Merge branches 'for-4.8/upstream-fixes', 'for-4.9/alps', 'for-4.9/hid-input', 'for...
[cascardo/linux.git] / drivers / hid / intel-ish-hid / ishtp / client.c
diff --git a/drivers/hid/intel-ish-hid/ishtp/client.c b/drivers/hid/intel-ish-hid/ishtp/client.c
new file mode 100644 (file)
index 0000000..aad6132
--- /dev/null
@@ -0,0 +1,1054 @@
+/*
+ * ISHTP client logic
+ *
+ * Copyright (c) 2003-2016, Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ */
+
+#include <linux/slab.h>
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include "hbm.h"
+#include "client.h"
+
+/**
+ * ishtp_read_list_flush() - Flush read queue
+ * @cl: ishtp client instance
+ *
+ * Used to remove all entries from read queue for a client
+ */
+static void ishtp_read_list_flush(struct ishtp_cl *cl)
+{
+       struct ishtp_cl_rb *rb;
+       struct ishtp_cl_rb *next;
+       unsigned long   flags;
+
+       spin_lock_irqsave(&cl->dev->read_list_spinlock, flags);
+       list_for_each_entry_safe(rb, next, &cl->dev->read_list.list, list)
+               if (rb->cl && ishtp_cl_cmp_id(cl, rb->cl)) {
+                       list_del(&rb->list);
+                       ishtp_io_rb_free(rb);
+               }
+       spin_unlock_irqrestore(&cl->dev->read_list_spinlock, flags);
+}
+
+/**
+ * ishtp_cl_flush_queues() - Flush all queues for a client
+ * @cl: ishtp client instance
+ *
+ * Used to remove all queues for a client. This is called when a client device
+ * needs reset due to error, S3 resume or during module removal
+ *
+ * Return: 0 on success else -EINVAL if device is NULL
+ */
+int ishtp_cl_flush_queues(struct ishtp_cl *cl)
+{
+       if (WARN_ON(!cl || !cl->dev))
+               return -EINVAL;
+
+       ishtp_read_list_flush(cl);
+
+       return 0;
+}
+EXPORT_SYMBOL(ishtp_cl_flush_queues);
+
+/**
+ * ishtp_cl_init() - Initialize all fields of a client device
+ * @cl: ishtp client instance
+ * @dev: ishtp device
+ *
+ * Initializes a client device fields: Init spinlocks, init queues etc.
+ * This function is called during new client creation
+ */
+static void ishtp_cl_init(struct ishtp_cl *cl, struct ishtp_device *dev)
+{
+       memset(cl, 0, sizeof(struct ishtp_cl));
+       init_waitqueue_head(&cl->wait_ctrl_res);
+       spin_lock_init(&cl->free_list_spinlock);
+       spin_lock_init(&cl->in_process_spinlock);
+       spin_lock_init(&cl->tx_list_spinlock);
+       spin_lock_init(&cl->tx_free_list_spinlock);
+       spin_lock_init(&cl->fc_spinlock);
+       INIT_LIST_HEAD(&cl->link);
+       cl->dev = dev;
+
+       INIT_LIST_HEAD(&cl->free_rb_list.list);
+       INIT_LIST_HEAD(&cl->tx_list.list);
+       INIT_LIST_HEAD(&cl->tx_free_list.list);
+       INIT_LIST_HEAD(&cl->in_process_list.list);
+
+       cl->rx_ring_size = CL_DEF_RX_RING_SIZE;
+       cl->tx_ring_size = CL_DEF_TX_RING_SIZE;
+
+       /* dma */
+       cl->last_tx_path = CL_TX_PATH_IPC;
+       cl->last_dma_acked = 1;
+       cl->last_dma_addr = NULL;
+       cl->last_ipc_acked = 1;
+}
+
+/**
+ * ishtp_cl_allocate() - allocates client structure and sets it up.
+ * @dev: ishtp device
+ *
+ * Allocate memory for new client device and call to initialize each field.
+ *
+ * Return: The allocated client instance or NULL on failure
+ */
+struct ishtp_cl *ishtp_cl_allocate(struct ishtp_device *dev)
+{
+       struct ishtp_cl *cl;
+
+       cl = kmalloc(sizeof(struct ishtp_cl), GFP_KERNEL);
+       if (!cl)
+               return NULL;
+
+       ishtp_cl_init(cl, dev);
+       return cl;
+}
+EXPORT_SYMBOL(ishtp_cl_allocate);
+
+/**
+ * ishtp_cl_free() - Frees a client device
+ * @cl: client device instance
+ *
+ * Frees a client device
+ */
+void   ishtp_cl_free(struct ishtp_cl *cl)
+{
+       struct ishtp_device *dev;
+       unsigned long flags;
+
+       if (!cl)
+               return;
+
+       dev = cl->dev;
+       if (!dev)
+               return;
+
+       spin_lock_irqsave(&dev->cl_list_lock, flags);
+       ishtp_cl_free_rx_ring(cl);
+       ishtp_cl_free_tx_ring(cl);
+       kfree(cl);
+       spin_unlock_irqrestore(&dev->cl_list_lock, flags);
+}
+EXPORT_SYMBOL(ishtp_cl_free);
+
+/**
+ * ishtp_cl_link() - Reserve a host id and link the client instance
+ * @cl: client device instance
+ * @id: host client id to use. It can be ISHTP_HOST_CLIENT_ID_ANY if any
+ *     id from the available can be used
+ *
+ *
+ * This allocates a single bit in the hostmap. This function will make sure
+ * that not many client sessions are opened at the same time. Once allocated
+ * the client device instance is added to the ishtp device in the current
+ * client list
+ *
+ * Return: 0 or error code on failure
+ */
+int ishtp_cl_link(struct ishtp_cl *cl, int id)
+{
+       struct ishtp_device *dev;
+       unsigned long   flags, flags_cl;
+       int     ret = 0;
+
+       if (WARN_ON(!cl || !cl->dev))
+               return -EINVAL;
+
+       dev = cl->dev;
+
+       spin_lock_irqsave(&dev->device_lock, flags);
+
+       if (dev->open_handle_count >= ISHTP_MAX_OPEN_HANDLE_COUNT) {
+               ret = -EMFILE;
+               goto unlock_dev;
+       }
+
+       /* If Id is not assigned get one*/
+       if (id == ISHTP_HOST_CLIENT_ID_ANY)
+               id = find_first_zero_bit(dev->host_clients_map,
+                       ISHTP_CLIENTS_MAX);
+
+       if (id >= ISHTP_CLIENTS_MAX) {
+               spin_unlock_irqrestore(&dev->device_lock, flags);
+               dev_err(&cl->device->dev, "id exceeded %d", ISHTP_CLIENTS_MAX);
+               return -ENOENT;
+       }
+
+       dev->open_handle_count++;
+       cl->host_client_id = id;
+       spin_lock_irqsave(&dev->cl_list_lock, flags_cl);
+       if (dev->dev_state != ISHTP_DEV_ENABLED) {
+               ret = -ENODEV;
+               goto unlock_cl;
+       }
+       list_add_tail(&cl->link, &dev->cl_list);
+       set_bit(id, dev->host_clients_map);
+       cl->state = ISHTP_CL_INITIALIZING;
+
+unlock_cl:
+       spin_unlock_irqrestore(&dev->cl_list_lock, flags_cl);
+unlock_dev:
+       spin_unlock_irqrestore(&dev->device_lock, flags);
+       return ret;
+}
+EXPORT_SYMBOL(ishtp_cl_link);
+
+/**
+ * ishtp_cl_unlink() - remove fw_cl from the client device list
+ * @cl: client device instance
+ *
+ * Remove a previously linked device to a ishtp device
+ */
+void ishtp_cl_unlink(struct ishtp_cl *cl)
+{
+       struct ishtp_device *dev;
+       struct ishtp_cl *pos;
+       unsigned long   flags;
+
+       /* don't shout on error exit path */
+       if (!cl || !cl->dev)
+               return;
+
+       dev = cl->dev;
+
+       spin_lock_irqsave(&dev->device_lock, flags);
+       if (dev->open_handle_count > 0) {
+               clear_bit(cl->host_client_id, dev->host_clients_map);
+               dev->open_handle_count--;
+       }
+       spin_unlock_irqrestore(&dev->device_lock, flags);
+
+       /*
+        * This checks that 'cl' is actually linked into device's structure,
+        * before attempting 'list_del'
+        */
+       spin_lock_irqsave(&dev->cl_list_lock, flags);
+       list_for_each_entry(pos, &dev->cl_list, link)
+               if (cl->host_client_id == pos->host_client_id) {
+                       list_del_init(&pos->link);
+                       break;
+               }
+       spin_unlock_irqrestore(&dev->cl_list_lock, flags);
+}
+EXPORT_SYMBOL(ishtp_cl_unlink);
+
+/**
+ * ishtp_cl_disconnect() - Send disconnect request to firmware
+ * @cl: client device instance
+ *
+ * Send a disconnect request for a client to firmware.
+ *
+ * Return: 0 if successful disconnect response from the firmware or error
+ * code on failure
+ */
+int ishtp_cl_disconnect(struct ishtp_cl *cl)
+{
+       struct ishtp_device *dev;
+       int err;
+
+       if (WARN_ON(!cl || !cl->dev))
+               return -ENODEV;
+
+       dev = cl->dev;
+
+       dev->print_log(dev, "%s() state %d\n", __func__, cl->state);
+
+       if (cl->state != ISHTP_CL_DISCONNECTING) {
+               dev->print_log(dev, "%s() Disconnect in progress\n", __func__);
+               return 0;
+       }
+
+       if (ishtp_hbm_cl_disconnect_req(dev, cl)) {
+               dev->print_log(dev, "%s() Failed to disconnect\n", __func__);
+               dev_err(&cl->device->dev, "failed to disconnect.\n");
+               return -ENODEV;
+       }
+
+       err = wait_event_interruptible_timeout(cl->wait_ctrl_res,
+                       (dev->dev_state != ISHTP_DEV_ENABLED ||
+                       cl->state == ISHTP_CL_DISCONNECTED),
+                       ishtp_secs_to_jiffies(ISHTP_CL_CONNECT_TIMEOUT));
+
+       /*
+        * If FW reset arrived, this will happen. Don't check cl->,
+        * as 'cl' may be freed already
+        */
+       if (dev->dev_state != ISHTP_DEV_ENABLED) {
+               dev->print_log(dev, "%s() dev_state != ISHTP_DEV_ENABLED\n",
+                              __func__);
+               return -ENODEV;
+       }
+
+       if (cl->state == ISHTP_CL_DISCONNECTED) {
+               dev->print_log(dev, "%s() successful\n", __func__);
+               return 0;
+       }
+
+       return -ENODEV;
+}
+EXPORT_SYMBOL(ishtp_cl_disconnect);
+
+/**
+ * ishtp_cl_is_other_connecting() - Check other client is connecting
+ * @cl: client device instance
+ *
+ * Checks if other client with the same fw client id is connecting
+ *
+ * Return: true if other client is connected else false
+ */
+static bool ishtp_cl_is_other_connecting(struct ishtp_cl *cl)
+{
+       struct ishtp_device *dev;
+       struct ishtp_cl *pos;
+       unsigned long   flags;
+
+       if (WARN_ON(!cl || !cl->dev))
+               return false;
+
+       dev = cl->dev;
+       spin_lock_irqsave(&dev->cl_list_lock, flags);
+       list_for_each_entry(pos, &dev->cl_list, link) {
+               if ((pos->state == ISHTP_CL_CONNECTING) && (pos != cl) &&
+                               cl->fw_client_id == pos->fw_client_id) {
+                       spin_unlock_irqrestore(&dev->cl_list_lock, flags);
+                       return true;
+               }
+       }
+       spin_unlock_irqrestore(&dev->cl_list_lock, flags);
+
+       return false;
+}
+
+/**
+ * ishtp_cl_connect() - Send connect request to firmware
+ * @cl: client device instance
+ *
+ * Send a connect request for a client to firmware. If successful it will
+ * RX and TX ring buffers
+ *
+ * Return: 0 if successful connect response from the firmware and able
+ * to bind and allocate ring buffers or error code on failure
+ */
+int ishtp_cl_connect(struct ishtp_cl *cl)
+{
+       struct ishtp_device *dev;
+       int rets;
+
+       if (WARN_ON(!cl || !cl->dev))
+               return -ENODEV;
+
+       dev = cl->dev;
+
+       dev->print_log(dev, "%s() current_state = %d\n", __func__, cl->state);
+
+       if (ishtp_cl_is_other_connecting(cl)) {
+               dev->print_log(dev, "%s() Busy\n", __func__);
+               return  -EBUSY;
+       }
+
+       if (ishtp_hbm_cl_connect_req(dev, cl)) {
+               dev->print_log(dev, "%s() HBM connect req fail\n", __func__);
+               return -ENODEV;
+       }
+
+       rets = wait_event_interruptible_timeout(cl->wait_ctrl_res,
+                               (dev->dev_state == ISHTP_DEV_ENABLED &&
+                               (cl->state == ISHTP_CL_CONNECTED ||
+                                cl->state == ISHTP_CL_DISCONNECTED)),
+                               ishtp_secs_to_jiffies(
+                                       ISHTP_CL_CONNECT_TIMEOUT));
+       /*
+        * If FW reset arrived, this will happen. Don't check cl->,
+        * as 'cl' may be freed already
+        */
+       if (dev->dev_state != ISHTP_DEV_ENABLED) {
+               dev->print_log(dev, "%s() dev_state != ISHTP_DEV_ENABLED\n",
+                              __func__);
+               return -EFAULT;
+       }
+
+       if (cl->state != ISHTP_CL_CONNECTED) {
+               dev->print_log(dev, "%s() state != ISHTP_CL_CONNECTED\n",
+                              __func__);
+               return -EFAULT;
+       }
+
+       rets = cl->status;
+       if (rets) {
+               dev->print_log(dev, "%s() Invalid status\n", __func__);
+               return rets;
+       }
+
+       rets = ishtp_cl_device_bind(cl);
+       if (rets) {
+               dev->print_log(dev, "%s() Bind error\n", __func__);
+               ishtp_cl_disconnect(cl);
+               return rets;
+       }
+
+       rets = ishtp_cl_alloc_rx_ring(cl);
+       if (rets) {
+               dev->print_log(dev, "%s() Alloc RX ring failed\n", __func__);
+               /* if failed allocation, disconnect */
+               ishtp_cl_disconnect(cl);
+               return rets;
+       }
+
+       rets = ishtp_cl_alloc_tx_ring(cl);
+       if (rets) {
+               dev->print_log(dev, "%s() Alloc TX ring failed\n", __func__);
+               /* if failed allocation, disconnect */
+               ishtp_cl_free_rx_ring(cl);
+               ishtp_cl_disconnect(cl);
+               return rets;
+       }
+
+       /* Upon successful connection and allocation, emit flow-control */
+       rets = ishtp_cl_read_start(cl);
+
+       dev->print_log(dev, "%s() successful\n", __func__);
+
+       return rets;
+}
+EXPORT_SYMBOL(ishtp_cl_connect);
+
+/**
+ * ishtp_cl_read_start() - Prepare to read client message
+ * @cl: client device instance
+ *
+ * Get a free buffer from pool of free read buffers and add to read buffer
+ * pool to add contents. Send a flow control request to firmware to be able
+ * send next message.
+ *
+ * Return: 0 if successful or error code on failure
+ */
+int ishtp_cl_read_start(struct ishtp_cl *cl)
+{
+       struct ishtp_device *dev;
+       struct ishtp_cl_rb *rb;
+       int rets;
+       int i;
+       unsigned long   flags;
+       unsigned long   dev_flags;
+
+       if (WARN_ON(!cl || !cl->dev))
+               return -ENODEV;
+
+       dev = cl->dev;
+
+       if (cl->state != ISHTP_CL_CONNECTED)
+               return -ENODEV;
+
+       if (dev->dev_state != ISHTP_DEV_ENABLED)
+               return -ENODEV;
+
+       i = ishtp_fw_cl_by_id(dev, cl->fw_client_id);
+       if (i < 0) {
+               dev_err(&cl->device->dev, "no such fw client %d\n",
+                       cl->fw_client_id);
+               return -ENODEV;
+       }
+
+       /* The current rb is the head of the free rb list */
+       spin_lock_irqsave(&cl->free_list_spinlock, flags);
+       if (list_empty(&cl->free_rb_list.list)) {
+               dev_warn(&cl->device->dev,
+                        "[ishtp-ish] Rx buffers pool is empty\n");
+               rets = -ENOMEM;
+               rb = NULL;
+               spin_unlock_irqrestore(&cl->free_list_spinlock, flags);
+               goto out;
+       }
+       rb = list_entry(cl->free_rb_list.list.next, struct ishtp_cl_rb, list);
+       list_del_init(&rb->list);
+       spin_unlock_irqrestore(&cl->free_list_spinlock, flags);
+
+       rb->cl = cl;
+       rb->buf_idx = 0;
+
+       INIT_LIST_HEAD(&rb->list);
+       rets = 0;
+
+       /*
+        * This must be BEFORE sending flow control -
+        * response in ISR may come too fast...
+        */
+       spin_lock_irqsave(&dev->read_list_spinlock, dev_flags);
+       list_add_tail(&rb->list, &dev->read_list.list);
+       spin_unlock_irqrestore(&dev->read_list_spinlock, dev_flags);
+       if (ishtp_hbm_cl_flow_control_req(dev, cl)) {
+               rets = -ENODEV;
+               goto out;
+       }
+out:
+       /* if ishtp_hbm_cl_flow_control_req failed, return rb to free list */
+       if (rets && rb) {
+               spin_lock_irqsave(&dev->read_list_spinlock, dev_flags);
+               list_del(&rb->list);
+               spin_unlock_irqrestore(&dev->read_list_spinlock, dev_flags);
+
+               spin_lock_irqsave(&cl->free_list_spinlock, flags);
+               list_add_tail(&rb->list, &cl->free_rb_list.list);
+               spin_unlock_irqrestore(&cl->free_list_spinlock, flags);
+       }
+       return rets;
+}
+
+/**
+ * ishtp_cl_send() - Send a message to firmware
+ * @cl: client device instance
+ * @buf: message buffer
+ * @length: length of message
+ *
+ * If the client is correct state to send message, this function gets a buffer
+ * from tx ring buffers, copy the message data and call to send the message
+ * using ishtp_cl_send_msg()
+ *
+ * Return: 0 if successful or error code on failure
+ */
+int ishtp_cl_send(struct ishtp_cl *cl, uint8_t *buf, size_t length)
+{
+       struct ishtp_device     *dev;
+       int     id;
+       struct ishtp_cl_tx_ring *cl_msg;
+       int     have_msg_to_send = 0;
+       unsigned long   tx_flags, tx_free_flags;
+
+       if (WARN_ON(!cl || !cl->dev))
+               return -ENODEV;
+
+       dev = cl->dev;
+
+       if (cl->state != ISHTP_CL_CONNECTED) {
+               ++cl->err_send_msg;
+               return -EPIPE;
+       }
+
+       if (dev->dev_state != ISHTP_DEV_ENABLED) {
+               ++cl->err_send_msg;
+               return -ENODEV;
+       }
+
+       /* Check if we have fw client device */
+       id = ishtp_fw_cl_by_id(dev, cl->fw_client_id);
+       if (id < 0) {
+               ++cl->err_send_msg;
+               return -ENOENT;
+       }
+
+       if (length > dev->fw_clients[id].props.max_msg_length) {
+               ++cl->err_send_msg;
+               return -EMSGSIZE;
+       }
+
+       /* No free bufs */
+       spin_lock_irqsave(&cl->tx_free_list_spinlock, tx_free_flags);
+       if (list_empty(&cl->tx_free_list.list)) {
+               spin_unlock_irqrestore(&cl->tx_free_list_spinlock,
+                       tx_free_flags);
+               ++cl->err_send_msg;
+               return  -ENOMEM;
+       }
+
+       cl_msg = list_first_entry(&cl->tx_free_list.list,
+               struct ishtp_cl_tx_ring, list);
+       if (!cl_msg->send_buf.data) {
+               spin_unlock_irqrestore(&cl->tx_free_list_spinlock,
+                       tx_free_flags);
+               return  -EIO;
+               /* Should not happen, as free list is pre-allocated */
+       }
+       /*
+        * This is safe, as 'length' is already checked for not exceeding
+        * max ISHTP message size per client
+        */
+       list_del_init(&cl_msg->list);
+       spin_unlock_irqrestore(&cl->tx_free_list_spinlock, tx_free_flags);
+       memcpy(cl_msg->send_buf.data, buf, length);
+       cl_msg->send_buf.size = length;
+       spin_lock_irqsave(&cl->tx_list_spinlock, tx_flags);
+       have_msg_to_send = !list_empty(&cl->tx_list.list);
+       list_add_tail(&cl_msg->list, &cl->tx_list.list);
+       spin_unlock_irqrestore(&cl->tx_list_spinlock, tx_flags);
+
+       if (!have_msg_to_send && cl->ishtp_flow_ctrl_creds > 0)
+               ishtp_cl_send_msg(dev, cl);
+
+       return  0;
+}
+EXPORT_SYMBOL(ishtp_cl_send);
+
+/**
+ * ishtp_cl_read_complete() - read complete
+ * @rb: Pointer to client request block
+ *
+ * If the message is completely received call ishtp_cl_bus_rx_event()
+ * to process message
+ */
+static void ishtp_cl_read_complete(struct ishtp_cl_rb *rb)
+{
+       unsigned long   flags;
+       int     schedule_work_flag = 0;
+       struct ishtp_cl *cl = rb->cl;
+
+       spin_lock_irqsave(&cl->in_process_spinlock, flags);
+       /*
+        * if in-process list is empty, then need to schedule
+        * the processing thread
+        */
+       schedule_work_flag = list_empty(&cl->in_process_list.list);
+       list_add_tail(&rb->list, &cl->in_process_list.list);
+       spin_unlock_irqrestore(&cl->in_process_spinlock, flags);
+
+       if (schedule_work_flag)
+               ishtp_cl_bus_rx_event(cl->device);
+}
+
+/**
+ * ipc_tx_callback() - IPC tx callback function
+ * @prm: Pointer to client device instance
+ *
+ * Send message over IPC either first time or on callback on previous message
+ * completion
+ */
+static void ipc_tx_callback(void *prm)
+{
+       struct ishtp_cl *cl = prm;
+       struct ishtp_cl_tx_ring *cl_msg;
+       size_t  rem;
+       struct ishtp_device     *dev = (cl ? cl->dev : NULL);
+       struct ishtp_msg_hdr    ishtp_hdr;
+       unsigned long   tx_flags, tx_free_flags;
+       unsigned char   *pmsg;
+
+       if (!dev)
+               return;
+
+       /*
+        * Other conditions if some critical error has
+        * occurred before this callback is called
+        */
+       if (dev->dev_state != ISHTP_DEV_ENABLED)
+               return;
+
+       if (cl->state != ISHTP_CL_CONNECTED)
+               return;
+
+       spin_lock_irqsave(&cl->tx_list_spinlock, tx_flags);
+       if (list_empty(&cl->tx_list.list)) {
+               spin_unlock_irqrestore(&cl->tx_list_spinlock, tx_flags);
+               return;
+       }
+
+       if (cl->ishtp_flow_ctrl_creds != 1 && !cl->sending) {
+               spin_unlock_irqrestore(&cl->tx_list_spinlock, tx_flags);
+               return;
+       }
+
+       if (!cl->sending) {
+               --cl->ishtp_flow_ctrl_creds;
+               cl->last_ipc_acked = 0;
+               cl->last_tx_path = CL_TX_PATH_IPC;
+               cl->sending = 1;
+       }
+
+       cl_msg = list_entry(cl->tx_list.list.next, struct ishtp_cl_tx_ring,
+                           list);
+       rem = cl_msg->send_buf.size - cl->tx_offs;
+
+       ishtp_hdr.host_addr = cl->host_client_id;
+       ishtp_hdr.fw_addr = cl->fw_client_id;
+       ishtp_hdr.reserved = 0;
+       pmsg = cl_msg->send_buf.data + cl->tx_offs;
+
+       if (rem <= dev->mtu) {
+               ishtp_hdr.length = rem;
+               ishtp_hdr.msg_complete = 1;
+               cl->sending = 0;
+               list_del_init(&cl_msg->list);   /* Must be before write */
+               spin_unlock_irqrestore(&cl->tx_list_spinlock, tx_flags);
+               /* Submit to IPC queue with no callback */
+               ishtp_write_message(dev, &ishtp_hdr, pmsg);
+               spin_lock_irqsave(&cl->tx_free_list_spinlock, tx_free_flags);
+               list_add_tail(&cl_msg->list, &cl->tx_free_list.list);
+               spin_unlock_irqrestore(&cl->tx_free_list_spinlock,
+                       tx_free_flags);
+       } else {
+               /* Send IPC fragment */
+               spin_unlock_irqrestore(&cl->tx_list_spinlock, tx_flags);
+               cl->tx_offs += dev->mtu;
+               ishtp_hdr.length = dev->mtu;
+               ishtp_hdr.msg_complete = 0;
+               ishtp_send_msg(dev, &ishtp_hdr, pmsg, ipc_tx_callback, cl);
+       }
+}
+
+/**
+ * ishtp_cl_send_msg_ipc() -Send message using IPC
+ * @dev: ISHTP device instance
+ * @cl: Pointer to client device instance
+ *
+ * Send message over IPC not using DMA
+ */
+static void ishtp_cl_send_msg_ipc(struct ishtp_device *dev,
+                                 struct ishtp_cl *cl)
+{
+       /* If last DMA message wasn't acked yet, leave this one in Tx queue */
+       if (cl->last_tx_path == CL_TX_PATH_DMA && cl->last_dma_acked == 0)
+               return;
+
+       cl->tx_offs = 0;
+       ipc_tx_callback(cl);
+       ++cl->send_msg_cnt_ipc;
+}
+
+/**
+ * ishtp_cl_send_msg_dma() -Send message using DMA
+ * @dev: ISHTP device instance
+ * @cl: Pointer to client device instance
+ *
+ * Send message using DMA
+ */
+static void ishtp_cl_send_msg_dma(struct ishtp_device *dev,
+       struct ishtp_cl *cl)
+{
+       struct ishtp_msg_hdr    hdr;
+       struct dma_xfer_hbm     dma_xfer;
+       unsigned char   *msg_addr;
+       int off;
+       struct ishtp_cl_tx_ring *cl_msg;
+       unsigned long tx_flags, tx_free_flags;
+
+       /* If last IPC message wasn't acked yet, leave this one in Tx queue */
+       if (cl->last_tx_path == CL_TX_PATH_IPC && cl->last_ipc_acked == 0)
+               return;
+
+       spin_lock_irqsave(&cl->tx_list_spinlock, tx_flags);
+       if (list_empty(&cl->tx_list.list)) {
+               spin_unlock_irqrestore(&cl->tx_list_spinlock, tx_flags);
+               return;
+       }
+
+       cl_msg = list_entry(cl->tx_list.list.next, struct ishtp_cl_tx_ring,
+               list);
+
+       msg_addr = ishtp_cl_get_dma_send_buf(dev, cl_msg->send_buf.size);
+       if (!msg_addr) {
+               spin_unlock_irqrestore(&cl->tx_list_spinlock, tx_flags);
+               if (dev->transfer_path == CL_TX_PATH_DEFAULT)
+                       ishtp_cl_send_msg_ipc(dev, cl);
+               return;
+       }
+
+       list_del_init(&cl_msg->list);   /* Must be before write */
+       spin_unlock_irqrestore(&cl->tx_list_spinlock, tx_flags);
+
+       --cl->ishtp_flow_ctrl_creds;
+       cl->last_dma_acked = 0;
+       cl->last_dma_addr = msg_addr;
+       cl->last_tx_path = CL_TX_PATH_DMA;
+
+       /* write msg to dma buf */
+       memcpy(msg_addr, cl_msg->send_buf.data, cl_msg->send_buf.size);
+
+       /* send dma_xfer hbm msg */
+       off = msg_addr - (unsigned char *)dev->ishtp_host_dma_tx_buf;
+       ishtp_hbm_hdr(&hdr, sizeof(struct dma_xfer_hbm));
+       dma_xfer.hbm = DMA_XFER;
+       dma_xfer.fw_client_id = cl->fw_client_id;
+       dma_xfer.host_client_id = cl->host_client_id;
+       dma_xfer.reserved = 0;
+       dma_xfer.msg_addr = dev->ishtp_host_dma_tx_buf_phys + off;
+       dma_xfer.msg_length = cl_msg->send_buf.size;
+       dma_xfer.reserved2 = 0;
+       ishtp_write_message(dev, &hdr, (unsigned char *)&dma_xfer);
+       spin_lock_irqsave(&cl->tx_free_list_spinlock, tx_free_flags);
+       list_add_tail(&cl_msg->list, &cl->tx_free_list.list);
+       spin_unlock_irqrestore(&cl->tx_free_list_spinlock, tx_free_flags);
+       ++cl->send_msg_cnt_dma;
+}
+
+/**
+ * ishtp_cl_send_msg() -Send message using DMA or IPC
+ * @dev: ISHTP device instance
+ * @cl: Pointer to client device instance
+ *
+ * Send message using DMA or IPC based on transfer_path
+ */
+void ishtp_cl_send_msg(struct ishtp_device *dev, struct ishtp_cl *cl)
+{
+       if (dev->transfer_path == CL_TX_PATH_DMA)
+               ishtp_cl_send_msg_dma(dev, cl);
+       else
+               ishtp_cl_send_msg_ipc(dev, cl);
+}
+
+/**
+ * recv_ishtp_cl_msg() -Receive client message
+ * @dev: ISHTP device instance
+ * @ishtp_hdr: Pointer to message header
+ *
+ * Receive and dispatch ISHTP client messages. This function executes in ISR
+ * context
+ */
+void recv_ishtp_cl_msg(struct ishtp_device *dev,
+                      struct ishtp_msg_hdr *ishtp_hdr)
+{
+       struct ishtp_cl *cl;
+       struct ishtp_cl_rb *rb;
+       struct ishtp_cl_rb *new_rb;
+       unsigned char *buffer = NULL;
+       struct ishtp_cl_rb *complete_rb = NULL;
+       unsigned long   dev_flags;
+       unsigned long   flags;
+       int     rb_count;
+
+       if (ishtp_hdr->reserved) {
+               dev_err(dev->devc, "corrupted message header.\n");
+               goto    eoi;
+       }
+
+       if (ishtp_hdr->length > IPC_PAYLOAD_SIZE) {
+               dev_err(dev->devc,
+                       "ISHTP message length in hdr exceeds IPC MTU\n");
+               goto    eoi;
+       }
+
+       spin_lock_irqsave(&dev->read_list_spinlock, dev_flags);
+       rb_count = -1;
+       list_for_each_entry(rb, &dev->read_list.list, list) {
+               ++rb_count;
+               cl = rb->cl;
+               if (!cl || !(cl->host_client_id == ishtp_hdr->host_addr &&
+                               cl->fw_client_id == ishtp_hdr->fw_addr) ||
+                               !(cl->state == ISHTP_CL_CONNECTED))
+                       continue;
+
+                /* If no Rx buffer is allocated, disband the rb */
+               if (rb->buffer.size == 0 || rb->buffer.data == NULL) {
+                       spin_unlock_irqrestore(&dev->read_list_spinlock,
+                               dev_flags);
+                       dev_err(&cl->device->dev,
+                               "Rx buffer is not allocated.\n");
+                       list_del(&rb->list);
+                       ishtp_io_rb_free(rb);
+                       cl->status = -ENOMEM;
+                       goto    eoi;
+               }
+
+               /*
+                * If message buffer overflown (exceeds max. client msg
+                * size, drop message and return to free buffer.
+                * Do we need to disconnect such a client? (We don't send
+                * back FC, so communication will be stuck anyway)
+                */
+               if (rb->buffer.size < ishtp_hdr->length + rb->buf_idx) {
+                       spin_unlock_irqrestore(&dev->read_list_spinlock,
+                               dev_flags);
+                       dev_err(&cl->device->dev,
+                               "message overflow. size %d len %d idx %ld\n",
+                               rb->buffer.size, ishtp_hdr->length,
+                               rb->buf_idx);
+                       list_del(&rb->list);
+                       ishtp_cl_io_rb_recycle(rb);
+                       cl->status = -EIO;
+                       goto    eoi;
+               }
+
+               buffer = rb->buffer.data + rb->buf_idx;
+               dev->ops->ishtp_read(dev, buffer, ishtp_hdr->length);
+
+               rb->buf_idx += ishtp_hdr->length;
+               if (ishtp_hdr->msg_complete) {
+                       /* Last fragment in message - it's complete */
+                       cl->status = 0;
+                       list_del(&rb->list);
+                       complete_rb = rb;
+
+                       --cl->out_flow_ctrl_creds;
+                       /*
+                        * the whole msg arrived, send a new FC, and add a new
+                        * rb buffer for the next coming msg
+                        */
+                       spin_lock_irqsave(&cl->free_list_spinlock, flags);
+
+                       if (!list_empty(&cl->free_rb_list.list)) {
+                               new_rb = list_entry(cl->free_rb_list.list.next,
+                                       struct ishtp_cl_rb, list);
+                               list_del_init(&new_rb->list);
+                               spin_unlock_irqrestore(&cl->free_list_spinlock,
+                                       flags);
+                               new_rb->cl = cl;
+                               new_rb->buf_idx = 0;
+                               INIT_LIST_HEAD(&new_rb->list);
+                               list_add_tail(&new_rb->list,
+                                       &dev->read_list.list);
+
+                               ishtp_hbm_cl_flow_control_req(dev, cl);
+                       } else {
+                               spin_unlock_irqrestore(&cl->free_list_spinlock,
+                                       flags);
+                       }
+               }
+               /* One more fragment in message (even if this was last) */
+               ++cl->recv_msg_num_frags;
+
+               /*
+                * We can safely break here (and in BH too),
+                * a single input message can go only to a single request!
+                */
+               break;
+       }
+
+       spin_unlock_irqrestore(&dev->read_list_spinlock, dev_flags);
+       /* If it's nobody's message, just read and discard it */
+       if (!buffer) {
+               uint8_t rd_msg_buf[ISHTP_RD_MSG_BUF_SIZE];
+
+               dev_err(dev->devc, "Dropped Rx msg - no request\n");
+               dev->ops->ishtp_read(dev, rd_msg_buf, ishtp_hdr->length);
+               goto    eoi;
+       }
+
+       if (complete_rb) {
+               getnstimeofday(&cl->ts_rx);
+               ++cl->recv_msg_cnt_ipc;
+               ishtp_cl_read_complete(complete_rb);
+       }
+eoi:
+       return;
+}
+
+/**
+ * recv_ishtp_cl_msg_dma() -Receive client message
+ * @dev: ISHTP device instance
+ * @msg: message pointer
+ * @hbm: hbm buffer
+ *
+ * Receive and dispatch ISHTP client messages using DMA. This function executes
+ * in ISR context
+ */
+void recv_ishtp_cl_msg_dma(struct ishtp_device *dev, void *msg,
+                          struct dma_xfer_hbm *hbm)
+{
+       struct ishtp_cl *cl;
+       struct ishtp_cl_rb *rb;
+       struct ishtp_cl_rb *new_rb;
+       unsigned char *buffer = NULL;
+       struct ishtp_cl_rb *complete_rb = NULL;
+       unsigned long   dev_flags;
+       unsigned long   flags;
+
+       spin_lock_irqsave(&dev->read_list_spinlock, dev_flags);
+       list_for_each_entry(rb, &dev->read_list.list, list) {
+               cl = rb->cl;
+               if (!cl || !(cl->host_client_id == hbm->host_client_id &&
+                               cl->fw_client_id == hbm->fw_client_id) ||
+                               !(cl->state == ISHTP_CL_CONNECTED))
+                       continue;
+
+               /*
+                * If no Rx buffer is allocated, disband the rb
+                */
+               if (rb->buffer.size == 0 || rb->buffer.data == NULL) {
+                       spin_unlock_irqrestore(&dev->read_list_spinlock,
+                               dev_flags);
+                       dev_err(&cl->device->dev,
+                               "response buffer is not allocated.\n");
+                       list_del(&rb->list);
+                       ishtp_io_rb_free(rb);
+                       cl->status = -ENOMEM;
+                       goto    eoi;
+               }
+
+               /*
+                * If message buffer overflown (exceeds max. client msg
+                * size, drop message and return to free buffer.
+                * Do we need to disconnect such a client? (We don't send
+                * back FC, so communication will be stuck anyway)
+                */
+               if (rb->buffer.size < hbm->msg_length) {
+                       spin_unlock_irqrestore(&dev->read_list_spinlock,
+                               dev_flags);
+                       dev_err(&cl->device->dev,
+                               "message overflow. size %d len %d idx %ld\n",
+                               rb->buffer.size, hbm->msg_length, rb->buf_idx);
+                       list_del(&rb->list);
+                       ishtp_cl_io_rb_recycle(rb);
+                       cl->status = -EIO;
+                       goto    eoi;
+               }
+
+               buffer = rb->buffer.data;
+               memcpy(buffer, msg, hbm->msg_length);
+               rb->buf_idx = hbm->msg_length;
+
+               /* Last fragment in message - it's complete */
+               cl->status = 0;
+               list_del(&rb->list);
+               complete_rb = rb;
+
+               --cl->out_flow_ctrl_creds;
+               /*
+                * the whole msg arrived, send a new FC, and add a new
+                * rb buffer for the next coming msg
+                */
+               spin_lock_irqsave(&cl->free_list_spinlock, flags);
+
+               if (!list_empty(&cl->free_rb_list.list)) {
+                       new_rb = list_entry(cl->free_rb_list.list.next,
+                               struct ishtp_cl_rb, list);
+                       list_del_init(&new_rb->list);
+                       spin_unlock_irqrestore(&cl->free_list_spinlock,
+                               flags);
+                       new_rb->cl = cl;
+                       new_rb->buf_idx = 0;
+                       INIT_LIST_HEAD(&new_rb->list);
+                       list_add_tail(&new_rb->list,
+                               &dev->read_list.list);
+
+                       ishtp_hbm_cl_flow_control_req(dev, cl);
+               } else {
+                       spin_unlock_irqrestore(&cl->free_list_spinlock,
+                               flags);
+               }
+
+               /* One more fragment in message (this is always last) */
+               ++cl->recv_msg_num_frags;
+
+               /*
+                * We can safely break here (and in BH too),
+                * a single input message can go only to a single request!
+                */
+               break;
+       }
+
+       spin_unlock_irqrestore(&dev->read_list_spinlock, dev_flags);
+       /* If it's nobody's message, just read and discard it */
+       if (!buffer) {
+               dev_err(dev->devc, "Dropped Rx (DMA) msg - no request\n");
+               goto    eoi;
+       }
+
+       if (complete_rb) {
+               getnstimeofday(&cl->ts_rx);
+               ++cl->recv_msg_cnt_dma;
+               ishtp_cl_read_complete(complete_rb);
+       }
+eoi:
+       return;
+}