update Rx */
+       for (i = 0; i < ENA_INTR_MAX_NUM_OF_LEVELS; i++)
+               intr_moder_tbl[i].intr_moder_interval /= intr_delay_resolution;
+
+       /* update Tx */
+       ena_dev->intr_moder_tx_interval /= intr_delay_resolution;
+}
+
+/*****************************************************************************/
+/*******************************      API       ******************************/
+/*****************************************************************************/
+
+int ena_com_execute_admin_command(struct ena_com_admin_queue *admin_queue,
+                                 struct ena_admin_aq_entry *cmd,
+                                 size_t cmd_size,
+                                 struct ena_admin_acq_entry *comp,
+                                 size_t comp_size)
+{
+       struct ena_comp_ctx *comp_ctx;
+       int ret;
+
+       comp_ctx = ena_com_submit_admin_cmd(admin_queue, cmd, cmd_size,
+                                           comp, comp_size);
+       if (unlikely(IS_ERR(comp_ctx))) {
+               pr_err("Failed to submit command [%ld]\n", PTR_ERR(comp_ctx));
+               return PTR_ERR(comp_ctx);
+       }
+
+       ret = ena_com_wait_and_process_admin_cq(comp_ctx, admin_queue);
+       if (unlikely(ret)) {
+               if (admin_queue->running_state)
+                       pr_err("Failed to process command. ret = %d\n", ret);
+               else
+                       pr_debug("Failed to process command. ret = %d\n", ret);
+       }
+       return ret;
+}
+
+int ena_com_create_io_cq(struct ena_com_dev *ena_dev,
+                        struct ena_com_io_cq *io_cq)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_admin_aq_create_cq_cmd create_cmd;
+       struct ena_admin_acq_create_cq_resp_desc cmd_completion;
+       int ret;
+
+       memset(&create_cmd, 0x0, sizeof(struct ena_admin_aq_create_cq_cmd));
+
+       create_cmd.aq_common_descriptor.opcode = ENA_ADMIN_CREATE_CQ;
+
+       create_cmd.cq_caps_2 |= (io_cq->cdesc_entry_size_in_bytes / 4) &
+               ENA_ADMIN_AQ_CREATE_CQ_CMD_CQ_ENTRY_SIZE_WORDS_MASK;
+       create_cmd.cq_caps_1 |=
+               ENA_ADMIN_AQ_CREATE_CQ_CMD_INTERRUPT_MODE_ENABLED_MASK;
+
+       create_cmd.msix_vector = io_cq->msix_vector;
+       create_cmd.cq_depth = io_cq->q_depth;
+
+       ret = ena_com_mem_addr_set(ena_dev,
+                                  &create_cmd.cq_ba,
+                                  io_cq->cdesc_addr.phys_addr);
+       if (unlikely(ret)) {
+               pr_err("memory address set failed\n");
+               return ret;
+       }
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&create_cmd,
+                                           sizeof(create_cmd),
+                                           (struct ena_admin_acq_entry *)&cmd_completion,
+                                           sizeof(cmd_completion));
+       if (unlikely(ret)) {
+               pr_err("Failed to create IO CQ. error: %d\n", ret);
+               return ret;
+       }
+
+       io_cq->idx = cmd_completion.cq_idx;
+
+       io_cq->unmask_reg = (u32 __iomem *)((uintptr_t)ena_dev->reg_bar +
+               cmd_completion.cq_interrupt_unmask_register_offset);
+
+       if (cmd_completion.cq_head_db_register_offset)
+               io_cq->cq_head_db_reg =
+                       (u32 __iomem *)((uintptr_t)ena_dev->reg_bar +
+                       cmd_completion.cq_head_db_register_offset);
+
+       if (cmd_completion.numa_node_register_offset)
+               io_cq->numa_node_cfg_reg =
+                       (u32 __iomem *)((uintptr_t)ena_dev->reg_bar +
+                       cmd_completion.numa_node_register_offset);
+
+       pr_debug("created cq[%u], depth[%u]\n", io_cq->idx, io_cq->q_depth);
+
+       return ret;
+}
+
+int ena_com_get_io_handlers(struct ena_com_dev *ena_dev, u16 qid,
+                           struct ena_com_io_sq **io_sq,
+                           struct ena_com_io_cq **io_cq)
+{
+       if (qid >= ENA_TOTAL_NUM_QUEUES) {
+               pr_err("Invalid queue number %d but the max is %d\n", qid,
+                      ENA_TOTAL_NUM_QUEUES);
+               return -EINVAL;
+       }
+
+       *io_sq = &ena_dev->io_sq_queues[qid];
+       *io_cq = &ena_dev->io_cq_queues[qid];
+
+       return 0;
+}
+
+void ena_com_abort_admin_commands(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_comp_ctx *comp_ctx;
+       u16 i;
+
+       if (!admin_queue->comp_ctx)
+               return;
+
+       for (i = 0; i < admin_queue->q_depth; i++) {
+               comp_ctx = get_comp_ctxt(admin_queue, i, false);
+               if (unlikely(!comp_ctx))
+                       break;
+
+               comp_ctx->status = ENA_CMD_ABORTED;
+
+               complete(&comp_ctx->wait_event);
+       }
+}
+
+void ena_com_wait_for_abort_completion(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       unsigned long flags;
+
+       spin_lock_irqsave(&admin_queue->q_lock, flags);
+       while (atomic_read(&admin_queue->outstanding_cmds) != 0) {
+               spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+               msleep(20);
+               spin_lock_irqsave(&admin_queue->q_lock, flags);
+       }
+       spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+}
+
+int ena_com_destroy_io_cq(struct ena_com_dev *ena_dev,
+                         struct ena_com_io_cq *io_cq)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_admin_aq_destroy_cq_cmd destroy_cmd;
+       struct ena_admin_acq_destroy_cq_resp_desc destroy_resp;
+       int ret;
+
+       memset(&destroy_cmd, 0x0, sizeof(struct ena_admin_aq_destroy_sq_cmd));
+
+       destroy_cmd.cq_idx = io_cq->idx;
+       destroy_cmd.aq_common_descriptor.opcode = ENA_ADMIN_DESTROY_CQ;
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&destroy_cmd,
+                                           sizeof(destroy_cmd),
+                                           (struct ena_admin_acq_entry *)&destroy_resp,
+                                           sizeof(destroy_resp));
+
+       if (unlikely(ret && (ret != -ENODEV)))
+               pr_err("Failed to destroy IO CQ. error: %d\n", ret);
+
+       return ret;
+}
+
+bool ena_com_get_admin_running_state(struct ena_com_dev *ena_dev)
+{
+       return ena_dev->admin_queue.running_state;
+}
+
+void ena_com_set_admin_running_state(struct ena_com_dev *ena_dev, bool state)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       unsigned long flags;
+
+       spin_lock_irqsave(&admin_queue->q_lock, flags);
+       ena_dev->admin_queue.running_state = state;
+       spin_unlock_irqrestore(&admin_queue->q_lock, flags);
+}
+
+void ena_com_admin_aenq_enable(struct ena_com_dev *ena_dev)
+{
+       u16 depth = ena_dev->aenq.q_depth;
+
+       WARN(ena_dev->aenq.head != depth, "Invalid AENQ state\n");
+
+       /* Init head_db to mark that all entries in the queue
+        * are initially available
+        */
+       writel(depth, ena_dev->reg_bar + ENA_REGS_AENQ_HEAD_DB_OFF);
+}
+
+int ena_com_set_aenq_config(struct ena_com_dev *ena_dev, u32 groups_flag)
+{
+       struct ena_com_admin_queue *admin_queue;
+       struct ena_admin_set_feat_cmd cmd;
+       struct ena_admin_set_feat_resp resp;
+       struct ena_admin_get_feat_resp get_resp;
+       int ret;
+
+       ret = ena_com_get_feature(ena_dev, &get_resp, ENA_ADMIN_AENQ_CONFIG);
+       if (ret) {
+               pr_info("Can't get aenq configuration\n");
+               return ret;
+       }
+
+       if ((get_resp.u.aenq.supported_groups & groups_flag) != groups_flag) {
+               pr_warn("Trying to set unsupported aenq events. supported flag: %x asked flag: %x\n",
+                       get_resp.u.aenq.supported_groups, groups_flag);
+               return -EPERM;
+       }
+
+       memset(&cmd, 0x0, sizeof(cmd));
+       admin_queue = &ena_dev->admin_queue;
+
+       cmd.aq_common_descriptor.opcode = ENA_ADMIN_SET_FEATURE;
+       cmd.aq_common_descriptor.flags = 0;
+       cmd.feat_common.feature_id = ENA_ADMIN_AENQ_CONFIG;
+       cmd.u.aenq.enabled_groups = groups_flag;
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&cmd,
+                                           sizeof(cmd),
+                                           (struct ena_admin_acq_entry *)&resp,
+                                           sizeof(resp));
+
+       if (unlikely(ret))
+               pr_err("Failed to config AENQ ret: %d\n", ret);
+
+       return ret;
+}
+
+int ena_com_get_dma_width(struct ena_com_dev *ena_dev)
+{
+       u32 caps = ena_com_reg_bar_read32(ena_dev, ENA_REGS_CAPS_OFF);
+       int width;
+
+       if (unlikely(caps == ENA_MMIO_READ_TIMEOUT)) {
+               pr_err("Reg read timeout occurred\n");
+               return -ETIME;
+       }
+
+       width = (caps & ENA_REGS_CAPS_DMA_ADDR_WIDTH_MASK) >>
+               ENA_REGS_CAPS_DMA_ADDR_WIDTH_SHIFT;
+
+       pr_debug("ENA dma width: %d\n", width);
+
+       if ((width < 32) || width > ENA_MAX_PHYS_ADDR_SIZE_BITS) {
+               pr_err("DMA width illegal value: %d\n", width);
+               return -EINVAL;
+       }
+
+       ena_dev->dma_addr_bits = width;
+
+       return width;
+}
+
+int ena_com_validate_version(struct ena_com_dev *ena_dev)
+{
+       u32 ver;
+       u32 ctrl_ver;
+       u32 ctrl_ver_masked;
+
+       /* Make sure the ENA version and the controller version are at least
+        * as the driver expects
+        */
+       ver = ena_com_reg_bar_read32(ena_dev, ENA_REGS_VERSION_OFF);
+       ctrl_ver = ena_com_reg_bar_read32(ena_dev,
+                                         ENA_REGS_CONTROLLER_VERSION_OFF);
+
+       if (unlikely((ver == ENA_MMIO_READ_TIMEOUT) ||
+                    (ctrl_ver == ENA_MMIO_READ_TIMEOUT))) {
+               pr_err("Reg read timeout occurred\n");
+               return -ETIME;
+       }
+
+       pr_info("ena device version: %d.%d\n",
+               (ver & ENA_REGS_VERSION_MAJOR_VERSION_MASK) >>
+                       ENA_REGS_VERSION_MAJOR_VERSION_SHIFT,
+               ver & ENA_REGS_VERSION_MINOR_VERSION_MASK);
+
+       if (ver < MIN_ENA_VER) {
+               pr_err("ENA version is lower than the minimal version the driver supports\n");
+               return -1;
+       }
+
+       pr_info("ena controller version: %d.%d.%d implementation version %d\n",
+               (ctrl_ver & ENA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_MASK) >>
+                       ENA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_SHIFT,
+               (ctrl_ver & ENA_REGS_CONTROLLER_VERSION_MINOR_VERSION_MASK) >>
+                       ENA_REGS_CONTROLLER_VERSION_MINOR_VERSION_SHIFT,
+               (ctrl_ver & ENA_REGS_CONTROLLER_VERSION_SUBMINOR_VERSION_MASK),
+               (ctrl_ver & ENA_REGS_CONTROLLER_VERSION_IMPL_ID_MASK) >>
+                       ENA_REGS_CONTROLLER_VERSION_IMPL_ID_SHIFT);
+
+       ctrl_ver_masked =
+               (ctrl_ver & ENA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_MASK) |
+               (ctrl_ver & ENA_REGS_CONTROLLER_VERSION_MINOR_VERSION_MASK) |
+               (ctrl_ver & ENA_REGS_CONTROLLER_VERSION_SUBMINOR_VERSION_MASK);
+
+       /* Validate the ctrl version without the implementation ID */
+       if (ctrl_ver_masked < MIN_ENA_CTRL_VER) {
+               pr_err("ENA ctrl version is lower than the minimal ctrl version the driver supports\n");
+               return -1;
+       }
+
+       return 0;
+}
+
+void ena_com_admin_destroy(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_com_admin_cq *cq = &admin_queue->cq;
+       struct ena_com_admin_sq *sq = &admin_queue->sq;
+       struct ena_com_aenq *aenq = &ena_dev->aenq;
+       u16 size;
+
+       if (admin_queue->comp_ctx)
+               devm_kfree(ena_dev->dmadev, admin_queue->comp_ctx);
+       admin_queue->comp_ctx = NULL;
+       size = ADMIN_SQ_SIZE(admin_queue->q_depth);
+       if (sq->entries)
+               dma_free_coherent(ena_dev->dmadev, size, sq->entries,
+                                 sq->dma_addr);
+       sq->entries = NULL;
+
+       size = ADMIN_CQ_SIZE(admin_queue->q_depth);
+       if (cq->entries)
+               dma_free_coherent(ena_dev->dmadev, size, cq->entries,
+                                 cq->dma_addr);
+       cq->entries = NULL;
+
+       size = ADMIN_AENQ_SIZE(aenq->q_depth);
+       if (ena_dev->aenq.entries)
+               dma_free_coherent(ena_dev->dmadev, size, aenq->entries,
+                                 aenq->dma_addr);
+       aenq->entries = NULL;
+}
+
+void ena_com_set_admin_polling_mode(struct ena_com_dev *ena_dev, bool polling)
+{
+       ena_dev->admin_queue.polling = polling;
+}
+
+int ena_com_mmio_reg_read_request_init(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_mmio_read *mmio_read = &ena_dev->mmio_read;
+
+       spin_lock_init(&mmio_read->lock);
+       mmio_read->read_resp =
+               dma_zalloc_coherent(ena_dev->dmadev,
+                                   sizeof(*mmio_read->read_resp),
+                                   &mmio_read->read_resp_dma_addr, GFP_KERNEL);
+       if (unlikely(!mmio_read->read_resp))
+               return -ENOMEM;
+
+       ena_com_mmio_reg_read_request_write_dev_addr(ena_dev);
+
+       mmio_read->read_resp->req_id = 0x0;
+       mmio_read->seq_num = 0x0;
+       mmio_read->readless_supported = true;
+
+       return 0;
+}
+
+void ena_com_set_mmio_read_mode(struct ena_com_dev *ena_dev, bool readless_supported)
+{
+       struct ena_com_mmio_read *mmio_read = &ena_dev->mmio_read;
+
+       mmio_read->readless_supported = readless_supported;
+}
+
+void ena_com_mmio_reg_read_request_destroy(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_mmio_read *mmio_read = &ena_dev->mmio_read;
+
+       writel(0x0, ena_dev->reg_bar + ENA_REGS_MMIO_RESP_LO_OFF);
+       writel(0x0, ena_dev->reg_bar + ENA_REGS_MMIO_RESP_HI_OFF);
+
+       dma_free_coherent(ena_dev->dmadev, sizeof(*mmio_read->read_resp),
+                         mmio_read->read_resp, mmio_read->read_resp_dma_addr);
+
+       mmio_read->read_resp = NULL;
+}
+
+void ena_com_mmio_reg_read_request_write_dev_addr(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_mmio_read *mmio_read = &ena_dev->mmio_read;
+       u32 addr_low, addr_high;
+
+       addr_low = ENA_DMA_ADDR_TO_UINT32_LOW(mmio_read->read_resp_dma_addr);
+       addr_high = ENA_DMA_ADDR_TO_UINT32_HIGH(mmio_read->read_resp_dma_addr);
+
+       writel(addr_low, ena_dev->reg_bar + ENA_REGS_MMIO_RESP_LO_OFF);
+       writel(addr_high, ena_dev->reg_bar + ENA_REGS_MMIO_RESP_HI_OFF);
+}
+
+int ena_com_admin_init(struct ena_com_dev *ena_dev,
+                      struct ena_aenq_handlers *aenq_handlers,
+                      bool init_spinlock)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       u32 aq_caps, acq_caps, dev_sts, addr_low, addr_high;
+       int ret;
+
+       dev_sts = ena_com_reg_bar_read32(ena_dev, ENA_REGS_DEV_STS_OFF);
+
+       if (unlikely(dev_sts == ENA_MMIO_READ_TIMEOUT)) {
+               pr_err("Reg read timeout occurred\n");
+               return -ETIME;
+       }
+
+       if (!(dev_sts & ENA_REGS_DEV_STS_READY_MASK)) {
+               pr_err("Device isn't ready, abort com init\n");
+               return -ENODEV;
+       }
+
+       admin_queue->q_depth = ENA_ADMIN_QUEUE_DEPTH;
+
+       admin_queue->q_dmadev = ena_dev->dmadev;
+       admin_queue->polling = false;
+       admin_queue->curr_cmd_id = 0;
+
+       atomic_set(&admin_queue->outstanding_cmds, 0);
+
+       if (init_spinlock)
+               spin_lock_init(&admin_queue->q_lock);
+
+       ret = ena_com_init_comp_ctxt(admin_queue);
+       if (ret)
+               goto error;
+
+       ret = ena_com_admin_init_sq(admin_queue);
+       if (ret)
+               goto error;
+
+       ret = ena_com_admin_init_cq(admin_queue);
+       if (ret)
+               goto error;
+
+       admin_queue->sq.db_addr = (u32 __iomem *)((uintptr_t)ena_dev->reg_bar +
+               ENA_REGS_AQ_DB_OFF);
+
+       addr_low = ENA_DMA_ADDR_TO_UINT32_LOW(admin_queue->sq.dma_addr);
+       addr_high = ENA_DMA_ADDR_TO_UINT32_HIGH(admin_queue->sq.dma_addr);
+
+       writel(addr_low, ena_dev->reg_bar + ENA_REGS_AQ_BASE_LO_OFF);
+       writel(addr_high, ena_dev->reg_bar + ENA_REGS_AQ_BASE_HI_OFF);
+
+       addr_low = ENA_DMA_ADDR_TO_UINT32_LOW(admin_queue->cq.dma_addr);
+       addr_high = ENA_DMA_ADDR_TO_UINT32_HIGH(admin_queue->cq.dma_addr);
+
+       writel(addr_low, ena_dev->reg_bar + ENA_REGS_ACQ_BASE_LO_OFF);
+       writel(addr_high, ena_dev->reg_bar + ENA_REGS_ACQ_BASE_HI_OFF);
+
+       aq_caps = 0;
+       aq_caps |= admin_queue->q_depth & ENA_REGS_AQ_CAPS_AQ_DEPTH_MASK;
+       aq_caps |= (sizeof(struct ena_admin_aq_entry) <<
+                       ENA_REGS_AQ_CAPS_AQ_ENTRY_SIZE_SHIFT) &
+                       ENA_REGS_AQ_CAPS_AQ_ENTRY_SIZE_MASK;
+
+       acq_caps = 0;
+       acq_caps |= admin_queue->q_depth & ENA_REGS_ACQ_CAPS_ACQ_DEPTH_MASK;
+       acq_caps |= (sizeof(struct ena_admin_acq_entry) <<
+               ENA_REGS_ACQ_CAPS_ACQ_ENTRY_SIZE_SHIFT) &
+               ENA_REGS_ACQ_CAPS_ACQ_ENTRY_SIZE_MASK;
+
+       writel(aq_caps, ena_dev->reg_bar + ENA_REGS_AQ_CAPS_OFF);
+       writel(acq_caps, ena_dev->reg_bar + ENA_REGS_ACQ_CAPS_OFF);
+       ret = ena_com_admin_init_aenq(ena_dev, aenq_handlers);
+       if (ret)
+               goto error;
+
+       admin_queue->running_state = true;
+
+       return 0;
+error:
+       ena_com_admin_destroy(ena_dev);
+
+       return ret;
+}
+
+int ena_com_create_io_queue(struct ena_com_dev *ena_dev,
+                           struct ena_com_create_io_ctx *ctx)
+{
+       struct ena_com_io_sq *io_sq;
+       struct ena_com_io_cq *io_cq;
+       int ret;
+
+       if (ctx->qid >= ENA_TOTAL_NUM_QUEUES) {
+               pr_err("Qid (%d) is bigger than max num of queues (%d)\n",
+                      ctx->qid, ENA_TOTAL_NUM_QUEUES);
+               return -EINVAL;
+       }
+
+       io_sq = &ena_dev->io_sq_queues[ctx->qid];
+       io_cq = &ena_dev->io_cq_queues[ctx->qid];
+
+       memset(io_sq, 0x0, sizeof(struct ena_com_io_sq));
+       memset(io_cq, 0x0, sizeof(struct ena_com_io_cq));
+
+       /* Init CQ */
+       io_cq->q_depth = ctx->queue_size;
+       io_cq->direction = ctx->direction;
+       io_cq->qid = ctx->qid;
+
+       io_cq->msix_vector = ctx->msix_vector;
+
+       io_sq->q_depth = ctx->queue_size;
+       io_sq->direction = ctx->direction;
+       io_sq->qid = ctx->qid;
+
+       io_sq->mem_queue_type = ctx->mem_queue_type;
+
+       if (ctx->direction == ENA_COM_IO_QUEUE_DIRECTION_TX)
+               /* header length is limited to 8 bits */
+               io_sq->tx_max_header_size =
+                       min_t(u32, ena_dev->tx_max_header_size, SZ_256);
+
+       ret = ena_com_init_io_sq(ena_dev, ctx, io_sq);
+       if (ret)
+               goto error;
+       ret = ena_com_init_io_cq(ena_dev, ctx, io_cq);
+       if (ret)
+               goto error;
+
+       ret = ena_com_create_io_cq(ena_dev, io_cq);
+       if (ret)
+               goto error;
+
+       ret = ena_com_create_io_sq(ena_dev, io_sq, io_cq->idx);
+       if (ret)
+               goto destroy_io_cq;
+
+       return 0;
+
+destroy_io_cq:
+       ena_com_destroy_io_cq(ena_dev, io_cq);
+error:
+       ena_com_io_queue_free(ena_dev, io_sq, io_cq);
+       return ret;
+}
+
+void ena_com_destroy_io_queue(struct ena_com_dev *ena_dev, u16 qid)
+{
+       struct ena_com_io_sq *io_sq;
+       struct ena_com_io_cq *io_cq;
+
+       if (qid >= ENA_TOTAL_NUM_QUEUES) {
+               pr_err("Qid (%d) is bigger than max num of queues (%d)\n", qid,
+                      ENA_TOTAL_NUM_QUEUES);
+               return;
+       }
+
+       io_sq = &ena_dev->io_sq_queues[qid];
+       io_cq = &ena_dev->io_cq_queues[qid];
+
+       ena_com_destroy_io_sq(ena_dev, io_sq);
+       ena_com_destroy_io_cq(ena_dev, io_cq);
+
+       ena_com_io_queue_free(ena_dev, io_sq, io_cq);
+}
+
+int ena_com_get_link_params(struct ena_com_dev *ena_dev,
+                           struct ena_admin_get_feat_resp *resp)
+{
+       return ena_com_get_feature(ena_dev, resp, ENA_ADMIN_LINK_CONFIG);
+}
+
+int ena_com_get_dev_attr_feat(struct ena_com_dev *ena_dev,
+                             struct ena_com_dev_get_features_ctx *get_feat_ctx)
+{
+       struct ena_admin_get_feat_resp get_resp;
+       int rc;
+
+       rc = ena_com_get_feature(ena_dev, &get_resp,
+                                ENA_ADMIN_DEVICE_ATTRIBUTES);
+       if (rc)
+               return rc;
+
+       memcpy(&get_feat_ctx->dev_attr, &get_resp.u.dev_attr,
+              sizeof(get_resp.u.dev_attr));
+       ena_dev->supported_features = get_resp.u.dev_attr.supported_features;
+
+       rc = ena_com_get_feature(ena_dev, &get_resp,
+                                ENA_ADMIN_MAX_QUEUES_NUM);
+       if (rc)
+               return rc;
+
+       memcpy(&get_feat_ctx->max_queues, &get_resp.u.max_queue,
+              sizeof(get_resp.u.max_queue));
+       ena_dev->tx_max_header_size = get_resp.u.max_queue.max_header_size;
+
+       rc = ena_com_get_feature(ena_dev, &get_resp,
+                                ENA_ADMIN_AENQ_CONFIG);
+       if (rc)
+               return rc;
+
+       memcpy(&get_feat_ctx->aenq, &get_resp.u.aenq,
+              sizeof(get_resp.u.aenq));
+
+       rc = ena_com_get_feature(ena_dev, &get_resp,
+                                ENA_ADMIN_STATELESS_OFFLOAD_CONFIG);
+       if (rc)
+               return rc;
+
+       memcpy(&get_feat_ctx->offload, &get_resp.u.offload,
+              sizeof(get_resp.u.offload));
+
+       return 0;
+}
+
+void ena_com_admin_q_comp_intr_handler(struct ena_com_dev *ena_dev)
+{
+       ena_com_handle_admin_completion(&ena_dev->admin_queue);
+}
+
+/* ena_handle_specific_aenq_event:
+ * return the handler that is relevant to the specific event group
+ */
+static ena_aenq_handler ena_com_get_specific_aenq_cb(struct ena_com_dev *dev,
+                                                    u16 group)
+{
+       struct ena_aenq_handlers *aenq_handlers = dev->aenq.aenq_handlers;
+
+       if ((group < ENA_MAX_HANDLERS) && aenq_handlers->handlers[group])
+               return aenq_handlers->handlers[group];
+
+       return aenq_handlers->unimplemented_handler;
+}
+
+/* ena_aenq_intr_handler:
+ * handles the aenq incoming events.
+ * pop events from the queue and apply the specific handler
+ */
+void ena_com_aenq_intr_handler(struct ena_com_dev *dev, void *data)
+{
+       struct ena_admin_aenq_entry *aenq_e;
+       struct ena_admin_aenq_common_desc *aenq_common;
+       struct ena_com_aenq *aenq  = &dev->aenq;
+       ena_aenq_handler handler_cb;
+       u16 masked_head, processed = 0;
+       u8 phase;
+
+       masked_head = aenq->head & (aenq->q_depth - 1);
+       phase = aenq->phase;
+       aenq_e = &aenq->entries[masked_head]; /* Get first entry */
+       aenq_common = &aenq_e->aenq_common_desc;
+
+       /* Go over all the events */
+       while ((aenq_common->flags & ENA_ADMIN_AENQ_COMMON_DESC_PHASE_MASK) ==
+              phase) {
+               pr_debug("AENQ! Group[%x] Syndrom[%x] timestamp: [%llus]\n",
+                        aenq_common->group, aenq_common->syndrom,
+                        (u64)aenq_common->timestamp_low +
+                                ((u64)aenq_common->timestamp_high << 32));
+
+               /* Handle specific event*/
+               handler_cb = ena_com_get_specific_aenq_cb(dev,
+                                                         aenq_common->group);
+               handler_cb(data, aenq_e); /* call the actual event handler*/
+
+               /* Get next event entry */
+               masked_head++;
+               processed++;
+
+               if (unlikely(masked_head == aenq->q_depth)) {
+                       masked_head = 0;
+                       phase = !phase;
+               }
+               aenq_e = &aenq->entries[masked_head];
+               aenq_common = &aenq_e->aenq_common_desc;
+       }
+
+       aenq->head += processed;
+       aenq->phase = phase;
+
+       /* Don't update aenq doorbell if there weren't any processed events */
+       if (!processed)
+               return;
+
+       /* write the aenq doorbell after all AENQ descriptors were read */
+       mb();
+       writel((u32)aenq->head, dev->reg_bar + ENA_REGS_AENQ_HEAD_DB_OFF);
+}
+
+int ena_com_dev_reset(struct ena_com_dev *ena_dev)
+{
+       u32 stat, timeout, cap, reset_val;
+       int rc;
+
+       stat = ena_com_reg_bar_read32(ena_dev, ENA_REGS_DEV_STS_OFF);
+       cap = ena_com_reg_bar_read32(ena_dev, ENA_REGS_CAPS_OFF);
+
+       if (unlikely((stat == ENA_MMIO_READ_TIMEOUT) ||
+                    (cap == ENA_MMIO_READ_TIMEOUT))) {
+               pr_err("Reg read32 timeout occurred\n");
+               return -ETIME;
+       }
+
+       if ((stat & ENA_REGS_DEV_STS_READY_MASK) == 0) {
+               pr_err("Device isn't ready, can't reset device\n");
+               return -EINVAL;
+       }
+
+       timeout = (cap & ENA_REGS_CAPS_RESET_TIMEOUT_MASK) >>
+                       ENA_REGS_CAPS_RESET_TIMEOUT_SHIFT;
+       if (timeout == 0) {
+               pr_err("Invalid timeout value\n");
+               return -EINVAL;
+       }
+
+       /* start reset */
+       reset_val = ENA_REGS_DEV_CTL_DEV_RESET_MASK;
+       writel(reset_val, ena_dev->reg_bar + ENA_REGS_DEV_CTL_OFF);
+
+       /* Write again the MMIO read request address */
+       ena_com_mmio_reg_read_request_write_dev_addr(ena_dev);
+
+       rc = wait_for_reset_state(ena_dev, timeout,
+                                 ENA_REGS_DEV_STS_RESET_IN_PROGRESS_MASK);
+       if (rc != 0) {
+               pr_err("Reset indication didn't turn on\n");
+               return rc;
+       }
+
+       /* reset done */
+       writel(0, ena_dev->reg_bar + ENA_REGS_DEV_CTL_OFF);
+       rc = wait_for_reset_state(ena_dev, timeout, 0);
+       if (rc != 0) {
+               pr_err("Reset indication didn't turn off\n");
+               return rc;
+       }
+
+       return 0;
+}
+
+static int ena_get_dev_stats(struct ena_com_dev *ena_dev,
+                            struct ena_com_stats_ctx *ctx,
+                            enum ena_admin_get_stats_type type)
+{
+       struct ena_admin_aq_get_stats_cmd *get_cmd = &ctx->get_cmd;
+       struct ena_admin_acq_get_stats_resp *get_resp = &ctx->get_resp;
+       struct ena_com_admin_queue *admin_queue;
+       int ret;
+
+       admin_queue = &ena_dev->admin_queue;
+
+       get_cmd->aq_common_descriptor.opcode = ENA_ADMIN_GET_STATS;
+       get_cmd->aq_common_descriptor.flags = 0;
+       get_cmd->type = type;
+
+       ret =  ena_com_execute_admin_command(admin_queue,
+                                            (struct ena_admin_aq_entry *)get_cmd,
+                                            sizeof(*get_cmd),
+                                            (struct ena_admin_acq_entry *)get_resp,
+                                            sizeof(*get_resp));
+
+       if (unlikely(ret))
+               pr_err("Failed to get stats. error: %d\n", ret);
+
+       return ret;
+}
+
+int ena_com_get_dev_basic_stats(struct ena_com_dev *ena_dev,
+                               struct ena_admin_basic_stats *stats)
+{
+       struct ena_com_stats_ctx ctx;
+       int ret;
+
+       memset(&ctx, 0x0, sizeof(ctx));
+       ret = ena_get_dev_stats(ena_dev, &ctx, ENA_ADMIN_GET_STATS_TYPE_BASIC);
+       if (likely(ret == 0))
+               memcpy(stats, &ctx.get_resp.basic_stats,
+                      sizeof(ctx.get_resp.basic_stats));
+
+       return ret;
+}
+
+int ena_com_set_dev_mtu(struct ena_com_dev *ena_dev, int mtu)
+{
+       struct ena_com_admin_queue *admin_queue;
+       struct ena_admin_set_feat_cmd cmd;
+       struct ena_admin_set_feat_resp resp;
+       int ret;
+
+       if (!ena_com_check_supported_feature_id(ena_dev, ENA_ADMIN_MTU)) {
+               pr_info("Feature %d isn't supported\n", ENA_ADMIN_MTU);
+               return -EPERM;
+       }
+
+       memset(&cmd, 0x0, sizeof(cmd));
+       admin_queue = &ena_dev->admin_queue;
+
+       cmd.aq_common_descriptor.opcode = ENA_ADMIN_SET_FEATURE;
+       cmd.aq_common_descriptor.flags = 0;
+       cmd.feat_common.feature_id = ENA_ADMIN_MTU;
+       cmd.u.mtu.mtu = mtu;
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&cmd,
+                                           sizeof(cmd),
+                                           (struct ena_admin_acq_entry *)&resp,
+                                           sizeof(resp));
+
+       if (unlikely(ret))
+               pr_err("Failed to set mtu %d. error: %d\n", mtu, ret);
+
+       return ret;
+}
+
+int ena_com_get_offload_settings(struct ena_com_dev *ena_dev,
+                                struct ena_admin_feature_offload_desc *offload)
+{
+       int ret;
+       struct ena_admin_get_feat_resp resp;
+
+       ret = ena_com_get_feature(ena_dev, &resp,
+                                 ENA_ADMIN_STATELESS_OFFLOAD_CONFIG);
+       if (unlikely(ret)) {
+               pr_err("Failed to get offload capabilities %d\n", ret);
+               return ret;
+       }
+
+       memcpy(offload, &resp.u.offload, sizeof(resp.u.offload));
+
+       return 0;
+}
+
+int ena_com_set_hash_function(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_set_feat_cmd cmd;
+       struct ena_admin_set_feat_resp resp;
+       struct ena_admin_get_feat_resp get_resp;
+       int ret;
+
+       if (!ena_com_check_supported_feature_id(ena_dev,
+                                               ENA_ADMIN_RSS_HASH_FUNCTION)) {
+               pr_info("Feature %d isn't supported\n",
+                       ENA_ADMIN_RSS_HASH_FUNCTION);
+               return -EPERM;
+       }
+
+       /* Validate hash function is supported */
+       ret = ena_com_get_feature(ena_dev, &get_resp,
+                                 ENA_ADMIN_RSS_HASH_FUNCTION);
+       if (unlikely(ret))
+               return ret;
+
+       if (get_resp.u.flow_hash_func.supported_func & (1 << rss->hash_func)) {
+               pr_err("Func hash %d isn't supported by device, abort\n",
+                      rss->hash_func);
+               return -EPERM;
+       }
+
+       memset(&cmd, 0x0, sizeof(cmd));
+
+       cmd.aq_common_descriptor.opcode = ENA_ADMIN_SET_FEATURE;
+       cmd.aq_common_descriptor.flags =
+               ENA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK;
+       cmd.feat_common.feature_id = ENA_ADMIN_RSS_HASH_FUNCTION;
+       cmd.u.flow_hash_func.init_val = rss->hash_init_val;
+       cmd.u.flow_hash_func.selected_func = 1 << rss->hash_func;
+
+       ret = ena_com_mem_addr_set(ena_dev,
+                                  &cmd.control_buffer.address,
+                                  rss->hash_key_dma_addr);
+       if (unlikely(ret)) {
+               pr_err("memory address set failed\n");
+               return ret;
+       }
+
+       cmd.control_buffer.length = sizeof(*rss->hash_key);
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&cmd,
+                                           sizeof(cmd),
+                                           (struct ena_admin_acq_entry *)&resp,
+                                           sizeof(resp));
+       if (unlikely(ret)) {
+               pr_err("Failed to set hash function %d. error: %d\n",
+                      rss->hash_func, ret);
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+int ena_com_fill_hash_function(struct ena_com_dev *ena_dev,
+                              enum ena_admin_hash_functions func,
+                              const u8 *key, u16 key_len, u32 init_val)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_get_feat_resp get_resp;
+       struct ena_admin_feature_rss_flow_hash_control *hash_key =
+               rss->hash_key;
+       int rc;
+
+       /* Make sure size is a mult of DWs */
+       if (unlikely(key_len & 0x3))
+               return -EINVAL;
+
+       rc = ena_com_get_feature_ex(ena_dev, &get_resp,
+                                   ENA_ADMIN_RSS_HASH_FUNCTION,
+                                   rss->hash_key_dma_addr,
+                                   sizeof(*rss->hash_key));
+       if (unlikely(rc))
+               return rc;
+
+       if (!((1 << func) & get_resp.u.flow_hash_func.supported_func)) {
+               pr_err("Flow hash function %d isn't supported\n", func);
+               return -EPERM;
+       }
+
+       switch (func) {
+       case ENA_ADMIN_TOEPLITZ:
+               if (key_len > sizeof(hash_key->key)) {
+                       pr_err("key len (%hu) is bigger than the max supported (%zu)\n",
+                              key_len, sizeof(hash_key->key));
+                       return -EINVAL;
+               }
+
+               memcpy(hash_key->key, key, key_len);
+               rss->hash_init_val = init_val;
+               hash_key->keys_num = key_len >> 2;
+               break;
+       case ENA_ADMIN_CRC32:
+               rss->hash_init_val = init_val;
+               break;
+       default:
+               pr_err("Invalid hash function (%d)\n", func);
+               return -EINVAL;
+       }
+
+       rc = ena_com_set_hash_function(ena_dev);
+
+       /* Restore the old function */
+       if (unlikely(rc))
+               ena_com_get_hash_function(ena_dev, NULL, NULL);
+
+       return rc;
+}
+
+int ena_com_get_hash_function(struct ena_com_dev *ena_dev,
+                             enum ena_admin_hash_functions *func,
+                             u8 *key)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_get_feat_resp get_resp;
+       struct ena_admin_feature_rss_flow_hash_control *hash_key =
+               rss->hash_key;
+       int rc;
+
+       rc = ena_com_get_feature_ex(ena_dev, &get_resp,
+                                   ENA_ADMIN_RSS_HASH_FUNCTION,
+                                   rss->hash_key_dma_addr,
+                                   sizeof(*rss->hash_key));
+       if (unlikely(rc))
+               return rc;
+
+       rss->hash_func = get_resp.u.flow_hash_func.selected_func;
+       if (func)
+               *func = rss->hash_func;
+
+       if (key)
+               memcpy(key, hash_key->key, (size_t)(hash_key->keys_num) << 2);
+
+       return 0;
+}
+
+int ena_com_get_hash_ctrl(struct ena_com_dev *ena_dev,
+                         enum ena_admin_flow_hash_proto proto,
+                         u16 *fields)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_get_feat_resp get_resp;
+       int rc;
+
+       rc = ena_com_get_feature_ex(ena_dev, &get_resp,
+                                   ENA_ADMIN_RSS_HASH_INPUT,
+                                   rss->hash_ctrl_dma_addr,
+                                   sizeof(*rss->hash_ctrl));
+       if (unlikely(rc))
+               return rc;
+
+       if (fields)
+               *fields = rss->hash_ctrl->selected_fields[proto].fields;
+
+       return 0;
+}
+
+int ena_com_set_hash_ctrl(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_feature_rss_hash_control *hash_ctrl = rss->hash_ctrl;
+       struct ena_admin_set_feat_cmd cmd;
+       struct ena_admin_set_feat_resp resp;
+       int ret;
+
+       if (!ena_com_check_supported_feature_id(ena_dev,
+                                               ENA_ADMIN_RSS_HASH_INPUT)) {
+               pr_info("Feature %d isn't supported\n", ENA_ADMIN_RSS_HASH_INPUT);
+               return -EPERM;
+       }
+
+       memset(&cmd, 0x0, sizeof(cmd));
+
+       cmd.aq_common_descriptor.opcode = ENA_ADMIN_SET_FEATURE;
+       cmd.aq_common_descriptor.flags =
+               ENA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK;
+       cmd.feat_common.feature_id = ENA_ADMIN_RSS_HASH_INPUT;
+       cmd.u.flow_hash_input.enabled_input_sort =
+               ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_L3_SORT_MASK |
+               ENA_ADMIN_FEATURE_RSS_FLOW_HASH_INPUT_L4_SORT_MASK;
+
+       ret = ena_com_mem_addr_set(ena_dev,
+                                  &cmd.control_buffer.address,
+                                  rss->hash_ctrl_dma_addr);
+       if (unlikely(ret)) {
+               pr_err("memory address set failed\n");
+               return ret;
+       }
+       cmd.control_buffer.length = sizeof(*hash_ctrl);
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&cmd,
+                                           sizeof(cmd),
+                                           (struct ena_admin_acq_entry *)&resp,
+                                           sizeof(resp));
+       if (unlikely(ret))
+               pr_err("Failed to set hash input. error: %d\n", ret);
+
+       return ret;
+}
+
+int ena_com_set_default_hash_ctrl(struct ena_com_dev *ena_dev)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_feature_rss_hash_control *hash_ctrl =
+               rss->hash_ctrl;
+       u16 available_fields = 0;
+       int rc, i;
+
+       /* Get the supported hash input */
+       rc = ena_com_get_hash_ctrl(ena_dev, 0, NULL);
+       if (unlikely(rc))
+               return rc;
+
+       hash_ctrl->selected_fields[ENA_ADMIN_RSS_TCP4].fields =
+               ENA_ADMIN_RSS_L3_SA | ENA_ADMIN_RSS_L3_DA |
+               ENA_ADMIN_RSS_L4_DP | ENA_ADMIN_RSS_L4_SP;
+
+       hash_ctrl->selected_fields[ENA_ADMIN_RSS_UDP4].fields =
+               ENA_ADMIN_RSS_L3_SA | ENA_ADMIN_RSS_L3_DA |
+               ENA_ADMIN_RSS_L4_DP | ENA_ADMIN_RSS_L4_SP;
+
+       hash_ctrl->selected_fields[ENA_ADMIN_RSS_TCP6].fields =
+               ENA_ADMIN_RSS_L3_SA | ENA_ADMIN_RSS_L3_DA |
+               ENA_ADMIN_RSS_L4_DP | ENA_ADMIN_RSS_L4_SP;
+
+       hash_ctrl->selected_fields[ENA_ADMIN_RSS_UDP6].fields =
+               ENA_ADMIN_RSS_L3_SA | ENA_ADMIN_RSS_L3_DA |
+               ENA_ADMIN_RSS_L4_DP | ENA_ADMIN_RSS_L4_SP;
+
+       hash_ctrl->selected_fields[ENA_ADMIN_RSS_IP4].fields =
+               ENA_ADMIN_RSS_L3_SA | ENA_ADMIN_RSS_L3_DA;
+
+       hash_ctrl->selected_fields[ENA_ADMIN_RSS_IP6].fields =
+               ENA_ADMIN_RSS_L3_SA | ENA_ADMIN_RSS_L3_DA;
+
+       hash_ctrl->selected_fields[ENA_ADMIN_RSS_IP4_FRAG].fields =
+               ENA_ADMIN_RSS_L3_SA | ENA_ADMIN_RSS_L3_DA;
+
+       hash_ctrl->selected_fields[ENA_ADMIN_RSS_IP4_FRAG].fields =
+               ENA_ADMIN_RSS_L2_DA | ENA_ADMIN_RSS_L2_SA;
+
+       for (i = 0; i < ENA_ADMIN_RSS_PROTO_NUM; i++) {
+               available_fields = hash_ctrl->selected_fields[i].fields &
+                               hash_ctrl->supported_fields[i].fields;
+               if (available_fields != hash_ctrl->selected_fields[i].fields) {
+                       pr_err("hash control doesn't support all the desire configuration. proto %x supported %x selected %x\n",
+                              i, hash_ctrl->supported_fields[i].fields,
+                              hash_ctrl->selected_fields[i].fields);
+                       return -EPERM;
+               }
+       }
+
+       rc = ena_com_set_hash_ctrl(ena_dev);
+
+       /* In case of failure, restore the old hash ctrl */
+       if (unlikely(rc))
+               ena_com_get_hash_ctrl(ena_dev, 0, NULL);
+
+       return rc;
+}
+
+int ena_com_fill_hash_ctrl(struct ena_com_dev *ena_dev,
+                          enum ena_admin_flow_hash_proto proto,
+                          u16 hash_fields)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_feature_rss_hash_control *hash_ctrl = rss->hash_ctrl;
+       u16 supported_fields;
+       int rc;
+
+       if (proto >= ENA_ADMIN_RSS_PROTO_NUM) {
+               pr_err("Invalid proto num (%u)\n", proto);
+               return -EINVAL;
+       }
+
+       /* Get the ctrl table */
+       rc = ena_com_get_hash_ctrl(ena_dev, proto, NULL);
+       if (unlikely(rc))
+               return rc;
+
+       /* Make sure all the fields are supported */
+       supported_fields = hash_ctrl->supported_fields[proto].fields;
+       if ((hash_fields & supported_fields) != hash_fields) {
+               pr_err("proto %d doesn't support the required fields %x. supports only: %x\n",
+                      proto, hash_fields, supported_fields);
+       }
+
+       hash_ctrl->selected_fields[proto].fields = hash_fields;
+
+       rc = ena_com_set_hash_ctrl(ena_dev);
+
+       /* In case of failure, restore the old hash ctrl */
+       if (unlikely(rc))
+               ena_com_get_hash_ctrl(ena_dev, 0, NULL);
+
+       return 0;
+}
+
+int ena_com_indirect_table_fill_entry(struct ena_com_dev *ena_dev,
+                                     u16 entry_idx, u16 entry_value)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+
+       if (unlikely(entry_idx >= (1 << rss->tbl_log_size)))
+               return -EINVAL;
+
+       if (unlikely((entry_value > ENA_TOTAL_NUM_QUEUES)))
+               return -EINVAL;
+
+       rss->host_rss_ind_tbl[entry_idx] = entry_value;
+
+       return 0;
+}
+
+int ena_com_indirect_table_set(struct ena_com_dev *ena_dev)
+{
+       struct ena_com_admin_queue *admin_queue = &ena_dev->admin_queue;
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_set_feat_cmd cmd;
+       struct ena_admin_set_feat_resp resp;
+       int ret;
+
+       if (!ena_com_check_supported_feature_id(
+                   ena_dev, ENA_ADMIN_RSS_REDIRECTION_TABLE_CONFIG)) {
+               pr_info("Feature %d isn't supported\n",
+                       ENA_ADMIN_RSS_REDIRECTION_TABLE_CONFIG);
+               return -EPERM;
+       }
+
+       ret = ena_com_ind_tbl_convert_to_device(ena_dev);
+       if (ret) {
+               pr_err("Failed to convert host indirection table to device table\n");
+               return ret;
+       }
+
+       memset(&cmd, 0x0, sizeof(cmd));
+
+       cmd.aq_common_descriptor.opcode = ENA_ADMIN_SET_FEATURE;
+       cmd.aq_common_descriptor.flags =
+               ENA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK;
+       cmd.feat_common.feature_id = ENA_ADMIN_RSS_REDIRECTION_TABLE_CONFIG;
+       cmd.u.ind_table.size = rss->tbl_log_size;
+       cmd.u.ind_table.inline_index = 0xFFFFFFFF;
+
+       ret = ena_com_mem_addr_set(ena_dev,
+                                  &cmd.control_buffer.address,
+                                  rss->rss_ind_tbl_dma_addr);
+       if (unlikely(ret)) {
+               pr_err("memory address set failed\n");
+               return ret;
+       }
+
+       cmd.control_buffer.length = (1ULL << rss->tbl_log_size) *
+               sizeof(struct ena_admin_rss_ind_table_entry);
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&cmd,
+                                           sizeof(cmd),
+                                           (struct ena_admin_acq_entry *)&resp,
+                                           sizeof(resp));
+
+       if (unlikely(ret))
+               pr_err("Failed to set indirect table. error: %d\n", ret);
+
+       return ret;
+}
+
+int ena_com_indirect_table_get(struct ena_com_dev *ena_dev, u32 *ind_tbl)
+{
+       struct ena_rss *rss = &ena_dev->rss;
+       struct ena_admin_get_feat_resp get_resp;
+       u32 tbl_size;
+       int i, rc;
+
+       tbl_size = (1ULL << rss->tbl_log_size) *
+               sizeof(struct ena_admin_rss_ind_table_entry);
+
+       rc = ena_com_get_feature_ex(ena_dev, &get_resp,
+                                   ENA_ADMIN_RSS_REDIRECTION_TABLE_CONFIG,
+                                   rss->rss_ind_tbl_dma_addr,
+                                   tbl_size);
+       if (unlikely(rc))
+               return rc;
+
+       if (!ind_tbl)
+               return 0;
+
+       rc = ena_com_ind_tbl_convert_from_device(ena_dev);
+       if (unlikely(rc))
+               return rc;
+
+       for (i = 0; i < (1 << rss->tbl_log_size); i++)
+               ind_tbl[i] = rss->host_rss_ind_tbl[i];
+
+       return 0;
+}
+
+int ena_com_rss_init(struct ena_com_dev *ena_dev, u16 indr_tbl_log_size)
+{
+       int rc;
+
+       memset(&ena_dev->rss, 0x0, sizeof(ena_dev->rss));
+
+       rc = ena_com_indirect_table_allocate(ena_dev, indr_tbl_log_size);
+       if (unlikely(rc))
+               goto err_indr_tbl;
+
+       rc = ena_com_hash_key_allocate(ena_dev);
+       if (unlikely(rc))
+               goto err_hash_key;
+
+       rc = ena_com_hash_ctrl_init(ena_dev);
+       if (unlikely(rc))
+               goto err_hash_ctrl;
+
+       return 0;
+
+err_hash_ctrl:
+       ena_com_hash_key_destroy(ena_dev);
+err_hash_key:
+       ena_com_indirect_table_destroy(ena_dev);
+err_indr_tbl:
+
+       return rc;
+}
+
+void ena_com_rss_destroy(struct ena_com_dev *ena_dev)
+{
+       ena_com_indirect_table_destroy(ena_dev);
+       ena_com_hash_key_destroy(ena_dev);
+       ena_com_hash_ctrl_destroy(ena_dev);
+
+       memset(&ena_dev->rss, 0x0, sizeof(ena_dev->rss));
+}
+
+int ena_com_allocate_host_info(struct ena_com_dev *ena_dev)
+{
+       struct ena_host_attribute *host_attr = &ena_dev->host_attr;
+
+       host_attr->host_info =
+               dma_zalloc_coherent(ena_dev->dmadev, SZ_4K,
+                                   &host_attr->host_info_dma_addr, GFP_KERNEL);
+       if (unlikely(!host_attr->host_info))
+               return -ENOMEM;
+
+       return 0;
+}
+
+int ena_com_allocate_debug_area(struct ena_com_dev *ena_dev,
+                               u32 debug_area_size)
+{
+       struct ena_host_attribute *host_attr = &ena_dev->host_attr;
+
+       host_attr->debug_area_virt_addr =
+               dma_zalloc_coherent(ena_dev->dmadev, debug_area_size,
+                                   &host_attr->debug_area_dma_addr, GFP_KERNEL);
+       if (unlikely(!host_attr->debug_area_virt_addr)) {
+               host_attr->debug_area_size = 0;
+               return -ENOMEM;
+       }
+
+       host_attr->debug_area_size = debug_area_size;
+
+       return 0;
+}
+
+void ena_com_delete_host_info(struct ena_com_dev *ena_dev)
+{
+       struct ena_host_attribute *host_attr = &ena_dev->host_attr;
+
+       if (host_attr->host_info) {
+               dma_free_coherent(ena_dev->dmadev, SZ_4K, host_attr->host_info,
+                                 host_attr->host_info_dma_addr);
+               host_attr->host_info = NULL;
+       }
+}
+
+void ena_com_delete_debug_area(struct ena_com_dev *ena_dev)
+{
+       struct ena_host_attribute *host_attr = &ena_dev->host_attr;
+
+       if (host_attr->debug_area_virt_addr) {
+               dma_free_coherent(ena_dev->dmadev, host_attr->debug_area_size,
+                                 host_attr->debug_area_virt_addr,
+                                 host_attr->debug_area_dma_addr);
+               host_attr->debug_area_virt_addr = NULL;
+       }
+}
+
+int ena_com_set_host_attributes(struct ena_com_dev *ena_dev)
+{
+       struct ena_host_attribute *host_attr = &ena_dev->host_attr;
+       struct ena_com_admin_queue *admin_queue;
+       struct ena_admin_set_feat_cmd cmd;
+       struct ena_admin_set_feat_resp resp;
+
+       int ret;
+
+       if (!ena_com_check_supported_feature_id(ena_dev,
+                                               ENA_ADMIN_HOST_ATTR_CONFIG)) {
+               pr_warn("Set host attribute isn't supported\n");
+               return -EPERM;
+       }
+
+       memset(&cmd, 0x0, sizeof(cmd));
+       admin_queue = &ena_dev->admin_queue;
+
+       cmd.aq_common_descriptor.opcode = ENA_ADMIN_SET_FEATURE;
+       cmd.feat_common.feature_id = ENA_ADMIN_HOST_ATTR_CONFIG;
+
+       ret = ena_com_mem_addr_set(ena_dev,
+                                  &cmd.u.host_attr.debug_ba,
+                                  host_attr->debug_area_dma_addr);
+       if (unlikely(ret)) {
+               pr_err("memory address set failed\n");
+               return ret;
+       }
+
+       ret = ena_com_mem_addr_set(ena_dev,
+                                  &cmd.u.host_attr.os_info_ba,
+                                  host_attr->host_info_dma_addr);
+       if (unlikely(ret)) {
+               pr_err("memory address set failed\n");
+               return ret;
+       }
+
+       cmd.u.host_attr.debug_area_size = host_attr->debug_area_size;
+
+       ret = ena_com_execute_admin_command(admin_queue,
+                                           (struct ena_admin_aq_entry *)&cmd,
+                                           sizeof(cmd),
+                                           (struct ena_admin_acq_entry *)&resp,
+                                           sizeof(resp));
+
+       if (unlikely(ret))
+               pr_err("Failed to set host attributes: %d\n", ret);
+
+       return ret;
+}
+
+/* Interrupt moderation */
+bool ena_com_interrupt_moderation_supported(struct ena_com_dev *ena_dev)
+{
+       return ena_com_check_supported_feature_id(ena_dev,
+                                                 ENA_ADMIN_INTERRUPT_MODERATION);
+}
+
+int ena_com_update_nonadaptive_moderation_interval_tx(struct ena_com_dev *ena_dev,
+                                                     u32 tx_coalesce_usecs)
+{
+       if (!ena_dev->intr_delay_resolution) {
+               pr_err("Illegal interrupt delay granularity value\n");
+               return -EFAULT;
+       }
+
+       ena_dev->intr_moder_tx_interval = tx_coalesce_usecs /
+               ena_dev->intr_delay_resolution;
+
+       return 0;
+}
+
+int ena_com_update_nonadaptive_moderation_interval_rx(struct ena_com_dev *ena_dev,
+                                                     u32 rx_coalesce_usecs)
+{
+       if (!ena_dev->intr_delay_resolution) {
+               pr_err("Illegal interrupt delay granularity value\n");
+               return -EFAULT;
+       }
+
+       /* We use LOWEST entry of moderation table for storing
+        * nonadaptive interrupt coalescing values
+        */
+       ena_dev->intr_moder_tbl[ENA_INTR_MODER_LOWEST].intr_moder_interval =
+               rx_coalesce_usecs / ena_dev->intr_delay_resolution;
+
+       return 0;
+}
+
+void ena_com_destroy_interrupt_moderation(struct ena_com_dev *ena_dev)
+{
+       if (ena_dev->intr_moder_tbl)
+               devm_kfree(ena_dev->dmadev, ena_dev->intr_moder_tbl);
+       ena_dev->intr_moder_tbl = NULL;
+}
+
+int ena_com_init_interrupt_moderation(struct ena_com_dev *ena_dev)
+{
+       struct ena_admin_get_feat_resp get_resp;
+       u16 delay_resolution;
+       int rc;
+
+       rc = ena_com_get_feature(ena_dev, &get_resp,
+                                ENA_ADMIN_INTERRUPT_MODERATION);
+
+       if (rc) {
+               if (rc == -EPERM) {
+                       pr_info("Feature %d isn't supported\n",
+                               ENA_ADMIN_INTERRUPT_MODERATION);
+                       rc = 0;
+               } else {
+                       pr_err("Failed to get interrupt moderation admin cmd. rc: %d\n",
+                              rc);
+               }
+
+               /* no moderation supported, disable adaptive support */
+               ena_com_disable_adaptive_moderation(ena_dev);
+               return rc;
+       }
+
+       rc = ena_com_init_interrupt_moderation_table(ena_dev);
+       if (rc)
+               goto err;
+
+       /* if moderation is supported by device we set adaptive moderation */
+       delay_resolution = get_resp.u.intr_moderation.intr_delay_resolution;
+       ena_com_update_intr_delay_resolution(ena_dev, delay_resolution);
+       ena_com_enable_adaptive_moderation(ena_dev);
+
+       return 0;
+err:
+       ena_com_destroy_interrupt_moderation(ena_dev);
+       return rc;
+}
+
+void ena_com_config_default_interrupt_moderation_table(struct ena_com_dev *ena_dev)
+{
+       struct ena_intr_moder_entry *intr_moder_tbl = ena_dev->intr_moder_tbl;
+
+       if (!intr_moder_tbl)
+               return;
+
+       intr_moder_tbl[ENA_INTR_MODER_LOWEST].intr_moder_interval =
+               ENA_INTR_LOWEST_USECS;
+       intr_moder_tbl[ENA_INTR_MODER_LOWEST].pkts_per_interval =
+               ENA_INTR_LOWEST_PKTS;
+       intr_moder_tbl[ENA_INTR_MODER_LOWEST].bytes_per_interval =
+               ENA_INTR_LOWEST_BYTES;
+
+       intr_moder_tbl[ENA_INTR_MODER_LOW].intr_moder_interval =
+               ENA_INTR_LOW_USECS;
+       intr_moder_tbl[ENA_INTR_MODER_LOW].pkts_per_interval =
+               ENA_INTR_LOW_PKTS;
+       intr_moder_tbl[ENA_INTR_MODER_LOW].bytes_per_interval =
+               ENA_INTR_LOW_BYTES;
+
+       intr_moder_tbl[ENA_INTR_MODER_MID].intr_moder_interval =
+               ENA_INTR_MID_USECS;
+       intr_moder_tbl[ENA_INTR_MODER_MID].pkts_per_interval =
+               ENA_INTR_MID_PKTS;
+       intr_moder_tbl[ENA_INTR_MODER_MID].bytes_per_interval =
+               ENA_INTR_MID_BYTES;
+
+       intr_moder_tbl[ENA_INTR_MODER_HIGH].intr_moder_interval =
+               ENA_INTR_HIGH_USECS;
+       intr_moder_tbl[ENA_INTR_MODER_HIGH].pkts_per_interval =
+               ENA_INTR_HIGH_PKTS;
+       intr_moder_tbl[ENA_INTR_MODER_HIGH].bytes_per_interval =
+               ENA_INTR_HIGH_BYTES;
+
+       intr_moder_tbl[ENA_INTR_MODER_HIGHEST].intr_moder_interval =
+               ENA_INTR_HIGHEST_USECS;
+       intr_moder_tbl[ENA_INTR_MODER_HIGHEST].pkts_per_interval =
+               ENA_INTR_HIGHEST_PKTS;
+       intr_moder_tbl[ENA_INTR_MODER_HIGHEST].bytes_per_interval =
+               ENA_INTR_HIGHEST_BYTES;
+}
+
+unsigned int ena_com_get_nonadaptive_moderation_interval_tx(struct ena_com_dev *ena_dev)
+{
+       return ena_dev->intr_moder_tx_interval;
+}
+
+unsigned int ena_com_get_nonadaptive_moderation_interval_rx(struct ena_com_dev *ena_dev)
+{
+       struct ena_intr_moder_entry *intr_moder_tbl = ena_dev->intr_moder_tbl;
+
+       if (intr_moder_tbl)
+               return intr_moder_tbl[ENA_INTR_MODER_LOWEST].intr_moder_interval;
+
+       return 0;
+}
+
+void ena_com_init_intr_moderation_entry(struct ena_com_dev *ena_dev,
+                                       enum ena_intr_moder_level level,
+                                       struct ena_intr_moder_entry *entry)
+{
+       struct ena_intr_moder_entry *intr_moder_tbl = ena_dev->intr_moder_tbl;
+
+       if (level >= ENA_INTR_MAX_NUM_OF_LEVELS)
+               return;
+
+       intr_moder_tbl[level].intr_moder_interval = entry->intr_moder_interval;
+       if (ena_dev->intr_delay_resolution)
+               intr_moder_tbl[level].intr_moder_interval /=
+                       ena_dev->intr_delay_resolution;
+       intr_moder_tbl[level].pkts_per_interval = entry->pkts_per_interval;
+
+       /* use hardcoded value until ethtool supports bytecount parameter */
+       if (entry->bytes_per_interval != ENA_INTR_BYTE_COUNT_NOT_SUPPORTED)
+               intr_moder_tbl[level].bytes_per_interval = entry->bytes_per_interval;
+}
+
+void ena_com_get_intr_moderation_entry(struct ena_com_dev *ena_dev,
+                                      enum ena_intr_moder_level level,
+                                      struct ena_intr_moder_entry *entry)
+{
+       struct ena_intr_moder_entry *intr_moder_tbl = ena_dev->intr_moder_tbl;
+
+       if (level >= ENA_INTR_MAX_NUM_OF_LEVELS)
+               return;
+
+       entry->intr_moder_interval = intr_moder_tbl[level].intr_moder_interval;
+       if (ena_dev->intr_delay_resolution)
+               entry->intr_moder_interval *= ena_dev->intr_delay_resolution;
+       entry->pkts_per_interval =
+       intr_moder_tbl[level].pkts_per_interval;
+       entry->bytes_per_interval = intr_moder_tbl[level].bytes_per_interval;
+}
diff --git a/drivers/net/ethernet/amazon/ena/ena_com.h b/drivers/net/ethernet/amazon/ena/ena_com.h
new file mode 100644 (file)
index 0000000..509d7b8
--- /dev/null
@@ -0,0 +1,1038 @@
+/*
+ * Copyright 2015 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef ENA_COM
+#define ENA_COM
+
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/gfp.h>
+#include <linux/sched.h>
+#include <linux/sizes.h>
+#include <linux/spinlock.h>
+#include <linux/types.h>
+#include <linux/wait.h>
+
+#include "ena_common_defs.h"
+#include "ena_admin_defs.h"
+#include "ena_eth_io_defs.h"
+#include "ena_regs_defs.h"
+
+#undef pr_fmt
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#define ENA_MAX_NUM_IO_QUEUES          128U
+/* We need to queues for each IO (on for Tx and one for Rx) */
+#define ENA_TOTAL_NUM_QUEUES           (2 * (ENA_MAX_NUM_IO_QUEUES))
+
+#define ENA_MAX_HANDLERS 256
+
+#define ENA_MAX_PHYS_ADDR_SIZE_BITS 48
+
+/* Unit in usec */
+#define ENA_REG_READ_TIMEOUT 200000
+
+#define ADMIN_SQ_SIZE(depth)   ((depth) * sizeof(struct ena_admin_aq_entry))
+#define ADMIN_CQ_SIZE(depth)   ((depth) * sizeof(struct ena_admin_acq_entry))
+#define ADMIN_AENQ_SIZE(depth) ((depth) * sizeof(struct ena_admin_aenq_entry))
+
+/*****************************************************************************/
+/*****************************************************************************/
+/* ENA adaptive interrupt moderation settings */
+
+#define ENA_INTR_LOWEST_USECS           (0)
+#define ENA_INTR_LOWEST_PKTS            (3)
+#define ENA_INTR_LOWEST_BYTES           (2 * 1524)
+
+#define ENA_INTR_LOW_USECS              (32)
+#define ENA_INTR_LOW_PKTS               (12)
+#define ENA_INTR_LOW_BYTES              (16 * 1024)
+
+#define ENA_INTR_MID_USECS              (80)
+#define ENA_INTR_MID_PKTS               (48)
+#define ENA_INTR_MID_BYTES              (64 * 1024)
+
+#define ENA_INTR_HIGH_USECS             (128)
+#define ENA_INTR_HIGH_PKTS              (96)
+#define ENA_INTR_HIGH_BYTES             (128 * 1024)
+
+#define ENA_INTR_HIGHEST_USECS          (192)
+#define ENA_INTR_HIGHEST_PKTS           (128)
+#define ENA_INTR_HIGHEST_BYTES          (192 * 1024)
+
+#define ENA_INTR_INITIAL_TX_INTERVAL_USECS             196
+#define ENA_INTR_INITIAL_RX_INTERVAL_USECS             4
+#define ENA_INTR_DELAY_OLD_VALUE_WEIGHT                        6
+#define ENA_INTR_DELAY_NEW_VALUE_WEIGHT                        4
+#define ENA_INTR_MODER_LEVEL_STRIDE                    2
+#define ENA_INTR_BYTE_COUNT_NOT_SUPPORTED              0xFFFFFF
+
+enum ena_intr_moder_level {
+       ENA_INTR_MODER_LOWEST = 0,
+       ENA_INTR_MODER_LOW,
+       ENA_INTR_MODER_MID,
+       ENA_INTR_MODER_HIGH,
+       ENA_INTR_MODER_HIGHEST,
+       ENA_INTR_MAX_NUM_OF_LEVELS,
+};
+
+struct ena_intr_moder_entry {
+       unsigned int intr_moder_interval;
+       unsigned int pkts_per_interval;
+       unsigned int bytes_per_interval;
+};
+
+enum queue_direction {
+       ENA_COM_IO_QUEUE_DIRECTION_TX,
+       ENA_COM_IO_QUEUE_DIRECTION_RX
+};
+
+struct ena_com_buf {
+       dma_addr_t paddr; /**< Buffer physical address */
+       u16 len; /**< Buffer length in bytes */
+};
+
+struct ena_com_rx_buf_info {
+       u16 len;
+       u16 req_id;
+};
+
+struct ena_com_io_desc_addr {
+       u8 __iomem *pbuf_dev_addr; /* LLQ address */
+       u8 *virt_addr;
+       dma_addr_t phys_addr;
+};
+
+struct ena_com_tx_meta {
+       u16 mss;
+       u16 l3_hdr_len;
+       u16 l3_hdr_offset;
+       u16 l4_hdr_len; /* In words */
+};
+
+struct ena_com_io_cq {
+       struct ena_com_io_desc_addr cdesc_addr;
+
+       /* Interrupt unmask register */
+       u32 __iomem *unmask_reg;
+
+       /* The completion queue head doorbell register */
+       u32 __iomem *cq_head_db_reg;
+
+       /* numa configuration register (for TPH) */
+       u32 __iomem *numa_node_cfg_reg;
+
+       /* The value to write to the above register to unmask
+        * the interrupt of this queue
+        */
+       u32 msix_vector;
+
+       enum queue_direction direction;
+
+       /* holds the number of cdesc of the current packet */
+       u16 cur_rx_pkt_cdesc_count;
+       /* save the firt cdesc idx of the current packet */
+       u16 cur_rx_pkt_cdesc_start_idx;
+
+       u16 q_depth;
+       /* Caller qid */
+       u16 qid;
+
+       /* Device queue index */
+       u16 idx;
+       u16 head;
+       u16 last_head_update;
+       u8 phase;
+       u8 cdesc_entry_size_in_bytes;
+
+} ____cacheline_aligned;
+
+struct ena_com_io_sq {
+       struct ena_com_io_desc_addr desc_addr;
+
+       u32 __iomem *db_addr;
+       u8 __iomem *header_addr;
+
+       enum queue_direction direction;
+       enum ena_admin_placement_policy_type mem_queue_type;
+
+       u32 msix_vector;
+       struct ena_com_tx_meta cached_tx_meta;
+
+       u16 q_depth;
+       u16 qid;
+
+       u16 idx;
+       u16 tail;
+       u16 next_to_comp;
+       u32 tx_max_header_size;
+       u8 phase;
+       u8 desc_entry_size;
+       u8 dma_addr_bits;
+} ____cacheline_aligned;
+
+struct ena_com_admin_cq {
+       struct ena_admin_acq_entry *entries;
+       dma_addr_t dma_addr;
+
+       u16 head;
+       u8 phase;
+};
+
+struct ena_com_admin_sq {
+       struct ena_admin_aq_entry *entries;
+       dma_addr_t dma_addr;
+
+       u32 __iomem *db_addr;
+
+       u16 head;
+       u16 tail;
+       u8 phase;
+
+};
+
+struct ena_com_stats_admin {
+       u32 aborted_cmd;
+       u32 submitted_cmd;
+       u32 completed_cmd;
+       u32 out_of_space;
+       u32 no_completion;
+};
+
+struct ena_com_admin_queue {
+       void *q_dmadev;
+       spinlock_t q_lock; /* spinlock for the admin queue */
+       struct ena_comp_ctx *comp_ctx;
+       u16 q_depth;
+       struct ena_com_admin_cq cq;
+       struct ena_com_admin_sq sq;
+
+       /* Indicate if the admin queue should poll for completion */
+       bool polling;
+
+       u16 curr_cmd_id;
+
+       /* Indicate that the ena was initialized and can
+        * process new admin commands
+        */
+       bool running_state;
+
+       /* Count the number of outstanding admin commands */
+       atomic_t outstanding_cmds;
+
+       struct ena_com_stats_admin stats;
+};
+
+struct ena_aenq_handlers;
+
+struct ena_com_aenq {
+       u16 head;
+       u8 phase;
+       struct ena_admin_aenq_entry *entries;
+       dma_addr_t dma_addr;
+       u16 q_depth;
+       struct ena_aenq_handlers *aenq_handlers;
+};
+
+struct ena_com_mmio_read {
+       struct ena_admin_ena_mmio_req_read_less_resp *read_resp;
+       dma_addr_t read_resp_dma_addr;
+       u16 seq_num;
+       bool readless_supported;
+       /* spin lock to ensure a single outstanding read */
+       spinlock_t lock;
+};
+
+struct ena_rss {
+       /* Indirect table */
+       u16 *host_rss_ind_tbl;
+       struct ena_admin_rss_ind_table_entry *rss_ind_tbl;
+       dma_addr_t rss_ind_tbl_dma_addr;
+       u16 tbl_log_size;
+
+       /* Hash key */
+       enum ena_admin_hash_functions hash_func;
+       struct ena_admin_feature_rss_flow_hash_control *hash_key;
+       dma_addr_t hash_key_dma_addr;
+       u32 hash_init_val;
+
+       /* Flow Control */
+       struct ena_admin_feature_rss_hash_control *hash_ctrl;
+       dma_addr_t hash_ctrl_dma_addr;
+
+};
+
+struct ena_host_attribute {
+       /* Debug area */
+       u8 *debug_area_virt_addr;
+       dma_addr_t debug_area_dma_addr;
+       u32 debug_area_size;
+
+       /* Host information */
+       struct ena_admin_host_info *host_info;
+       dma_addr_t host_info_dma_addr;
+};
+
+/* Each ena_dev is a PCI function. */
+struct ena_com_dev {
+       struct ena_com_admin_queue admin_queue;
+       struct ena_com_aenq aenq;
+       struct ena_com_io_cq io_cq_queues[ENA_TOTAL_NUM_QUEUES];
+       struct ena_com_io_sq io_sq_queues[ENA_TOTAL_NUM_QUEUES];
+       u8 __iomem *reg_bar;
+       void __iomem *mem_bar;
+       void *dmadev;
+
+       enum ena_admin_placement_policy_type tx_mem_queue_type;
+       u32 tx_max_header_size;
+       u16 stats_func; /* Selected function for extended statistic dump */
+       u16 stats_queue; /* Selected queue for extended statistic dump */
+
+       struct ena_com_mmio_read mmio_read;
+
+       struct ena_rss rss;
+       u32 supported_features;
+       u32 dma_addr_bits;
+
+       struct ena_host_attribute host_attr;
+       bool adaptive_coalescing;
+       u16 intr_delay_resolution;
+       u32 intr_moder_tx_interval;
+       struct ena_intr_moder_entry *intr_moder_tbl;
+};
+
+struct ena_com_dev_get_features_ctx {
+       struct ena_admin_queue_feature_desc max_queues;
+       struct ena_admin_device_attr_feature_desc dev_attr;
+       struct ena_admin_feature_aenq_desc aenq;
+       struct ena_admin_feature_offload_desc offload;
+};
+
+struct ena_com_create_io_ctx {
+       enum ena_admin_placement_policy_type mem_queue_type;
+       enum queue_direction direction;
+       int numa_node;
+       u32 msix_vector;
+       u16 queue_size;
+       u16 qid;
+};
+
+typedef void (*ena_aenq_handler)(void *data,
+       struct ena_admin_aenq_entry *aenq_e);
+
+/* Holds aenq handlers. Indexed by AENQ event group */
+struct ena_aenq_handlers {
+       ena_aenq_handler handlers[ENA_MAX_HANDLERS];
+       ena_aenq_handler unimplemented_handler;
+};
+
+/*****************************************************************************/
+/*****************************************************************************/
+
+/* ena_com_mmio_reg_read_request_init - Init the mmio reg read mechanism
+ * @ena_dev: ENA communication layer struct
+ *
+ * Initialize the register read mechanism.
+ *
+ * @note: This method must be the first stage in the initialization sequence.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_mmio_reg_read_request_init(struct ena_com_dev *ena_dev);
+
+/* ena_com_set_mmio_read_mode - Enable/disable the mmio reg read mechanism
+ * @ena_dev: ENA communication layer struct
+ * @readless_supported: readless mode (enable/disable)
+ */
+void ena_com_set_mmio_read_mode(struct ena_com_dev *ena_dev,
+                               bool readless_supported);
+
+/* ena_com_mmio_reg_read_request_write_dev_addr - Write the mmio reg read return
+ * value physical address.
+ * @ena_dev: ENA communication layer struct
+ */
+void ena_com_mmio_reg_read_request_write_dev_addr(struct ena_com_dev *ena_dev);
+
+/* ena_com_mmio_reg_read_request_destroy - Destroy the mmio reg read mechanism
+ * @ena_dev: ENA communication layer struct
+ */
+void ena_com_mmio_reg_read_request_destroy(struct ena_com_dev *ena_dev);
+
+/* ena_com_admin_init - Init the admin and the async queues
+ * @ena_dev: ENA communication layer struct
+ * @aenq_handlers: Those handlers to be called upon event.
+ * @init_spinlock: Indicate if this method should init the admin spinlock or
+ * the spinlock was init before (for example, in a case of FLR).
+ *
+ * Initialize the admin submission and completion queues.
+ * Initialize the asynchronous events notification queues.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_admin_init(struct ena_com_dev *ena_dev,
+                      struct ena_aenq_handlers *aenq_handlers,
+                      bool init_spinlock);
+
+/* ena_com_admin_destroy - Destroy the admin and the async events queues.
+ * @ena_dev: ENA communication layer struct
+ *
+ * @note: Before calling this method, the caller must validate that the device
+ * won't send any additional admin completions/aenq.
+ * To achieve that, a FLR is recommended.
+ */
+void ena_com_admin_destroy(struct ena_com_dev *ena_dev);
+
+/* ena_com_dev_reset - Perform device FLR to the device.
+ * @ena_dev: ENA communication layer struct
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_dev_reset(struct ena_com_dev *ena_dev);
+
+/* ena_com_create_io_queue - Create io queue.
+ * @ena_dev: ENA communication layer struct
+ * @ctx - create context structure
+ *
+ * Create the submission and the completion queues.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_create_io_queue(struct ena_com_dev *ena_dev,
+                           struct ena_com_create_io_ctx *ctx);
+
+/* ena_com_destroy_io_queue - Destroy IO queue with the queue id - qid.
+ * @ena_dev: ENA communication layer struct
+ * @qid - the caller virtual queue id.
+ */
+void ena_com_destroy_io_queue(struct ena_com_dev *ena_dev, u16 qid);
+
+/* ena_com_get_io_handlers - Return the io queue handlers
+ * @ena_dev: ENA communication layer struct
+ * @qid - the caller virtual queue id.
+ * @io_sq - IO submission queue handler
+ * @io_cq - IO completion queue handler.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_get_io_handlers(struct ena_com_dev *ena_dev, u16 qid,
+                           struct ena_com_io_sq **io_sq,
+                           struct ena_com_io_cq **io_cq);
+
+/* ena_com_admin_aenq_enable - ENAble asynchronous event notifications
+ * @ena_dev: ENA communication layer struct
+ *
+ * After this method, aenq event can be received via AENQ.
+ */
+void ena_com_admin_aenq_enable(struct ena_com_dev *ena_dev);
+
+/* ena_com_set_admin_running_state - Set the state of the admin queue
+ * @ena_dev: ENA communication layer struct
+ *
+ * Change the state of the admin queue (enable/disable)
+ */
+void ena_com_set_admin_running_state(struct ena_com_dev *ena_dev, bool state);
+
+/* ena_com_get_admin_running_state - Get the admin queue state
+ * @ena_dev: ENA communication layer struct
+ *
+ * Retrieve the state of the admin queue (enable/disable)
+ *
+ * @return - current polling mode (enable/disable)
+ */
+bool ena_com_get_admin_running_state(struct ena_com_dev *ena_dev);
+
+/* ena_com_set_admin_polling_mode - Set the admin completion queue polling mode
+ * @ena_dev: ENA communication layer struct
+ * @polling: ENAble/Disable polling mode
+ *
+ * Set the admin completion mode.
+ */
+void ena_com_set_admin_polling_mode(struct ena_com_dev *ena_dev, bool polling);
+
+/* ena_com_set_admin_polling_mode - Get the admin completion queue polling mode
+ * @ena_dev: ENA communication layer struct
+ *
+ * Get the admin completion mode.
+ * If polling mode is on, ena_com_execute_admin_command will perform a
+ * polling on the admin completion queue for the commands completion,
+ * otherwise it will wait on wait event.
+ *
+ * @return state
+ */
+bool ena_com_get_ena_admin_polling_mode(struct ena_com_dev *ena_dev);
+
+/* ena_com_admin_q_comp_intr_handler - admin queue interrupt handler
+ * @ena_dev: ENA communication layer struct
+ *
+ * This method go over the admin completion queue and wake up all the pending
+ * threads that wait on the commands wait event.
+ *
+ * @note: Should be called after MSI-X interrupt.
+ */
+void ena_com_admin_q_comp_intr_handler(struct ena_com_dev *ena_dev);
+
+/* ena_com_aenq_intr_handler - AENQ interrupt handler
+ * @ena_dev: ENA communication layer struct
+ *
+ * This method go over the async event notification queue and call the proper
+ * aenq handler.
+ */
+void ena_com_aenq_intr_handler(struct ena_com_dev *dev, void *data);
+
+/* ena_com_abort_admin_commands - Abort all the outstanding admin commands.
+ * @ena_dev: ENA communication layer struct
+ *
+ * This method aborts all the outstanding admin commands.
+ * The caller should then call ena_com_wait_for_abort_completion to make sure
+ * all the commands were completed.
+ */
+void ena_com_abort_admin_commands(struct ena_com_dev *ena_dev);
+
+/* ena_com_wait_for_abort_completion - Wait for admin commands abort.
+ * @ena_dev: ENA communication layer struct
+ *
+ * This method wait until all the outstanding admin commands will be completed.
+ */
+void ena_com_wait_for_abort_completion(struct ena_com_dev *ena_dev);
+
+/* ena_com_validate_version - Validate the device parameters
+ * @ena_dev: ENA communication layer struct
+ *
+ * This method validate the device parameters are the same as the saved
+ * parameters in ena_dev.
+ * This method is useful after device reset, to validate the device mac address
+ * and the device offloads are the same as before the reset.
+ *
+ * @return - 0 on success negative value otherwise.
+ */
+int ena_com_validate_version(struct ena_com_dev *ena_dev);
+
+/* ena_com_get_link_params - Retrieve physical link parameters.
+ * @ena_dev: ENA communication layer struct
+ * @resp: Link parameters
+ *
+ * Retrieve the physical link parameters,
+ * like speed, auto-negotiation and full duplex support.
+ *
+ * @return - 0 on Success negative value otherwise.
+ */
+int ena_com_get_link_params(struct ena_com_dev *ena_dev,
+                           struct ena_admin_get_feat_resp *resp);
+
+/* ena_com_get_dma_width - Retrieve physical dma address width the device
+ * supports.
+ * @ena_dev: ENA communication layer struct
+ *
+ * Retrieve the maximum physical address bits the device can handle.
+ *
+ * @return: > 0 on Success and negative value otherwise.
+ */
+int ena_com_get_dma_width(struct ena_com_dev *ena_dev);
+
+/* ena_com_set_aenq_config - Set aenq groups configurations
+ * @ena_dev: ENA communication layer struct
+ * @groups flag: bit fields flags of enum ena_admin_aenq_group.
+ *
+ * Configure which aenq event group the driver would like to receive.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_set_aenq_config(struct ena_com_dev *ena_dev, u32 groups_flag);
+
+/* ena_com_get_dev_attr_feat - Get device features
+ * @ena_dev: ENA communication layer struct
+ * @get_feat_ctx: returned context that contain the get features.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_get_dev_attr_feat(struct ena_com_dev *ena_dev,
+                             struct ena_com_dev_get_features_ctx *get_feat_ctx);
+
+/* ena_com_get_dev_basic_stats - Get device basic statistics
+ * @ena_dev: ENA communication layer struct
+ * @stats: stats return value
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_get_dev_basic_stats(struct ena_com_dev *ena_dev,
+                               struct ena_admin_basic_stats *stats);
+
+/* ena_com_set_dev_mtu - Configure the device mtu.
+ * @ena_dev: ENA communication layer struct
+ * @mtu: mtu value
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_set_dev_mtu(struct ena_com_dev *ena_dev, int mtu);
+
+/* ena_com_get_offload_settings - Retrieve the device offloads capabilities
+ * @ena_dev: ENA communication layer struct
+ * @offlad: offload return value
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_get_offload_settings(struct ena_com_dev *ena_dev,
+                                struct ena_admin_feature_offload_desc *offload);
+
+/* ena_com_rss_init - Init RSS
+ * @ena_dev: ENA communication layer struct
+ * @log_size: indirection log size
+ *
+ * Allocate RSS/RFS resources.
+ * The caller then can configure rss using ena_com_set_hash_function,
+ * ena_com_set_hash_ctrl and ena_com_indirect_table_set.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_rss_init(struct ena_com_dev *ena_dev, u16 log_size);
+
+/* ena_com_rss_destroy - Destroy rss
+ * @ena_dev: ENA communication layer struct
+ *
+ * Free all the RSS/RFS resources.
+ */
+void ena_com_rss_destroy(struct ena_com_dev *ena_dev);
+
+/* ena_com_fill_hash_function - Fill RSS hash function
+ * @ena_dev: ENA communication layer struct
+ * @func: The hash function (Toeplitz or crc)
+ * @key: Hash key (for toeplitz hash)
+ * @key_len: key length (max length 10 DW)
+ * @init_val: initial value for the hash function
+ *
+ * Fill the ena_dev resources with the desire hash function, hash key, key_len
+ * and key initial value (if needed by the hash function).
+ * To flush the key into the device the caller should call
+ * ena_com_set_hash_function.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_fill_hash_function(struct ena_com_dev *ena_dev,
+                              enum ena_admin_hash_functions func,
+                              const u8 *key, u16 key_len, u32 init_val);
+
+/* ena_com_set_hash_function - Flush the hash function and it dependencies to
+ * the device.
+ * @ena_dev: ENA communication layer struct
+ *
+ * Flush the hash function and it dependencies (key, key length and
+ * initial value) if needed.
+ *
+ * @note: Prior to this method the caller should call ena_com_fill_hash_function
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_set_hash_function(struct ena_com_dev *ena_dev);
+
+/* ena_com_get_hash_function - Retrieve the hash function and the hash key
+ * from the device.
+ * @ena_dev: ENA communication layer struct
+ * @func: hash function
+ * @key: hash key
+ *
+ * Retrieve the hash function and the hash key from the device.
+ *
+ * @note: If the caller called ena_com_fill_hash_function but didn't flash
+ * it to the device, the new configuration will be lost.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_get_hash_function(struct ena_com_dev *ena_dev,
+                             enum ena_admin_hash_functions *func,
+                             u8 *key);
+
+/* ena_com_fill_hash_ctrl - Fill RSS hash control
+ * @ena_dev: ENA communication layer struct.
+ * @proto: The protocol to configure.
+ * @hash_fields: bit mask of ena_admin_flow_hash_fields
+ *
+ * Fill the ena_dev resources with the desire hash control (the ethernet
+ * fields that take part of the hash) for a specific protocol.
+ * To flush the hash control to the device, the caller should call
+ * ena_com_set_hash_ctrl.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_fill_hash_ctrl(struct ena_com_dev *ena_dev,
+                          enum ena_admin_flow_hash_proto proto,
+                          u16 hash_fields);
+
+/* ena_com_set_hash_ctrl - Flush the hash control resources to the device.
+ * @ena_dev: ENA communication layer struct
+ *
+ * Flush the hash control (the ethernet fields that take part of the hash)
+ *
+ * @note: Prior to this method the caller should call ena_com_fill_hash_ctrl.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_set_hash_ctrl(struct ena_com_dev *ena_dev);
+
+/* ena_com_get_hash_ctrl - Retrieve the hash control from the device.
+ * @ena_dev: ENA communication layer struct
+ * @proto: The protocol to retrieve.
+ * @fields: bit mask of ena_admin_flow_hash_fields.
+ *
+ * Retrieve the hash control from the device.
+ *
+ * @note, If the caller called ena_com_fill_hash_ctrl but didn't flash
+ * it to the device, the new configuration will be lost.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_get_hash_ctrl(struct ena_com_dev *ena_dev,
+                         enum ena_admin_flow_hash_proto proto,
+                         u16 *fields);
+
+/* ena_com_set_default_hash_ctrl - Set the hash control to a default
+ * configuration.
+ * @ena_dev: ENA communication layer struct
+ *
+ * Fill the ena_dev resources with the default hash control configuration.
+ * To flush the hash control to the device, the caller should call
+ * ena_com_set_hash_ctrl.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_set_default_hash_ctrl(struct ena_com_dev *ena_dev);
+
+/* ena_com_indirect_table_fill_entry - Fill a single entry in the RSS
+ * indirection table
+ * @ena_dev: ENA communication layer struct.
+ * @entry_idx - indirection table entry.
+ * @entry_value - redirection value
+ *
+ * Fill a single entry of the RSS indirection table in the ena_dev resources.
+ * To flush the indirection table to the device, the called should call
+ * ena_com_indirect_table_set.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_indirect_table_fill_entry(struct ena_com_dev *ena_dev,
+                                     u16 entry_idx, u16 entry_value);
+
+/* ena_com_indirect_table_set - Flush the indirection table to the device.
+ * @ena_dev: ENA communication layer struct
+ *
+ * Flush the indirection hash control to the device.
+ * Prior to this method the caller should call ena_com_indirect_table_fill_entry
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_indirect_table_set(struct ena_com_dev *ena_dev);
+
+/* ena_com_indirect_table_get - Retrieve the indirection table from the device.
+ * @ena_dev: ENA communication layer struct
+ * @ind_tbl: indirection table
+ *
+ * Retrieve the RSS indirection table from the device.
+ *
+ * @note: If the caller called ena_com_indirect_table_fill_entry but didn't flash
+ * it to the device, the new configuration will be lost.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_indirect_table_get(struct ena_com_dev *ena_dev, u32 *ind_tbl);
+
+/* ena_com_allocate_host_info - Allocate host info resources.
+ * @ena_dev: ENA communication layer struct
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_allocate_host_info(struct ena_com_dev *ena_dev);
+
+/* ena_com_allocate_debug_area - Allocate debug area.
+ * @ena_dev: ENA communication layer struct
+ * @debug_area_size - debug area size.
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_allocate_debug_area(struct ena_com_dev *ena_dev,
+                               u32 debug_area_size);
+
+/* ena_com_delete_debug_area - Free the debug area resources.
+ * @ena_dev: ENA communication layer struct
+ *
+ * Free the allocate debug area.
+ */
+void ena_com_delete_debug_area(struct ena_com_dev *ena_dev);
+
+/* ena_com_delete_host_info - Free the host info resources.
+ * @ena_dev: ENA communication layer struct
+ *
+ * Free the allocate host info.
+ */
+void ena_com_delete_host_info(struct ena_com_dev *ena_dev);
+
+/* ena_com_set_host_attributes - Update the device with the host
+ * attributes (debug area and host info) base address.
+ * @ena_dev: ENA communication layer struct
+ *
+ * @return: 0 on Success and negative value otherwise.
+ */
+int ena_com_set_host_attributes(struct ena_com_dev *ena_dev);
+
+/* ena_com_create_io_cq - Create io completion queue.
+ * @ena_dev: ENA communication layer struct
+ * @io_cq - io completion queue handler
+
+ * Create IO completion queue.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_create_io_cq(struct ena_com_dev *ena_dev,
+                        struct ena_com_io_cq *io_cq);
+
+/* ena_com_destroy_io_cq - Destroy io completion queue.
+ * @ena_dev: ENA communication layer struct
+ * @io_cq - io completion queue handler
+
+ * Destroy IO completion queue.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_destroy_io_cq(struct ena_com_dev *ena_dev,
+                         struct ena_com_io_cq *io_cq);
+
+/* ena_com_execute_admin_command - Execute admin command
+ * @admin_queue: admin queue.
+ * @cmd: the admin command to execute.
+ * @cmd_size: the command size.
+ * @cmd_completion: command completion return value.
+ * @cmd_comp_size: command completion size.
+
+ * Submit an admin command and then wait until the device will return a
+ * completion.
+ * The completion will be copyed into cmd_comp.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_execute_admin_command(struct ena_com_admin_queue *admin_queue,
+                                 struct ena_admin_aq_entry *cmd,
+                                 size_t cmd_size,
+                                 struct ena_admin_acq_entry *cmd_comp,
+                                 size_t cmd_comp_size);
+
+/* ena_com_init_interrupt_moderation - Init interrupt moderation
+ * @ena_dev: ENA communication layer struct
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_init_interrupt_moderation(struct ena_com_dev *ena_dev);
+
+/* ena_com_destroy_interrupt_moderation - Destroy interrupt moderation resources
+ * @ena_dev: ENA communication layer struct
+ */
+void ena_com_destroy_interrupt_moderation(struct ena_com_dev *ena_dev);
+
+/* ena_com_interrupt_moderation_supported - Return if interrupt moderation
+ * capability is supported by the device.
+ *
+ * @return - supported or not.
+ */
+bool ena_com_interrupt_moderation_supported(struct ena_com_dev *ena_dev);
+
+/* ena_com_config_default_interrupt_moderation_table - Restore the interrupt
+ * moderation table back to the default parameters.
+ * @ena_dev: ENA communication layer struct
+ */
+void ena_com_config_default_interrupt_moderation_table(struct ena_com_dev *ena_dev);
+
+/* ena_com_update_nonadaptive_moderation_interval_tx - Update the
+ * non-adaptive interval in Tx direction.
+ * @ena_dev: ENA communication layer struct
+ * @tx_coalesce_usecs: Interval in usec.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_update_nonadaptive_moderation_interval_tx(struct ena_com_dev *ena_dev,
+                                                     u32 tx_coalesce_usecs);
+
+/* ena_com_update_nonadaptive_moderation_interval_rx - Update the
+ * non-adaptive interval in Rx direction.
+ * @ena_dev: ENA communication layer struct
+ * @rx_coalesce_usecs: Interval in usec.
+ *
+ * @return - 0 on success, negative value on failure.
+ */
+int ena_com_update_nonadaptive_moderation_interval_rx(struct ena_com_dev *ena_dev,
+                                                     u32 rx_coalesce_usecs);
+
+/* ena_com_get_nonadaptive_moderation_interval_tx - Retrieve the
+ * non-adaptive interval in Tx direction.
+ * @ena_dev: ENA communication layer struct
+ *
+ * @return - interval in usec
+ */
+unsigned int ena_com_get_nonadaptive_moderation_interval_tx(struct ena_com_dev *ena_dev);
+
+/* ena_com_get_nonadaptive_moderation_interval_rx - Retrieve the
+ * non-adaptive interval in Rx direction.
+ * @ena_dev: ENA communication layer struct
+ *
+ * @return - interval in usec
+ */
+unsigned int ena_com_get_nonadaptive_moderation_interval_rx(struct ena_com_dev *ena_dev);
+
+/* ena_com_init_intr_moderation_entry - Update a single entry in the interrupt
+ * moderation table.
+ * @ena_dev: ENA communication layer struct
+ * @level: Interrupt moderation table level
+ * @entry: Entry value
+ *
+ * Update a single entry in the interrupt moderation table.
+ */
+void ena_com_init_intr_moderation_entry(struct ena_com_dev *ena_dev,
+                                       enum ena_intr_moder_level level,
+                                       struct ena_intr_moder_entry *entry);
+
+/* ena_com_get_intr_moderation_entry - Init ena_intr_moder_entry.
+ * @ena_dev: ENA communication layer struct
+ * @level: Interrupt moderation table level
+ * @entry: Entry to fill.
+ *
+ * Initialize the entry according to the adaptive interrupt moderation table.
+ */
+void ena_com_get_intr_moderation_entry(struct ena_com_dev *ena_dev,
+                                      enum ena_intr_moder_level level,
+                                      struct ena_intr_moder_entry *entry);
+
+static inline bool ena_com_get_adaptive_moderation_enabled(struct ena_com_dev *ena_dev)
+{
+       return ena_dev->adaptive_coalescing;
+}
+
+static inline void ena_com_enable_adaptive_moderation(struct ena_com_dev *ena_dev)
+{
+       ena_dev->adaptive_coalescing = true;
+}
+
+static inline void ena_com_disable_adaptive_moderation(struct ena_com_dev *ena_dev)
+{
+       ena_dev->adaptive_coalescing = false;
+}
+
+/* ena_com_calculate_interrupt_delay - Calculate new interrupt delay
+ * @ena_dev: ENA communication layer struct
+ * @pkts: Number of packets since the last update
+ * @bytes: Number of bytes received since the last update.
+ * @smoothed_interval: Returned interval
+ * @moder_tbl_idx: Current table level as input update new level as return
+ * value.
+ */
+static inline void ena_com_calculate_interrupt_delay(struct ena_com_dev *ena_dev,
+                                                    unsigned int pkts,
+                                                    unsigned int bytes,
+                                                    unsigned int *smoothed_interval,
+                                                    unsigned int *moder_tbl_idx)
+{
+       enum ena_intr_moder_level curr_moder_idx, new_moder_idx;
+       struct ena_intr_moder_entry *curr_moder_entry;
+       struct ena_intr_moder_entry *pred_moder_entry;
+       struct ena_intr_moder_entry *new_moder_entry;
+       struct ena_intr_moder_entry *intr_moder_tbl = ena_dev->intr_moder_tbl;
+       unsigned int interval;
+
+       /* We apply adaptive moderation on Rx path only.
+        * Tx uses static interrupt moderation.
+        */
+       if (!pkts || !bytes)
+               /* Tx interrupt, or spurious interrupt,
+                * in both cases we just use same delay values
+                */
+               return;
+
+       curr_moder_idx = (enum ena_intr_moder_level)(*moder_tbl_idx);
+       if (unlikely(curr_moder_idx >= ENA_INTR_MAX_NUM_OF_LEVELS)) {
+               pr_err("Wrong moderation index %u\n", curr_moder_idx);
+               return;
+       }
+
+       curr_moder_entry = &intr_moder_tbl[curr_moder_idx];
+       new_moder_idx = curr_moder_idx;
+
+       if (curr_moder_idx == ENA_INTR_MODER_LOWEST) {
+               if ((pkts > curr_moder_entry->pkts_per_interval) ||
+                   (bytes > curr_moder_entry->bytes_per_interval))
+                       new_moder_idx =
+                               (enum ena_intr_moder_level)(curr_moder_idx + ENA_INTR_MODER_LEVEL_STRIDE);
+       } else {
+               pred_moder_entry = &intr_moder_tbl[curr_moder_idx - ENA_INTR_MODER_LEVEL_STRIDE];
+
+               if ((pkts <= pred_moder_entry->pkts_per_interval) ||
+                   (bytes <= pred_moder_entry->bytes_per_interval))
+                       new_moder_idx =
+                               (enum ena_intr_moder_level)(curr_moder_idx - ENA_INTR_MODER_LEVEL_STRIDE);
+               else if ((pkts > curr_moder_entry->pkts_per_interval) ||
+                        (bytes > curr_moder_entry->bytes_per_interval)) {
+                       if (curr_moder_idx != ENA_INTR_MODER_HIGHEST)
+                               new_moder_idx =
+                                       (enum ena_intr_moder_level)(curr_moder_idx + ENA_INTR_MODER_LEVEL_STRIDE);
+               }
+       }
+       new_moder_entry = &intr_moder_tbl[new_moder_idx];
+
+       interval = new_moder_entry->intr_moder_interval;
+       *smoothed_interval = (
+               (interval * ENA_INTR_DELAY_NEW_VALUE_WEIGHT +
+               ENA_INTR_DELAY_OLD_VALUE_WEIGHT * (*smoothed_interval)) + 5) /
+               10;
+
+       *moder_tbl_idx = new_moder_idx;
+}
+
+/* ena_com_update_intr_reg - Prepare interrupt register
+ * @intr_reg: interrupt register to update.
+ * @rx_delay_interval: Rx interval in usecs
+ * @tx_delay_interval: Tx interval in usecs
+ * @unmask: unask enable/disable
+ *
+ * Prepare interrupt update register with the supplied parameters.
+ */
+static inline void ena_com_update_intr_reg(struct ena_eth_io_intr_reg *intr_reg,
+                                          u32 rx_delay_interval,
+                                          u32 tx_delay_interval,
+                                          bool unmask)
+{
+       intr_reg->intr_control = 0;
+       intr_reg->intr_control |= rx_delay_interval &
+               ENA_ETH_IO_INTR_REG_RX_INTR_DELAY_MASK;
+
+       intr_reg->intr_control |=
+               (tx_delay_interval << ENA_ETH_IO_INTR_REG_TX_INTR_DELAY_SHIFT)
+               & ENA_ETH_IO_INTR_REG_TX_INTR_DELAY_MASK;
+
+       if (unmask)
+               intr_reg->intr_control |= ENA_ETH_IO_INTR_REG_INTR_UNMASK_MASK;
+}
+
+#endif /* !(ENA_COM) */
diff --git a/drivers/net/ethernet/amazon/ena/ena_common_defs.h b/drivers/net/ethernet/amazon/ena/ena_common_defs.h
new file mode 100644 (file)
index 0000000..bb8d736
--- /dev/null
@@ -0,0 +1,48 @@
+/*
+ * Copyright 2015 - 2016 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _ENA_COMMON_H_
+#define _ENA_COMMON_H_
+
+#define ENA_COMMON_SPEC_VERSION_MAJOR  0 /*  */
+#define ENA_COMMON_SPEC_VERSION_MINOR  10 /*  */
+
+/* ENA operates with 48-bit memory addresses. ena_mem_addr_t */
+struct ena_common_mem_addr {
+       u32 mem_addr_low;
+
+       u16 mem_addr_high;
+
+       /* MBZ */
+       u16 reserved16;
+};
+
+#endif /*_ENA_COMMON_H_ */
diff --git a/drivers/net/ethernet/amazon/ena/ena_eth_com.c b/drivers/net/ethernet/amazon/ena/ena_eth_com.c
new file mode 100644 (file)
index 0000000..539c536
--- /dev/null
@@ -0,0 +1,501 @@
+/*
+ * Copyright 2015 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "ena_eth_com.h"
+
+static inline struct ena_eth_io_rx_cdesc_base *ena_com_get_next_rx_cdesc(
+       struct ena_com_io_cq *io_cq)
+{
+       struct ena_eth_io_rx_cdesc_base *cdesc;
+       u16 expected_phase, head_masked;
+       u16 desc_phase;
+
+       head_masked = io_cq->head & (io_cq->q_depth - 1);
+       expected_phase = io_cq->phase;
+
+       cdesc = (struct ena_eth_io_rx_cdesc_base *)(io_cq->cdesc_addr.virt_addr
+                       + (head_masked * io_cq->cdesc_entry_size_in_bytes));
+
+       desc_phase = (cdesc->status & ENA_ETH_IO_RX_CDESC_BASE_PHASE_MASK) >>
+                       ENA_ETH_IO_RX_CDESC_BASE_PHASE_SHIFT;
+
+       if (desc_phase != expected_phase)
+               return NULL;
+
+       return cdesc;
+}
+
+static inline void ena_com_cq_inc_head(struct ena_com_io_cq *io_cq)
+{
+       io_cq->head++;
+
+       /* Switch phase bit in case of wrap around */
+       if (unlikely((io_cq->head & (io_cq->q_depth - 1)) == 0))
+               io_cq->phase ^= 1;
+}
+
+static inline void *get_sq_desc(struct ena_com_io_sq *io_sq)
+{
+       u16 tail_masked;
+       u32 offset;
+
+       tail_masked = io_sq->tail & (io_sq->q_depth - 1);
+
+       offset = tail_masked * io_sq->desc_entry_size;
+
+       return (void *)((uintptr_t)io_sq->desc_addr.virt_addr + offset);
+}
+
+static inline void ena_com_copy_curr_sq_desc_to_dev(struct ena_com_io_sq *io_sq)
+{
+       u16 tail_masked = io_sq->tail & (io_sq->q_depth - 1);
+       u32 offset = tail_masked * io_sq->desc_entry_size;
+
+       /* In case this queue isn't a LLQ */
+       if (io_sq->mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
+               return;
+
+       memcpy_toio(io_sq->desc_addr.pbuf_dev_addr + offset,
+                   io_sq->desc_addr.virt_addr + offset,
+                   io_sq->desc_entry_size);
+}
+
+static inline void ena_com_sq_update_tail(struct ena_com_io_sq *io_sq)
+{
+       io_sq->tail++;
+
+       /* Switch phase bit in case of wrap around */
+       if (unlikely((io_sq->tail & (io_sq->q_depth - 1)) == 0))
+               io_sq->phase ^= 1;
+}
+
+static inline int ena_com_write_header(struct ena_com_io_sq *io_sq,
+                                      u8 *head_src, u16 header_len)
+{
+       u16 tail_masked = io_sq->tail & (io_sq->q_depth - 1);
+       u8 __iomem *dev_head_addr =
+               io_sq->header_addr + (tail_masked * io_sq->tx_max_header_size);
+
+       if (io_sq->mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
+               return 0;
+
+       if (unlikely(!io_sq->header_addr)) {
+               pr_err("Push buffer header ptr is NULL\n");
+               return -EINVAL;
+       }
+
+       memcpy_toio(dev_head_addr, head_src, header_len);
+
+       return 0;
+}
+
+static inline struct ena_eth_io_rx_cdesc_base *
+       ena_com_rx_cdesc_idx_to_ptr(struct ena_com_io_cq *io_cq, u16 idx)
+{
+       idx &= (io_cq->q_depth - 1);
+       return (struct ena_eth_io_rx_cdesc_base *)
+               ((uintptr_t)io_cq->cdesc_addr.virt_addr +
+               idx * io_cq->cdesc_entry_size_in_bytes);
+}
+
+static inline u16 ena_com_cdesc_rx_pkt_get(struct ena_com_io_cq *io_cq,
+                                          u16 *first_cdesc_idx)
+{
+       struct ena_eth_io_rx_cdesc_base *cdesc;
+       u16 count = 0, head_masked;
+       u32 last = 0;
+
+       do {
+               cdesc = ena_com_get_next_rx_cdesc(io_cq);
+               if (!cdesc)
+                       break;
+
+               ena_com_cq_inc_head(io_cq);
+               count++;
+               last = (cdesc->status & ENA_ETH_IO_RX_CDESC_BASE_LAST_MASK) >>
+                       ENA_ETH_IO_RX_CDESC_BASE_LAST_SHIFT;
+       } while (!last);
+
+       if (last) {
+               *first_cdesc_idx = io_cq->cur_rx_pkt_cdesc_start_idx;
+               count += io_cq->cur_rx_pkt_cdesc_count;
+
+               head_masked = io_cq->head & (io_cq->q_depth - 1);
+
+               io_cq->cur_rx_pkt_cdesc_count = 0;
+               io_cq->cur_rx_pkt_cdesc_start_idx = head_masked;
+
+               pr_debug("ena q_id: %d packets were completed. first desc idx %u descs# %d\n",
+                        io_cq->qid, *first_cdesc_idx, count);
+       } else {
+               io_cq->cur_rx_pkt_cdesc_count += count;
+               count = 0;
+       }
+
+       return count;
+}
+
+static inline bool ena_com_meta_desc_changed(struct ena_com_io_sq *io_sq,
+                                            struct ena_com_tx_ctx *ena_tx_ctx)
+{
+       int rc;
+
+       if (ena_tx_ctx->meta_valid) {
+               rc = memcmp(&io_sq->cached_tx_meta,
+                           &ena_tx_ctx->ena_meta,
+                           sizeof(struct ena_com_tx_meta));
+
+               if (unlikely(rc != 0))
+                       return true;
+       }
+
+       return false;
+}
+
+static inline void ena_com_create_and_store_tx_meta_desc(struct ena_com_io_sq *io_sq,
+                                                        struct ena_com_tx_ctx *ena_tx_ctx)
+{
+       struct ena_eth_io_tx_meta_desc *meta_desc = NULL;
+       struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
+
+       meta_desc = get_sq_desc(io_sq);
+       memset(meta_desc, 0x0, sizeof(struct ena_eth_io_tx_meta_desc));
+
+       meta_desc->len_ctrl |= ENA_ETH_IO_TX_META_DESC_META_DESC_MASK;
+
+       meta_desc->len_ctrl |= ENA_ETH_IO_TX_META_DESC_EXT_VALID_MASK;
+
+       /* bits 0-9 of the mss */
+       meta_desc->word2 |= (ena_meta->mss <<
+               ENA_ETH_IO_TX_META_DESC_MSS_LO_SHIFT) &
+               ENA_ETH_IO_TX_META_DESC_MSS_LO_MASK;
+       /* bits 10-13 of the mss */
+       meta_desc->len_ctrl |= ((ena_meta->mss >> 10) <<
+               ENA_ETH_IO_TX_META_DESC_MSS_HI_SHIFT) &
+               ENA_ETH_IO_TX_META_DESC_MSS_HI_MASK;
+
+       /* Extended meta desc */
+       meta_desc->len_ctrl |= ENA_ETH_IO_TX_META_DESC_ETH_META_TYPE_MASK;
+       meta_desc->len_ctrl |= ENA_ETH_IO_TX_META_DESC_META_STORE_MASK;
+       meta_desc->len_ctrl |= (io_sq->phase <<
+               ENA_ETH_IO_TX_META_DESC_PHASE_SHIFT) &
+               ENA_ETH_IO_TX_META_DESC_PHASE_MASK;
+
+       meta_desc->len_ctrl |= ENA_ETH_IO_TX_META_DESC_FIRST_MASK;
+       meta_desc->word2 |= ena_meta->l3_hdr_len &
+               ENA_ETH_IO_TX_META_DESC_L3_HDR_LEN_MASK;
+       meta_desc->word2 |= (ena_meta->l3_hdr_offset <<
+               ENA_ETH_IO_TX_META_DESC_L3_HDR_OFF_SHIFT) &
+               ENA_ETH_IO_TX_META_DESC_L3_HDR_OFF_MASK;
+
+       meta_desc->word2 |= (ena_meta->l4_hdr_len <<
+               ENA_ETH_IO_TX_META_DESC_L4_HDR_LEN_IN_WORDS_SHIFT) &
+               ENA_ETH_IO_TX_META_DESC_L4_HDR_LEN_IN_WORDS_MASK;
+
+       meta_desc->len_ctrl |= ENA_ETH_IO_TX_META_DESC_META_STORE_MASK;
+
+       /* Cached the meta desc */
+       memcpy(&io_sq->cached_tx_meta, ena_meta,
+              sizeof(struct ena_com_tx_meta));
+
+       ena_com_copy_curr_sq_desc_to_dev(io_sq);
+       ena_com_sq_update_tail(io_sq);
+}
+
+static inline void ena_com_rx_set_flags(struct ena_com_rx_ctx *ena_rx_ctx,
+                                       struct ena_eth_io_rx_cdesc_base *cdesc)
+{
+       ena_rx_ctx->l3_proto = cdesc->status &
+               ENA_ETH_IO_RX_CDESC_BASE_L3_PROTO_IDX_MASK;
+       ena_rx_ctx->l4_proto =
+               (cdesc->status & ENA_ETH_IO_RX_CDESC_BASE_L4_PROTO_IDX_MASK) >>
+               ENA_ETH_IO_RX_CDESC_BASE_L4_PROTO_IDX_SHIFT;
+       ena_rx_ctx->l3_csum_err =
+               (cdesc->status & ENA_ETH_IO_RX_CDESC_BASE_L3_CSUM_ERR_MASK) >>
+               ENA_ETH_IO_RX_CDESC_BASE_L3_CSUM_ERR_SHIFT;
+       ena_rx_ctx->l4_csum_err =
+               (cdesc->status & ENA_ETH_IO_RX_CDESC_BASE_L4_CSUM_ERR_MASK) >>
+               ENA_ETH_IO_RX_CDESC_BASE_L4_CSUM_ERR_SHIFT;
+       ena_rx_ctx->hash = cdesc->hash;
+       ena_rx_ctx->frag =
+               (cdesc->status & ENA_ETH_IO_RX_CDESC_BASE_IPV4_FRAG_MASK) >>
+               ENA_ETH_IO_RX_CDESC_BASE_IPV4_FRAG_SHIFT;
+
+       pr_debug("ena_rx_ctx->l3_proto %d ena_rx_ctx->l4_proto %d\nena_rx_ctx->l3_csum_err %d ena_rx_ctx->l4_csum_err %d\nhash frag %d frag: %d cdesc_status: %x\n",
+                ena_rx_ctx->l3_proto, ena_rx_ctx->l4_proto,
+                ena_rx_ctx->l3_csum_err, ena_rx_ctx->l4_csum_err,
+                ena_rx_ctx->hash, ena_rx_ctx->frag, cdesc->status);
+}
+
+/*****************************************************************************/
+/*****************************     API      **********************************/
+/*****************************************************************************/
+
+int ena_com_prepare_tx(struct ena_com_io_sq *io_sq,
+                      struct ena_com_tx_ctx *ena_tx_ctx,
+                      int *nb_hw_desc)
+{
+       struct ena_eth_io_tx_desc *desc = NULL;
+       struct ena_com_buf *ena_bufs = ena_tx_ctx->ena_bufs;
+       void *push_header = ena_tx_ctx->push_header;
+       u16 header_len = ena_tx_ctx->header_len;
+       u16 num_bufs = ena_tx_ctx->num_bufs;
+       int total_desc, i, rc;
+       bool have_meta;
+       u64 addr_hi;
+
+       WARN(io_sq->direction != ENA_COM_IO_QUEUE_DIRECTION_TX, "wrong Q type");
+
+       /* num_bufs +1 for potential meta desc */
+       if (ena_com_sq_empty_space(io_sq) < (num_bufs + 1)) {
+               pr_err("Not enough space in the tx queue\n");
+               return -ENOMEM;
+       }
+
+       if (unlikely(header_len > io_sq->tx_max_header_size)) {
+               pr_err("header size is too large %d max header: %d\n",
+                      header_len, io_sq->tx_max_header_size);
+               return -EINVAL;
+       }
+
+       /* start with pushing the header (if needed) */
+       rc = ena_com_write_header(io_sq, push_header, header_len);
+       if (unlikely(rc))
+               return rc;
+
+       have_meta = ena_tx_ctx->meta_valid && ena_com_meta_desc_changed(io_sq,
+                       ena_tx_ctx);
+       if (have_meta)
+               ena_com_create_and_store_tx_meta_desc(io_sq, ena_tx_ctx);
+
+       /* If the caller doesn't want send packets */
+       if (unlikely(!num_bufs && !header_len)) {
+               *nb_hw_desc = have_meta ? 0 : 1;
+               return 0;
+       }
+
+       desc = get_sq_desc(io_sq);
+       memset(desc, 0x0, sizeof(struct ena_eth_io_tx_desc));
+
+       /* Set first desc when we don't have meta descriptor */
+       if (!have_meta)
+               desc->len_ctrl |= ENA_ETH_IO_TX_DESC_FIRST_MASK;
+
+       desc->buff_addr_hi_hdr_sz |= (header_len <<
+               ENA_ETH_IO_TX_DESC_HEADER_LENGTH_SHIFT) &
+               ENA_ETH_IO_TX_DESC_HEADER_LENGTH_MASK;
+       desc->len_ctrl |= (io_sq->phase << ENA_ETH_IO_TX_DESC_PHASE_SHIFT) &
+               ENA_ETH_IO_TX_DESC_PHASE_MASK;
+
+       desc->len_ctrl |= ENA_ETH_IO_TX_DESC_COMP_REQ_MASK;
+
+       /* Bits 0-9 */
+       desc->meta_ctrl |= (ena_tx_ctx->req_id <<
+               ENA_ETH_IO_TX_DESC_REQ_ID_LO_SHIFT) &
+               ENA_ETH_IO_TX_DESC_REQ_ID_LO_MASK;
+
+       desc->meta_ctrl |= (ena_tx_ctx->df <<
+               ENA_ETH_IO_TX_DESC_DF_SHIFT) &
+               ENA_ETH_IO_TX_DESC_DF_MASK;
+
+       /* Bits 10-15 */
+       desc->len_ctrl |= ((ena_tx_ctx->req_id >> 10) <<
+               ENA_ETH_IO_TX_DESC_REQ_ID_HI_SHIFT) &
+               ENA_ETH_IO_TX_DESC_REQ_ID_HI_MASK;
+
+       if (ena_tx_ctx->meta_valid) {
+               desc->meta_ctrl |= (ena_tx_ctx->tso_enable <<
+                       ENA_ETH_IO_TX_DESC_TSO_EN_SHIFT) &
+                       ENA_ETH_IO_TX_DESC_TSO_EN_MASK;
+               desc->meta_ctrl |= ena_tx_ctx->l3_proto &
+                       ENA_ETH_IO_TX_DESC_L3_PROTO_IDX_MASK;
+               desc->meta_ctrl |= (ena_tx_ctx->l4_proto <<
+                       ENA_ETH_IO_TX_DESC_L4_PROTO_IDX_SHIFT) &
+                       ENA_ETH_IO_TX_DESC_L4_PROTO_IDX_MASK;
+               desc->meta_ctrl |= (ena_tx_ctx->l3_csum_enable <<
+                       ENA_ETH_IO_TX_DESC_L3_CSUM_EN_SHIFT) &
+                       ENA_ETH_IO_TX_DESC_L3_CSUM_EN_MASK;
+               desc->meta_ctrl |= (ena_tx_ctx->l4_csum_enable <<
+                       ENA_ETH_IO_TX_DESC_L4_CSUM_EN_SHIFT) &
+                       ENA_ETH_IO_TX_DESC_L4_CSUM_EN_MASK;
+               desc->meta_ctrl |= (ena_tx_ctx->l4_csum_partial <<
+                       ENA_ETH_IO_TX_DESC_L4_CSUM_PARTIAL_SHIFT) &
+                       ENA_ETH_IO_TX_DESC_L4_CSUM_PARTIAL_MASK;
+       }
+
+       for (i = 0; i < num_bufs; i++) {
+               /* The first desc share the same desc as the header */
+               if (likely(i != 0)) {
+                       ena_com_copy_curr_sq_desc_to_dev(io_sq);
+                       ena_com_sq_update_tail(io_sq);
+
+                       desc = get_sq_desc(io_sq);
+                       memset(desc, 0x0, sizeof(struct ena_eth_io_tx_desc));
+
+                       desc->len_ctrl |= (io_sq->phase <<
+                               ENA_ETH_IO_TX_DESC_PHASE_SHIFT) &
+                               ENA_ETH_IO_TX_DESC_PHASE_MASK;
+               }
+
+               desc->len_ctrl |= ena_bufs->len &
+                       ENA_ETH_IO_TX_DESC_LENGTH_MASK;
+
+               addr_hi = ((ena_bufs->paddr &
+                       GENMASK_ULL(io_sq->dma_addr_bits - 1, 32)) >> 32);
+
+               desc->buff_addr_lo = (u32)ena_bufs->paddr;
+               desc->buff_addr_hi_hdr_sz |= addr_hi &
+                       ENA_ETH_IO_TX_DESC_ADDR_HI_MASK;
+               ena_bufs++;
+       }
+
+       /* set the last desc indicator */
+       desc->len_ctrl |= ENA_ETH_IO_TX_DESC_LAST_MASK;
+
+       ena_com_copy_curr_sq_desc_to_dev(io_sq);
+
+       ena_com_sq_update_tail(io_sq);
+
+       total_desc = max_t(u16, num_bufs, 1);
+       total_desc += have_meta ? 1 : 0;
+
+       *nb_hw_desc = total_desc;
+       return 0;
+}
+
+int ena_com_rx_pkt(struct ena_com_io_cq *io_cq,
+                  struct ena_com_io_sq *io_sq,
+                  struct ena_com_rx_ctx *ena_rx_ctx)
+{
+       struct ena_com_rx_buf_info *ena_buf = &ena_rx_ctx->ena_bufs[0];
+       struct ena_eth_io_rx_cdesc_base *cdesc = NULL;
+       u16 cdesc_idx = 0;
+       u16 nb_hw_desc;
+       u16 i;
+
+       WARN(io_cq->direction != ENA_COM_IO_QUEUE_DIRECTION_RX, "wrong Q type");
+
+       nb_hw_desc = ena_com_cdesc_rx_pkt_get(io_cq, &cdesc_idx);
+       if (nb_hw_desc == 0) {
+               ena_rx_ctx->descs = nb_hw_desc;
+               return 0;
+       }
+
+       pr_debug("fetch rx packet: queue %d completed desc: %d\n", io_cq->qid,
+                nb_hw_desc);
+
+       if (unlikely(nb_hw_desc > ena_rx_ctx->max_bufs)) {
+               pr_err("Too many RX cdescs (%d) > MAX(%d)\n", nb_hw_desc,
+                      ena_rx_ctx->max_bufs);
+               return -ENOSPC;
+       }
+
+       for (i = 0; i < nb_hw_desc; i++) {
+               cdesc = ena_com_rx_cdesc_idx_to_ptr(io_cq, cdesc_idx + i);
+
+               ena_buf->len = cdesc->length;
+               ena_buf->req_id = cdesc->req_id;
+               ena_buf++;
+       }
+
+       /* Update SQ head ptr */
+       io_sq->next_to_comp += nb_hw_desc;
+
+       pr_debug("[%s][QID#%d] Updating SQ head to: %d\n", __func__, io_sq->qid,
+                io_sq->next_to_comp);
+
+       /* Get rx flags from the last pkt */
+       ena_com_rx_set_flags(ena_rx_ctx, cdesc);
+
+       ena_rx_ctx->descs = nb_hw_desc;
+       return 0;
+}
+
+int ena_com_add_single_rx_desc(struct ena_com_io_sq *io_sq,
+                              struct ena_com_buf *ena_buf,
+                              u16 req_id)
+{
+       struct ena_eth_io_rx_desc *desc;
+
+       WARN(io_sq->direction != ENA_COM_IO_QUEUE_DIRECTION_RX, "wrong Q type");
+
+       if (unlikely(ena_com_sq_empty_space(io_sq) == 0))
+               return -ENOSPC;
+
+       desc = get_sq_desc(io_sq);
+       memset(desc, 0x0, sizeof(struct ena_eth_io_rx_desc));
+
+       desc->length = ena_buf->len;
+
+       desc->ctrl |= ENA_ETH_IO_RX_DESC_FIRST_MASK;
+       desc->ctrl |= ENA_ETH_IO_RX_DESC_LAST_MASK;
+       desc->ctrl |= io_sq->phase & ENA_ETH_IO_RX_DESC_PHASE_MASK;
+       desc->ctrl |= ENA_ETH_IO_RX_DESC_COMP_REQ_MASK;
+
+       desc->req_id = req_id;
+
+       desc->buff_addr_lo = (u32)ena_buf->paddr;
+       desc->buff_addr_hi =
+               ((ena_buf->paddr & GENMASK_ULL(io_sq->dma_addr_bits - 1, 32)) >> 32);
+
+       ena_com_sq_update_tail(io_sq);
+
+       return 0;
+}
+
+int ena_com_tx_comp_req_id_get(struct ena_com_io_cq *io_cq, u16 *req_id)
+{
+       u8 expected_phase, cdesc_phase;
+       struct ena_eth_io_tx_cdesc *cdesc;
+       u16 masked_head;
+
+       masked_head = io_cq->head & (io_cq->q_depth - 1);
+       expected_phase = io_cq->phase;
+
+       cdesc = (struct ena_eth_io_tx_cdesc *)
+               ((uintptr_t)io_cq->cdesc_addr.virt_addr +
+               (masked_head * io_cq->cdesc_entry_size_in_bytes));
+
+       /* When the current completion descriptor phase isn't the same as the
+        * expected, it mean that the device still didn't update
+        * this completion.
+        */
+       cdesc_phase = cdesc->flags & ENA_ETH_IO_TX_CDESC_PHASE_MASK;
+       if (cdesc_phase != expected_phase)
+               return -EAGAIN;
+
+       ena_com_cq_inc_head(io_cq);
+
+       *req_id = cdesc->req_id;
+
+       return 0;
+}
diff --git a/drivers/net/ethernet/amazon/ena/ena_eth_com.h b/drivers/net/ethernet/amazon/ena/ena_eth_com.h
new file mode 100644 (file)
index 0000000..bb53c3a
--- /dev/null
@@ -0,0 +1,160 @@
+/*
+ * Copyright 2015 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef ENA_ETH_COM_H_
+#define ENA_ETH_COM_H_
+
+#include "ena_com.h"
+
+/* head update threshold in units of (queue size / ENA_COMP_HEAD_THRESH) */
+#define ENA_COMP_HEAD_THRESH 4
+
+struct ena_com_tx_ctx {
+       struct ena_com_tx_meta ena_meta;
+       struct ena_com_buf *ena_bufs;
+       /* For LLQ, header buffer - pushed to the device mem space */
+       void *push_header;
+
+       enum ena_eth_io_l3_proto_index l3_proto;
+       enum ena_eth_io_l4_proto_index l4_proto;
+       u16 num_bufs;
+       u16 req_id;
+       /* For regular queue, indicate the size of the header
+        * For LLQ, indicate the size of the pushed buffer
+        */
+       u16 header_len;
+
+       u8 meta_valid;
+       u8 tso_enable;
+       u8 l3_csum_enable;
+       u8 l4_csum_enable;
+       u8 l4_csum_partial;
+       u8 df; /* Don't fragment */
+};
+
+struct ena_com_rx_ctx {
+       struct ena_com_rx_buf_info *ena_bufs;
+       enum ena_eth_io_l3_proto_index l3_proto;
+       enum ena_eth_io_l4_proto_index l4_proto;
+       bool l3_csum_err;
+       bool l4_csum_err;
+       /* fragmented packet */
+       bool frag;
+       u32 hash;
+       u16 descs;
+       int max_bufs;
+};
+
+int ena_com_prepare_tx(struct ena_com_io_sq *io_sq,
+                      struct ena_com_tx_ctx *ena_tx_ctx,
+                      int *nb_hw_desc);
+
+int ena_com_rx_pkt(struct ena_com_io_cq *io_cq,
+                  struct ena_com_io_sq *io_sq,
+                  struct ena_com_rx_ctx *ena_rx_ctx);
+
+int ena_com_add_single_rx_desc(struct ena_com_io_sq *io_sq,
+                              struct ena_com_buf *ena_buf,
+                              u16 req_id);
+
+int ena_com_tx_comp_req_id_get(struct ena_com_io_cq *io_cq, u16 *req_id);
+
+static inline void ena_com_unmask_intr(struct ena_com_io_cq *io_cq,
+                                      struct ena_eth_io_intr_reg *intr_reg)
+{
+       writel(intr_reg->intr_control, io_cq->unmask_reg);
+}
+
+static inline int ena_com_sq_empty_space(struct ena_com_io_sq *io_sq)
+{
+       u16 tail, next_to_comp, cnt;
+
+       next_to_comp = io_sq->next_to_comp;
+       tail = io_sq->tail;
+       cnt = tail - next_to_comp;
+
+       return io_sq->q_depth - 1 - cnt;
+}
+
+static inline int ena_com_write_sq_doorbell(struct ena_com_io_sq *io_sq)
+{
+       u16 tail;
+
+       tail = io_sq->tail;
+
+       pr_debug("write submission queue doorbell for queue: %d tail: %d\n",
+                io_sq->qid, tail);
+
+       writel(tail, io_sq->db_addr);
+
+       return 0;
+}
+
+static inline int ena_com_update_dev_comp_head(struct ena_com_io_cq *io_cq)
+{
+       u16 unreported_comp, head;
+       bool need_update;
+
+       head = io_cq->head;
+       unreported_comp = head - io_cq->last_head_update;
+       need_update = unreported_comp > (io_cq->q_depth / ENA_COMP_HEAD_THRESH);
+
+       if (io_cq->cq_head_db_reg && need_update) {
+               pr_debug("Write completion queue doorbell for queue %d: head: %d\n",
+                        io_cq->qid, head);
+               writel(head, io_cq->cq_head_db_reg);
+               io_cq->last_head_update = head;
+       }
+
+       return 0;
+}
+
+static inline void ena_com_update_numa_node(struct ena_com_io_cq *io_cq,
+                                           u8 numa_node)
+{
+       struct ena_eth_io_numa_node_cfg_reg numa_cfg;
+
+       if (!io_cq->numa_node_cfg_reg)
+               return;
+
+       numa_cfg.numa_cfg = (numa_node & ENA_ETH_IO_NUMA_NODE_CFG_REG_NUMA_MASK)
+               | ENA_ETH_IO_NUMA_NODE_CFG_REG_ENABLED_MASK;
+
+       writel(numa_cfg.numa_cfg, io_cq->numa_node_cfg_reg);
+}
+
+static inline void ena_com_comp_ack(struct ena_com_io_sq *io_sq, u16 elem)
+{
+       io_sq->next_to_comp += elem;
+}
+
+#endif /* ENA_ETH_COM_H_ */
diff --git a/drivers/net/ethernet/amazon/ena/ena_eth_io_defs.h b/drivers/net/ethernet/amazon/ena/ena_eth_io_defs.h
new file mode 100644 (file)
index 0000000..f320c58
--- /dev/null
@@ -0,0 +1,416 @@
+/*
+ * Copyright 2015 - 2016 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _ENA_ETH_IO_H_
+#define _ENA_ETH_IO_H_
+
+enum ena_eth_io_l3_proto_index {
+       ENA_ETH_IO_L3_PROTO_UNKNOWN     = 0,
+
+       ENA_ETH_IO_L3_PROTO_IPV4        = 8,
+
+       ENA_ETH_IO_L3_PROTO_IPV6        = 11,
+
+       ENA_ETH_IO_L3_PROTO_FCOE        = 21,
+
+       ENA_ETH_IO_L3_PROTO_ROCE        = 22,
+};
+
+enum ena_eth_io_l4_proto_index {
+       ENA_ETH_IO_L4_PROTO_UNKNOWN             = 0,
+
+       ENA_ETH_IO_L4_PROTO_TCP                 = 12,
+
+       ENA_ETH_IO_L4_PROTO_UDP                 = 13,
+
+       ENA_ETH_IO_L4_PROTO_ROUTEABLE_ROCE      = 23,
+};
+
+struct ena_eth_io_tx_desc {
+       /* 15:0 : length - Buffer length in bytes, must
+        *    include any packet trailers that the ENA supposed
+        *    to update like End-to-End CRC, Authentication GMAC
+        *    etc. This length must not include the
+        *    'Push_Buffer' length. This length must not include
+        *    the 4-byte added in the end for 802.3 Ethernet FCS
+        * 21:16 : req_id_hi - Request ID[15:10]
+        * 22 : reserved22 - MBZ
+        * 23 : meta_desc - MBZ
+        * 24 : phase
+        * 25 : reserved1 - MBZ
+        * 26 : first - Indicates first descriptor in
+        *    transaction
+        * 27 : last - Indicates last descriptor in
+        *    transaction
+        * 28 : comp_req - Indicates whether completion
+        *    should be posted, after packet is transmitted.
+        *    Valid only for first descriptor
+        * 30:29 : reserved29 - MBZ
+        * 31 : reserved31 - MBZ
+        */
+       u32 len_ctrl;
+
+       /* 3:0 : l3_proto_idx - L3 protocol. This field
+        *    required when l3_csum_en,l3_csum or tso_en are set.
+        * 4 : DF - IPv4 DF, must be 0 if packet is IPv4 and
+        *    DF flags of the IPv4 header is 0. Otherwise must
+        *    be set to 1
+        * 6:5 : reserved5
+        * 7 : tso_en - Enable TSO, For TCP only.
+        * 12:8 : l4_proto_idx - L4 protocol. This field need
+        *    to be set when l4_csum_en or tso_en are set.
+        * 13 : l3_csum_en - enable IPv4 header checksum.
+        * 14 : l4_csum_en - enable TCP/UDP checksum.
+        * 15 : ethernet_fcs_dis - when set, the controller
+        *    will not append the 802.3 Ethernet Frame Check
+        *    Sequence to the packet
+        * 16 : reserved16
+        * 17 : l4_csum_partial - L4 partial checksum. when
+        *    set to 0, the ENA calculates the L4 checksum,
+        *    where the Destination Address required for the
+        *    TCP/UDP pseudo-header is taken from the actual
+        *    packet L3 header. when set to 1, the ENA doesn't
+        *    calculate the sum of the pseudo-header, instead,
+        *    the checksum field of the L4 is used instead. When
+        *    TSO enabled, the checksum of the pseudo-header
+        *    must not include the tcp length field. L4 partial
+        *    checksum should be used for IPv6 packet that
+        *    contains Routing Headers.
+        * 20:18 : reserved18 - MBZ
+        * 21 : reserved21 - MBZ
+        * 31:22 : req_id_lo - Request ID[9:0]
+        */
+       u32 meta_ctrl;
+
+       u32 buff_addr_lo;
+
+       /* address high and header size
+        * 15:0 : addr_hi - Buffer Pointer[47:32]
+        * 23:16 : reserved16_w2
+        * 31:24 : header_length - Header length. For Low
+        *    Latency Queues, this fields indicates the number
+        *    of bytes written to the headers' memory. For
+        *    normal queues, if packet is TCP or UDP, and longer
+        *    than max_header_size, then this field should be
+        *    set to the sum of L4 header offset and L4 header
+        *    size(without options), otherwise, this field
+        *    should be set to 0. For both modes, this field
+        *    must not exceed the max_header_size.
+        *    max_header_size value is reported by the Max
+        *    Queues Feature descriptor
+        */
+       u32 buff_addr_hi_hdr_sz;
+};
+
+struct ena_eth_io_tx_meta_desc {
+       /* 9:0 : req_id_lo - Request ID[9:0]
+        * 11:10 : reserved10 - MBZ
+        * 12 : reserved12 - MBZ
+        * 13 : reserved13 - MBZ
+        * 14 : ext_valid - if set, offset fields in Word2
+        *    are valid Also MSS High in Word 0 and bits [31:24]
+        *    in Word 3
+        * 15 : reserved15
+        * 19:16 : mss_hi
+        * 20 : eth_meta_type - 0: Tx Metadata Descriptor, 1:
+        *    Extended Metadata Descriptor
+        * 21 : meta_store - Store extended metadata in queue
+        *    cache
+        * 22 : reserved22 - MBZ
+        * 23 : meta_desc - MBO
+        * 24 : phase
+        * 25 : reserved25 - MBZ
+        * 26 : first - Indicates first descriptor in
+        *    transaction
+        * 27 : last - Indicates last descriptor in
+        *    transaction
+        * 28 : comp_req - Indicates whether completion
+        *    should be posted, after packet is transmitted.
+        *    Valid only for first descriptor
+        * 30:29 : reserved29 - MBZ
+        * 31 : reserved31 - MBZ
+        */
+       u32 len_ctrl;
+
+       /* 5:0 : req_id_hi
+        * 31:6 : reserved6 - MBZ
+        */
+       u32 word1;
+
+       /* 7:0 : l3_hdr_len
+        * 15:8 : l3_hdr_off
+        * 21:16 : l4_hdr_len_in_words - counts the L4 header
+        *    length in words. there is an explicit assumption
+        *    that L4 header appears right after L3 header and
+        *    L4 offset is based on l3_hdr_off+l3_hdr_len
+        * 31:22 : mss_lo
+        */
+       u32 word2;
+
+       u32 reserved;
+};
+
+struct ena_eth_io_tx_cdesc {
+       /* Request ID[15:0] */
+       u16 req_id;
+
+       u8 status;
+
+       /* flags
+        * 0 : phase
+        * 7:1 : reserved1
+        */
+       u8 flags;
+
+       u16 sub_qid;
+
+       u16 sq_head_idx;
+};
+
+struct ena_eth_io_rx_desc {
+       /* In bytes. 0 means 64KB */
+       u16 length;
+
+       /* MBZ */
+       u8 reserved2;
+
+       /* 0 : phase
+        * 1 : reserved1 - MBZ
+        * 2 : first - Indicates first descriptor in
+        *    transaction
+        * 3 : last - Indicates last descriptor in transaction
+        * 4 : comp_req
+        * 5 : reserved5 - MBO
+        * 7:6 : reserved6 - MBZ
+        */
+       u8 ctrl;
+
+       u16 req_id;
+
+       /* MBZ */
+       u16 reserved6;
+
+       u32 buff_addr_lo;
+
+       u16 buff_addr_hi;
+
+       /* MBZ */
+       u16 reserved16_w3;
+};
+
+/* 4-word format Note: all ethernet parsing information are valid only when
+ * last=1
+ */
+struct ena_eth_io_rx_cdesc_base {
+       /* 4:0 : l3_proto_idx
+        * 6:5 : src_vlan_cnt
+        * 7 : reserved7 - MBZ
+        * 12:8 : l4_proto_idx
+        * 13 : l3_csum_err - when set, either the L3
+        *    checksum error detected, or, the controller didn't
+        *    validate the checksum. This bit is valid only when
+        *    l3_proto_idx indicates IPv4 packet
+        * 14 : l4_csum_err - when set, either the L4
+        *    checksum error detected, or, the controller didn't
+        *    validate the checksum. This bit is valid only when
+        *    l4_proto_idx indicates TCP/UDP packet, and,
+        *    ipv4_frag is not set
+        * 15 : ipv4_frag - Indicates IPv4 fragmented packet
+        * 23:16 : reserved16
+        * 24 : phase
+        * 25 : l3_csum2 - second checksum engine result
+        * 26 : first - Indicates first descriptor in
+        *    transaction
+        * 27 : last - Indicates last descriptor in
+        *    transaction
+        * 29:28 : reserved28
+        * 30 : buffer - 0: Metadata descriptor. 1: Buffer
+        *    Descriptor was used
+        * 31 : reserved31
+        */
+       u32 status;
+
+       u16 length;
+
+       u16 req_id;
+
+       /* 32-bit hash result */
+       u32 hash;
+
+       u16 sub_qid;
+
+       u16 reserved;
+};
+
+/* 8-word format */
+struct ena_eth_io_rx_cdesc_ext {
+       struct ena_eth_io_rx_cdesc_base base;
+
+       u32 buff_addr_lo;
+
+       u16 buff_addr_hi;
+
+       u16 reserved16;
+
+       u32 reserved_w6;
+
+       u32 reserved_w7;
+};
+
+struct ena_eth_io_intr_reg {
+       /* 14:0 : rx_intr_delay
+        * 29:15 : tx_intr_delay
+        * 30 : intr_unmask
+        * 31 : reserved
+        */
+       u32 intr_control;
+};
+
+struct ena_eth_io_numa_node_cfg_reg {
+       /* 7:0 : numa
+        * 30:8 : reserved
+        * 31 : enabled
+        */
+       u32 numa_cfg;
+};
+
+/* tx_desc */
+#define ENA_ETH_IO_TX_DESC_LENGTH_MASK GENMASK(15, 0)
+#define ENA_ETH_IO_TX_DESC_REQ_ID_HI_SHIFT 16
+#define ENA_ETH_IO_TX_DESC_REQ_ID_HI_MASK GENMASK(21, 16)
+#define ENA_ETH_IO_TX_DESC_META_DESC_SHIFT 23
+#define ENA_ETH_IO_TX_DESC_META_DESC_MASK BIT(23)
+#define ENA_ETH_IO_TX_DESC_PHASE_SHIFT 24
+#define ENA_ETH_IO_TX_DESC_PHASE_MASK BIT(24)
+#define ENA_ETH_IO_TX_DESC_FIRST_SHIFT 26
+#define ENA_ETH_IO_TX_DESC_FIRST_MASK BIT(26)
+#define ENA_ETH_IO_TX_DESC_LAST_SHIFT 27
+#define ENA_ETH_IO_TX_DESC_LAST_MASK BIT(27)
+#define ENA_ETH_IO_TX_DESC_COMP_REQ_SHIFT 28
+#define ENA_ETH_IO_TX_DESC_COMP_REQ_MASK BIT(28)
+#define ENA_ETH_IO_TX_DESC_L3_PROTO_IDX_MASK GENMASK(3, 0)
+#define ENA_ETH_IO_TX_DESC_DF_SHIFT 4
+#define ENA_ETH_IO_TX_DESC_DF_MASK BIT(4)
+#define ENA_ETH_IO_TX_DESC_TSO_EN_SHIFT 7
+#define ENA_ETH_IO_TX_DESC_TSO_EN_MASK BIT(7)
+#define ENA_ETH_IO_TX_DESC_L4_PROTO_IDX_SHIFT 8
+#define ENA_ETH_IO_TX_DESC_L4_PROTO_IDX_MASK GENMASK(12, 8)
+#define ENA_ETH_IO_TX_DESC_L3_CSUM_EN_SHIFT 13
+#define ENA_ETH_IO_TX_DESC_L3_CSUM_EN_MASK BIT(13)
+#define ENA_ETH_IO_TX_DESC_L4_CSUM_EN_SHIFT 14
+#define ENA_ETH_IO_TX_DESC_L4_CSUM_EN_MASK BIT(14)
+#define ENA_ETH_IO_TX_DESC_ETHERNET_FCS_DIS_SHIFT 15
+#define ENA_ETH_IO_TX_DESC_ETHERNET_FCS_DIS_MASK BIT(15)
+#define ENA_ETH_IO_TX_DESC_L4_CSUM_PARTIAL_SHIFT 17
+#define ENA_ETH_IO_TX_DESC_L4_CSUM_PARTIAL_MASK BIT(17)
+#define ENA_ETH_IO_TX_DESC_REQ_ID_LO_SHIFT 22
+#define ENA_ETH_IO_TX_DESC_REQ_ID_LO_MASK GENMASK(31, 22)
+#define ENA_ETH_IO_TX_DESC_ADDR_HI_MASK GENMASK(15, 0)
+#define ENA_ETH_IO_TX_DESC_HEADER_LENGTH_SHIFT 24
+#define ENA_ETH_IO_TX_DESC_HEADER_LENGTH_MASK GENMASK(31, 24)
+
+/* tx_meta_desc */
+#define ENA_ETH_IO_TX_META_DESC_REQ_ID_LO_MASK GENMASK(9, 0)
+#define ENA_ETH_IO_TX_META_DESC_EXT_VALID_SHIFT 14
+#define ENA_ETH_IO_TX_META_DESC_EXT_VALID_MASK BIT(14)
+#define ENA_ETH_IO_TX_META_DESC_MSS_HI_SHIFT 16
+#define ENA_ETH_IO_TX_META_DESC_MSS_HI_MASK GENMASK(19, 16)
+#define ENA_ETH_IO_TX_META_DESC_ETH_META_TYPE_SHIFT 20
+#define ENA_ETH_IO_TX_META_DESC_ETH_META_TYPE_MASK BIT(20)
+#define ENA_ETH_IO_TX_META_DESC_META_STORE_SHIFT 21
+#define ENA_ETH_IO_TX_META_DESC_META_STORE_MASK BIT(21)
+#define ENA_ETH_IO_TX_META_DESC_META_DESC_SHIFT 23
+#define ENA_ETH_IO_TX_META_DESC_META_DESC_MASK BIT(23)
+#define ENA_ETH_IO_TX_META_DESC_PHASE_SHIFT 24
+#define ENA_ETH_IO_TX_META_DESC_PHASE_MASK BIT(24)
+#define ENA_ETH_IO_TX_META_DESC_FIRST_SHIFT 26
+#define ENA_ETH_IO_TX_META_DESC_FIRST_MASK BIT(26)
+#define ENA_ETH_IO_TX_META_DESC_LAST_SHIFT 27
+#define ENA_ETH_IO_TX_META_DESC_LAST_MASK BIT(27)
+#define ENA_ETH_IO_TX_META_DESC_COMP_REQ_SHIFT 28
+#define ENA_ETH_IO_TX_META_DESC_COMP_REQ_MASK BIT(28)
+#define ENA_ETH_IO_TX_META_DESC_REQ_ID_HI_MASK GENMASK(5, 0)
+#define ENA_ETH_IO_TX_META_DESC_L3_HDR_LEN_MASK GENMASK(7, 0)
+#define ENA_ETH_IO_TX_META_DESC_L3_HDR_OFF_SHIFT 8
+#define ENA_ETH_IO_TX_META_DESC_L3_HDR_OFF_MASK GENMASK(15, 8)
+#define ENA_ETH_IO_TX_META_DESC_L4_HDR_LEN_IN_WORDS_SHIFT 16
+#define ENA_ETH_IO_TX_META_DESC_L4_HDR_LEN_IN_WORDS_MASK GENMASK(21, 16)
+#define ENA_ETH_IO_TX_META_DESC_MSS_LO_SHIFT 22
+#define ENA_ETH_IO_TX_META_DESC_MSS_LO_MASK GENMASK(31, 22)
+
+/* tx_cdesc */
+#define ENA_ETH_IO_TX_CDESC_PHASE_MASK BIT(0)
+
+/* rx_desc */
+#define ENA_ETH_IO_RX_DESC_PHASE_MASK BIT(0)
+#define ENA_ETH_IO_RX_DESC_FIRST_SHIFT 2
+#define ENA_ETH_IO_RX_DESC_FIRST_MASK BIT(2)
+#define ENA_ETH_IO_RX_DESC_LAST_SHIFT 3
+#define ENA_ETH_IO_RX_DESC_LAST_MASK BIT(3)
+#define ENA_ETH_IO_RX_DESC_COMP_REQ_SHIFT 4
+#define ENA_ETH_IO_RX_DESC_COMP_REQ_MASK BIT(4)
+
+/* rx_cdesc_base */
+#define ENA_ETH_IO_RX_CDESC_BASE_L3_PROTO_IDX_MASK GENMASK(4, 0)
+#define ENA_ETH_IO_RX_CDESC_BASE_SRC_VLAN_CNT_SHIFT 5
+#define ENA_ETH_IO_RX_CDESC_BASE_SRC_VLAN_CNT_MASK GENMASK(6, 5)
+#define ENA_ETH_IO_RX_CDESC_BASE_L4_PROTO_IDX_SHIFT 8
+#define ENA_ETH_IO_RX_CDESC_BASE_L4_PROTO_IDX_MASK GENMASK(12, 8)
+#define ENA_ETH_IO_RX_CDESC_BASE_L3_CSUM_ERR_SHIFT 13
+#define ENA_ETH_IO_RX_CDESC_BASE_L3_CSUM_ERR_MASK BIT(13)
+#define ENA_ETH_IO_RX_CDESC_BASE_L4_CSUM_ERR_SHIFT 14
+#define ENA_ETH_IO_RX_CDESC_BASE_L4_CSUM_ERR_MASK BIT(14)
+#define ENA_ETH_IO_RX_CDESC_BASE_IPV4_FRAG_SHIFT 15
+#define ENA_ETH_IO_RX_CDESC_BASE_IPV4_FRAG_MASK BIT(15)
+#define ENA_ETH_IO_RX_CDESC_BASE_PHASE_SHIFT 24
+#define ENA_ETH_IO_RX_CDESC_BASE_PHASE_MASK BIT(24)
+#define ENA_ETH_IO_RX_CDESC_BASE_L3_CSUM2_SHIFT 25
+#define ENA_ETH_IO_RX_CDESC_BASE_L3_CSUM2_MASK BIT(25)
+#define ENA_ETH_IO_RX_CDESC_BASE_FIRST_SHIFT 26
+#define ENA_ETH_IO_RX_CDESC_BASE_FIRST_MASK BIT(26)
+#define ENA_ETH_IO_RX_CDESC_BASE_LAST_SHIFT 27
+#define ENA_ETH_IO_RX_CDESC_BASE_LAST_MASK BIT(27)
+#define ENA_ETH_IO_RX_CDESC_BASE_BUFFER_SHIFT 30
+#define ENA_ETH_IO_RX_CDESC_BASE_BUFFER_MASK BIT(30)
+
+/* intr_reg */
+#define ENA_ETH_IO_INTR_REG_RX_INTR_DELAY_MASK GENMASK(14, 0)
+#define ENA_ETH_IO_INTR_REG_TX_INTR_DELAY_SHIFT 15
+#define ENA_ETH_IO_INTR_REG_TX_INTR_DELAY_MASK GENMASK(29, 15)
+#define ENA_ETH_IO_INTR_REG_INTR_UNMASK_SHIFT 30
+#define ENA_ETH_IO_INTR_REG_INTR_UNMASK_MASK BIT(30)
+
+/* numa_node_cfg_reg */
+#define ENA_ETH_IO_NUMA_NODE_CFG_REG_NUMA_MASK GENMASK(7, 0)
+#define ENA_ETH_IO_NUMA_NODE_CFG_REG_ENABLED_SHIFT 31
+#define ENA_ETH_IO_NUMA_NODE_CFG_REG_ENABLED_MASK BIT(31)
+
+#endif /*_ENA_ETH_IO_H_ */
diff --git a/drivers/net/ethernet/amazon/ena/ena_ethtool.c b/drivers/net/ethernet/amazon/ena/ena_ethtool.c
new file mode 100644 (file)
index 0000000..67b2338
--- /dev/null
@@ -0,0 +1,895 @@
+/*
+ * Copyright 2015 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+
+#include "ena_netdev.h"
+
+struct ena_stats {
+       char name[ETH_GSTRING_LEN];
+       int stat_offset;
+};
+
+#define ENA_STAT_ENA_COM_ENTRY(stat) { \
+       .name = #stat, \
+       .stat_offset = offsetof(struct ena_com_stats_admin, stat) \
+}
+
+#define ENA_STAT_ENTRY(stat, stat_type) { \
+       .name = #stat, \
+       .stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
+}
+
+#define ENA_STAT_RX_ENTRY(stat) \
+       ENA_STAT_ENTRY(stat, rx)
+
+#define ENA_STAT_TX_ENTRY(stat) \
+       ENA_STAT_ENTRY(stat, tx)
+
+#define ENA_STAT_GLOBAL_ENTRY(stat) \
+       ENA_STAT_ENTRY(stat, dev)
+
+static const struct ena_stats ena_stats_global_strings[] = {
+       ENA_STAT_GLOBAL_ENTRY(tx_timeout),
+       ENA_STAT_GLOBAL_ENTRY(io_suspend),
+       ENA_STAT_GLOBAL_ENTRY(io_resume),
+       ENA_STAT_GLOBAL_ENTRY(wd_expired),
+       ENA_STAT_GLOBAL_ENTRY(interface_up),
+       ENA_STAT_GLOBAL_ENTRY(interface_down),
+       ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
+};
+
+static const struct ena_stats ena_stats_tx_strings[] = {
+       ENA_STAT_TX_ENTRY(cnt),
+       ENA_STAT_TX_ENTRY(bytes),
+       ENA_STAT_TX_ENTRY(queue_stop),
+       ENA_STAT_TX_ENTRY(queue_wakeup),
+       ENA_STAT_TX_ENTRY(dma_mapping_err),
+       ENA_STAT_TX_ENTRY(linearize),
+       ENA_STAT_TX_ENTRY(linearize_failed),
+       ENA_STAT_TX_ENTRY(napi_comp),
+       ENA_STAT_TX_ENTRY(tx_poll),
+       ENA_STAT_TX_ENTRY(doorbells),
+       ENA_STAT_TX_ENTRY(prepare_ctx_err),
+       ENA_STAT_TX_ENTRY(missing_tx_comp),
+       ENA_STAT_TX_ENTRY(bad_req_id),
+};
+
+static const struct ena_stats ena_stats_rx_strings[] = {
+       ENA_STAT_RX_ENTRY(cnt),
+       ENA_STAT_RX_ENTRY(bytes),
+       ENA_STAT_RX_ENTRY(refil_partial),
+       ENA_STAT_RX_ENTRY(bad_csum),
+       ENA_STAT_RX_ENTRY(page_alloc_fail),
+       ENA_STAT_RX_ENTRY(skb_alloc_fail),
+       ENA_STAT_RX_ENTRY(dma_mapping_err),
+       ENA_STAT_RX_ENTRY(bad_desc_num),
+       ENA_STAT_RX_ENTRY(rx_copybreak_pkt),
+};
+
+static const struct ena_stats ena_stats_ena_com_strings[] = {
+       ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
+       ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
+       ENA_STAT_ENA_COM_ENTRY(completed_cmd),
+       ENA_STAT_ENA_COM_ENTRY(out_of_space),
+       ENA_STAT_ENA_COM_ENTRY(no_completion),
+};
+
+#define ENA_STATS_ARRAY_GLOBAL ARRAY_SIZE(ena_stats_global_strings)
+#define ENA_STATS_ARRAY_TX     ARRAY_SIZE(ena_stats_tx_strings)
+#define ENA_STATS_ARRAY_RX     ARRAY_SIZE(ena_stats_rx_strings)
+#define ENA_STATS_ARRAY_ENA_COM        ARRAY_SIZE(ena_stats_ena_com_strings)
+
+static void ena_safe_update_stat(u64 *src, u64 *dst,
+                                struct u64_stats_sync *syncp)
+{
+       unsigned int start;
+
+       do {
+               start = u64_stats_fetch_begin_irq(syncp);
+               *(dst) = *src;
+       } while (u64_stats_fetch_retry_irq(syncp, start));
+}
+
+static void ena_queue_stats(struct ena_adapter *adapter, u64 **data)
+{
+       const struct ena_stats *ena_stats;
+       struct ena_ring *ring;
+
+       u64 *ptr;
+       int i, j;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               /* Tx stats */
+               ring = &adapter->tx_ring[i];
+
+               for (j = 0; j < ENA_STATS_ARRAY_TX; j++) {
+                       ena_stats = &ena_stats_tx_strings[j];
+
+                       ptr = (u64 *)((uintptr_t)&ring->tx_stats +
+                               (uintptr_t)ena_stats->stat_offset);
+
+                       ena_safe_update_stat(ptr, (*data)++, &ring->syncp);
+               }
+
+               /* Rx stats */
+               ring = &adapter->rx_ring[i];
+
+               for (j = 0; j < ENA_STATS_ARRAY_RX; j++) {
+                       ena_stats = &ena_stats_rx_strings[j];
+
+                       ptr = (u64 *)((uintptr_t)&ring->rx_stats +
+                               (uintptr_t)ena_stats->stat_offset);
+
+                       ena_safe_update_stat(ptr, (*data)++, &ring->syncp);
+               }
+       }
+}
+
+static void ena_dev_admin_queue_stats(struct ena_adapter *adapter, u64 **data)
+{
+       const struct ena_stats *ena_stats;
+       u32 *ptr;
+       int i;
+
+       for (i = 0; i < ENA_STATS_ARRAY_ENA_COM; i++) {
+               ena_stats = &ena_stats_ena_com_strings[i];
+
+               ptr = (u32 *)((uintptr_t)&adapter->ena_dev->admin_queue.stats +
+                       (uintptr_t)ena_stats->stat_offset);
+
+               *(*data)++ = *ptr;
+       }
+}
+
+static void ena_get_ethtool_stats(struct net_device *netdev,
+                                 struct ethtool_stats *stats,
+                                 u64 *data)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       const struct ena_stats *ena_stats;
+       u64 *ptr;
+       int i;
+
+       for (i = 0; i < ENA_STATS_ARRAY_GLOBAL; i++) {
+               ena_stats = &ena_stats_global_strings[i];
+
+               ptr = (u64 *)((uintptr_t)&adapter->dev_stats +
+                       (uintptr_t)ena_stats->stat_offset);
+
+               ena_safe_update_stat(ptr, data++, &adapter->syncp);
+       }
+
+       ena_queue_stats(adapter, &data);
+       ena_dev_admin_queue_stats(adapter, &data);
+}
+
+int ena_get_sset_count(struct net_device *netdev, int sset)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+
+       if (sset != ETH_SS_STATS)
+               return -EOPNOTSUPP;
+
+       return  adapter->num_queues * (ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX)
+               + ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
+}
+
+static void ena_queue_strings(struct ena_adapter *adapter, u8 **data)
+{
+       const struct ena_stats *ena_stats;
+       int i, j;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               /* Tx stats */
+               for (j = 0; j < ENA_STATS_ARRAY_TX; j++) {
+                       ena_stats = &ena_stats_tx_strings[j];
+
+                       snprintf(*data, ETH_GSTRING_LEN,
+                                "queue_%u_tx_%s", i, ena_stats->name);
+                        (*data) += ETH_GSTRING_LEN;
+               }
+               /* Rx stats */
+               for (j = 0; j < ENA_STATS_ARRAY_RX; j++) {
+                       ena_stats = &ena_stats_rx_strings[j];
+
+                       snprintf(*data, ETH_GSTRING_LEN,
+                                "queue_%u_rx_%s", i, ena_stats->name);
+                       (*data) += ETH_GSTRING_LEN;
+               }
+       }
+}
+
+static void ena_com_dev_strings(u8 **data)
+{
+       const struct ena_stats *ena_stats;
+       int i;
+
+       for (i = 0; i < ENA_STATS_ARRAY_ENA_COM; i++) {
+               ena_stats = &ena_stats_ena_com_strings[i];
+
+               snprintf(*data, ETH_GSTRING_LEN,
+                        "ena_admin_q_%s", ena_stats->name);
+               (*data) += ETH_GSTRING_LEN;
+       }
+}
+
+static void ena_get_strings(struct net_device *netdev, u32 sset, u8 *data)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       const struct ena_stats *ena_stats;
+       int i;
+
+       if (sset != ETH_SS_STATS)
+               return;
+
+       for (i = 0; i < ENA_STATS_ARRAY_GLOBAL; i++) {
+               ena_stats = &ena_stats_global_strings[i];
+
+               memcpy(data, ena_stats->name, ETH_GSTRING_LEN);
+               data += ETH_GSTRING_LEN;
+       }
+
+       ena_queue_strings(adapter, &data);
+       ena_com_dev_strings(&data);
+}
+
+static int ena_get_link_ksettings(struct net_device *netdev,
+                                 struct ethtool_link_ksettings *link_ksettings)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       struct ena_admin_get_feature_link_desc *link;
+       struct ena_admin_get_feat_resp feat_resp;
+       int rc;
+
+       rc = ena_com_get_link_params(ena_dev, &feat_resp);
+       if (rc)
+               return rc;
+
+       link = &feat_resp.u.link;
+       link_ksettings->base.speed = link->speed;
+
+       if (link->flags & ENA_ADMIN_GET_FEATURE_LINK_DESC_AUTONEG_MASK) {
+               ethtool_link_ksettings_add_link_mode(link_ksettings,
+                                                    supported, Autoneg);
+               ethtool_link_ksettings_add_link_mode(link_ksettings,
+                                                    supported, Autoneg);
+       }
+
+       link_ksettings->base.autoneg =
+               (link->flags & ENA_ADMIN_GET_FEATURE_LINK_DESC_AUTONEG_MASK) ?
+               AUTONEG_ENABLE : AUTONEG_DISABLE;
+
+       link_ksettings->base.duplex = DUPLEX_FULL;
+
+       return 0;
+}
+
+static int ena_get_coalesce(struct net_device *net_dev,
+                           struct ethtool_coalesce *coalesce)
+{
+       struct ena_adapter *adapter = netdev_priv(net_dev);
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       struct ena_intr_moder_entry intr_moder_entry;
+
+       if (!ena_com_interrupt_moderation_supported(ena_dev)) {
+               /* the devie doesn't support interrupt moderation */
+               return -EOPNOTSUPP;
+       }
+       coalesce->tx_coalesce_usecs =
+               ena_com_get_nonadaptive_moderation_interval_tx(ena_dev) /
+                       ena_dev->intr_delay_resolution;
+       if (!ena_com_get_adaptive_moderation_enabled(ena_dev)) {
+               coalesce->rx_coalesce_usecs =
+                       ena_com_get_nonadaptive_moderation_interval_rx(ena_dev)
+                       / ena_dev->intr_delay_resolution;
+       } else {
+               ena_com_get_intr_moderation_entry(adapter->ena_dev, ENA_INTR_MODER_LOWEST, &intr_moder_entry);
+               coalesce->rx_coalesce_usecs_low = intr_moder_entry.intr_moder_interval;
+               coalesce->rx_max_coalesced_frames_low = intr_moder_entry.pkts_per_interval;
+
+               ena_com_get_intr_moderation_entry(adapter->ena_dev, ENA_INTR_MODER_MID, &intr_moder_entry);
+               coalesce->rx_coalesce_usecs = intr_moder_entry.intr_moder_interval;
+               coalesce->rx_max_coalesced_frames = intr_moder_entry.pkts_per_interval;
+
+               ena_com_get_intr_moderation_entry(adapter->ena_dev, ENA_INTR_MODER_HIGHEST, &intr_moder_entry);
+               coalesce->rx_coalesce_usecs_high = intr_moder_entry.intr_moder_interval;
+               coalesce->rx_max_coalesced_frames_high = intr_moder_entry.pkts_per_interval;
+       }
+       coalesce->use_adaptive_rx_coalesce =
+               ena_com_get_adaptive_moderation_enabled(ena_dev);
+
+       return 0;
+}
+
+static void ena_update_tx_rings_intr_moderation(struct ena_adapter *adapter)
+{
+       unsigned int val;
+       int i;
+
+       val = ena_com_get_nonadaptive_moderation_interval_tx(adapter->ena_dev);
+
+       for (i = 0; i < adapter->num_queues; i++)
+               adapter->tx_ring[i].smoothed_interval = val;
+}
+
+static int ena_set_coalesce(struct net_device *net_dev,
+                           struct ethtool_coalesce *coalesce)
+{
+       struct ena_adapter *adapter = netdev_priv(net_dev);
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       struct ena_intr_moder_entry intr_moder_entry;
+       int rc;
+
+       if (!ena_com_interrupt_moderation_supported(ena_dev)) {
+               /* the devie doesn't support interrupt moderation */
+               return -EOPNOTSUPP;
+       }
+
+       if (coalesce->rx_coalesce_usecs_irq ||
+           coalesce->rx_max_coalesced_frames_irq ||
+           coalesce->tx_coalesce_usecs_irq ||
+           coalesce->tx_max_coalesced_frames ||
+           coalesce->tx_max_coalesced_frames_irq ||
+           coalesce->stats_block_coalesce_usecs ||
+           coalesce->use_adaptive_tx_coalesce ||
+           coalesce->pkt_rate_low ||
+           coalesce->tx_coalesce_usecs_low ||
+           coalesce->tx_max_coalesced_frames_low ||
+           coalesce->pkt_rate_high ||
+           coalesce->tx_coalesce_usecs_high ||
+           coalesce->tx_max_coalesced_frames_high ||
+           coalesce->rate_sample_interval)
+               return -EINVAL;
+
+       rc = ena_com_update_nonadaptive_moderation_interval_tx(ena_dev,
+                                                              coalesce->tx_coalesce_usecs);
+       if (rc)
+               return rc;
+
+       ena_update_tx_rings_intr_moderation(adapter);
+
+       if (ena_com_get_adaptive_moderation_enabled(ena_dev)) {
+               if (!coalesce->use_adaptive_rx_coalesce) {
+                       ena_com_disable_adaptive_moderation(ena_dev);
+                       rc = ena_com_update_nonadaptive_moderation_interval_rx(ena_dev,
+                                                                              coalesce->rx_coalesce_usecs);
+                       return rc;
+               }
+       } else { /* was in non-adaptive mode */
+               if (coalesce->use_adaptive_rx_coalesce) {
+                       ena_com_enable_adaptive_moderation(ena_dev);
+               } else {
+                       rc = ena_com_update_nonadaptive_moderation_interval_rx(ena_dev,
+                                                                              coalesce->rx_coalesce_usecs);
+                       return rc;
+               }
+       }
+
+       intr_moder_entry.intr_moder_interval = coalesce->rx_coalesce_usecs_low;
+       intr_moder_entry.pkts_per_interval = coalesce->rx_max_coalesced_frames_low;
+       intr_moder_entry.bytes_per_interval = ENA_INTR_BYTE_COUNT_NOT_SUPPORTED;
+       ena_com_init_intr_moderation_entry(adapter->ena_dev, ENA_INTR_MODER_LOWEST, &intr_moder_entry);
+
+       intr_moder_entry.intr_moder_interval = coalesce->rx_coalesce_usecs;
+       intr_moder_entry.pkts_per_interval = coalesce->rx_max_coalesced_frames;
+       intr_moder_entry.bytes_per_interval = ENA_INTR_BYTE_COUNT_NOT_SUPPORTED;
+       ena_com_init_intr_moderation_entry(adapter->ena_dev, ENA_INTR_MODER_MID, &intr_moder_entry);
+
+       intr_moder_entry.intr_moder_interval = coalesce->rx_coalesce_usecs_high;
+       intr_moder_entry.pkts_per_interval = coalesce->rx_max_coalesced_frames_high;
+       intr_moder_entry.bytes_per_interval = ENA_INTR_BYTE_COUNT_NOT_SUPPORTED;
+       ena_com_init_intr_moderation_entry(adapter->ena_dev, ENA_INTR_MODER_HIGHEST, &intr_moder_entry);
+
+       return 0;
+}
+
+static u32 ena_get_msglevel(struct net_device *netdev)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+
+       return adapter->msg_enable;
+}
+
+static void ena_set_msglevel(struct net_device *netdev, u32 value)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+
+       adapter->msg_enable = value;
+}
+
+static void ena_get_drvinfo(struct net_device *dev,
+                           struct ethtool_drvinfo *info)
+{
+       struct ena_adapter *adapter = netdev_priv(dev);
+
+       strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
+       strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
+       strlcpy(info->bus_info, pci_name(adapter->pdev),
+               sizeof(info->bus_info));
+}
+
+static void ena_get_ringparam(struct net_device *netdev,
+                             struct ethtool_ringparam *ring)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       struct ena_ring *tx_ring = &adapter->tx_ring[0];
+       struct ena_ring *rx_ring = &adapter->rx_ring[0];
+
+       ring->rx_max_pending = rx_ring->ring_size;
+       ring->tx_max_pending = tx_ring->ring_size;
+       ring->rx_pending = rx_ring->ring_size;
+       ring->tx_pending = tx_ring->ring_size;
+}
+
+static u32 ena_flow_hash_to_flow_type(u16 hash_fields)
+{
+       u32 data = 0;
+
+       if (hash_fields & ENA_ADMIN_RSS_L2_DA)
+               data |= RXH_L2DA;
+
+       if (hash_fields & ENA_ADMIN_RSS_L3_DA)
+               data |= RXH_IP_DST;
+
+       if (hash_fields & ENA_ADMIN_RSS_L3_SA)
+               data |= RXH_IP_SRC;
+
+       if (hash_fields & ENA_ADMIN_RSS_L4_DP)
+               data |= RXH_L4_B_2_3;
+
+       if (hash_fields & ENA_ADMIN_RSS_L4_SP)
+               data |= RXH_L4_B_0_1;
+
+       return data;
+}
+
+static u16 ena_flow_data_to_flow_hash(u32 hash_fields)
+{
+       u16 data = 0;
+
+       if (hash_fields & RXH_L2DA)
+               data |= ENA_ADMIN_RSS_L2_DA;
+
+       if (hash_fields & RXH_IP_DST)
+               data |= ENA_ADMIN_RSS_L3_DA;
+
+       if (hash_fields & RXH_IP_SRC)
+               data |= ENA_ADMIN_RSS_L3_SA;
+
+       if (hash_fields & RXH_L4_B_2_3)
+               data |= ENA_ADMIN_RSS_L4_DP;
+
+       if (hash_fields & RXH_L4_B_0_1)
+               data |= ENA_ADMIN_RSS_L4_SP;
+
+       return data;
+}
+
+static int ena_get_rss_hash(struct ena_com_dev *ena_dev,
+                           struct ethtool_rxnfc *cmd)
+{
+       enum ena_admin_flow_hash_proto proto;
+       u16 hash_fields;
+       int rc;
+
+       cmd->data = 0;
+
+       switch (cmd->flow_type) {
+       case TCP_V4_FLOW:
+               proto = ENA_ADMIN_RSS_TCP4;
+               break;
+       case UDP_V4_FLOW:
+               proto = ENA_ADMIN_RSS_UDP4;
+               break;
+       case TCP_V6_FLOW:
+               proto = ENA_ADMIN_RSS_TCP6;
+               break;
+       case UDP_V6_FLOW:
+               proto = ENA_ADMIN_RSS_UDP6;
+               break;
+       case IPV4_FLOW:
+               proto = ENA_ADMIN_RSS_IP4;
+               break;
+       case IPV6_FLOW:
+               proto = ENA_ADMIN_RSS_IP6;
+               break;
+       case ETHER_FLOW:
+               proto = ENA_ADMIN_RSS_NOT_IP;
+               break;
+       case AH_V4_FLOW:
+       case ESP_V4_FLOW:
+       case AH_V6_FLOW:
+       case ESP_V6_FLOW:
+       case SCTP_V4_FLOW:
+       case AH_ESP_V4_FLOW:
+               return -EOPNOTSUPP;
+       default:
+               return -EINVAL;
+       }
+
+       rc = ena_com_get_hash_ctrl(ena_dev, proto, &hash_fields);
+       if (rc) {
+               /* If device don't have permission, return unsupported */
+               if (rc == -EPERM)
+                       rc = -EOPNOTSUPP;
+               return rc;
+       }
+
+       cmd->data = ena_flow_hash_to_flow_type(hash_fields);
+
+       return 0;
+}
+
+static int ena_set_rss_hash(struct ena_com_dev *ena_dev,
+                           struct ethtool_rxnfc *cmd)
+{
+       enum ena_admin_flow_hash_proto proto;
+       u16 hash_fields;
+
+       switch (cmd->flow_type) {
+       case TCP_V4_FLOW:
+               proto = ENA_ADMIN_RSS_TCP4;
+               break;
+       case UDP_V4_FLOW:
+               proto = ENA_ADMIN_RSS_UDP4;
+               break;
+       case TCP_V6_FLOW:
+               proto = ENA_ADMIN_RSS_TCP6;
+               break;
+       case UDP_V6_FLOW:
+               proto = ENA_ADMIN_RSS_UDP6;
+               break;
+       case IPV4_FLOW:
+               proto = ENA_ADMIN_RSS_IP4;
+               break;
+       case IPV6_FLOW:
+               proto = ENA_ADMIN_RSS_IP6;
+               break;
+       case ETHER_FLOW:
+               proto = ENA_ADMIN_RSS_NOT_IP;
+               break;
+       case AH_V4_FLOW:
+       case ESP_V4_FLOW:
+       case AH_V6_FLOW:
+       case ESP_V6_FLOW:
+       case SCTP_V4_FLOW:
+       case AH_ESP_V4_FLOW:
+               return -EOPNOTSUPP;
+       default:
+               return -EINVAL;
+       }
+
+       hash_fields = ena_flow_data_to_flow_hash(cmd->data);
+
+       return ena_com_fill_hash_ctrl(ena_dev, proto, hash_fields);
+}
+
+static int ena_set_rxnfc(struct net_device *netdev, struct ethtool_rxnfc *info)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       int rc = 0;
+
+       switch (info->cmd) {
+       case ETHTOOL_SRXFH:
+               rc = ena_set_rss_hash(adapter->ena_dev, info);
+               break;
+       case ETHTOOL_SRXCLSRLDEL:
+       case ETHTOOL_SRXCLSRLINS:
+       default:
+               netif_err(adapter, drv, netdev,
+                         "Command parameter %d is not supported\n", info->cmd);
+               rc = -EOPNOTSUPP;
+       }
+
+       return (rc == -EPERM) ? -EOPNOTSUPP : rc;
+}
+
+static int ena_get_rxnfc(struct net_device *netdev, struct ethtool_rxnfc *info,
+                        u32 *rules)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       int rc = 0;
+
+       switch (info->cmd) {
+       case ETHTOOL_GRXRINGS:
+               info->data = adapter->num_queues;
+               rc = 0;
+               break;
+       case ETHTOOL_GRXFH:
+               rc = ena_get_rss_hash(adapter->ena_dev, info);
+               break;
+       case ETHTOOL_GRXCLSRLCNT:
+       case ETHTOOL_GRXCLSRULE:
+       case ETHTOOL_GRXCLSRLALL:
+       default:
+               netif_err(adapter, drv, netdev,
+                         "Command parameter %d is not supported\n", info->cmd);
+               rc = -EOPNOTSUPP;
+       }
+
+       return (rc == -EPERM) ? -EOPNOTSUPP : rc;
+}
+
+static u32 ena_get_rxfh_indir_size(struct net_device *netdev)
+{
+       return ENA_RX_RSS_TABLE_SIZE;
+}
+
+static u32 ena_get_rxfh_key_size(struct net_device *netdev)
+{
+       return ENA_HASH_KEY_SIZE;
+}
+
+static int ena_get_rxfh(struct net_device *netdev, u32 *indir, u8 *key,
+                       u8 *hfunc)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       enum ena_admin_hash_functions ena_func;
+       u8 func;
+       int rc;
+
+       rc = ena_com_indirect_table_get(adapter->ena_dev, indir);
+       if (rc)
+               return rc;
+
+       rc = ena_com_get_hash_function(adapter->ena_dev, &ena_func, key);
+       if (rc)
+               return rc;
+
+       switch (ena_func) {
+       case ENA_ADMIN_TOEPLITZ:
+               func = ETH_RSS_HASH_TOP;
+               break;
+       case ENA_ADMIN_CRC32:
+               func = ETH_RSS_HASH_XOR;
+               break;
+       default:
+               netif_err(adapter, drv, netdev,
+                         "Command parameter is not supported\n");
+               return -EOPNOTSUPP;
+       }
+
+       if (hfunc)
+               *hfunc = func;
+
+       return rc;
+}
+
+static int ena_set_rxfh(struct net_device *netdev, const u32 *indir,
+                       const u8 *key, const u8 hfunc)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       enum ena_admin_hash_functions func;
+       int rc, i;
+
+       if (indir) {
+               for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
+                       rc = ena_com_indirect_table_fill_entry(ena_dev,
+                                                              ENA_IO_RXQ_IDX(indir[i]),
+                                                              i);
+                       if (unlikely(rc)) {
+                               netif_err(adapter, drv, netdev,
+                                         "Cannot fill indirect table (index is too large)\n");
+                               return rc;
+                       }
+               }
+
+               rc = ena_com_indirect_table_set(ena_dev);
+               if (rc) {
+                       netif_err(adapter, drv, netdev,
+                                 "Cannot set indirect table\n");
+                       return rc == -EPERM ? -EOPNOTSUPP : rc;
+               }
+       }
+
+       switch (hfunc) {
+       case ETH_RSS_HASH_TOP:
+               func = ENA_ADMIN_TOEPLITZ;
+               break;
+       case ETH_RSS_HASH_XOR:
+               func = ENA_ADMIN_CRC32;
+               break;
+       default:
+               netif_err(adapter, drv, netdev, "Unsupported hfunc %d\n",
+                         hfunc);
+               return -EOPNOTSUPP;
+       }
+
+       if (key) {
+               rc = ena_com_fill_hash_function(ena_dev, func, key,
+                                               ENA_HASH_KEY_SIZE,
+                                               0xFFFFFFFF);
+               if (unlikely(rc)) {
+                       netif_err(adapter, drv, netdev, "Cannot fill key\n");
+                       return rc == -EPERM ? -EOPNOTSUPP : rc;
+               }
+       }
+
+       return 0;
+}
+
+static void ena_get_channels(struct net_device *netdev,
+                            struct ethtool_channels *channels)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+
+       channels->max_rx = ENA_MAX_NUM_IO_QUEUES;
+       channels->max_tx = ENA_MAX_NUM_IO_QUEUES;
+       channels->max_other = 0;
+       channels->max_combined = 0;
+       channels->rx_count = adapter->num_queues;
+       channels->tx_count = adapter->num_queues;
+       channels->other_count = 0;
+       channels->combined_count = 0;
+}
+
+static int ena_get_tunable(struct net_device *netdev,
+                          const struct ethtool_tunable *tuna, void *data)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       int ret = 0;
+
+       switch (tuna->id) {
+       case ETHTOOL_RX_COPYBREAK:
+               *(u32 *)data = adapter->rx_copybreak;
+               break;
+       default:
+               ret = -EINVAL;
+               break;
+       }
+
+       return ret;
+}
+
+static int ena_set_tunable(struct net_device *netdev,
+                          const struct ethtool_tunable *tuna,
+                          const void *data)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       int ret = 0;
+       u32 len;
+
+       switch (tuna->id) {
+       case ETHTOOL_RX_COPYBREAK:
+               len = *(u32 *)data;
+               if (len > adapter->netdev->mtu) {
+                       ret = -EINVAL;
+                       break;
+               }
+               adapter->rx_copybreak = len;
+               break;
+       default:
+               ret = -EINVAL;
+               break;
+       }
+
+       return ret;
+}
+
+static const struct ethtool_ops ena_ethtool_ops = {
+       .get_link_ksettings     = ena_get_link_ksettings,
+       .get_drvinfo            = ena_get_drvinfo,
+       .get_msglevel           = ena_get_msglevel,
+       .set_msglevel           = ena_set_msglevel,
+       .get_link               = ethtool_op_get_link,
+       .get_coalesce           = ena_get_coalesce,
+       .set_coalesce           = ena_set_coalesce,
+       .get_ringparam          = ena_get_ringparam,
+       .get_sset_count         = ena_get_sset_count,
+       .get_strings            = ena_get_strings,
+       .get_ethtool_stats      = ena_get_ethtool_stats,
+       .get_rxnfc              = ena_get_rxnfc,
+       .set_rxnfc              = ena_set_rxnfc,
+       .get_rxfh_indir_size    = ena_get_rxfh_indir_size,
+       .get_rxfh_key_size      = ena_get_rxfh_key_size,
+       .get_rxfh               = ena_get_rxfh,
+       .set_rxfh               = ena_set_rxfh,
+       .get_channels           = ena_get_channels,
+       .get_tunable            = ena_get_tunable,
+       .set_tunable            = ena_set_tunable,
+};
+
+void ena_set_ethtool_ops(struct net_device *netdev)
+{
+       netdev->ethtool_ops = &ena_ethtool_ops;
+}
+
+static void ena_dump_stats_ex(struct ena_adapter *adapter, u8 *buf)
+{
+       struct net_device *netdev = adapter->netdev;
+       u8 *strings_buf;
+       u64 *data_buf;
+       int strings_num;
+       int i, rc;
+
+       strings_num = ena_get_sset_count(netdev, ETH_SS_STATS);
+       if (strings_num <= 0) {
+               netif_err(adapter, drv, netdev, "Can't get stats num\n");
+               return;
+       }
+
+       strings_buf = devm_kzalloc(&adapter->pdev->dev,
+                                  strings_num * ETH_GSTRING_LEN,
+                                  GFP_ATOMIC);
+       if (!strings_buf) {
+               netif_err(adapter, drv, netdev,
+                         "failed to alloc strings_buf\n");
+               return;
+       }
+
+       data_buf = devm_kzalloc(&adapter->pdev->dev,
+                               strings_num * sizeof(u64),
+                               GFP_ATOMIC);
+       if (!data_buf) {
+               netif_err(adapter, drv, netdev,
+                         "failed to allocate data buf\n");
+               devm_kfree(&adapter->pdev->dev, strings_buf);
+               return;
+       }
+
+       ena_get_strings(netdev, ETH_SS_STATS, strings_buf);
+       ena_get_ethtool_stats(netdev, NULL, data_buf);
+
+       /* If there is a buffer, dump stats, otherwise print them to dmesg */
+       if (buf)
+               for (i = 0; i < strings_num; i++) {
+                       rc = snprintf(buf, ETH_GSTRING_LEN + sizeof(u64),
+                                     "%s %llu\n",
+                                     strings_buf + i * ETH_GSTRING_LEN,
+                                     data_buf[i]);
+                       buf += rc;
+               }
+       else
+               for (i = 0; i < strings_num; i++)
+                       netif_err(adapter, drv, netdev, "%s: %llu\n",
+                                 strings_buf + i * ETH_GSTRING_LEN,
+                                 data_buf[i]);
+
+       devm_kfree(&adapter->pdev->dev, strings_buf);
+       devm_kfree(&adapter->pdev->dev, data_buf);
+}
+
+void ena_dump_stats_to_buf(struct ena_adapter *adapter, u8 *buf)
+{
+       if (!buf)
+               return;
+
+       ena_dump_stats_ex(adapter, buf);
+}
+
+void ena_dump_stats_to_dmesg(struct ena_adapter *adapter)
+{
+       ena_dump_stats_ex(adapter, NULL);
+}
diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.c b/drivers/net/ethernet/amazon/ena/ena_netdev.c
new file mode 100644 (file)
index 0000000..94790df
--- /dev/null
@@ -0,0 +1,3280 @@
+/*
+ * Copyright 2015 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#ifdef CONFIG_RFS_ACCEL
+#include <linux/cpu_rmap.h>
+#endif /* CONFIG_RFS_ACCEL */
+#include <linux/ethtool.h>
+#include <linux/if_vlan.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/numa.h>
+#include <linux/pci.h>
+#include <linux/utsname.h>
+#include <linux/version.h>
+#include <linux/vmalloc.h>
+#include <net/ip.h>
+
+#include "ena_netdev.h"
+#include "ena_pci_id_tbl.h"
+
+static char version[] = DEVICE_NAME " v" DRV_MODULE_VERSION "\n";
+
+MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
+MODULE_DESCRIPTION(DEVICE_NAME);
+MODULE_LICENSE("GPL");
+MODULE_VERSION(DRV_MODULE_VERSION);
+
+/* Time in jiffies before concluding the transmitter is hung. */
+#define TX_TIMEOUT  (5 * HZ)
+
+#define ENA_NAPI_BUDGET 64
+
+#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
+               NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
+static int debug = -1;
+module_param(debug, int, 0);
+MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
+
+static struct ena_aenq_handlers aenq_handlers;
+
+static struct workqueue_struct *ena_wq;
+
+MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
+
+static int ena_rss_init_default(struct ena_adapter *adapter);
+
+static void ena_tx_timeout(struct net_device *dev)
+{
+       struct ena_adapter *adapter = netdev_priv(dev);
+
+       u64_stats_update_begin(&adapter->syncp);
+       adapter->dev_stats.tx_timeout++;
+       u64_stats_update_end(&adapter->syncp);
+
+       netif_err(adapter, tx_err, dev, "Transmit time out\n");
+
+       /* Change the state of the device to trigger reset */
+       set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
+}
+
+static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
+{
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++)
+               adapter->rx_ring[i].mtu = mtu;
+}
+
+static int ena_change_mtu(struct net_device *dev, int new_mtu)
+{
+       struct ena_adapter *adapter = netdev_priv(dev);
+       int ret;
+
+       if ((new_mtu > adapter->max_mtu) || (new_mtu < ENA_MIN_MTU)) {
+               netif_err(adapter, drv, dev,
+                         "Invalid MTU setting. new_mtu: %d\n", new_mtu);
+
+               return -EINVAL;
+       }
+
+       ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
+       if (!ret) {
+               netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
+               update_rx_ring_mtu(adapter, new_mtu);
+               dev->mtu = new_mtu;
+       } else {
+               netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
+                         new_mtu);
+       }
+
+       return ret;
+}
+
+static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
+{
+#ifdef CONFIG_RFS_ACCEL
+       u32 i;
+       int rc;
+
+       adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_queues);
+       if (!adapter->netdev->rx_cpu_rmap)
+               return -ENOMEM;
+       for (i = 0; i < adapter->num_queues; i++) {
+               int irq_idx = ENA_IO_IRQ_IDX(i);
+
+               rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
+                                     adapter->msix_entries[irq_idx].vector);
+               if (rc) {
+                       free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
+                       adapter->netdev->rx_cpu_rmap = NULL;
+                       return rc;
+               }
+       }
+#endif /* CONFIG_RFS_ACCEL */
+       return 0;
+}
+
+static void ena_init_io_rings_common(struct ena_adapter *adapter,
+                                    struct ena_ring *ring, u16 qid)
+{
+       ring->qid = qid;
+       ring->pdev = adapter->pdev;
+       ring->dev = &adapter->pdev->dev;
+       ring->netdev = adapter->netdev;
+       ring->napi = &adapter->ena_napi[qid].napi;
+       ring->adapter = adapter;
+       ring->ena_dev = adapter->ena_dev;
+       ring->per_napi_packets = 0;
+       ring->per_napi_bytes = 0;
+       ring->cpu = 0;
+       u64_stats_init(&ring->syncp);
+}
+
+static void ena_init_io_rings(struct ena_adapter *adapter)
+{
+       struct ena_com_dev *ena_dev;
+       struct ena_ring *txr, *rxr;
+       int i;
+
+       ena_dev = adapter->ena_dev;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               txr = &adapter->tx_ring[i];
+               rxr = &adapter->rx_ring[i];
+
+               /* TX/RX common ring state */
+               ena_init_io_rings_common(adapter, txr, i);
+               ena_init_io_rings_common(adapter, rxr, i);
+
+               /* TX specific ring state */
+               txr->ring_size = adapter->tx_ring_size;
+               txr->tx_max_header_size = ena_dev->tx_max_header_size;
+               txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
+               txr->sgl_size = adapter->max_tx_sgl_size;
+               txr->smoothed_interval =
+                       ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
+
+               /* RX specific ring state */
+               rxr->ring_size = adapter->rx_ring_size;
+               rxr->rx_copybreak = adapter->rx_copybreak;
+               rxr->sgl_size = adapter->max_rx_sgl_size;
+               rxr->smoothed_interval =
+                       ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
+       }
+}
+
+/* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
+ * @adapter: network interface device structure
+ * @qid: queue index
+ *
+ * Return 0 on success, negative on failure
+ */
+static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
+{
+       struct ena_ring *tx_ring = &adapter->tx_ring[qid];
+       struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
+       int size, i, node;
+
+       if (tx_ring->tx_buffer_info) {
+               netif_err(adapter, ifup,
+                         adapter->netdev, "tx_buffer_info info is not NULL");
+               return -EEXIST;
+       }
+
+       size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
+       node = cpu_to_node(ena_irq->cpu);
+
+       tx_ring->tx_buffer_info = vzalloc_node(size, node);
+       if (!tx_ring->tx_buffer_info) {
+               tx_ring->tx_buffer_info = vzalloc(size);
+               if (!tx_ring->tx_buffer_info)
+                       return -ENOMEM;
+       }
+
+       size = sizeof(u16) * tx_ring->ring_size;
+       tx_ring->free_tx_ids = vzalloc_node(size, node);
+       if (!tx_ring->free_tx_ids) {
+               tx_ring->free_tx_ids = vzalloc(size);
+               if (!tx_ring->free_tx_ids) {
+                       vfree(tx_ring->tx_buffer_info);
+                       return -ENOMEM;
+               }
+       }
+
+       /* Req id ring for TX out of order completions */
+       for (i = 0; i < tx_ring->ring_size; i++)
+               tx_ring->free_tx_ids[i] = i;
+
+       /* Reset tx statistics */
+       memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
+
+       tx_ring->next_to_use = 0;
+       tx_ring->next_to_clean = 0;
+       tx_ring->cpu = ena_irq->cpu;
+       return 0;
+}
+
+/* ena_free_tx_resources - Free I/O Tx Resources per Queue
+ * @adapter: network interface device structure
+ * @qid: queue index
+ *
+ * Free all transmit software resources
+ */
+static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
+{
+       struct ena_ring *tx_ring = &adapter->tx_ring[qid];
+
+       vfree(tx_ring->tx_buffer_info);
+       tx_ring->tx_buffer_info = NULL;
+
+       vfree(tx_ring->free_tx_ids);
+       tx_ring->free_tx_ids = NULL;
+}
+
+/* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
+ * @adapter: private structure
+ *
+ * Return 0 on success, negative on failure
+ */
+static int ena_setup_all_tx_resources(struct ena_adapter *adapter)
+{
+       int i, rc = 0;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               rc = ena_setup_tx_resources(adapter, i);
+               if (rc)
+                       goto err_setup_tx;
+       }
+
+       return 0;
+
+err_setup_tx:
+
+       netif_err(adapter, ifup, adapter->netdev,
+                 "Tx queue %d: allocation failed\n", i);
+
+       /* rewind the index freeing the rings as we go */
+       while (i--)
+               ena_free_tx_resources(adapter, i);
+       return rc;
+}
+
+/* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
+ * @adapter: board private structure
+ *
+ * Free all transmit software resources
+ */
+static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
+{
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++)
+               ena_free_tx_resources(adapter, i);
+}
+
+/* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
+ * @adapter: network interface device structure
+ * @qid: queue index
+ *
+ * Returns 0 on success, negative on failure
+ */
+static int ena_setup_rx_resources(struct ena_adapter *adapter,
+                                 u32 qid)
+{
+       struct ena_ring *rx_ring = &adapter->rx_ring[qid];
+       struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
+       int size, node;
+
+       if (rx_ring->rx_buffer_info) {
+               netif_err(adapter, ifup, adapter->netdev,
+                         "rx_buffer_info is not NULL");
+               return -EEXIST;
+       }
+
+       /* alloc extra element so in rx path
+        * we can always prefetch rx_info + 1
+        */
+       size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
+       node = cpu_to_node(ena_irq->cpu);
+
+       rx_ring->rx_buffer_info = vzalloc_node(size, node);
+       if (!rx_ring->rx_buffer_info) {
+               rx_ring->rx_buffer_info = vzalloc(size);
+               if (!rx_ring->rx_buffer_info)
+                       return -ENOMEM;
+       }
+
+       /* Reset rx statistics */
+       memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
+
+       rx_ring->next_to_clean = 0;
+       rx_ring->next_to_use = 0;
+       rx_ring->cpu = ena_irq->cpu;
+
+       return 0;
+}
+
+/* ena_free_rx_resources - Free I/O Rx Resources
+ * @adapter: network interface device structure
+ * @qid: queue index
+ *
+ * Free all receive software resources
+ */
+static void ena_free_rx_resources(struct ena_adapter *adapter,
+                                 u32 qid)
+{
+       struct ena_ring *rx_ring = &adapter->rx_ring[qid];
+
+       vfree(rx_ring->rx_buffer_info);
+       rx_ring->rx_buffer_info = NULL;
+}
+
+/* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
+ * @adapter: board private structure
+ *
+ * Return 0 on success, negative on failure
+ */
+static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
+{
+       int i, rc = 0;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               rc = ena_setup_rx_resources(adapter, i);
+               if (rc)
+                       goto err_setup_rx;
+       }
+
+       return 0;
+
+err_setup_rx:
+
+       netif_err(adapter, ifup, adapter->netdev,
+                 "Rx queue %d: allocation failed\n", i);
+
+       /* rewind the index freeing the rings as we go */
+       while (i--)
+               ena_free_rx_resources(adapter, i);
+       return rc;
+}
+
+/* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
+ * @adapter: board private structure
+ *
+ * Free all receive software resources
+ */
+static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
+{
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++)
+               ena_free_rx_resources(adapter, i);
+}
+
+static inline int ena_alloc_rx_page(struct ena_ring *rx_ring,
+                                   struct ena_rx_buffer *rx_info, gfp_t gfp)
+{
+       struct ena_com_buf *ena_buf;
+       struct page *page;
+       dma_addr_t dma;
+
+       /* if previous allocated page is not used */
+       if (unlikely(rx_info->page))
+               return 0;
+
+       page = alloc_page(gfp);
+       if (unlikely(!page)) {
+               u64_stats_update_begin(&rx_ring->syncp);
+               rx_ring->rx_stats.page_alloc_fail++;
+               u64_stats_update_end(&rx_ring->syncp);
+               return -ENOMEM;
+       }
+
+       dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE,
+                          DMA_FROM_DEVICE);
+       if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
+               u64_stats_update_begin(&rx_ring->syncp);
+               rx_ring->rx_stats.dma_mapping_err++;
+               u64_stats_update_end(&rx_ring->syncp);
+
+               __free_page(page);
+               return -EIO;
+       }
+       netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
+                 "alloc page %p, rx_info %p\n", page, rx_info);
+
+       rx_info->page = page;
+       rx_info->page_offset = 0;
+       ena_buf = &rx_info->ena_buf;
+       ena_buf->paddr = dma;
+       ena_buf->len = PAGE_SIZE;
+
+       return 0;
+}
+
+static void ena_free_rx_page(struct ena_ring *rx_ring,
+                            struct ena_rx_buffer *rx_info)
+{
+       struct page *page = rx_info->page;
+       struct ena_com_buf *ena_buf = &rx_info->ena_buf;
+
+       if (unlikely(!page)) {
+               netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
+                          "Trying to free unallocated buffer\n");
+               return;
+       }
+
+       dma_unmap_page(rx_ring->dev, ena_buf->paddr, PAGE_SIZE,
+                      DMA_FROM_DEVICE);
+
+       __free_page(page);
+       rx_info->page = NULL;
+}
+
+static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
+{
+       u16 next_to_use;
+       u32 i;
+       int rc;
+
+       next_to_use = rx_ring->next_to_use;
+
+       for (i = 0; i < num; i++) {
+               struct ena_rx_buffer *rx_info =
+                       &rx_ring->rx_buffer_info[next_to_use];
+
+               rc = ena_alloc_rx_page(rx_ring, rx_info,
+                                      __GFP_COLD | GFP_ATOMIC | __GFP_COMP);
+               if (unlikely(rc < 0)) {
+                       netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
+                                  "failed to alloc buffer for rx queue %d\n",
+                                  rx_ring->qid);
+                       break;
+               }
+               rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
+                                               &rx_info->ena_buf,
+                                               next_to_use);
+               if (unlikely(rc)) {
+                       netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
+                                  "failed to add buffer for rx queue %d\n",
+                                  rx_ring->qid);
+                       break;
+               }
+               next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
+                                                  rx_ring->ring_size);
+       }
+
+       if (unlikely(i < num)) {
+               u64_stats_update_begin(&rx_ring->syncp);
+               rx_ring->rx_stats.refil_partial++;
+               u64_stats_update_end(&rx_ring->syncp);
+               netdev_warn(rx_ring->netdev,
+                           "refilled rx qid %d with only %d buffers (from %d)\n",
+                           rx_ring->qid, i, num);
+       }
+
+       if (likely(i)) {
+               /* Add memory barrier to make sure the desc were written before
+                * issue a doorbell
+                */
+               wmb();
+               ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
+       }
+
+       rx_ring->next_to_use = next_to_use;
+
+       return i;
+}
+
+static void ena_free_rx_bufs(struct ena_adapter *adapter,
+                            u32 qid)
+{
+       struct ena_ring *rx_ring = &adapter->rx_ring[qid];
+       u32 i;
+
+       for (i = 0; i < rx_ring->ring_size; i++) {
+               struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
+
+               if (rx_info->page)
+                       ena_free_rx_page(rx_ring, rx_info);
+       }
+}
+
+/* ena_refill_all_rx_bufs - allocate all queues Rx buffers
+ * @adapter: board private structure
+ *
+ */
+static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
+{
+       struct ena_ring *rx_ring;
+       int i, rc, bufs_num;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               rx_ring = &adapter->rx_ring[i];
+               bufs_num = rx_ring->ring_size - 1;
+               rc = ena_refill_rx_bufs(rx_ring, bufs_num);
+
+               if (unlikely(rc != bufs_num))
+                       netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
+                                  "refilling Queue %d failed. allocated %d buffers from: %d\n",
+                                  i, rc, bufs_num);
+       }
+}
+
+static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
+{
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++)
+               ena_free_rx_bufs(adapter, i);
+}
+
+/* ena_free_tx_bufs - Free Tx Buffers per Queue
+ * @tx_ring: TX ring for which buffers be freed
+ */
+static void ena_free_tx_bufs(struct ena_ring *tx_ring)
+{
+       u32 i;
+
+       for (i = 0; i < tx_ring->ring_size; i++) {
+               struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
+               struct ena_com_buf *ena_buf;
+               int nr_frags;
+               int j;
+
+               if (!tx_info->skb)
+                       continue;
+
+               netdev_notice(tx_ring->netdev,
+                             "free uncompleted tx skb qid %d idx 0x%x\n",
+                             tx_ring->qid, i);
+
+               ena_buf = tx_info->bufs;
+               dma_unmap_single(tx_ring->dev,
+                                ena_buf->paddr,
+                                ena_buf->len,
+                                DMA_TO_DEVICE);
+
+               /* unmap remaining mapped pages */
+               nr_frags = tx_info->num_of_bufs - 1;
+               for (j = 0; j < nr_frags; j++) {
+                       ena_buf++;
+                       dma_unmap_page(tx_ring->dev,
+                                      ena_buf->paddr,
+                                      ena_buf->len,
+                                      DMA_TO_DEVICE);
+               }
+
+               dev_kfree_skb_any(tx_info->skb);
+       }
+       netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
+                                                 tx_ring->qid));
+}
+
+static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
+{
+       struct ena_ring *tx_ring;
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               tx_ring = &adapter->tx_ring[i];
+               ena_free_tx_bufs(tx_ring);
+       }
+}
+
+static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
+{
+       u16 ena_qid;
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               ena_qid = ENA_IO_TXQ_IDX(i);
+               ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
+       }
+}
+
+static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
+{
+       u16 ena_qid;
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               ena_qid = ENA_IO_RXQ_IDX(i);
+               ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
+       }
+}
+
+static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
+{
+       ena_destroy_all_tx_queues(adapter);
+       ena_destroy_all_rx_queues(adapter);
+}
+
+static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
+{
+       struct ena_tx_buffer *tx_info = NULL;
+
+       if (likely(req_id < tx_ring->ring_size)) {
+               tx_info = &tx_ring->tx_buffer_info[req_id];
+               if (likely(tx_info->skb))
+                       return 0;
+       }
+
+       if (tx_info)
+               netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
+                         "tx_info doesn't have valid skb\n");
+       else
+               netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
+                         "Invalid req_id: %hu\n", req_id);
+
+       u64_stats_update_begin(&tx_ring->syncp);
+       tx_ring->tx_stats.bad_req_id++;
+       u64_stats_update_end(&tx_ring->syncp);
+
+       /* Trigger device reset */
+       set_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags);
+       return -EFAULT;
+}
+
+static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
+{
+       struct netdev_queue *txq;
+       bool above_thresh;
+       u32 tx_bytes = 0;
+       u32 total_done = 0;
+       u16 next_to_clean;
+       u16 req_id;
+       int tx_pkts = 0;
+       int rc;
+
+       next_to_clean = tx_ring->next_to_clean;
+       txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
+
+       while (tx_pkts < budget) {
+               struct ena_tx_buffer *tx_info;
+               struct sk_buff *skb;
+               struct ena_com_buf *ena_buf;
+               int i, nr_frags;
+
+               rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
+                                               &req_id);
+               if (rc)
+                       break;
+
+               rc = validate_tx_req_id(tx_ring, req_id);
+               if (rc)
+                       break;
+
+               tx_info = &tx_ring->tx_buffer_info[req_id];
+               skb = tx_info->skb;
+
+               /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
+               prefetch(&skb->end);
+
+               tx_info->skb = NULL;
+               tx_info->last_jiffies = 0;
+
+               if (likely(tx_info->num_of_bufs != 0)) {
+                       ena_buf = tx_info->bufs;
+
+                       dma_unmap_single(tx_ring->dev,
+                                        dma_unmap_addr(ena_buf, paddr),
+                                        dma_unmap_len(ena_buf, len),
+                                        DMA_TO_DEVICE);
+
+                       /* unmap remaining mapped pages */
+                       nr_frags = tx_info->num_of_bufs - 1;
+                       for (i = 0; i < nr_frags; i++) {
+                               ena_buf++;
+                               dma_unmap_page(tx_ring->dev,
+                                              dma_unmap_addr(ena_buf, paddr),
+                                              dma_unmap_len(ena_buf, len),
+                                              DMA_TO_DEVICE);
+                       }
+               }
+
+               netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
+                         "tx_poll: q %d skb %p completed\n", tx_ring->qid,
+                         skb);
+
+               tx_bytes += skb->len;
+               dev_kfree_skb(skb);
+               tx_pkts++;
+               total_done += tx_info->tx_descs;
+
+               tx_ring->free_tx_ids[next_to_clean] = req_id;
+               next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
+                                                    tx_ring->ring_size);
+       }
+
+       tx_ring->next_to_clean = next_to_clean;
+       ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
+       ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
+
+       netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
+
+       netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
+                 "tx_poll: q %d done. total pkts: %d\n",
+                 tx_ring->qid, tx_pkts);
+
+       /* need to make the rings circular update visible to
+        * ena_start_xmit() before checking for netif_queue_stopped().
+        */
+       smp_mb();
+
+       above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
+               ENA_TX_WAKEUP_THRESH;
+       if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
+               __netif_tx_lock(txq, smp_processor_id());
+               above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
+                       ENA_TX_WAKEUP_THRESH;
+               if (netif_tx_queue_stopped(txq) && above_thresh) {
+                       netif_tx_wake_queue(txq);
+                       u64_stats_update_begin(&tx_ring->syncp);
+                       tx_ring->tx_stats.queue_wakeup++;
+                       u64_stats_update_end(&tx_ring->syncp);
+               }
+               __netif_tx_unlock(txq);
+       }
+
+       tx_ring->per_napi_bytes += tx_bytes;
+       tx_ring->per_napi_packets += tx_pkts;
+
+       return tx_pkts;
+}
+
+static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
+                                 struct ena_com_rx_buf_info *ena_bufs,
+                                 u32 descs,
+                                 u16 *next_to_clean)
+{
+       struct sk_buff *skb;
+       struct ena_rx_buffer *rx_info =
+               &rx_ring->rx_buffer_info[*next_to_clean];
+       u32 len;
+       u32 buf = 0;
+       void *va;
+
+       len = ena_bufs[0].len;
+       if (unlikely(!rx_info->page)) {
+               netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
+                         "Page is NULL\n");
+               return NULL;
+       }
+
+       netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
+                 "rx_info %p page %p\n",
+                 rx_info, rx_info->page);
+
+       /* save virt address of first buffer */
+       va = page_address(rx_info->page) + rx_info->page_offset;
+       prefetch(va + NET_IP_ALIGN);
+
+       if (len <= rx_ring->rx_copybreak) {
+               skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
+                                               rx_ring->rx_copybreak);
+               if (unlikely(!skb)) {
+                       u64_stats_update_begin(&rx_ring->syncp);
+                       rx_ring->rx_stats.skb_alloc_fail++;
+                       u64_stats_update_end(&rx_ring->syncp);
+                       netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
+                                 "Failed to allocate skb\n");
+                       return NULL;
+               }
+
+               netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
+                         "rx allocated small packet. len %d. data_len %d\n",
+                         skb->len, skb->data_len);
+
+               /* sync this buffer for CPU use */
+               dma_sync_single_for_cpu(rx_ring->dev,
+                                       dma_unmap_addr(&rx_info->ena_buf, paddr),
+                                       len,
+                                       DMA_FROM_DEVICE);
+               skb_copy_to_linear_data(skb, va, len);
+               dma_sync_single_for_device(rx_ring->dev,
+                                          dma_unmap_addr(&rx_info->ena_buf, paddr),
+                                          len,
+                                          DMA_FROM_DEVICE);
+
+               skb_put(skb, len);
+               skb->protocol = eth_type_trans(skb, rx_ring->netdev);
+               *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
+                                                    rx_ring->ring_size);
+               return skb;
+       }
+
+       skb = napi_get_frags(rx_ring->napi);
+       if (unlikely(!skb)) {
+               netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
+                         "Failed allocating skb\n");
+               u64_stats_update_begin(&rx_ring->syncp);
+               rx_ring->rx_stats.skb_alloc_fail++;
+               u64_stats_update_end(&rx_ring->syncp);
+               return NULL;
+       }
+
+       do {
+               dma_unmap_page(rx_ring->dev,
+                              dma_unmap_addr(&rx_info->ena_buf, paddr),
+                              PAGE_SIZE, DMA_FROM_DEVICE);
+
+               skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
+                               rx_info->page_offset, len, PAGE_SIZE);
+
+               netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
+                         "rx skb updated. len %d. data_len %d\n",
+                         skb->len, skb->data_len);
+
+               rx_info->page = NULL;
+               *next_to_clean =
+                       ENA_RX_RING_IDX_NEXT(*next_to_clean,
+                                            rx_ring->ring_size);
+               if (likely(--descs == 0))
+                       break;
+               rx_info = &rx_ring->rx_buffer_info[*next_to_clean];
+               len = ena_bufs[++buf].len;
+       } while (1);
+
+       return skb;
+}
+
+/* ena_rx_checksum - indicate in skb if hw indicated a good cksum
+ * @adapter: structure containing adapter specific data
+ * @ena_rx_ctx: received packet context/metadata
+ * @skb: skb currently being received and modified
+ */
+static inline void ena_rx_checksum(struct ena_ring *rx_ring,
+                                  struct ena_com_rx_ctx *ena_rx_ctx,
+                                  struct sk_buff *skb)
+{
+       /* Rx csum disabled */
+       if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
+               skb->ip_summed = CHECKSUM_NONE;
+               return;
+       }
+
+       /* For fragmented packets the checksum isn't valid */
+       if (ena_rx_ctx->frag) {
+               skb->ip_summed = CHECKSUM_NONE;
+               return;
+       }
+
+       /* if IP and error */
+       if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
+                    (ena_rx_ctx->l3_csum_err))) {
+               /* ipv4 checksum error */
+               skb->ip_summed = CHECKSUM_NONE;
+               u64_stats_update_begin(&rx_ring->syncp);
+               rx_ring->rx_stats.bad_csum++;
+               u64_stats_update_end(&rx_ring->syncp);
+               netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
+                         "RX IPv4 header checksum error\n");
+               return;
+       }
+
+       /* if TCP/UDP */
+       if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
+                  (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
+               if (unlikely(ena_rx_ctx->l4_csum_err)) {
+                       /* TCP/UDP checksum error */
+                       u64_stats_update_begin(&rx_ring->syncp);
+                       rx_ring->rx_stats.bad_csum++;
+                       u64_stats_update_end(&rx_ring->syncp);
+                       netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
+                                 "RX L4 checksum error\n");
+                       skb->ip_summed = CHECKSUM_NONE;
+                       return;
+               }
+
+               skb->ip_summed = CHECKSUM_UNNECESSARY;
+       }
+}
+
+static void ena_set_rx_hash(struct ena_ring *rx_ring,
+                           struct ena_com_rx_ctx *ena_rx_ctx,
+                           struct sk_buff *skb)
+{
+       enum pkt_hash_types hash_type;
+
+       if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
+               if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
+                          (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
+
+                       hash_type = PKT_HASH_TYPE_L4;
+               else
+                       hash_type = PKT_HASH_TYPE_NONE;
+
+               /* Override hash type if the packet is fragmented */
+               if (ena_rx_ctx->frag)
+                       hash_type = PKT_HASH_TYPE_NONE;
+
+               skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
+       }
+}
+
+/* ena_clean_rx_irq - Cleanup RX irq
+ * @rx_ring: RX ring to clean
+ * @napi: napi handler
+ * @budget: how many packets driver is allowed to clean
+ *
+ * Returns the number of cleaned buffers.
+ */
+static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
+                           u32 budget)
+{
+       u16 next_to_clean = rx_ring->next_to_clean;
+       u32 res_budget, work_done;
+
+       struct ena_com_rx_ctx ena_rx_ctx;
+       struct ena_adapter *adapter;
+       struct sk_buff *skb;
+       int refill_required;
+       int refill_threshold;
+       int rc = 0;
+       int total_len = 0;
+       int rx_copybreak_pkt = 0;
+
+       netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
+                 "%s qid %d\n", __func__, rx_ring->qid);
+       res_budget = budget;
+
+       do {
+               ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
+               ena_rx_ctx.max_bufs = rx_ring->sgl_size;
+               ena_rx_ctx.descs = 0;
+               rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
+                                   rx_ring->ena_com_io_sq,
+                                   &ena_rx_ctx);
+               if (unlikely(rc))
+                       goto error;
+
+               if (unlikely(ena_rx_ctx.descs == 0))
+                       break;
+
+               netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
+                         "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
+                         rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
+                         ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
+
+               /* allocate skb and fill it */
+               skb = ena_rx_skb(rx_ring, rx_ring->ena_bufs, ena_rx_ctx.descs,
+                                &next_to_clean);
+
+               /* exit if we failed to retrieve a buffer */
+               if (unlikely(!skb)) {
+                       next_to_clean = ENA_RX_RING_IDX_ADD(next_to_clean,
+                                                           ena_rx_ctx.descs,
+                                                           rx_ring->ring_size);
+                       break;
+               }
+
+               ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
+
+               ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
+
+               skb_record_rx_queue(skb, rx_ring->qid);
+
+               if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
+                       total_len += rx_ring->ena_bufs[0].len;
+                       rx_copybreak_pkt++;
+                       napi_gro_receive(napi, skb);
+               } else {
+                       total_len += skb->len;
+                       napi_gro_frags(napi);
+               }
+
+               res_budget--;
+       } while (likely(res_budget));
+
+       work_done = budget - res_budget;
+       rx_ring->per_napi_bytes += total_len;
+       rx_ring->per_napi_packets += work_done;
+       u64_stats_update_begin(&rx_ring->syncp);
+       rx_ring->rx_stats.bytes += total_len;
+       rx_ring->rx_stats.cnt += work_done;
+       rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
+       u64_stats_update_end(&rx_ring->syncp);
+
+       rx_ring->next_to_clean = next_to_clean;
+
+       refill_required = ena_com_sq_empty_space(rx_ring->ena_com_io_sq);
+       refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER;
+
+       /* Optimization, try to batch new rx buffers */
+       if (refill_required > refill_threshold) {
+               ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
+               ena_refill_rx_bufs(rx_ring, refill_required);
+       }
+
+       return work_done;
+
+error:
+       adapter = netdev_priv(rx_ring->netdev);
+
+       u64_stats_update_begin(&rx_ring->syncp);
+       rx_ring->rx_stats.bad_desc_num++;
+       u64_stats_update_end(&rx_ring->syncp);
+
+       /* Too many desc from the device. Trigger reset */
+       set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
+
+       return 0;
+}
+
+inline void ena_adjust_intr_moderation(struct ena_ring *rx_ring,
+                                      struct ena_ring *tx_ring)
+{
+       /* We apply adaptive moderation on Rx path only.
+        * Tx uses static interrupt moderation.
+        */
+       ena_com_calculate_interrupt_delay(rx_ring->ena_dev,
+                                         rx_ring->per_napi_packets,
+                                         rx_ring->per_napi_bytes,
+                                         &rx_ring->smoothed_interval,
+                                         &rx_ring->moder_tbl_idx);
+
+       /* Reset per napi packets/bytes */
+       tx_ring->per_napi_packets = 0;
+       tx_ring->per_napi_bytes = 0;
+       rx_ring->per_napi_packets = 0;
+       rx_ring->per_napi_bytes = 0;
+}
+
+static inline void ena_update_ring_numa_node(struct ena_ring *tx_ring,
+                                            struct ena_ring *rx_ring)
+{
+       int cpu = get_cpu();
+       int numa_node;
+
+       /* Check only one ring since the 2 rings are running on the same cpu */
+       if (likely(tx_ring->cpu == cpu))
+               goto out;
+
+       numa_node = cpu_to_node(cpu);
+       put_cpu();
+
+       if (numa_node != NUMA_NO_NODE) {
+               ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
+               ena_com_update_numa_node(rx_ring->ena_com_io_cq, numa_node);
+       }
+
+       tx_ring->cpu = cpu;
+       rx_ring->cpu = cpu;
+
+       return;
+out:
+       put_cpu();
+}
+
+static int ena_io_poll(struct napi_struct *napi, int budget)
+{
+       struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
+       struct ena_ring *tx_ring, *rx_ring;
+       struct ena_eth_io_intr_reg intr_reg;
+
+       u32 tx_work_done;
+       u32 rx_work_done;
+       int tx_budget;
+       int napi_comp_call = 0;
+       int ret;
+
+       tx_ring = ena_napi->tx_ring;
+       rx_ring = ena_napi->rx_ring;
+
+       tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
+
+       if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
+               napi_complete_done(napi, 0);
+               return 0;
+       }
+
+       tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
+       rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
+
+       if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
+               napi_complete_done(napi, rx_work_done);
+
+               napi_comp_call = 1;
+               /* Tx and Rx share the same interrupt vector */
+               if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
+                       ena_adjust_intr_moderation(rx_ring, tx_ring);
+
+               /* Update intr register: rx intr delay, tx intr delay and
+                * interrupt unmask
+                */
+               ena_com_update_intr_reg(&intr_reg,
+                                       rx_ring->smoothed_interval,
+                                       tx_ring->smoothed_interval,
+                                       true);
+
+               /* It is a shared MSI-X. Tx and Rx CQ have pointer to it.
+                * So we use one of them to reach the intr reg
+                */
+               ena_com_unmask_intr(rx_ring->ena_com_io_cq, &intr_reg);
+
+               ena_update_ring_numa_node(tx_ring, rx_ring);
+
+               ret = rx_work_done;
+       } else {
+               ret = budget;
+       }
+
+       u64_stats_update_begin(&tx_ring->syncp);
+       tx_ring->tx_stats.napi_comp += napi_comp_call;
+       tx_ring->tx_stats.tx_poll++;
+       u64_stats_update_end(&tx_ring->syncp);
+
+       return ret;
+}
+
+static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
+{
+       struct ena_adapter *adapter = (struct ena_adapter *)data;
+
+       ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
+
+       /* Don't call the aenq handler before probe is done */
+       if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
+               ena_com_aenq_intr_handler(adapter->ena_dev, data);
+
+       return IRQ_HANDLED;
+}
+
+/* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
+ * @irq: interrupt number
+ * @data: pointer to a network interface private napi device structure
+ */
+static irqreturn_t ena_intr_msix_io(int irq, void *data)
+{
+       struct ena_napi *ena_napi = data;
+
+       napi_schedule(&ena_napi->napi);
+
+       return IRQ_HANDLED;
+}
+
+static int ena_enable_msix(struct ena_adapter *adapter, int num_queues)
+{
+       int i, msix_vecs, rc;
+
+       if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
+               netif_err(adapter, probe, adapter->netdev,
+                         "Error, MSI-X is already enabled\n");
+               return -EPERM;
+       }
+
+       /* Reserved the max msix vectors we might need */
+       msix_vecs = ENA_MAX_MSIX_VEC(num_queues);
+
+       netif_dbg(adapter, probe, adapter->netdev,
+                 "trying to enable MSI-X, vectors %d\n", msix_vecs);
+
+       adapter->msix_entries = vzalloc(msix_vecs * sizeof(struct msix_entry));
+
+       if (!adapter->msix_entries)
+               return -ENOMEM;
+
+       for (i = 0; i < msix_vecs; i++)
+               adapter->msix_entries[i].entry = i;
+
+       rc = pci_enable_msix(adapter->pdev, adapter->msix_entries, msix_vecs);
+       if (rc != 0) {
+               netif_err(adapter, probe, adapter->netdev,
+                         "Failed to enable MSI-X, vectors %d rc %d\n",
+                         msix_vecs, rc);
+               return -ENOSPC;
+       }
+
+       netif_dbg(adapter, probe, adapter->netdev, "enable MSI-X, vectors %d\n",
+                 msix_vecs);
+
+       if (msix_vecs >= 1) {
+               if (ena_init_rx_cpu_rmap(adapter))
+                       netif_warn(adapter, probe, adapter->netdev,
+                                  "Failed to map IRQs to CPUs\n");
+       }
+
+       adapter->msix_vecs = msix_vecs;
+       set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
+
+       return 0;
+}
+
+static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
+{
+       u32 cpu;
+
+       snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
+                ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
+                pci_name(adapter->pdev));
+       adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
+               ena_intr_msix_mgmnt;
+       adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
+       adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
+               adapter->msix_entries[ENA_MGMNT_IRQ_IDX].vector;
+       cpu = cpumask_first(cpu_online_mask);
+       adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
+       cpumask_set_cpu(cpu,
+                       &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
+}
+
+static void ena_setup_io_intr(struct ena_adapter *adapter)
+{
+       struct net_device *netdev;
+       int irq_idx, i, cpu;
+
+       netdev = adapter->netdev;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               irq_idx = ENA_IO_IRQ_IDX(i);
+               cpu = i % num_online_cpus();
+
+               snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
+                        "%s-Tx-Rx-%d", netdev->name, i);
+               adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
+               adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
+               adapter->irq_tbl[irq_idx].vector =
+                       adapter->msix_entries[irq_idx].vector;
+               adapter->irq_tbl[irq_idx].cpu = cpu;
+
+               cpumask_set_cpu(cpu,
+                               &adapter->irq_tbl[irq_idx].affinity_hint_mask);
+       }
+}
+
+static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
+{
+       unsigned long flags = 0;
+       struct ena_irq *irq;
+       int rc;
+
+       irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
+       rc = request_irq(irq->vector, irq->handler, flags, irq->name,
+                        irq->data);
+       if (rc) {
+               netif_err(adapter, probe, adapter->netdev,
+                         "failed to request admin irq\n");
+               return rc;
+       }
+
+       netif_dbg(adapter, probe, adapter->netdev,
+                 "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
+                 irq->affinity_hint_mask.bits[0], irq->vector);
+
+       irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
+
+       return rc;
+}
+
+static int ena_request_io_irq(struct ena_adapter *adapter)
+{
+       unsigned long flags = 0;
+       struct ena_irq *irq;
+       int rc = 0, i, k;
+
+       if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
+               netif_err(adapter, ifup, adapter->netdev,
+                         "Failed to request I/O IRQ: MSI-X is not enabled\n");
+               return -EINVAL;
+       }
+
+       for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
+               irq = &adapter->irq_tbl[i];
+               rc = request_irq(irq->vector, irq->handler, flags, irq->name,
+                                irq->data);
+               if (rc) {
+                       netif_err(adapter, ifup, adapter->netdev,
+                                 "Failed to request I/O IRQ. index %d rc %d\n",
+                                  i, rc);
+                       goto err;
+               }
+
+               netif_dbg(adapter, ifup, adapter->netdev,
+                         "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
+                         i, irq->affinity_hint_mask.bits[0], irq->vector);
+
+               irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
+       }
+
+       return rc;
+
+err:
+       for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
+               irq = &adapter->irq_tbl[k];
+               free_irq(irq->vector, irq->data);
+       }
+
+       return rc;
+}
+
+static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
+{
+       struct ena_irq *irq;
+
+       irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
+       synchronize_irq(irq->vector);
+       irq_set_affinity_hint(irq->vector, NULL);
+       free_irq(irq->vector, irq->data);
+}
+
+static void ena_free_io_irq(struct ena_adapter *adapter)
+{
+       struct ena_irq *irq;
+       int i;
+
+#ifdef CONFIG_RFS_ACCEL
+       if (adapter->msix_vecs >= 1) {
+               free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
+               adapter->netdev->rx_cpu_rmap = NULL;
+       }
+#endif /* CONFIG_RFS_ACCEL */
+
+       for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
+               irq = &adapter->irq_tbl[i];
+               irq_set_affinity_hint(irq->vector, NULL);
+               free_irq(irq->vector, irq->data);
+       }
+}
+
+static void ena_disable_msix(struct ena_adapter *adapter)
+{
+       if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
+               pci_disable_msix(adapter->pdev);
+
+       if (adapter->msix_entries)
+               vfree(adapter->msix_entries);
+       adapter->msix_entries = NULL;
+}
+
+static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
+{
+       int i;
+
+       if (!netif_running(adapter->netdev))
+               return;
+
+       for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++)
+               synchronize_irq(adapter->irq_tbl[i].vector);
+}
+
+static void ena_del_napi(struct ena_adapter *adapter)
+{
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++)
+               netif_napi_del(&adapter->ena_napi[i].napi);
+}
+
+static void ena_init_napi(struct ena_adapter *adapter)
+{
+       struct ena_napi *napi;
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               napi = &adapter->ena_napi[i];
+
+               netif_napi_add(adapter->netdev,
+                              &adapter->ena_napi[i].napi,
+                              ena_io_poll,
+                              ENA_NAPI_BUDGET);
+               napi->rx_ring = &adapter->rx_ring[i];
+               napi->tx_ring = &adapter->tx_ring[i];
+               napi->qid = i;
+       }
+}
+
+static void ena_napi_disable_all(struct ena_adapter *adapter)
+{
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++)
+               napi_disable(&adapter->ena_napi[i].napi);
+}
+
+static void ena_napi_enable_all(struct ena_adapter *adapter)
+{
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++)
+               napi_enable(&adapter->ena_napi[i].napi);
+}
+
+static void ena_restore_ethtool_params(struct ena_adapter *adapter)
+{
+       adapter->tx_usecs = 0;
+       adapter->rx_usecs = 0;
+       adapter->tx_frames = 1;
+       adapter->rx_frames = 1;
+}
+
+/* Configure the Rx forwarding */
+static int ena_rss_configure(struct ena_adapter *adapter)
+{
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       int rc;
+
+       /* In case the RSS table wasn't initialized by probe */
+       if (!ena_dev->rss.tbl_log_size) {
+               rc = ena_rss_init_default(adapter);
+               if (rc && (rc != -EPERM)) {
+                       netif_err(adapter, ifup, adapter->netdev,
+                                 "Failed to init RSS rc: %d\n", rc);
+                       return rc;
+               }
+       }
+
+       /* Set indirect table */
+       rc = ena_com_indirect_table_set(ena_dev);
+       if (unlikely(rc && rc != -EPERM))
+               return rc;
+
+       /* Configure hash function (if supported) */
+       rc = ena_com_set_hash_function(ena_dev);
+       if (unlikely(rc && (rc != -EPERM)))
+               return rc;
+
+       /* Configure hash inputs (if supported) */
+       rc = ena_com_set_hash_ctrl(ena_dev);
+       if (unlikely(rc && (rc != -EPERM)))
+               return rc;
+
+       return 0;
+}
+
+static int ena_up_complete(struct ena_adapter *adapter)
+{
+       int rc, i;
+
+       rc = ena_rss_configure(adapter);
+       if (rc)
+               return rc;
+
+       ena_init_napi(adapter);
+
+       ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
+
+       ena_refill_all_rx_bufs(adapter);
+
+       /* enable transmits */
+       netif_tx_start_all_queues(adapter->netdev);
+
+       ena_restore_ethtool_params(adapter);
+
+       ena_napi_enable_all(adapter);
+
+       /* schedule napi in case we had pending packets
+        * from the last time we disable napi
+        */
+       for (i = 0; i < adapter->num_queues; i++)
+               napi_schedule(&adapter->ena_napi[i].napi);
+
+       return 0;
+}
+
+static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
+{
+       struct ena_com_create_io_ctx ctx = { 0 };
+       struct ena_com_dev *ena_dev;
+       struct ena_ring *tx_ring;
+       u32 msix_vector;
+       u16 ena_qid;
+       int rc;
+
+       ena_dev = adapter->ena_dev;
+
+       tx_ring = &adapter->tx_ring[qid];
+       msix_vector = ENA_IO_IRQ_IDX(qid);
+       ena_qid = ENA_IO_TXQ_IDX(qid);
+
+       ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
+       ctx.qid = ena_qid;
+       ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
+       ctx.msix_vector = msix_vector;
+       ctx.queue_size = adapter->tx_ring_size;
+       ctx.numa_node = cpu_to_node(tx_ring->cpu);
+
+       rc = ena_com_create_io_queue(ena_dev, &ctx);
+       if (rc) {
+               netif_err(adapter, ifup, adapter->netdev,
+                         "Failed to create I/O TX queue num %d rc: %d\n",
+                         qid, rc);
+               return rc;
+       }
+
+       rc = ena_com_get_io_handlers(ena_dev, ena_qid,
+                                    &tx_ring->ena_com_io_sq,
+                                    &tx_ring->ena_com_io_cq);
+       if (rc) {
+               netif_err(adapter, ifup, adapter->netdev,
+                         "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
+                         qid, rc);
+               ena_com_destroy_io_queue(ena_dev, ena_qid);
+       }
+
+       ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
+       return rc;
+}
+
+static int ena_create_all_io_tx_queues(struct ena_adapter *adapter)
+{
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       int rc, i;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               rc = ena_create_io_tx_queue(adapter, i);
+               if (rc)
+                       goto create_err;
+       }
+
+       return 0;
+
+create_err:
+       while (i--)
+               ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
+
+       return rc;
+}
+
+static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
+{
+       struct ena_com_dev *ena_dev;
+       struct ena_com_create_io_ctx ctx = { 0 };
+       struct ena_ring *rx_ring;
+       u32 msix_vector;
+       u16 ena_qid;
+       int rc;
+
+       ena_dev = adapter->ena_dev;
+
+       rx_ring = &adapter->rx_ring[qid];
+       msix_vector = ENA_IO_IRQ_IDX(qid);
+       ena_qid = ENA_IO_RXQ_IDX(qid);
+
+       ctx.qid = ena_qid;
+       ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
+       ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
+       ctx.msix_vector = msix_vector;
+       ctx.queue_size = adapter->rx_ring_size;
+       ctx.numa_node = cpu_to_node(rx_ring->cpu);
+
+       rc = ena_com_create_io_queue(ena_dev, &ctx);
+       if (rc) {
+               netif_err(adapter, ifup, adapter->netdev,
+                         "Failed to create I/O RX queue num %d rc: %d\n",
+                         qid, rc);
+               return rc;
+       }
+
+       rc = ena_com_get_io_handlers(ena_dev, ena_qid,
+                                    &rx_ring->ena_com_io_sq,
+                                    &rx_ring->ena_com_io_cq);
+       if (rc) {
+               netif_err(adapter, ifup, adapter->netdev,
+                         "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
+                         qid, rc);
+               ena_com_destroy_io_queue(ena_dev, ena_qid);
+       }
+
+       ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
+
+       return rc;
+}
+
+static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
+{
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       int rc, i;
+
+       for (i = 0; i < adapter->num_queues; i++) {
+               rc = ena_create_io_rx_queue(adapter, i);
+               if (rc)
+                       goto create_err;
+       }
+
+       return 0;
+
+create_err:
+       while (i--)
+               ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
+
+       return rc;
+}
+
+static int ena_up(struct ena_adapter *adapter)
+{
+       int rc;
+
+       netdev_dbg(adapter->netdev, "%s\n", __func__);
+
+       ena_setup_io_intr(adapter);
+
+       rc = ena_request_io_irq(adapter);
+       if (rc)
+               goto err_req_irq;
+
+       /* allocate transmit descriptors */
+       rc = ena_setup_all_tx_resources(adapter);
+       if (rc)
+               goto err_setup_tx;
+
+       /* allocate receive descriptors */
+       rc = ena_setup_all_rx_resources(adapter);
+       if (rc)
+               goto err_setup_rx;
+
+       /* Create TX queues */
+       rc = ena_create_all_io_tx_queues(adapter);
+       if (rc)
+               goto err_create_tx_queues;
+
+       /* Create RX queues */
+       rc = ena_create_all_io_rx_queues(adapter);
+       if (rc)
+               goto err_create_rx_queues;
+
+       rc = ena_up_complete(adapter);
+       if (rc)
+               goto err_up;
+
+       if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
+               netif_carrier_on(adapter->netdev);
+
+       u64_stats_update_begin(&adapter->syncp);
+       adapter->dev_stats.interface_up++;
+       u64_stats_update_end(&adapter->syncp);
+
+       set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
+
+       return rc;
+
+err_up:
+       ena_destroy_all_rx_queues(adapter);
+err_create_rx_queues:
+       ena_destroy_all_tx_queues(adapter);
+err_create_tx_queues:
+       ena_free_all_io_rx_resources(adapter);
+err_setup_rx:
+       ena_free_all_io_tx_resources(adapter);
+err_setup_tx:
+       ena_free_io_irq(adapter);
+err_req_irq:
+
+       return rc;
+}
+
+static void ena_down(struct ena_adapter *adapter)
+{
+       netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
+
+       clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
+
+       u64_stats_update_begin(&adapter->syncp);
+       adapter->dev_stats.interface_down++;
+       u64_stats_update_end(&adapter->syncp);
+
+       /* After this point the napi handler won't enable the tx queue */
+       ena_napi_disable_all(adapter);
+       netif_carrier_off(adapter->netdev);
+       netif_tx_disable(adapter->netdev);
+
+       /* After destroy the queue there won't be any new interrupts */
+       ena_destroy_all_io_queues(adapter);
+
+       ena_disable_io_intr_sync(adapter);
+       ena_free_io_irq(adapter);
+       ena_del_napi(adapter);
+
+       ena_free_all_tx_bufs(adapter);
+       ena_free_all_rx_bufs(adapter);
+       ena_free_all_io_tx_resources(adapter);
+       ena_free_all_io_rx_resources(adapter);
+}
+
+/* ena_open - Called when a network interface is made active
+ * @netdev: network interface device structure
+ *
+ * Returns 0 on success, negative value on failure
+ *
+ * The open entry point is called when a network interface is made
+ * active by the system (IFF_UP).  At this point all resources needed
+ * for transmit and receive operations are allocated, the interrupt
+ * handler is registered with the OS, the watchdog timer is started,
+ * and the stack is notified that the interface is ready.
+ */
+static int ena_open(struct net_device *netdev)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       int rc;
+
+       /* Notify the stack of the actual queue counts. */
+       rc = netif_set_real_num_tx_queues(netdev, adapter->num_queues);
+       if (rc) {
+               netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
+               return rc;
+       }
+
+       rc = netif_set_real_num_rx_queues(netdev, adapter->num_queues);
+       if (rc) {
+               netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
+               return rc;
+       }
+
+       rc = ena_up(adapter);
+       if (rc)
+               return rc;
+
+       return rc;
+}
+
+/* ena_close - Disables a network interface
+ * @netdev: network interface device structure
+ *
+ * Returns 0, this is not allowed to fail
+ *
+ * The close entry point is called when an interface is de-activated
+ * by the OS.  The hardware is still under the drivers control, but
+ * needs to be disabled.  A global MAC reset is issued to stop the
+ * hardware, and all transmit and receive resources are freed.
+ */
+static int ena_close(struct net_device *netdev)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+
+       netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
+
+       if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
+               ena_down(adapter);
+
+       return 0;
+}
+
+static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
+{
+       u32 mss = skb_shinfo(skb)->gso_size;
+       struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
+       u8 l4_protocol = 0;
+
+       if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
+               ena_tx_ctx->l4_csum_enable = 1;
+               if (mss) {
+                       ena_tx_ctx->tso_enable = 1;
+                       ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
+                       ena_tx_ctx->l4_csum_partial = 0;
+               } else {
+                       ena_tx_ctx->tso_enable = 0;
+                       ena_meta->l4_hdr_len = 0;
+                       ena_tx_ctx->l4_csum_partial = 1;
+               }
+
+               switch (ip_hdr(skb)->version) {
+               case IPVERSION:
+                       ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
+                       if (ip_hdr(skb)->frag_off & htons(IP_DF))
+                               ena_tx_ctx->df = 1;
+                       if (mss)
+                               ena_tx_ctx->l3_csum_enable = 1;
+                       l4_protocol = ip_hdr(skb)->protocol;
+                       break;
+               case 6:
+                       ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
+                       l4_protocol = ipv6_hdr(skb)->nexthdr;
+                       break;
+               default:
+                       break;
+               }
+
+               if (l4_protocol == IPPROTO_TCP)
+                       ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
+               else
+                       ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
+
+               ena_meta->mss = mss;
+               ena_meta->l3_hdr_len = skb_network_header_len(skb);
+               ena_meta->l3_hdr_offset = skb_network_offset(skb);
+               ena_tx_ctx->meta_valid = 1;
+
+       } else {
+               ena_tx_ctx->meta_valid = 0;
+       }
+}
+
+static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
+                                      struct sk_buff *skb)
+{
+       int num_frags, header_len, rc;
+
+       num_frags = skb_shinfo(skb)->nr_frags;
+       header_len = skb_headlen(skb);
+
+       if (num_frags < tx_ring->sgl_size)
+               return 0;
+
+       if ((num_frags == tx_ring->sgl_size) &&
+           (header_len < tx_ring->tx_max_header_size))
+               return 0;
+
+       u64_stats_update_begin(&tx_ring->syncp);
+       tx_ring->tx_stats.linearize++;
+       u64_stats_update_end(&tx_ring->syncp);
+
+       rc = skb_linearize(skb);
+       if (unlikely(rc)) {
+               u64_stats_update_begin(&tx_ring->syncp);
+               tx_ring->tx_stats.linearize_failed++;
+               u64_stats_update_end(&tx_ring->syncp);
+       }
+
+       return rc;
+}
+
+/* Called with netif_tx_lock. */
+static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct ena_adapter *adapter = netdev_priv(dev);
+       struct ena_tx_buffer *tx_info;
+       struct ena_com_tx_ctx ena_tx_ctx;
+       struct ena_ring *tx_ring;
+       struct netdev_queue *txq;
+       struct ena_com_buf *ena_buf;
+       void *push_hdr;
+       u32 len, last_frag;
+       u16 next_to_use;
+       u16 req_id;
+       u16 push_len;
+       u16 header_len;
+       dma_addr_t dma;
+       int qid, rc, nb_hw_desc;
+       int i = -1;
+
+       netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
+       /*  Determine which tx ring we will be placed on */
+       qid = skb_get_queue_mapping(skb);
+       tx_ring = &adapter->tx_ring[qid];
+       txq = netdev_get_tx_queue(dev, qid);
+
+       rc = ena_check_and_linearize_skb(tx_ring, skb);
+       if (unlikely(rc))
+               goto error_drop_packet;
+
+       skb_tx_timestamp(skb);
+       len = skb_headlen(skb);
+
+       next_to_use = tx_ring->next_to_use;
+       req_id = tx_ring->free_tx_ids[next_to_use];
+       tx_info = &tx_ring->tx_buffer_info[req_id];
+       tx_info->num_of_bufs = 0;
+
+       WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
+       ena_buf = tx_info->bufs;
+       tx_info->skb = skb;
+
+       if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
+               /* prepared the push buffer */
+               push_len = min_t(u32, len, tx_ring->tx_max_header_size);
+               header_len = push_len;
+               push_hdr = skb->data;
+       } else {
+               push_len = 0;
+               header_len = min_t(u32, len, tx_ring->tx_max_header_size);
+               push_hdr = NULL;
+       }
+
+       netif_dbg(adapter, tx_queued, dev,
+                 "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
+                 push_hdr, push_len);
+
+       if (len > push_len) {
+               dma = dma_map_single(tx_ring->dev, skb->data + push_len,
+                                    len - push_len, DMA_TO_DEVICE);
+               if (dma_mapping_error(tx_ring->dev, dma))
+                       goto error_report_dma_error;
+
+               ena_buf->paddr = dma;
+               ena_buf->len = len - push_len;
+
+               ena_buf++;
+               tx_info->num_of_bufs++;
+       }
+
+       last_frag = skb_shinfo(skb)->nr_frags;
+
+       for (i = 0; i < last_frag; i++) {
+               const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+
+               len = skb_frag_size(frag);
+               dma = skb_frag_dma_map(tx_ring->dev, frag, 0, len,
+                                      DMA_TO_DEVICE);
+               if (dma_mapping_error(tx_ring->dev, dma))
+                       goto error_report_dma_error;
+
+               ena_buf->paddr = dma;
+               ena_buf->len = len;
+               ena_buf++;
+       }
+
+       tx_info->num_of_bufs += last_frag;
+
+       memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
+       ena_tx_ctx.ena_bufs = tx_info->bufs;
+       ena_tx_ctx.push_header = push_hdr;
+       ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
+       ena_tx_ctx.req_id = req_id;
+       ena_tx_ctx.header_len = header_len;
+
+       /* set flags and meta data */
+       ena_tx_csum(&ena_tx_ctx, skb);
+
+       /* prepare the packet's descriptors to dma engine */
+       rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
+                               &nb_hw_desc);
+
+       if (unlikely(rc)) {
+               netif_err(adapter, tx_queued, dev,
+                         "failed to prepare tx bufs\n");
+               u64_stats_update_begin(&tx_ring->syncp);
+               tx_ring->tx_stats.queue_stop++;
+               tx_ring->tx_stats.prepare_ctx_err++;
+               u64_stats_update_end(&tx_ring->syncp);
+               netif_tx_stop_queue(txq);
+               goto error_unmap_dma;
+       }
+
+       netdev_tx_sent_queue(txq, skb->len);
+
+       u64_stats_update_begin(&tx_ring->syncp);
+       tx_ring->tx_stats.cnt++;
+       tx_ring->tx_stats.bytes += skb->len;
+       u64_stats_update_end(&tx_ring->syncp);
+
+       tx_info->tx_descs = nb_hw_desc;
+       tx_info->last_jiffies = jiffies;
+
+       tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
+               tx_ring->ring_size);
+
+       /* This WMB is aimed to:
+        * 1 - perform smp barrier before reading next_to_completion
+        * 2 - make sure the desc were written before trigger DB
+        */
+       wmb();
+
+       /* stop the queue when no more space available, the packet can have up
+        * to sgl_size + 2. one for the meta descriptor and one for header
+        * (if the header is larger than tx_max_header_size).
+        */
+       if (unlikely(ena_com_sq_empty_space(tx_ring->ena_com_io_sq) <
+                    (tx_ring->sgl_size + 2))) {
+               netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
+                         __func__, qid);
+
+               netif_tx_stop_queue(txq);
+               u64_stats_update_begin(&tx_ring->syncp);
+               tx_ring->tx_stats.queue_stop++;
+               u64_stats_update_end(&tx_ring->syncp);
+
+               /* There is a rare condition where this function decide to
+                * stop the queue but meanwhile clean_tx_irq updates
+                * next_to_completion and terminates.
+                * The queue will remain stopped forever.
+                * To solve this issue this function perform rmb, check
+                * the wakeup condition and wake up the queue if needed.
+                */
+               smp_rmb();
+
+               if (ena_com_sq_empty_space(tx_ring->ena_com_io_sq)
+                               > ENA_TX_WAKEUP_THRESH) {
+                       netif_tx_wake_queue(txq);
+                       u64_stats_update_begin(&tx_ring->syncp);
+                       tx_ring->tx_stats.queue_wakeup++;
+                       u64_stats_update_end(&tx_ring->syncp);
+               }
+       }
+
+       if (netif_xmit_stopped(txq) || !skb->xmit_more) {
+               /* trigger the dma engine */
+               ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
+               u64_stats_update_begin(&tx_ring->syncp);
+               tx_ring->tx_stats.doorbells++;
+               u64_stats_update_end(&tx_ring->syncp);
+       }
+
+       return NETDEV_TX_OK;
+
+error_report_dma_error:
+       u64_stats_update_begin(&tx_ring->syncp);
+       tx_ring->tx_stats.dma_mapping_err++;
+       u64_stats_update_end(&tx_ring->syncp);
+       netdev_warn(adapter->netdev, "failed to map skb\n");
+
+       tx_info->skb = NULL;
+
+error_unmap_dma:
+       if (i >= 0) {
+               /* save value of frag that failed */
+               last_frag = i;
+
+               /* start back at beginning and unmap skb */
+               tx_info->skb = NULL;
+               ena_buf = tx_info->bufs;
+               dma_unmap_single(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
+                                dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
+
+               /* unmap remaining mapped pages */
+               for (i = 0; i < last_frag; i++) {
+                       ena_buf++;
+                       dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
+                                      dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
+               }
+       }
+
+error_drop_packet:
+
+       dev_kfree_skb(skb);
+       return NETDEV_TX_OK;
+}
+
+#ifdef CONFIG_NET_POLL_CONTROLLER
+static void ena_netpoll(struct net_device *netdev)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       int i;
+
+       for (i = 0; i < adapter->num_queues; i++)
+               napi_schedule(&adapter->ena_napi[i].napi);
+}
+#endif /* CONFIG_NET_POLL_CONTROLLER */
+
+static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
+                           void *accel_priv, select_queue_fallback_t fallback)
+{
+       u16 qid;
+       /* we suspect that this is good for in--kernel network services that
+        * want to loop incoming skb rx to tx in normal user generated traffic,
+        * most probably we will not get to this
+        */
+       if (skb_rx_queue_recorded(skb))
+               qid = skb_get_rx_queue(skb);
+       else
+               qid = fallback(dev, skb);
+
+       return qid;
+}
+
+static void ena_config_host_info(struct ena_com_dev *ena_dev)
+{
+       struct ena_admin_host_info *host_info;
+       int rc;
+
+       /* Allocate only the host info */
+       rc = ena_com_allocate_host_info(ena_dev);
+       if (rc) {
+               pr_err("Cannot allocate host info\n");
+               return;
+       }
+
+       host_info = ena_dev->host_attr.host_info;
+
+       host_info->os_type = ENA_ADMIN_OS_LINUX;
+       host_info->kernel_ver = LINUX_VERSION_CODE;
+       strncpy(host_info->kernel_ver_str, utsname()->version,
+               sizeof(host_info->kernel_ver_str) - 1);
+       host_info->os_dist = 0;
+       strncpy(host_info->os_dist_str, utsname()->release,
+               sizeof(host_info->os_dist_str) - 1);
+       host_info->driver_version =
+               (DRV_MODULE_VER_MAJOR) |
+               (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
+               (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
+
+       rc = ena_com_set_host_attributes(ena_dev);
+       if (rc) {
+               if (rc == -EPERM)
+                       pr_warn("Cannot set host attributes\n");
+               else
+                       pr_err("Cannot set host attributes\n");
+
+               goto err;
+       }
+
+       return;
+
+err:
+       ena_com_delete_host_info(ena_dev);
+}
+
+static void ena_config_debug_area(struct ena_adapter *adapter)
+{
+       u32 debug_area_size;
+       int rc, ss_count;
+
+       ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
+       if (ss_count <= 0) {
+               netif_err(adapter, drv, adapter->netdev,
+                         "SS count is negative\n");
+               return;
+       }
+
+       /* allocate 32 bytes for each string and 64bit for the value */
+       debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
+
+       rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
+       if (rc) {
+               pr_err("Cannot allocate debug area\n");
+               return;
+       }
+
+       rc = ena_com_set_host_attributes(adapter->ena_dev);
+       if (rc) {
+               if (rc == -EPERM)
+                       netif_warn(adapter, drv, adapter->netdev,
+                                  "Cannot set host attributes\n");
+               else
+                       netif_err(adapter, drv, adapter->netdev,
+                                 "Cannot set host attributes\n");
+               goto err;
+       }
+
+       return;
+err:
+       ena_com_delete_debug_area(adapter->ena_dev);
+}
+
+static struct rtnl_link_stats64 *ena_get_stats64(struct net_device *netdev,
+                                                struct rtnl_link_stats64 *stats)
+{
+       struct ena_adapter *adapter = netdev_priv(netdev);
+       struct ena_admin_basic_stats ena_stats;
+       int rc;
+
+       if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
+               return NULL;
+
+       rc = ena_com_get_dev_basic_stats(adapter->ena_dev, &ena_stats);
+       if (rc)
+               return NULL;
+
+       stats->tx_bytes = ((u64)ena_stats.tx_bytes_high << 32) |
+               ena_stats.tx_bytes_low;
+       stats->rx_bytes = ((u64)ena_stats.rx_bytes_high << 32) |
+               ena_stats.rx_bytes_low;
+
+       stats->rx_packets = ((u64)ena_stats.rx_pkts_high << 32) |
+               ena_stats.rx_pkts_low;
+       stats->tx_packets = ((u64)ena_stats.tx_pkts_high << 32) |
+               ena_stats.tx_pkts_low;
+
+       stats->rx_dropped = ((u64)ena_stats.rx_drops_high << 32) |
+               ena_stats.rx_drops_low;
+
+       stats->multicast = 0;
+       stats->collisions = 0;
+
+       stats->rx_length_errors = 0;
+       stats->rx_crc_errors = 0;
+       stats->rx_frame_errors = 0;
+       stats->rx_fifo_errors = 0;
+       stats->rx_missed_errors = 0;
+       stats->tx_window_errors = 0;
+
+       stats->rx_errors = 0;
+       stats->tx_errors = 0;
+
+       return stats;
+}
+
+static const struct net_device_ops ena_netdev_ops = {
+       .ndo_open               = ena_open,
+       .ndo_stop               = ena_close,
+       .ndo_start_xmit         = ena_start_xmit,
+       .ndo_select_queue       = ena_select_queue,
+       .ndo_get_stats64        = ena_get_stats64,
+       .ndo_tx_timeout         = ena_tx_timeout,
+       .ndo_change_mtu         = ena_change_mtu,
+       .ndo_set_mac_address    = NULL,
+       .ndo_validate_addr      = eth_validate_addr,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = ena_netpoll,
+#endif /* CONFIG_NET_POLL_CONTROLLER */
+};
+
+static void ena_device_io_suspend(struct work_struct *work)
+{
+       struct ena_adapter *adapter =
+               container_of(work, struct ena_adapter, suspend_io_task);
+       struct net_device *netdev = adapter->netdev;
+
+       /* ena_napi_disable_all disables only the IO handling.
+        * We are still subject to AENQ keep alive watchdog.
+        */
+       u64_stats_update_begin(&adapter->syncp);
+       adapter->dev_stats.io_suspend++;
+       u64_stats_update_begin(&adapter->syncp);
+       ena_napi_disable_all(adapter);
+       netif_tx_lock(netdev);
+       netif_device_detach(netdev);
+       netif_tx_unlock(netdev);
+}
+
+static void ena_device_io_resume(struct work_struct *work)
+{
+       struct ena_adapter *adapter =
+               container_of(work, struct ena_adapter, resume_io_task);
+       struct net_device *netdev = adapter->netdev;
+
+       u64_stats_update_begin(&adapter->syncp);
+       adapter->dev_stats.io_resume++;
+       u64_stats_update_end(&adapter->syncp);
+
+       netif_device_attach(netdev);
+       ena_napi_enable_all(adapter);
+}
+
+static int ena_device_validate_params(struct ena_adapter *adapter,
+                                     struct ena_com_dev_get_features_ctx *get_feat_ctx)
+{
+       struct net_device *netdev = adapter->netdev;
+       int rc;
+
+       rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
+                             adapter->mac_addr);
+       if (!rc) {
+               netif_err(adapter, drv, netdev,
+                         "Error, mac address are different\n");
+               return -EINVAL;
+       }
+
+       if ((get_feat_ctx->max_queues.max_cq_num < adapter->num_queues) ||
+           (get_feat_ctx->max_queues.max_sq_num < adapter->num_queues)) {
+               netif_err(adapter, drv, netdev,
+                         "Error, device doesn't support enough queues\n");
+               return -EINVAL;
+       }
+
+       if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
+               netif_err(adapter, drv, netdev,
+                         "Error, device max mtu is smaller than netdev MTU\n");
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
+                          struct ena_com_dev_get_features_ctx *get_feat_ctx,
+                          bool *wd_state)
+{
+       struct device *dev = &pdev->dev;
+       bool readless_supported;
+       u32 aenq_groups;
+       int dma_width;
+       int rc;
+
+       rc = ena_com_mmio_reg_read_request_init(ena_dev);
+       if (rc) {
+               dev_err(dev, "failed to init mmio read less\n");
+               return rc;
+       }
+
+       /* The PCIe configuration space revision id indicate if mmio reg
+        * read is disabled
+        */
+       readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
+       ena_com_set_mmio_read_mode(ena_dev, readless_supported);
+
+       rc = ena_com_dev_reset(ena_dev);
+       if (rc) {
+               dev_err(dev, "Can not reset device\n");
+               goto err_mmio_read_less;
+       }
+
+       rc = ena_com_validate_version(ena_dev);
+       if (rc) {
+               dev_err(dev, "device version is too low\n");
+               goto err_mmio_read_less;
+       }
+
+       dma_width = ena_com_get_dma_width(ena_dev);
+       if (dma_width < 0) {
+               dev_err(dev, "Invalid dma width value %d", dma_width);
+               goto err_mmio_read_less;
+       }
+
+       rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
+       if (rc) {
+               dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
+               goto err_mmio_read_less;
+       }
+
+       rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
+       if (rc) {
+               dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
+                       rc);
+               goto err_mmio_read_less;
+       }
+
+       /* ENA admin level init */
+       rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
+       if (rc) {
+               dev_err(dev,
+                       "Can not initialize ena admin queue with device\n");
+               goto err_mmio_read_less;
+       }
+
+       /* To enable the msix interrupts the driver needs to know the number
+        * of queues. So the driver uses polling mode to retrieve this
+        * information
+        */
+       ena_com_set_admin_polling_mode(ena_dev, true);
+
+       /* Get Device Attributes*/
+       rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
+       if (rc) {
+               dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
+               goto err_admin_init;
+       }
+
+       /* Try to turn all the available aenq groups */
+       aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
+               BIT(ENA_ADMIN_FATAL_ERROR) |
+               BIT(ENA_ADMIN_WARNING) |
+               BIT(ENA_ADMIN_NOTIFICATION) |
+               BIT(ENA_ADMIN_KEEP_ALIVE);
+
+       aenq_groups &= get_feat_ctx->aenq.supported_groups;
+
+       rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
+       if (rc) {
+               dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
+               goto err_admin_init;
+       }
+
+       *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
+
+       ena_config_host_info(ena_dev);
+
+       return 0;
+
+err_admin_init:
+       ena_com_admin_destroy(ena_dev);
+err_mmio_read_less:
+       ena_com_mmio_reg_read_request_destroy(ena_dev);
+
+       return rc;
+}
+
+static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
+                                                   int io_vectors)
+{
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       struct device *dev = &adapter->pdev->dev;
+       int rc;
+
+       rc = ena_enable_msix(adapter, io_vectors);
+       if (rc) {
+               dev_err(dev, "Can not reserve msix vectors\n");
+               return rc;
+       }
+
+       ena_setup_mgmnt_intr(adapter);
+
+       rc = ena_request_mgmnt_irq(adapter);
+       if (rc) {
+               dev_err(dev, "Can not setup management interrupts\n");
+               goto err_disable_msix;
+       }
+
+       ena_com_set_admin_polling_mode(ena_dev, false);
+
+       ena_com_admin_aenq_enable(ena_dev);
+
+       return 0;
+
+err_disable_msix:
+       ena_disable_msix(adapter);
+
+       return rc;
+}
+
+static void ena_fw_reset_device(struct work_struct *work)
+{
+       struct ena_com_dev_get_features_ctx get_feat_ctx;
+       struct ena_adapter *adapter =
+               container_of(work, struct ena_adapter, reset_task);
+       struct net_device *netdev = adapter->netdev;
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       struct pci_dev *pdev = adapter->pdev;
+       bool dev_up, wd_state;
+       int rc;
+
+       del_timer_sync(&adapter->timer_service);
+
+       rtnl_lock();
+
+       dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
+       ena_com_set_admin_running_state(ena_dev, false);
+
+       /* After calling ena_close the tx queues and the napi
+        * are disabled so no one can interfere or touch the
+        * data structures
+        */
+       ena_close(netdev);
+
+       rc = ena_com_dev_reset(ena_dev);
+       if (rc) {
+               dev_err(&pdev->dev, "Device reset failed\n");
+               goto err;
+       }
+
+       ena_free_mgmnt_irq(adapter);
+
+       ena_disable_msix(adapter);
+
+       ena_com_abort_admin_commands(ena_dev);
+
+       ena_com_wait_for_abort_completion(ena_dev);
+
+       ena_com_admin_destroy(ena_dev);
+
+       ena_com_mmio_reg_read_request_destroy(ena_dev);
+
+       /* Finish with the destroy part. Start the init part */
+
+       rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
+       if (rc) {
+               dev_err(&pdev->dev, "Can not initialize device\n");
+               goto err;
+       }
+       adapter->wd_state = wd_state;
+
+       rc = ena_device_validate_params(adapter, &get_feat_ctx);
+       if (rc) {
+               dev_err(&pdev->dev, "Validation of device parameters failed\n");
+               goto err_device_destroy;
+       }
+
+       rc = ena_enable_msix_and_set_admin_interrupts(adapter,
+                                                     adapter->num_queues);
+       if (rc) {
+               dev_err(&pdev->dev, "Enable MSI-X failed\n");
+               goto err_device_destroy;
+       }
+       /* If the interface was up before the reset bring it up */
+       if (dev_up) {
+               rc = ena_up(adapter);
+               if (rc) {
+                       dev_err(&pdev->dev, "Failed to create I/O queues\n");
+                       goto err_disable_msix;
+               }
+       }
+
+       mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
+
+       rtnl_unlock();
+
+       dev_err(&pdev->dev, "Device reset completed successfully\n");
+
+       return;
+err_disable_msix:
+       ena_free_mgmnt_irq(adapter);
+       ena_disable_msix(adapter);
+err_device_destroy:
+       ena_com_admin_destroy(ena_dev);
+err:
+       rtnl_unlock();
+
+       dev_err(&pdev->dev,
+               "Reset attempt failed. Can not reset the device\n");
+}
+
+static void check_for_missing_tx_completions(struct ena_adapter *adapter)
+{
+       struct ena_tx_buffer *tx_buf;
+       unsigned long last_jiffies;
+       struct ena_ring *tx_ring;
+       int i, j, budget;
+       u32 missed_tx;
+
+       /* Make sure the driver doesn't turn the device in other process */
+       smp_rmb();
+
+       if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
+               return;
+
+       budget = ENA_MONITORED_TX_QUEUES;
+
+       for (i = adapter->last_monitored_tx_qid; i < adapter->num_queues; i++) {
+               tx_ring = &adapter->tx_ring[i];
+
+               for (j = 0; j < tx_ring->ring_size; j++) {
+                       tx_buf = &tx_ring->tx_buffer_info[j];
+                       last_jiffies = tx_buf->last_jiffies;
+                       if (unlikely(last_jiffies && time_is_before_jiffies(last_jiffies + TX_TIMEOUT))) {
+                               netif_notice(adapter, tx_err, adapter->netdev,
+                                            "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
+                                            tx_ring->qid, j);
+
+                               u64_stats_update_begin(&tx_ring->syncp);
+                               missed_tx = tx_ring->tx_stats.missing_tx_comp++;
+                               u64_stats_update_end(&tx_ring->syncp);
+
+                               /* Clear last jiffies so the lost buffer won't
+                                * be counted twice.
+                                */
+                               tx_buf->last_jiffies = 0;
+
+                               if (unlikely(missed_tx > MAX_NUM_OF_TIMEOUTED_PACKETS)) {
+                                       netif_err(adapter, tx_err, adapter->netdev,
+                                                 "The number of lost tx completion is above the threshold (%d > %d). Reset the device\n",
+                                                 missed_tx, MAX_NUM_OF_TIMEOUTED_PACKETS);
+                                       set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
+                               }
+                       }
+               }
+
+               budget--;
+               if (!budget)
+                       break;
+       }
+
+       adapter->last_monitored_tx_qid = i % adapter->num_queues;
+}
+
+/* Check for keep alive expiration */
+static void check_for_missing_keep_alive(struct ena_adapter *adapter)
+{
+       unsigned long keep_alive_expired;
+
+       if (!adapter->wd_state)
+               return;
+
+       keep_alive_expired = round_jiffies(adapter->last_keep_alive_jiffies
+                                          + ENA_DEVICE_KALIVE_TIMEOUT);
+       if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
+               netif_err(adapter, drv, adapter->netdev,
+                         "Keep alive watchdog timeout.\n");
+               u64_stats_update_begin(&adapter->syncp);
+               adapter->dev_stats.wd_expired++;
+               u64_stats_update_end(&adapter->syncp);
+               set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
+       }
+}
+
+static void check_for_admin_com_state(struct ena_adapter *adapter)
+{
+       if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
+               netif_err(adapter, drv, adapter->netdev,
+                         "ENA admin queue is not in running state!\n");
+               u64_stats_update_begin(&adapter->syncp);
+               adapter->dev_stats.admin_q_pause++;
+               u64_stats_update_end(&adapter->syncp);
+               set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
+       }
+}
+
+static void ena_update_host_info(struct ena_admin_host_info *host_info,
+                                struct net_device *netdev)
+{
+       host_info->supported_network_features[0] =
+               netdev->features & GENMASK_ULL(31, 0);
+       host_info->supported_network_features[1] =
+               (netdev->features & GENMASK_ULL(63, 32)) >> 32;
+}
+
+static void ena_timer_service(unsigned long data)
+{
+       struct ena_adapter *adapter = (struct ena_adapter *)data;
+       u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
+       struct ena_admin_host_info *host_info =
+               adapter->ena_dev->host_attr.host_info;
+
+       check_for_missing_keep_alive(adapter);
+
+       check_for_admin_com_state(adapter);
+
+       check_for_missing_tx_completions(adapter);
+
+       if (debug_area)
+               ena_dump_stats_to_buf(adapter, debug_area);
+
+       if (host_info)
+               ena_update_host_info(host_info, adapter->netdev);
+
+       if (unlikely(test_and_clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
+               netif_err(adapter, drv, adapter->netdev,
+                         "Trigger reset is on\n");
+               ena_dump_stats_to_dmesg(adapter);
+               queue_work(ena_wq, &adapter->reset_task);
+               return;
+       }
+
+       /* Reset the timer */
+       mod_timer(&adapter->timer_service, jiffies + HZ);
+}
+
+static int ena_calc_io_queue_num(struct pci_dev *pdev,
+                                struct ena_com_dev *ena_dev,
+                                struct ena_com_dev_get_features_ctx *get_feat_ctx)
+{
+       int io_sq_num, io_queue_num;
+
+       /* In case of LLQ use the llq number in the get feature cmd */
+       if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
+               io_sq_num = get_feat_ctx->max_queues.max_llq_num;
+
+               if (io_sq_num == 0) {
+                       dev_err(&pdev->dev,
+                               "Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
+
+                       ena_dev->tx_mem_queue_type =
+                               ENA_ADMIN_PLACEMENT_POLICY_HOST;
+                       io_sq_num = get_feat_ctx->max_queues.max_sq_num;
+               }
+       } else {
+               io_sq_num = get_feat_ctx->max_queues.max_sq_num;
+       }
+
+       io_queue_num = min_t(int, num_possible_cpus(), ENA_MAX_NUM_IO_QUEUES);
+       io_queue_num = min_t(int, io_queue_num, io_sq_num);
+       io_queue_num = min_t(int, io_queue_num,
+                            get_feat_ctx->max_queues.max_cq_num);
+       /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
+       io_queue_num = min_t(int, io_queue_num, pci_msix_vec_count(pdev) - 1);
+       if (unlikely(!io_queue_num)) {
+               dev_err(&pdev->dev, "The device doesn't have io queues\n");
+               return -EFAULT;
+       }
+
+       return io_queue_num;
+}
+
+static int ena_set_push_mode(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
+                            struct ena_com_dev_get_features_ctx *get_feat_ctx)
+{
+       bool has_mem_bar;
+
+       has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
+
+       /* Enable push mode if device supports LLQ */
+       if (has_mem_bar && (get_feat_ctx->max_queues.max_llq_num > 0))
+               ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
+       else
+               ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
+
+       return 0;
+}
+
+static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
+                                struct net_device *netdev)
+{
+       netdev_features_t dev_features = 0;
+
+       /* Set offload features */
+       if (feat->offload.tx &
+               ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
+               dev_features |= NETIF_F_IP_CSUM;
+
+       if (feat->offload.tx &
+               ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
+               dev_features |= NETIF_F_IPV6_CSUM;
+
+       if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
+               dev_features |= NETIF_F_TSO;
+
+       if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
+               dev_features |= NETIF_F_TSO6;
+
+       if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
+               dev_features |= NETIF_F_TSO_ECN;
+
+       if (feat->offload.rx_supported &
+               ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
+               dev_features |= NETIF_F_RXCSUM;
+
+       if (feat->offload.rx_supported &
+               ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
+               dev_features |= NETIF_F_RXCSUM;
+
+       netdev->features =
+               dev_features |
+               NETIF_F_SG |
+               NETIF_F_NTUPLE |
+               NETIF_F_RXHASH |
+               NETIF_F_HIGHDMA;
+
+       netdev->hw_features |= netdev->features;
+       netdev->vlan_features |= netdev->features;
+}
+
+static void ena_set_conf_feat_params(struct ena_adapter *adapter,
+                                    struct ena_com_dev_get_features_ctx *feat)
+{
+       struct net_device *netdev = adapter->netdev;
+
+       /* Copy mac address */
+       if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
+               eth_hw_addr_random(netdev);
+               ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
+       } else {
+               ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
+               ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
+       }
+
+       /* Set offload features */
+       ena_set_dev_offloads(feat, netdev);
+
+       adapter->max_mtu = feat->dev_attr.max_mtu;
+}
+
+static int ena_rss_init_default(struct ena_adapter *adapter)
+{
+       struct ena_com_dev *ena_dev = adapter->ena_dev;
+       struct device *dev = &adapter->pdev->dev;
+       int rc, i;
+       u32 val;
+
+       rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
+       if (unlikely(rc)) {
+               dev_err(dev, "Cannot init indirect table\n");
+               goto err_rss_init;
+       }
+
+       for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
+               val = ethtool_rxfh_indir_default(i, adapter->num_queues);
+               rc = ena_com_indirect_table_fill_entry(ena_dev, i,
+                                                      ENA_IO_RXQ_IDX(val));
+               if (unlikely(rc && (rc != -EPERM))) {
+                       dev_err(dev, "Cannot fill indirect table\n");
+                       goto err_fill_indir;
+               }
+       }
+
+       rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
+                                       ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
+       if (unlikely(rc && (rc != -EPERM))) {
+               dev_err(dev, "Cannot fill hash function\n");
+               goto err_fill_indir;
+       }
+
+       rc = ena_com_set_default_hash_ctrl(ena_dev);
+       if (unlikely(rc && (rc != -EPERM))) {
+               dev_err(dev, "Cannot fill hash control\n");
+               goto err_fill_indir;
+       }
+
+       return 0;
+
+err_fill_indir:
+       ena_com_rss_destroy(ena_dev);
+err_rss_init:
+
+       return rc;
+}
+
+static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
+{
+       int release_bars;
+
+       release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
+       pci_release_selected_regions(pdev, release_bars);
+}
+
+static int ena_calc_queue_size(struct pci_dev *pdev,
+                              struct ena_com_dev *ena_dev,
+                              u16 *max_tx_sgl_size,
+                              u16 *max_rx_sgl_size,
+                              struct ena_com_dev_get_features_ctx *get_feat_ctx)
+{
+       u32 queue_size = ENA_DEFAULT_RING_SIZE;
+
+       queue_size = min_t(u32, queue_size,
+                          get_feat_ctx->max_queues.max_cq_depth);
+       queue_size = min_t(u32, queue_size,
+                          get_feat_ctx->max_queues.max_sq_depth);
+
+       if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
+               queue_size = min_t(u32, queue_size,
+                                  get_feat_ctx->max_queues.max_llq_depth);
+
+       queue_size = rounddown_pow_of_two(queue_size);
+
+       if (unlikely(!queue_size)) {
+               dev_err(&pdev->dev, "Invalid queue size\n");
+               return -EFAULT;
+       }
+
+       *max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
+                                get_feat_ctx->max_queues.max_packet_tx_descs);
+       *max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
+                                get_feat_ctx->max_queues.max_packet_rx_descs);
+
+       return queue_size;
+}
+
+/* ena_probe - Device Initialization Routine
+ * @pdev: PCI device information struct
+ * @ent: entry in ena_pci_tbl
+ *
+ * Returns 0 on success, negative on failure
+ *
+ * ena_probe initializes an adapter identified by a pci_dev structure.
+ * The OS initialization, configuring of the adapter private structure,
+ * and a hardware reset occur.
+ */
+static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+       struct ena_com_dev_get_features_ctx get_feat_ctx;
+       static int version_printed;
+       struct net_device *netdev;
+       struct ena_adapter *adapter;
+       struct ena_com_dev *ena_dev = NULL;
+       static int adapters_found;
+       int io_queue_num, bars, rc;
+       int queue_size;
+       u16 tx_sgl_size = 0;
+       u16 rx_sgl_size = 0;
+       bool wd_state;
+
+       dev_dbg(&pdev->dev, "%s\n", __func__);
+
+       if (version_printed++ == 0)
+               dev_info(&pdev->dev, "%s", version);
+
+       rc = pci_enable_device_mem(pdev);
+       if (rc) {
+               dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
+               return rc;
+       }
+
+       pci_set_master(pdev);
+
+       ena_dev = vzalloc(sizeof(*ena_dev));
+       if (!ena_dev) {
+               rc = -ENOMEM;
+               goto err_disable_device;
+       }
+
+       bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
+       rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
+       if (rc) {
+               dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
+                       rc);
+               goto err_free_ena_dev;
+       }
+
+       ena_dev->reg_bar = ioremap(pci_resource_start(pdev, ENA_REG_BAR),
+                                  pci_resource_len(pdev, ENA_REG_BAR));
+       if (!ena_dev->reg_bar) {
+               dev_err(&pdev->dev, "failed to remap regs bar\n");
+               rc = -EFAULT;
+               goto err_free_region;
+       }
+
+       ena_dev->dmadev = &pdev->dev;
+
+       rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
+       if (rc) {
+               dev_err(&pdev->dev, "ena device init failed\n");
+               if (rc == -ETIME)
+                       rc = -EPROBE_DEFER;
+               goto err_free_region;
+       }
+
+       rc = ena_set_push_mode(pdev, ena_dev, &get_feat_ctx);
+       if (rc) {
+               dev_err(&pdev->dev, "Invalid module param(push_mode)\n");
+               goto err_device_destroy;
+       }
+
+       if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
+               ena_dev->mem_bar = ioremap_wc(pci_resource_start(pdev, ENA_MEM_BAR),
+                                             pci_resource_len(pdev, ENA_MEM_BAR));
+               if (!ena_dev->mem_bar) {
+                       rc = -EFAULT;
+                       goto err_device_destroy;
+               }
+       }
+
+       /* initial Tx interrupt delay, Assumes 1 usec granularity.
+       * Updated during device initialization with the real granularity
+       */
+       ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
+       io_queue_num = ena_calc_io_queue_num(pdev, ena_dev, &get_feat_ctx);
+       queue_size = ena_calc_queue_size(pdev, ena_dev, &tx_sgl_size,
+                                        &rx_sgl_size, &get_feat_ctx);
+       if ((queue_size <= 0) || (io_queue_num <= 0)) {
+               rc = -EFAULT;
+               goto err_device_destroy;
+       }
+
+       dev_info(&pdev->dev, "creating %d io queues. queue size: %d\n",
+                io_queue_num, queue_size);
+
+       /* dev zeroed in init_etherdev */
+       netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), io_queue_num);
+       if (!netdev) {
+               dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
+               rc = -ENOMEM;
+               goto err_device_destroy;
+       }
+
+       SET_NETDEV_DEV(netdev, &pdev->dev);
+
+       adapter = netdev_priv(netdev);
+       pci_set_drvdata(pdev, adapter);
+
+       adapter->ena_dev = ena_dev;
+       adapter->netdev = netdev;
+       adapter->pdev = pdev;
+
+       ena_set_conf_feat_params(adapter, &get_feat_ctx);
+
+       adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
+
+       adapter->tx_ring_size = queue_size;
+       adapter->rx_ring_size = queue_size;
+
+       adapter->max_tx_sgl_size = tx_sgl_size;
+       adapter->max_rx_sgl_size = rx_sgl_size;
+
+       adapter->num_queues = io_queue_num;
+       adapter->last_monitored_tx_qid = 0;
+
+       adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
+       adapter->wd_state = wd_state;
+
+       snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
+
+       rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
+       if (rc) {
+               dev_err(&pdev->dev,
+                       "Failed to query interrupt moderation feature\n");
+               goto err_netdev_destroy;
+       }
+       ena_init_io_rings(adapter);
+
+       netdev->netdev_ops = &ena_netdev_ops;
+       netdev->watchdog_timeo = TX_TIMEOUT;
+       ena_set_ethtool_ops(netdev);
+
+       netdev->priv_flags |= IFF_UNICAST_FLT;
+
+       u64_stats_init(&adapter->syncp);
+
+       rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
+       if (rc) {
+               dev_err(&pdev->dev,
+                       "Failed to enable and set the admin interrupts\n");
+               goto err_worker_destroy;
+       }
+       rc = ena_rss_init_default(adapter);
+       if (rc && (rc != -EPERM)) {
+               dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
+               goto err_free_msix;
+       }
+
+       ena_config_debug_area(adapter);
+
+       memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
+
+       netif_carrier_off(netdev);
+
+       rc = register_netdev(netdev);
+       if (rc) {
+               dev_err(&pdev->dev, "Cannot register net device\n");
+               goto err_rss;
+       }
+
+       INIT_WORK(&adapter->suspend_io_task, ena_device_io_suspend);
+       INIT_WORK(&adapter->resume_io_task, ena_device_io_resume);
+       INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
+
+       adapter->last_keep_alive_jiffies = jiffies;
+
+       init_timer(&adapter->timer_service);
+       adapter->timer_service.expires = round_jiffies(jiffies + HZ);
+       adapter->timer_service.function = ena_timer_service;
+       adapter->timer_service.data = (unsigned long)adapter;
+
+       add_timer(&adapter->timer_service);
+
+       dev_info(&pdev->dev, "%s found at mem %lx, mac addr %pM Queues %d\n",
+                DEVICE_NAME, (long)pci_resource_start(pdev, 0),
+                netdev->dev_addr, io_queue_num);
+
+       set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
+
+       adapters_found++;
+
+       return 0;
+
+err_rss:
+       ena_com_delete_debug_area(ena_dev);
+       ena_com_rss_destroy(ena_dev);
+err_free_msix:
+       ena_com_dev_reset(ena_dev);
+       ena_free_mgmnt_irq(adapter);
+       ena_disable_msix(adapter);
+err_worker_destroy:
+       ena_com_destroy_interrupt_moderation(ena_dev);
+       del_timer(&adapter->timer_service);
+       cancel_work_sync(&adapter->suspend_io_task);
+       cancel_work_sync(&adapter->resume_io_task);
+err_netdev_destroy:
+       free_netdev(netdev);
+err_device_destroy:
+       ena_com_delete_host_info(ena_dev);
+       ena_com_admin_destroy(ena_dev);
+err_free_region:
+       ena_release_bars(ena_dev, pdev);
+err_free_ena_dev:
+       pci_set_drvdata(pdev, NULL);
+       vfree(ena_dev);
+err_disable_device:
+       pci_disable_device(pdev);
+       return rc;
+}
+
+/*****************************************************************************/
+static int ena_sriov_configure(struct pci_dev *dev, int numvfs)
+{
+       int rc;
+
+       if (numvfs > 0) {
+               rc = pci_enable_sriov(dev, numvfs);
+               if (rc != 0) {
+                       dev_err(&dev->dev,
+                               "pci_enable_sriov failed to enable: %d vfs with the error: %d\n",
+                               numvfs, rc);
+                       return rc;
+               }
+
+               return numvfs;
+       }
+
+       if (numvfs == 0) {
+               pci_disable_sriov(dev);
+               return 0;
+       }
+
+       return -EINVAL;
+}
+
+/*****************************************************************************/
+/*****************************************************************************/
+
+/* ena_remove - Device Removal Routine
+ * @pdev: PCI device information struct
+ *
+ * ena_remove is called by the PCI subsystem to alert the driver
+ * that it should release a PCI device.
+ */
+static void ena_remove(struct pci_dev *pdev)
+{
+       struct ena_adapter *adapter = pci_get_drvdata(pdev);
+       struct ena_com_dev *ena_dev;
+       struct net_device *netdev;
+
+       if (!adapter)
+               /* This device didn't load properly and it's resources
+                * already released, nothing to do
+                */
+               return;
+
+       ena_dev = adapter->ena_dev;
+       netdev = adapter->netdev;
+
+#ifdef CONFIG_RFS_ACCEL
+       if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
+               free_irq_cpu_rmap(netdev->rx_cpu_rmap);
+               netdev->rx_cpu_rmap = NULL;
+       }
+#endif /* CONFIG_RFS_ACCEL */
+
+       unregister_netdev(netdev);
+       del_timer_sync(&adapter->timer_service);
+
+       cancel_work_sync(&adapter->reset_task);
+
+       cancel_work_sync(&adapter->suspend_io_task);
+
+       cancel_work_sync(&adapter->resume_io_task);
+
+       ena_com_dev_reset(ena_dev);
+
+       ena_free_mgmnt_irq(adapter);
+
+       ena_disable_msix(adapter);
+
+       free_netdev(netdev);
+
+       ena_com_mmio_reg_read_request_destroy(ena_dev);
+
+       ena_com_abort_admin_commands(ena_dev);
+
+       ena_com_wait_for_abort_completion(ena_dev);
+
+       ena_com_admin_destroy(ena_dev);
+
+       ena_com_rss_destroy(ena_dev);
+
+       ena_com_delete_debug_area(ena_dev);
+
+       ena_com_delete_host_info(ena_dev);
+
+       ena_release_bars(ena_dev, pdev);
+
+       pci_set_drvdata(pdev, NULL);
+
+       pci_disable_device(pdev);
+
+       ena_com_destroy_interrupt_moderation(ena_dev);
+
+       vfree(ena_dev);
+}
+
+static struct pci_driver ena_pci_driver = {
+       .name           = DRV_MODULE_NAME,
+       .id_table       = ena_pci_tbl,
+       .probe          = ena_probe,
+       .remove         = ena_remove,
+       .sriov_configure = ena_sriov_configure,
+};
+
+static int __init ena_init(void)
+{
+       pr_info("%s", version);
+
+       ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
+       if (!ena_wq) {
+               pr_err("Failed to create workqueue\n");
+               return -ENOMEM;
+       }
+
+       return pci_register_driver(&ena_pci_driver);
+}
+
+static void __exit ena_cleanup(void)
+{
+       pci_unregister_driver(&ena_pci_driver);
+
+       if (ena_wq) {
+               destroy_workqueue(ena_wq);
+               ena_wq = NULL;
+       }
+}
+
+/******************************************************************************
+ ******************************** AENQ Handlers *******************************
+ *****************************************************************************/
+/* ena_update_on_link_change:
+ * Notify the network interface about the change in link status
+ */
+static void ena_update_on_link_change(void *adapter_data,
+                                     struct ena_admin_aenq_entry *aenq_e)
+{
+       struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
+       struct ena_admin_aenq_link_change_desc *aenq_desc =
+               (struct ena_admin_aenq_link_change_desc *)aenq_e;
+       int status = aenq_desc->flags &
+               ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
+
+       if (status) {
+               netdev_dbg(adapter->netdev, "%s\n", __func__);
+               set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
+               netif_carrier_on(adapter->netdev);
+       } else {
+               clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
+               netif_carrier_off(adapter->netdev);
+       }
+}
+
+static void ena_keep_alive_wd(void *adapter_data,
+                             struct ena_admin_aenq_entry *aenq_e)
+{
+       struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
+
+       adapter->last_keep_alive_jiffies = jiffies;
+}
+
+static void ena_notification(void *adapter_data,
+                            struct ena_admin_aenq_entry *aenq_e)
+{
+       struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
+
+       WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
+            "Invalid group(%x) expected %x\n",
+            aenq_e->aenq_common_desc.group,
+            ENA_ADMIN_NOTIFICATION);
+
+       switch (aenq_e->aenq_common_desc.syndrom) {
+       case ENA_ADMIN_SUSPEND:
+               /* Suspend just the IO queues.
+                * We deliberately don't suspend admin so the timer and
+                * the keep_alive events should remain.
+                */
+               queue_work(ena_wq, &adapter->suspend_io_task);
+               break;
+       case ENA_ADMIN_RESUME:
+               queue_work(ena_wq, &adapter->resume_io_task);
+               break;
+       default:
+               netif_err(adapter, drv, adapter->netdev,
+                         "Invalid aenq notification link state %d\n",
+                         aenq_e->aenq_common_desc.syndrom);
+       }
+}
+
+/* This handler will called for unknown event group or unimplemented handlers*/
+static void unimplemented_aenq_handler(void *data,
+                                      struct ena_admin_aenq_entry *aenq_e)
+{
+       struct ena_adapter *adapter = (struct ena_adapter *)data;
+
+       netif_err(adapter, drv, adapter->netdev,
+                 "Unknown event was received or event with unimplemented handler\n");
+}
+
+static struct ena_aenq_handlers aenq_handlers = {
+       .handlers = {
+               [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
+               [ENA_ADMIN_NOTIFICATION] = ena_notification,
+               [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
+       },
+       .unimplemented_handler = unimplemented_aenq_handler
+};
+
+module_init(ena_init);
+module_exit(ena_cleanup);
diff --git a/drivers/net/ethernet/amazon/ena/ena_netdev.h b/drivers/net/ethernet/amazon/ena/ena_netdev.h
new file mode 100644 (file)
index 0000000..69d7e9e
--- /dev/null
@@ -0,0 +1,324 @@
+/*
+ * Copyright 2015 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef ENA_H
+#define ENA_H
+
+#include <linux/bitops.h>
+#include <linux/etherdevice.h>
+#include <linux/inetdevice.h>
+#include <linux/interrupt.h>
+#include <linux/netdevice.h>
+#include <linux/skbuff.h>
+
+#include "ena_com.h"
+#include "ena_eth_com.h"
+
+#define DRV_MODULE_VER_MAJOR   1
+#define DRV_MODULE_VER_MINOR   0
+#define DRV_MODULE_VER_SUBMINOR 2
+
+#define DRV_MODULE_NAME                "ena"
+#ifndef DRV_MODULE_VERSION
+#define DRV_MODULE_VERSION \
+       __stringify(DRV_MODULE_VER_MAJOR) "."   \
+       __stringify(DRV_MODULE_VER_MINOR) "."   \
+       __stringify(DRV_MODULE_VER_SUBMINOR)
+#endif
+
+#define DEVICE_NAME    "Elastic Network Adapter (ENA)"
+
+/* 1 for AENQ + ADMIN */
+#define ENA_MAX_MSIX_VEC(io_queues)    (1 + (io_queues))
+
+#define ENA_REG_BAR                    0
+#define ENA_MEM_BAR                    2
+#define ENA_BAR_MASK (BIT(ENA_REG_BAR) | BIT(ENA_MEM_BAR))
+
+#define ENA_DEFAULT_RING_SIZE  (1024)
+
+#define ENA_TX_WAKEUP_THRESH           (MAX_SKB_FRAGS + 2)
+#define ENA_DEFAULT_RX_COPYBREAK       (128 - NET_IP_ALIGN)
+
+/* limit the buffer size to 600 bytes to handle MTU changes from very
+ * small to very large, in which case the number of buffers per packet
+ * could exceed ENA_PKT_MAX_BUFS
+ */
+#define ENA_DEFAULT_MIN_RX_BUFF_ALLOC_SIZE 600
+
+#define ENA_MIN_MTU            128
+
+#define ENA_NAME_MAX_LEN       20
+#define ENA_IRQNAME_SIZE       40
+
+#define ENA_PKT_MAX_BUFS       19
+
+#define ENA_RX_RSS_TABLE_LOG_SIZE  7
+#define ENA_RX_RSS_TABLE_SIZE  (1 << ENA_RX_RSS_TABLE_LOG_SIZE)
+
+#define ENA_HASH_KEY_SIZE      40
+
+/* The number of tx packet completions that will be handled each NAPI poll
+ * cycle is ring_size / ENA_TX_POLL_BUDGET_DIVIDER.
+ */
+#define ENA_TX_POLL_BUDGET_DIVIDER     4
+
+/* Refill Rx queue when number of available descriptors is below
+ * QUEUE_SIZE / ENA_RX_REFILL_THRESH_DIVIDER
+ */
+#define ENA_RX_REFILL_THRESH_DIVIDER   8
+
+/* Number of queues to check for missing queues per timer service */
+#define ENA_MONITORED_TX_QUEUES        4
+/* Max timeout packets before device reset */
+#define MAX_NUM_OF_TIMEOUTED_PACKETS 32
+
+#define ENA_TX_RING_IDX_NEXT(idx, ring_size) (((idx) + 1) & ((ring_size) - 1))
+
+#define ENA_RX_RING_IDX_NEXT(idx, ring_size) (((idx) + 1) & ((ring_size) - 1))
+#define ENA_RX_RING_IDX_ADD(idx, n, ring_size) \
+       (((idx) + (n)) & ((ring_size) - 1))
+
+#define ENA_IO_TXQ_IDX(q)      (2 * (q))
+#define ENA_IO_RXQ_IDX(q)      (2 * (q) + 1)
+
+#define ENA_MGMNT_IRQ_IDX              0
+#define ENA_IO_IRQ_FIRST_IDX           1
+#define ENA_IO_IRQ_IDX(q)              (ENA_IO_IRQ_FIRST_IDX + (q))
+
+/* ENA device should send keep alive msg every 1 sec.
+ * We wait for 3 sec just to be on the safe side.
+ */
+#define ENA_DEVICE_KALIVE_TIMEOUT      (3 * HZ)
+
+#define ENA_MMIO_DISABLE_REG_READ      BIT(0)
+
+struct ena_irq {
+       irq_handler_t handler;
+       void *data;
+       int cpu;
+       u32 vector;
+       cpumask_t affinity_hint_mask;
+       char name[ENA_IRQNAME_SIZE];
+};
+
+struct ena_napi {
+       struct napi_struct napi ____cacheline_aligned;
+       struct ena_ring *tx_ring;
+       struct ena_ring *rx_ring;
+       u32 qid;
+};
+
+struct ena_tx_buffer {
+       struct sk_buff *skb;
+       /* num of ena desc for this specific skb
+        * (includes data desc and metadata desc)
+        */
+       u32 tx_descs;
+       /* num of buffers used by this skb */
+       u32 num_of_bufs;
+       /* Save the last jiffies to detect missing tx packets */
+       unsigned long last_jiffies;
+       struct ena_com_buf bufs[ENA_PKT_MAX_BUFS];
+} ____cacheline_aligned;
+
+struct ena_rx_buffer {
+       struct sk_buff *skb;
+       struct page *page;
+       u32 page_offset;
+       struct ena_com_buf ena_buf;
+} ____cacheline_aligned;
+
+struct ena_stats_tx {
+       u64 cnt;
+       u64 bytes;
+       u64 queue_stop;
+       u64 prepare_ctx_err;
+       u64 queue_wakeup;
+       u64 dma_mapping_err;
+       u64 linearize;
+       u64 linearize_failed;
+       u64 napi_comp;
+       u64 tx_poll;
+       u64 doorbells;
+       u64 missing_tx_comp;
+       u64 bad_req_id;
+};
+
+struct ena_stats_rx {
+       u64 cnt;
+       u64 bytes;
+       u64 refil_partial;
+       u64 bad_csum;
+       u64 page_alloc_fail;
+       u64 skb_alloc_fail;
+       u64 dma_mapping_err;
+       u64 bad_desc_num;
+       u64 rx_copybreak_pkt;
+};
+
+struct ena_ring {
+       /* Holds the empty requests for TX out of order completions */
+       u16 *free_tx_ids;
+       union {
+               struct ena_tx_buffer *tx_buffer_info;
+               struct ena_rx_buffer *rx_buffer_info;
+       };
+
+       /* cache ptr to avoid using the adapter */
+       struct device *dev;
+       struct pci_dev *pdev;
+       struct napi_struct *napi;
+       struct net_device *netdev;
+       struct ena_com_dev *ena_dev;
+       struct ena_adapter *adapter;
+       struct ena_com_io_cq *ena_com_io_cq;
+       struct ena_com_io_sq *ena_com_io_sq;
+
+       u16 next_to_use;
+       u16 next_to_clean;
+       u16 rx_copybreak;
+       u16 qid;
+       u16 mtu;
+       u16 sgl_size;
+
+       /* The maximum header length the device can handle */
+       u8 tx_max_header_size;
+
+       /* cpu for TPH */
+       int cpu;
+        /* number of tx/rx_buffer_info's entries */
+       int ring_size;
+
+       enum ena_admin_placement_policy_type tx_mem_queue_type;
+
+       struct ena_com_rx_buf_info ena_bufs[ENA_PKT_MAX_BUFS];
+       u32  smoothed_interval;
+       u32  per_napi_packets;
+       u32  per_napi_bytes;
+       enum ena_intr_moder_level moder_tbl_idx;
+       struct u64_stats_sync syncp;
+       union {
+               struct ena_stats_tx tx_stats;
+               struct ena_stats_rx rx_stats;
+       };
+} ____cacheline_aligned;
+
+struct ena_stats_dev {
+       u64 tx_timeout;
+       u64 io_suspend;
+       u64 io_resume;
+       u64 wd_expired;
+       u64 interface_up;
+       u64 interface_down;
+       u64 admin_q_pause;
+};
+
+enum ena_flags_t {
+       ENA_FLAG_DEVICE_RUNNING,
+       ENA_FLAG_DEV_UP,
+       ENA_FLAG_LINK_UP,
+       ENA_FLAG_MSIX_ENABLED,
+       ENA_FLAG_TRIGGER_RESET
+};
+
+/* adapter specific private data structure */
+struct ena_adapter {
+       struct ena_com_dev *ena_dev;
+       /* OS defined structs */
+       struct net_device *netdev;
+       struct pci_dev *pdev;
+
+       /* rx packets that shorter that this len will be copied to the skb
+        * header
+        */
+       u32 rx_copybreak;
+       u32 max_mtu;
+
+       int num_queues;
+
+       struct msix_entry *msix_entries;
+       int msix_vecs;
+
+       u32 tx_usecs, rx_usecs; /* interrupt moderation */
+       u32 tx_frames, rx_frames; /* interrupt moderation */
+
+       u32 tx_ring_size;
+       u32 rx_ring_size;
+
+       u32 msg_enable;
+
+       u16 max_tx_sgl_size;
+       u16 max_rx_sgl_size;
+
+       u8 mac_addr[ETH_ALEN];
+
+       char name[ENA_NAME_MAX_LEN];
+
+       unsigned long flags;
+       /* TX */
+       struct ena_ring tx_ring[ENA_MAX_NUM_IO_QUEUES]
+               ____cacheline_aligned_in_smp;
+
+       /* RX */
+       struct ena_ring rx_ring[ENA_MAX_NUM_IO_QUEUES]
+               ____cacheline_aligned_in_smp;
+
+       struct ena_napi ena_napi[ENA_MAX_NUM_IO_QUEUES];
+
+       struct ena_irq irq_tbl[ENA_MAX_MSIX_VEC(ENA_MAX_NUM_IO_QUEUES)];
+
+       /* timer service */
+       struct work_struct reset_task;
+       struct work_struct suspend_io_task;
+       struct work_struct resume_io_task;
+       struct timer_list timer_service;
+
+       bool wd_state;
+       unsigned long last_keep_alive_jiffies;
+
+       struct u64_stats_sync syncp;
+       struct ena_stats_dev dev_stats;
+
+       /* last queue index that was checked for uncompleted tx packets */
+       u32 last_monitored_tx_qid;
+};
+
+void ena_set_ethtool_ops(struct net_device *netdev);
+
+void ena_dump_stats_to_dmesg(struct ena_adapter *adapter);
+
+void ena_dump_stats_to_buf(struct ena_adapter *adapter, u8 *buf);
+
+int ena_get_sset_count(struct net_device *netdev, int sset);
+
+#endif /* !(ENA_H) */
diff --git a/drivers/net/ethernet/amazon/ena/ena_pci_id_tbl.h b/drivers/net/ethernet/amazon/ena/ena_pci_id_tbl.h
new file mode 100644 (file)
index 0000000..f80d2a4
--- /dev/null
@@ -0,0 +1,67 @@
+/*
+ * Copyright 2015 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef ENA_PCI_ID_TBL_H_
+#define ENA_PCI_ID_TBL_H_
+
+#ifndef PCI_VENDOR_ID_AMAZON
+#define PCI_VENDOR_ID_AMAZON 0x1d0f
+#endif
+
+#ifndef PCI_DEV_ID_ENA_PF
+#define PCI_DEV_ID_ENA_PF      0x0ec2
+#endif
+
+#ifndef PCI_DEV_ID_ENA_LLQ_PF
+#define PCI_DEV_ID_ENA_LLQ_PF  0x1ec2
+#endif
+
+#ifndef PCI_DEV_ID_ENA_VF
+#define PCI_DEV_ID_ENA_VF      0xec20
+#endif
+
+#ifndef PCI_DEV_ID_ENA_LLQ_VF
+#define PCI_DEV_ID_ENA_LLQ_VF  0xec21
+#endif
+
+#define ENA_PCI_ID_TABLE_ENTRY(devid) \
+       {PCI_DEVICE(PCI_VENDOR_ID_AMAZON, devid)},
+
+static const struct pci_device_id ena_pci_tbl[] = {
+       ENA_PCI_ID_TABLE_ENTRY(PCI_DEV_ID_ENA_PF)
+       ENA_PCI_ID_TABLE_ENTRY(PCI_DEV_ID_ENA_LLQ_PF)
+       ENA_PCI_ID_TABLE_ENTRY(PCI_DEV_ID_ENA_VF)
+       ENA_PCI_ID_TABLE_ENTRY(PCI_DEV_ID_ENA_LLQ_VF)
+       { }
+};
+
+#endif /* ENA_PCI_ID_TBL_H_ */
diff --git a/drivers/net/ethernet/amazon/ena/ena_regs_defs.h b/drivers/net/ethernet/amazon/ena/ena_regs_defs.h
new file mode 100644 (file)
index 0000000..26097a2
--- /dev/null
@@ -0,0 +1,133 @@
+/*
+ * Copyright 2015 - 2016 Amazon.com, Inc. or its affiliates.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _ENA_REGS_H_
+#define _ENA_REGS_H_
+
+/* ena_registers offsets */
+#define ENA_REGS_VERSION_OFF           0x0
+#define ENA_REGS_CONTROLLER_VERSION_OFF                0x4
+#define ENA_REGS_CAPS_OFF              0x8
+#define ENA_REGS_CAPS_EXT_OFF          0xc
+#define ENA_REGS_AQ_BASE_LO_OFF                0x10
+#define ENA_REGS_AQ_BASE_HI_OFF                0x14
+#define ENA_REGS_AQ_CAPS_OFF           0x18
+#define ENA_REGS_ACQ_BASE_LO_OFF               0x20
+#define ENA_REGS_ACQ_BASE_HI_OFF               0x24
+#define ENA_REGS_ACQ_CAPS_OFF          0x28
+#define ENA_REGS_AQ_DB_OFF             0x2c
+#define ENA_REGS_ACQ_TAIL_OFF          0x30
+#define ENA_REGS_AENQ_CAPS_OFF         0x34
+#define ENA_REGS_AENQ_BASE_LO_OFF              0x38
+#define ENA_REGS_AENQ_BASE_HI_OFF              0x3c
+#define ENA_REGS_AENQ_HEAD_DB_OFF              0x40
+#define ENA_REGS_AENQ_TAIL_OFF         0x44
+#define ENA_REGS_INTR_MASK_OFF         0x4c
+#define ENA_REGS_DEV_CTL_OFF           0x54
+#define ENA_REGS_DEV_STS_OFF           0x58
+#define ENA_REGS_MMIO_REG_READ_OFF             0x5c
+#define ENA_REGS_MMIO_RESP_LO_OFF              0x60
+#define ENA_REGS_MMIO_RESP_HI_OFF              0x64
+#define ENA_REGS_RSS_IND_ENTRY_UPDATE_OFF              0x68
+
+/* version register */
+#define ENA_REGS_VERSION_MINOR_VERSION_MASK            0xff
+#define ENA_REGS_VERSION_MAJOR_VERSION_SHIFT           8
+#define ENA_REGS_VERSION_MAJOR_VERSION_MASK            0xff00
+
+/* controller_version register */
+#define ENA_REGS_CONTROLLER_VERSION_SUBMINOR_VERSION_MASK              0xff
+#define ENA_REGS_CONTROLLER_VERSION_MINOR_VERSION_SHIFT                8
+#define ENA_REGS_CONTROLLER_VERSION_MINOR_VERSION_MASK         0xff00
+#define ENA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_SHIFT                16
+#define ENA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_MASK         0xff0000
+#define ENA_REGS_CONTROLLER_VERSION_IMPL_ID_SHIFT              24
+#define ENA_REGS_CONTROLLER_VERSION_IMPL_ID_MASK               0xff000000
+
+/* caps register */
+#define ENA_REGS_CAPS_CONTIGUOUS_QUEUE_REQUIRED_MASK           0x1
+#define ENA_REGS_CAPS_RESET_TIMEOUT_SHIFT              1
+#define ENA_REGS_CAPS_RESET_TIMEOUT_MASK               0x3e
+#define ENA_REGS_CAPS_DMA_ADDR_WIDTH_SHIFT             8
+#define ENA_REGS_CAPS_DMA_ADDR_WIDTH_MASK              0xff00
+
+/* aq_caps register */
+#define ENA_REGS_AQ_CAPS_AQ_DEPTH_MASK         0xffff
+#define ENA_REGS_AQ_CAPS_AQ_ENTRY_SIZE_SHIFT           16
+#define ENA_REGS_AQ_CAPS_AQ_ENTRY_SIZE_MASK            0xffff0000
+
+/* acq_caps register */
+#define ENA_REGS_ACQ_CAPS_ACQ_DEPTH_MASK               0xffff
+#define ENA_REGS_ACQ_CAPS_ACQ_ENTRY_SIZE_SHIFT         16
+#define ENA_REGS_ACQ_CAPS_ACQ_ENTRY_SIZE_MASK          0xffff0000
+
+/* aenq_caps register */
+#define ENA_REGS_AENQ_CAPS_AENQ_DEPTH_MASK             0xffff
+#define ENA_REGS_AENQ_CAPS_AENQ_ENTRY_SIZE_SHIFT               16
+#define ENA_REGS_AENQ_CAPS_AENQ_ENTRY_SIZE_MASK                0xffff0000
+
+/* dev_ctl register */
+#define ENA_REGS_DEV_CTL_DEV_RESET_MASK                0x1
+#define ENA_REGS_DEV_CTL_AQ_RESTART_SHIFT              1
+#define ENA_REGS_DEV_CTL_AQ_RESTART_MASK               0x2
+#define ENA_REGS_DEV_CTL_QUIESCENT_SHIFT               2
+#define ENA_REGS_DEV_CTL_QUIESCENT_MASK                0x4
+#define ENA_REGS_DEV_CTL_IO_RESUME_SHIFT               3
+#define ENA_REGS_DEV_CTL_IO_RESUME_MASK                0x8
+
+/* dev_sts register */
+#define ENA_REGS_DEV_STS_READY_MASK            0x1
+#define ENA_REGS_DEV_STS_AQ_RESTART_IN_PROGRESS_SHIFT          1
+#define ENA_REGS_DEV_STS_AQ_RESTART_IN_PROGRESS_MASK           0x2
+#define ENA_REGS_DEV_STS_AQ_RESTART_FINISHED_SHIFT             2
+#define ENA_REGS_DEV_STS_AQ_RESTART_FINISHED_MASK              0x4
+#define ENA_REGS_DEV_STS_RESET_IN_PROGRESS_SHIFT               3
+#define ENA_REGS_DEV_STS_RESET_IN_PROGRESS_MASK                0x8
+#define ENA_REGS_DEV_STS_RESET_FINISHED_SHIFT          4
+#define ENA_REGS_DEV_STS_RESET_FINISHED_MASK           0x10
+#define ENA_REGS_DEV_STS_FATAL_ERROR_SHIFT             5
+#define ENA_REGS_DEV_STS_FATAL_ERROR_MASK              0x20
+#define ENA_REGS_DEV_STS_QUIESCENT_STATE_IN_PROGRESS_SHIFT             6
+#define ENA_REGS_DEV_STS_QUIESCENT_STATE_IN_PROGRESS_MASK              0x40
+#define ENA_REGS_DEV_STS_QUIESCENT_STATE_ACHIEVED_SHIFT                7
+#define ENA_REGS_DEV_STS_QUIESCENT_STATE_ACHIEVED_MASK         0x80
+
+/* mmio_reg_read register */
+#define ENA_REGS_MMIO_REG_READ_REQ_ID_MASK             0xffff
+#define ENA_REGS_MMIO_REG_READ_REG_OFF_SHIFT           16
+#define ENA_REGS_MMIO_REG_READ_REG_OFF_MASK            0xffff0000
+
+/* rss_ind_entry_update register */
+#define ENA_REGS_RSS_IND_ENTRY_UPDATE_INDEX_MASK               0xffff
+#define ENA_REGS_RSS_IND_ENTRY_UPDATE_CQ_IDX_SHIFT             16
+#define ENA_REGS_RSS_IND_ENTRY_UPDATE_CQ_IDX_MASK              0xffff0000
+
+#endif /*_ENA_REGS_H_ */