2 * This file is part of the Chelsio T4 PCI-E SR-IOV Virtual Function Ethernet
5 * Copyright (c) 2009-2010 Chelsio Communications, Inc. All rights reserved.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38 #include <linux/module.h>
39 #include <linux/moduleparam.h>
40 #include <linux/init.h>
41 #include <linux/pci.h>
42 #include <linux/dma-mapping.h>
43 #include <linux/netdevice.h>
44 #include <linux/etherdevice.h>
45 #include <linux/debugfs.h>
46 #include <linux/ethtool.h>
48 #include "t4vf_common.h"
49 #include "t4vf_defs.h"
51 #include "../cxgb4/t4_regs.h"
52 #include "../cxgb4/t4_msg.h"
55 * Generic information about the driver.
57 #define DRV_VERSION "2.0.0-ko"
58 #define DRV_DESC "Chelsio T4/T5 Virtual Function (VF) Network Driver"
66 * Default ethtool "message level" for adapters.
68 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
69 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
70 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
72 static int dflt_msg_enable = DFLT_MSG_ENABLE;
74 module_param(dflt_msg_enable, int, 0644);
75 MODULE_PARM_DESC(dflt_msg_enable,
76 "default adapter ethtool message level bitmap");
79 * The driver uses the best interrupt scheme available on a platform in the
80 * order MSI-X then MSI. This parameter determines which of these schemes the
81 * driver may consider as follows:
83 * msi = 2: choose from among MSI-X and MSI
84 * msi = 1: only consider MSI interrupts
86 * Note that unlike the Physical Function driver, this Virtual Function driver
87 * does _not_ support legacy INTx interrupts (this limitation is mandated by
88 * the PCI-E SR-IOV standard).
92 #define MSI_DEFAULT MSI_MSIX
94 static int msi = MSI_DEFAULT;
96 module_param(msi, int, 0644);
97 MODULE_PARM_DESC(msi, "whether to use MSI-X or MSI");
100 * Fundamental constants.
101 * ======================
105 MAX_TXQ_ENTRIES = 16384,
106 MAX_RSPQ_ENTRIES = 16384,
107 MAX_RX_BUFFERS = 16384,
109 MIN_TXQ_ENTRIES = 32,
110 MIN_RSPQ_ENTRIES = 128,
114 * For purposes of manipulating the Free List size we need to
115 * recognize that Free Lists are actually Egress Queues (the host
116 * produces free buffers which the hardware consumes), Egress Queues
117 * indices are all in units of Egress Context Units bytes, and free
118 * list entries are 64-bit PCI DMA addresses. And since the state of
119 * the Producer Index == the Consumer Index implies an EMPTY list, we
120 * always have at least one Egress Unit's worth of Free List entries
121 * unused. See sge.c for more details ...
123 EQ_UNIT = SGE_EQ_IDXSIZE,
124 FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64),
125 MIN_FL_RESID = FL_PER_EQ_UNIT,
129 * Global driver state.
130 * ====================
133 static struct dentry *cxgb4vf_debugfs_root;
136 * OS "Callback" functions.
137 * ========================
141 * The link status has changed on the indicated "port" (Virtual Interface).
143 void t4vf_os_link_changed(struct adapter *adapter, int pidx, int link_ok)
145 struct net_device *dev = adapter->port[pidx];
148 * If the port is disabled or the current recorded "link up"
149 * status matches the new status, just return.
151 if (!netif_running(dev) || link_ok == netif_carrier_ok(dev))
155 * Tell the OS that the link status has changed and print a short
156 * informative message on the console about the event.
161 const struct port_info *pi = netdev_priv(dev);
163 netif_carrier_on(dev);
165 switch (pi->link_cfg.speed) {
187 switch (pi->link_cfg.fc) {
196 case PAUSE_RX|PAUSE_TX:
205 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s, fc);
207 netif_carrier_off(dev);
208 netdev_info(dev, "link down\n");
213 * Net device operations.
214 * ======================
221 * Perform the MAC and PHY actions needed to enable a "port" (Virtual
224 static int link_start(struct net_device *dev)
227 struct port_info *pi = netdev_priv(dev);
230 * We do not set address filters and promiscuity here, the stack does
231 * that step explicitly. Enable vlan accel.
233 ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, 1,
236 ret = t4vf_change_mac(pi->adapter, pi->viid,
237 pi->xact_addr_filt, dev->dev_addr, true);
239 pi->xact_addr_filt = ret;
245 * We don't need to actually "start the link" itself since the
246 * firmware will do that for us when the first Virtual Interface
247 * is enabled on a port.
250 ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true);
255 * Name the MSI-X interrupts.
257 static void name_msix_vecs(struct adapter *adapter)
259 int namelen = sizeof(adapter->msix_info[0].desc) - 1;
265 snprintf(adapter->msix_info[MSIX_FW].desc, namelen,
266 "%s-FWeventq", adapter->name);
267 adapter->msix_info[MSIX_FW].desc[namelen] = 0;
272 for_each_port(adapter, pidx) {
273 struct net_device *dev = adapter->port[pidx];
274 const struct port_info *pi = netdev_priv(dev);
277 for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) {
278 snprintf(adapter->msix_info[msi].desc, namelen,
279 "%s-%d", dev->name, qs);
280 adapter->msix_info[msi].desc[namelen] = 0;
286 * Request all of our MSI-X resources.
288 static int request_msix_queue_irqs(struct adapter *adapter)
290 struct sge *s = &adapter->sge;
296 err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix,
297 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq);
305 for_each_ethrxq(s, rxq) {
306 err = request_irq(adapter->msix_info[msi].vec,
307 t4vf_sge_intr_msix, 0,
308 adapter->msix_info[msi].desc,
309 &s->ethrxq[rxq].rspq);
318 free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq);
319 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
324 * Free our MSI-X resources.
326 static void free_msix_queue_irqs(struct adapter *adapter)
328 struct sge *s = &adapter->sge;
331 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
333 for_each_ethrxq(s, rxq)
334 free_irq(adapter->msix_info[msi++].vec,
335 &s->ethrxq[rxq].rspq);
339 * Turn on NAPI and start up interrupts on a response queue.
341 static void qenable(struct sge_rspq *rspq)
343 napi_enable(&rspq->napi);
346 * 0-increment the Going To Sleep register to start the timer and
349 t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
351 SEINTARM(rspq->intr_params) |
352 INGRESSQID(rspq->cntxt_id));
356 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
358 static void enable_rx(struct adapter *adapter)
361 struct sge *s = &adapter->sge;
363 for_each_ethrxq(s, rxq)
364 qenable(&s->ethrxq[rxq].rspq);
365 qenable(&s->fw_evtq);
368 * The interrupt queue doesn't use NAPI so we do the 0-increment of
369 * its Going To Sleep register here to get it started.
371 if (adapter->flags & USING_MSI)
372 t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
374 SEINTARM(s->intrq.intr_params) |
375 INGRESSQID(s->intrq.cntxt_id));
380 * Wait until all NAPI handlers are descheduled.
382 static void quiesce_rx(struct adapter *adapter)
384 struct sge *s = &adapter->sge;
387 for_each_ethrxq(s, rxq)
388 napi_disable(&s->ethrxq[rxq].rspq.napi);
389 napi_disable(&s->fw_evtq.napi);
393 * Response queue handler for the firmware event queue.
395 static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp,
396 const struct pkt_gl *gl)
399 * Extract response opcode and get pointer to CPL message body.
401 struct adapter *adapter = rspq->adapter;
402 u8 opcode = ((const struct rss_header *)rsp)->opcode;
403 void *cpl = (void *)(rsp + 1);
408 * We've received an asynchronous message from the firmware.
410 const struct cpl_fw6_msg *fw_msg = cpl;
411 if (fw_msg->type == FW6_TYPE_CMD_RPL)
412 t4vf_handle_fw_rpl(adapter, fw_msg->data);
417 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
419 const struct cpl_sge_egr_update *p = (void *)(rsp + 3);
420 opcode = G_CPL_OPCODE(ntohl(p->opcode_qid));
421 if (opcode != CPL_SGE_EGR_UPDATE) {
422 dev_err(adapter->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
430 case CPL_SGE_EGR_UPDATE: {
432 * We've received an Egress Queue Status Update message. We
433 * get these, if the SGE is configured to send these when the
434 * firmware passes certain points in processing our TX
435 * Ethernet Queue or if we make an explicit request for one.
436 * We use these updates to determine when we may need to
437 * restart a TX Ethernet Queue which was stopped for lack of
438 * free TX Queue Descriptors ...
440 const struct cpl_sge_egr_update *p = cpl;
441 unsigned int qid = EGR_QID(be32_to_cpu(p->opcode_qid));
442 struct sge *s = &adapter->sge;
444 struct sge_eth_txq *txq;
448 * Perform sanity checking on the Queue ID to make sure it
449 * really refers to one of our TX Ethernet Egress Queues which
450 * is active and matches the queue's ID. None of these error
451 * conditions should ever happen so we may want to either make
452 * them fatal and/or conditionalized under DEBUG.
454 eq_idx = EQ_IDX(s, qid);
455 if (unlikely(eq_idx >= MAX_EGRQ)) {
456 dev_err(adapter->pdev_dev,
457 "Egress Update QID %d out of range\n", qid);
460 tq = s->egr_map[eq_idx];
461 if (unlikely(tq == NULL)) {
462 dev_err(adapter->pdev_dev,
463 "Egress Update QID %d TXQ=NULL\n", qid);
466 txq = container_of(tq, struct sge_eth_txq, q);
467 if (unlikely(tq->abs_id != qid)) {
468 dev_err(adapter->pdev_dev,
469 "Egress Update QID %d refers to TXQ %d\n",
475 * Restart a stopped TX Queue which has less than half of its
479 netif_tx_wake_queue(txq->txq);
484 dev_err(adapter->pdev_dev,
485 "unexpected CPL %#x on FW event queue\n", opcode);
492 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
493 * to use and initializes them. We support multiple "Queue Sets" per port if
494 * we have MSI-X, otherwise just one queue set per port.
496 static int setup_sge_queues(struct adapter *adapter)
498 struct sge *s = &adapter->sge;
502 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
505 bitmap_zero(s->starving_fl, MAX_EGRQ);
508 * If we're using MSI interrupt mode we need to set up a "forwarded
509 * interrupt" queue which we'll set up with our MSI vector. The rest
510 * of the ingress queues will be set up to forward their interrupts to
511 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
512 * the intrq's queue ID as the interrupt forwarding queue for the
513 * subsequent calls ...
515 if (adapter->flags & USING_MSI) {
516 err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false,
517 adapter->port[0], 0, NULL, NULL);
519 goto err_free_queues;
523 * Allocate our ingress queue for asynchronous firmware messages.
525 err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0],
526 MSIX_FW, NULL, fwevtq_handler);
528 goto err_free_queues;
531 * Allocate each "port"'s initial Queue Sets. These can be changed
532 * later on ... up to the point where any interface on the adapter is
533 * brought up at which point lots of things get nailed down
537 for_each_port(adapter, pidx) {
538 struct net_device *dev = adapter->port[pidx];
539 struct port_info *pi = netdev_priv(dev);
540 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
541 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
544 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
545 err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false,
547 &rxq->fl, t4vf_ethrx_handler);
549 goto err_free_queues;
551 err = t4vf_sge_alloc_eth_txq(adapter, txq, dev,
552 netdev_get_tx_queue(dev, qs),
553 s->fw_evtq.cntxt_id);
555 goto err_free_queues;
558 memset(&rxq->stats, 0, sizeof(rxq->stats));
563 * Create the reverse mappings for the queues.
565 s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id;
566 s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id;
567 IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq;
568 for_each_port(adapter, pidx) {
569 struct net_device *dev = adapter->port[pidx];
570 struct port_info *pi = netdev_priv(dev);
571 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
572 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
575 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
576 IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq;
577 EQ_MAP(s, txq->q.abs_id) = &txq->q;
580 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
581 * for Free Lists but since all of the Egress Queues
582 * (including Free Lists) have Relative Queue IDs
583 * which are computed as Absolute - Base Queue ID, we
584 * can synthesize the Absolute Queue IDs for the Free
585 * Lists. This is useful for debugging purposes when
586 * we want to dump Queue Contexts via the PF Driver.
588 rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base;
589 EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl;
595 t4vf_free_sge_resources(adapter);
600 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
601 * queues. We configure the RSS CPU lookup table to distribute to the number
602 * of HW receive queues, and the response queue lookup table to narrow that
603 * down to the response queues actually configured for each "port" (Virtual
604 * Interface). We always configure the RSS mapping for all ports since the
605 * mapping table has plenty of entries.
607 static int setup_rss(struct adapter *adapter)
611 for_each_port(adapter, pidx) {
612 struct port_info *pi = adap2pinfo(adapter, pidx);
613 struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
614 u16 rss[MAX_PORT_QSETS];
617 for (qs = 0; qs < pi->nqsets; qs++)
618 rss[qs] = rxq[qs].rspq.abs_id;
620 err = t4vf_config_rss_range(adapter, pi->viid,
621 0, pi->rss_size, rss, pi->nqsets);
626 * Perform Global RSS Mode-specific initialization.
628 switch (adapter->params.rss.mode) {
629 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL:
631 * If Tunnel All Lookup isn't specified in the global
632 * RSS Configuration, then we need to specify a
633 * default Ingress Queue for any ingress packets which
634 * aren't hashed. We'll use our first ingress queue
637 if (!adapter->params.rss.u.basicvirtual.tnlalllookup) {
638 union rss_vi_config config;
639 err = t4vf_read_rss_vi_config(adapter,
644 config.basicvirtual.defaultq =
646 err = t4vf_write_rss_vi_config(adapter,
660 * Bring the adapter up. Called whenever we go from no "ports" open to having
661 * one open. This function performs the actions necessary to make an adapter
662 * operational, such as completing the initialization of HW modules, and
663 * enabling interrupts. Must be called with the rtnl lock held. (Note that
664 * this is called "cxgb_up" in the PF Driver.)
666 static int adapter_up(struct adapter *adapter)
671 * If this is the first time we've been called, perform basic
672 * adapter setup. Once we've done this, many of our adapter
673 * parameters can no longer be changed ...
675 if ((adapter->flags & FULL_INIT_DONE) == 0) {
676 err = setup_sge_queues(adapter);
679 err = setup_rss(adapter);
681 t4vf_free_sge_resources(adapter);
685 if (adapter->flags & USING_MSIX)
686 name_msix_vecs(adapter);
687 adapter->flags |= FULL_INIT_DONE;
691 * Acquire our interrupt resources. We only support MSI-X and MSI.
693 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
694 if (adapter->flags & USING_MSIX)
695 err = request_msix_queue_irqs(adapter);
697 err = request_irq(adapter->pdev->irq,
698 t4vf_intr_handler(adapter), 0,
699 adapter->name, adapter);
701 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
707 * Enable NAPI ingress processing and return success.
710 t4vf_sge_start(adapter);
715 * Bring the adapter down. Called whenever the last "port" (Virtual
716 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
719 static void adapter_down(struct adapter *adapter)
722 * Free interrupt resources.
724 if (adapter->flags & USING_MSIX)
725 free_msix_queue_irqs(adapter);
727 free_irq(adapter->pdev->irq, adapter);
730 * Wait for NAPI handlers to finish.
736 * Start up a net device.
738 static int cxgb4vf_open(struct net_device *dev)
741 struct port_info *pi = netdev_priv(dev);
742 struct adapter *adapter = pi->adapter;
745 * If this is the first interface that we're opening on the "adapter",
746 * bring the "adapter" up now.
748 if (adapter->open_device_map == 0) {
749 err = adapter_up(adapter);
755 * Note that this interface is up and start everything up ...
757 netif_set_real_num_tx_queues(dev, pi->nqsets);
758 err = netif_set_real_num_rx_queues(dev, pi->nqsets);
761 err = link_start(dev);
765 netif_tx_start_all_queues(dev);
766 set_bit(pi->port_id, &adapter->open_device_map);
770 if (adapter->open_device_map == 0)
771 adapter_down(adapter);
776 * Shut down a net device. This routine is called "cxgb_close" in the PF
779 static int cxgb4vf_stop(struct net_device *dev)
781 struct port_info *pi = netdev_priv(dev);
782 struct adapter *adapter = pi->adapter;
784 netif_tx_stop_all_queues(dev);
785 netif_carrier_off(dev);
786 t4vf_enable_vi(adapter, pi->viid, false, false);
787 pi->link_cfg.link_ok = 0;
789 clear_bit(pi->port_id, &adapter->open_device_map);
790 if (adapter->open_device_map == 0)
791 adapter_down(adapter);
796 * Translate our basic statistics into the standard "ifconfig" statistics.
798 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
800 struct t4vf_port_stats stats;
801 struct port_info *pi = netdev2pinfo(dev);
802 struct adapter *adapter = pi->adapter;
803 struct net_device_stats *ns = &dev->stats;
806 spin_lock(&adapter->stats_lock);
807 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
808 spin_unlock(&adapter->stats_lock);
810 memset(ns, 0, sizeof(*ns));
814 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
815 stats.tx_ucast_bytes + stats.tx_offload_bytes);
816 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
817 stats.tx_ucast_frames + stats.tx_offload_frames);
818 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
819 stats.rx_ucast_bytes);
820 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
821 stats.rx_ucast_frames);
822 ns->multicast = stats.rx_mcast_frames;
823 ns->tx_errors = stats.tx_drop_frames;
824 ns->rx_errors = stats.rx_err_frames;
830 * Collect up to maxaddrs worth of a netdevice's unicast addresses, starting
831 * at a specified offset within the list, into an array of addrss pointers and
832 * return the number collected.
834 static inline unsigned int collect_netdev_uc_list_addrs(const struct net_device *dev,
837 unsigned int maxaddrs)
839 unsigned int index = 0;
840 unsigned int naddr = 0;
841 const struct netdev_hw_addr *ha;
843 for_each_dev_addr(dev, ha)
844 if (index++ >= offset) {
845 addr[naddr++] = ha->addr;
846 if (naddr >= maxaddrs)
853 * Collect up to maxaddrs worth of a netdevice's multicast addresses, starting
854 * at a specified offset within the list, into an array of addrss pointers and
855 * return the number collected.
857 static inline unsigned int collect_netdev_mc_list_addrs(const struct net_device *dev,
860 unsigned int maxaddrs)
862 unsigned int index = 0;
863 unsigned int naddr = 0;
864 const struct netdev_hw_addr *ha;
866 netdev_for_each_mc_addr(ha, dev)
867 if (index++ >= offset) {
868 addr[naddr++] = ha->addr;
869 if (naddr >= maxaddrs)
876 * Configure the exact and hash address filters to handle a port's multicast
877 * and secondary unicast MAC addresses.
879 static int set_addr_filters(const struct net_device *dev, bool sleep)
884 unsigned int offset, naddr;
887 const struct port_info *pi = netdev_priv(dev);
889 /* first do the secondary unicast addresses */
890 for (offset = 0; ; offset += naddr) {
891 naddr = collect_netdev_uc_list_addrs(dev, addr, offset,
896 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
897 naddr, addr, NULL, &uhash, sleep);
904 /* next set up the multicast addresses */
905 for (offset = 0; ; offset += naddr) {
906 naddr = collect_netdev_mc_list_addrs(dev, addr, offset,
911 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
912 naddr, addr, NULL, &mhash, sleep);
918 return t4vf_set_addr_hash(pi->adapter, pi->viid, uhash != 0,
919 uhash | mhash, sleep);
923 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
924 * If @mtu is -1 it is left unchanged.
926 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
929 struct port_info *pi = netdev_priv(dev);
931 ret = set_addr_filters(dev, sleep_ok);
933 ret = t4vf_set_rxmode(pi->adapter, pi->viid, -1,
934 (dev->flags & IFF_PROMISC) != 0,
935 (dev->flags & IFF_ALLMULTI) != 0,
941 * Set the current receive modes on the device.
943 static void cxgb4vf_set_rxmode(struct net_device *dev)
945 /* unfortunately we can't return errors to the stack */
946 set_rxmode(dev, -1, false);
950 * Find the entry in the interrupt holdoff timer value array which comes
951 * closest to the specified interrupt holdoff value.
953 static int closest_timer(const struct sge *s, int us)
955 int i, timer_idx = 0, min_delta = INT_MAX;
957 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
958 int delta = us - s->timer_val[i];
961 if (delta < min_delta) {
969 static int closest_thres(const struct sge *s, int thres)
971 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
973 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
974 delta = thres - s->counter_val[i];
977 if (delta < min_delta) {
986 * Return a queue's interrupt hold-off time in us. 0 means no timer.
988 static unsigned int qtimer_val(const struct adapter *adapter,
989 const struct sge_rspq *rspq)
991 unsigned int timer_idx = QINTR_TIMER_IDX_GET(rspq->intr_params);
993 return timer_idx < SGE_NTIMERS
994 ? adapter->sge.timer_val[timer_idx]
999 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1000 * @adapter: the adapter
1001 * @rspq: the RX response queue
1002 * @us: the hold-off time in us, or 0 to disable timer
1003 * @cnt: the hold-off packet count, or 0 to disable counter
1005 * Sets an RX response queue's interrupt hold-off time and packet count.
1006 * At least one of the two needs to be enabled for the queue to generate
1009 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
1010 unsigned int us, unsigned int cnt)
1012 unsigned int timer_idx;
1015 * If both the interrupt holdoff timer and count are specified as
1016 * zero, default to a holdoff count of 1 ...
1018 if ((us | cnt) == 0)
1022 * If an interrupt holdoff count has been specified, then find the
1023 * closest configured holdoff count and use that. If the response
1024 * queue has already been created, then update its queue context
1031 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1032 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1033 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1035 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1036 FW_PARAMS_PARAM_YZ(rspq->cntxt_id);
1037 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1041 rspq->pktcnt_idx = pktcnt_idx;
1045 * Compute the closest holdoff timer index from the supplied holdoff
1048 timer_idx = (us == 0
1049 ? SGE_TIMER_RSTRT_CNTR
1050 : closest_timer(&adapter->sge, us));
1053 * Update the response queue's interrupt coalescing parameters and
1056 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
1057 (cnt > 0 ? QINTR_CNT_EN : 0));
1062 * Return a version number to identify the type of adapter. The scheme is:
1063 * - bits 0..9: chip version
1064 * - bits 10..15: chip revision
1066 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1069 * Chip version 4, revision 0x3f (cxgb4vf).
1071 return CHELSIO_CHIP_VERSION(adapter->params.chip) | (0x3f << 10);
1075 * Execute the specified ioctl command.
1077 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1083 * The VF Driver doesn't have access to any of the other
1084 * common Ethernet device ioctl()'s (like reading/writing
1085 * PHY registers, etc.
1096 * Change the device's MTU.
1098 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1101 struct port_info *pi = netdev_priv(dev);
1103 /* accommodate SACK */
1107 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1108 -1, -1, -1, -1, true);
1114 static netdev_features_t cxgb4vf_fix_features(struct net_device *dev,
1115 netdev_features_t features)
1118 * Since there is no support for separate rx/tx vlan accel
1119 * enable/disable make sure tx flag is always in same state as rx.
1121 if (features & NETIF_F_HW_VLAN_CTAG_RX)
1122 features |= NETIF_F_HW_VLAN_CTAG_TX;
1124 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
1129 static int cxgb4vf_set_features(struct net_device *dev,
1130 netdev_features_t features)
1132 struct port_info *pi = netdev_priv(dev);
1133 netdev_features_t changed = dev->features ^ features;
1135 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1136 t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1,
1137 features & NETIF_F_HW_VLAN_CTAG_TX, 0);
1143 * Change the devices MAC address.
1145 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1148 struct sockaddr *addr = _addr;
1149 struct port_info *pi = netdev_priv(dev);
1151 if (!is_valid_ether_addr(addr->sa_data))
1152 return -EADDRNOTAVAIL;
1154 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1155 addr->sa_data, true);
1159 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1160 pi->xact_addr_filt = ret;
1164 #ifdef CONFIG_NET_POLL_CONTROLLER
1166 * Poll all of our receive queues. This is called outside of normal interrupt
1169 static void cxgb4vf_poll_controller(struct net_device *dev)
1171 struct port_info *pi = netdev_priv(dev);
1172 struct adapter *adapter = pi->adapter;
1174 if (adapter->flags & USING_MSIX) {
1175 struct sge_eth_rxq *rxq;
1178 rxq = &adapter->sge.ethrxq[pi->first_qset];
1179 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1180 t4vf_sge_intr_msix(0, &rxq->rspq);
1184 t4vf_intr_handler(adapter)(0, adapter);
1189 * Ethtool operations.
1190 * ===================
1192 * Note that we don't support any ethtool operations which change the physical
1193 * state of the port to which we're linked.
1197 * Return current port link settings.
1199 static int cxgb4vf_get_settings(struct net_device *dev,
1200 struct ethtool_cmd *cmd)
1202 const struct port_info *pi = netdev_priv(dev);
1204 cmd->supported = pi->link_cfg.supported;
1205 cmd->advertising = pi->link_cfg.advertising;
1206 ethtool_cmd_speed_set(cmd,
1207 netif_carrier_ok(dev) ? pi->link_cfg.speed : -1);
1208 cmd->duplex = DUPLEX_FULL;
1210 cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1211 cmd->phy_address = pi->port_id;
1212 cmd->transceiver = XCVR_EXTERNAL;
1213 cmd->autoneg = pi->link_cfg.autoneg;
1220 * Return our driver information.
1222 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1223 struct ethtool_drvinfo *drvinfo)
1225 struct adapter *adapter = netdev2adap(dev);
1227 strlcpy(drvinfo->driver, KBUILD_MODNAME, sizeof(drvinfo->driver));
1228 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
1229 strlcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)),
1230 sizeof(drvinfo->bus_info));
1231 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1232 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1233 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.fwrev),
1234 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.fwrev),
1235 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.fwrev),
1236 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.fwrev),
1237 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.tprev),
1238 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.tprev),
1239 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.tprev),
1240 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.tprev));
1244 * Return current adapter message level.
1246 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1248 return netdev2adap(dev)->msg_enable;
1252 * Set current adapter message level.
1254 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1256 netdev2adap(dev)->msg_enable = msglevel;
1260 * Return the device's current Queue Set ring size parameters along with the
1261 * allowed maximum values. Since ethtool doesn't understand the concept of
1262 * multi-queue devices, we just return the current values associated with the
1265 static void cxgb4vf_get_ringparam(struct net_device *dev,
1266 struct ethtool_ringparam *rp)
1268 const struct port_info *pi = netdev_priv(dev);
1269 const struct sge *s = &pi->adapter->sge;
1271 rp->rx_max_pending = MAX_RX_BUFFERS;
1272 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1273 rp->rx_jumbo_max_pending = 0;
1274 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1276 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1277 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1278 rp->rx_jumbo_pending = 0;
1279 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1283 * Set the Queue Set ring size parameters for the device. Again, since
1284 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1285 * apply these new values across all of the Queue Sets associated with the
1286 * device -- after vetting them of course!
1288 static int cxgb4vf_set_ringparam(struct net_device *dev,
1289 struct ethtool_ringparam *rp)
1291 const struct port_info *pi = netdev_priv(dev);
1292 struct adapter *adapter = pi->adapter;
1293 struct sge *s = &adapter->sge;
1296 if (rp->rx_pending > MAX_RX_BUFFERS ||
1297 rp->rx_jumbo_pending ||
1298 rp->tx_pending > MAX_TXQ_ENTRIES ||
1299 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1300 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1301 rp->rx_pending < MIN_FL_ENTRIES ||
1302 rp->tx_pending < MIN_TXQ_ENTRIES)
1305 if (adapter->flags & FULL_INIT_DONE)
1308 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1309 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1310 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1311 s->ethtxq[qs].q.size = rp->tx_pending;
1317 * Return the interrupt holdoff timer and count for the first Queue Set on the
1318 * device. Our extension ioctl() (the cxgbtool interface) allows the
1319 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1321 static int cxgb4vf_get_coalesce(struct net_device *dev,
1322 struct ethtool_coalesce *coalesce)
1324 const struct port_info *pi = netdev_priv(dev);
1325 const struct adapter *adapter = pi->adapter;
1326 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1328 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1329 coalesce->rx_max_coalesced_frames =
1330 ((rspq->intr_params & QINTR_CNT_EN)
1331 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1337 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1338 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1339 * the interrupt holdoff timer on any of the device's Queue Sets.
1341 static int cxgb4vf_set_coalesce(struct net_device *dev,
1342 struct ethtool_coalesce *coalesce)
1344 const struct port_info *pi = netdev_priv(dev);
1345 struct adapter *adapter = pi->adapter;
1347 return set_rxq_intr_params(adapter,
1348 &adapter->sge.ethrxq[pi->first_qset].rspq,
1349 coalesce->rx_coalesce_usecs,
1350 coalesce->rx_max_coalesced_frames);
1354 * Report current port link pause parameter settings.
1356 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1357 struct ethtool_pauseparam *pauseparam)
1359 struct port_info *pi = netdev_priv(dev);
1361 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1362 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1363 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1367 * Identify the port by blinking the port's LED.
1369 static int cxgb4vf_phys_id(struct net_device *dev,
1370 enum ethtool_phys_id_state state)
1373 struct port_info *pi = netdev_priv(dev);
1375 if (state == ETHTOOL_ID_ACTIVE)
1377 else if (state == ETHTOOL_ID_INACTIVE)
1382 return t4vf_identify_port(pi->adapter, pi->viid, val);
1386 * Port stats maintained per queue of the port.
1388 struct queue_port_stats {
1399 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1400 * these need to match the order of statistics returned by
1401 * t4vf_get_port_stats().
1403 static const char stats_strings[][ETH_GSTRING_LEN] = {
1405 * These must match the layout of the t4vf_port_stats structure.
1407 "TxBroadcastBytes ",
1408 "TxBroadcastFrames ",
1409 "TxMulticastBytes ",
1410 "TxMulticastFrames ",
1416 "RxBroadcastBytes ",
1417 "RxBroadcastFrames ",
1418 "RxMulticastBytes ",
1419 "RxMulticastFrames ",
1425 * These are accumulated per-queue statistics and must match the
1426 * order of the fields in the queue_port_stats structure.
1438 * Return the number of statistics in the specified statistics set.
1440 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1444 return ARRAY_SIZE(stats_strings);
1452 * Return the strings for the specified statistics set.
1454 static void cxgb4vf_get_strings(struct net_device *dev,
1460 memcpy(data, stats_strings, sizeof(stats_strings));
1466 * Small utility routine to accumulate queue statistics across the queues of
1469 static void collect_sge_port_stats(const struct adapter *adapter,
1470 const struct port_info *pi,
1471 struct queue_port_stats *stats)
1473 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1474 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1477 memset(stats, 0, sizeof(*stats));
1478 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1479 stats->tso += txq->tso;
1480 stats->tx_csum += txq->tx_cso;
1481 stats->rx_csum += rxq->stats.rx_cso;
1482 stats->vlan_ex += rxq->stats.vlan_ex;
1483 stats->vlan_ins += txq->vlan_ins;
1484 stats->lro_pkts += rxq->stats.lro_pkts;
1485 stats->lro_merged += rxq->stats.lro_merged;
1490 * Return the ETH_SS_STATS statistics set.
1492 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1493 struct ethtool_stats *stats,
1496 struct port_info *pi = netdev2pinfo(dev);
1497 struct adapter *adapter = pi->adapter;
1498 int err = t4vf_get_port_stats(adapter, pi->pidx,
1499 (struct t4vf_port_stats *)data);
1501 memset(data, 0, sizeof(struct t4vf_port_stats));
1503 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1504 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1508 * Return the size of our register map.
1510 static int cxgb4vf_get_regs_len(struct net_device *dev)
1512 return T4VF_REGMAP_SIZE;
1516 * Dump a block of registers, start to end inclusive, into a buffer.
1518 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1519 unsigned int start, unsigned int end)
1521 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1523 for ( ; start <= end; start += sizeof(u32)) {
1525 * Avoid reading the Mailbox Control register since that
1526 * can trigger a Mailbox Ownership Arbitration cycle and
1527 * interfere with communication with the firmware.
1529 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1532 *bp++ = t4_read_reg(adapter, start);
1537 * Copy our entire register map into the provided buffer.
1539 static void cxgb4vf_get_regs(struct net_device *dev,
1540 struct ethtool_regs *regs,
1543 struct adapter *adapter = netdev2adap(dev);
1545 regs->version = mk_adap_vers(adapter);
1548 * Fill in register buffer with our register map.
1550 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1552 reg_block_dump(adapter, regbuf,
1553 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1554 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1555 reg_block_dump(adapter, regbuf,
1556 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1557 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1559 /* T5 adds new registers in the PL Register map.
1561 reg_block_dump(adapter, regbuf,
1562 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1563 T4VF_PL_BASE_ADDR + (is_t4(adapter->params.chip)
1564 ? A_PL_VF_WHOAMI : A_PL_VF_REVISION));
1565 reg_block_dump(adapter, regbuf,
1566 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1567 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1569 reg_block_dump(adapter, regbuf,
1570 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1571 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1575 * Report current Wake On LAN settings.
1577 static void cxgb4vf_get_wol(struct net_device *dev,
1578 struct ethtool_wolinfo *wol)
1582 memset(&wol->sopass, 0, sizeof(wol->sopass));
1586 * TCP Segmentation Offload flags which we support.
1588 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1590 static const struct ethtool_ops cxgb4vf_ethtool_ops = {
1591 .get_settings = cxgb4vf_get_settings,
1592 .get_drvinfo = cxgb4vf_get_drvinfo,
1593 .get_msglevel = cxgb4vf_get_msglevel,
1594 .set_msglevel = cxgb4vf_set_msglevel,
1595 .get_ringparam = cxgb4vf_get_ringparam,
1596 .set_ringparam = cxgb4vf_set_ringparam,
1597 .get_coalesce = cxgb4vf_get_coalesce,
1598 .set_coalesce = cxgb4vf_set_coalesce,
1599 .get_pauseparam = cxgb4vf_get_pauseparam,
1600 .get_link = ethtool_op_get_link,
1601 .get_strings = cxgb4vf_get_strings,
1602 .set_phys_id = cxgb4vf_phys_id,
1603 .get_sset_count = cxgb4vf_get_sset_count,
1604 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1605 .get_regs_len = cxgb4vf_get_regs_len,
1606 .get_regs = cxgb4vf_get_regs,
1607 .get_wol = cxgb4vf_get_wol,
1611 * /sys/kernel/debug/cxgb4vf support code and data.
1612 * ================================================
1616 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1620 static int sge_qinfo_show(struct seq_file *seq, void *v)
1622 struct adapter *adapter = seq->private;
1623 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1624 int qs, r = (uintptr_t)v - 1;
1627 seq_putc(seq, '\n');
1629 #define S3(fmt_spec, s, v) \
1631 seq_printf(seq, "%-12s", s); \
1632 for (qs = 0; qs < n; ++qs) \
1633 seq_printf(seq, " %16" fmt_spec, v); \
1634 seq_putc(seq, '\n'); \
1636 #define S(s, v) S3("s", s, v)
1637 #define T(s, v) S3("u", s, txq[qs].v)
1638 #define R(s, v) S3("u", s, rxq[qs].v)
1640 if (r < eth_entries) {
1641 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1642 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1643 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1645 S("QType:", "Ethernet");
1647 (rxq[qs].rspq.netdev
1648 ? rxq[qs].rspq.netdev->name
1651 (rxq[qs].rspq.netdev
1652 ? ((struct port_info *)
1653 netdev_priv(rxq[qs].rspq.netdev))->port_id
1655 T("TxQ ID:", q.abs_id);
1656 T("TxQ size:", q.size);
1657 T("TxQ inuse:", q.in_use);
1658 T("TxQ PIdx:", q.pidx);
1659 T("TxQ CIdx:", q.cidx);
1660 R("RspQ ID:", rspq.abs_id);
1661 R("RspQ size:", rspq.size);
1662 R("RspQE size:", rspq.iqe_len);
1663 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1664 S3("u", "Intr pktcnt:",
1665 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1666 R("RspQ CIdx:", rspq.cidx);
1667 R("RspQ Gen:", rspq.gen);
1668 R("FL ID:", fl.abs_id);
1669 R("FL size:", fl.size - MIN_FL_RESID);
1670 R("FL avail:", fl.avail);
1671 R("FL PIdx:", fl.pidx);
1672 R("FL CIdx:", fl.cidx);
1678 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1680 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1681 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1682 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1683 qtimer_val(adapter, evtq));
1684 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1685 adapter->sge.counter_val[evtq->pktcnt_idx]);
1686 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1687 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1688 } else if (r == 1) {
1689 const struct sge_rspq *intrq = &adapter->sge.intrq;
1691 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1692 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1693 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1694 qtimer_val(adapter, intrq));
1695 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1696 adapter->sge.counter_val[intrq->pktcnt_idx]);
1697 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1698 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1710 * Return the number of "entries" in our "file". We group the multi-Queue
1711 * sections with QPL Queue Sets per "entry". The sections of the output are:
1713 * Ethernet RX/TX Queue Sets
1714 * Firmware Event Queue
1715 * Forwarded Interrupt Queue (if in MSI mode)
1717 static int sge_queue_entries(const struct adapter *adapter)
1719 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1720 ((adapter->flags & USING_MSI) != 0);
1723 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1725 int entries = sge_queue_entries(seq->private);
1727 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1730 static void sge_queue_stop(struct seq_file *seq, void *v)
1734 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
1736 int entries = sge_queue_entries(seq->private);
1739 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1742 static const struct seq_operations sge_qinfo_seq_ops = {
1743 .start = sge_queue_start,
1744 .next = sge_queue_next,
1745 .stop = sge_queue_stop,
1746 .show = sge_qinfo_show
1749 static int sge_qinfo_open(struct inode *inode, struct file *file)
1751 int res = seq_open(file, &sge_qinfo_seq_ops);
1754 struct seq_file *seq = file->private_data;
1755 seq->private = inode->i_private;
1760 static const struct file_operations sge_qinfo_debugfs_fops = {
1761 .owner = THIS_MODULE,
1762 .open = sge_qinfo_open,
1764 .llseek = seq_lseek,
1765 .release = seq_release,
1769 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
1773 static int sge_qstats_show(struct seq_file *seq, void *v)
1775 struct adapter *adapter = seq->private;
1776 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1777 int qs, r = (uintptr_t)v - 1;
1780 seq_putc(seq, '\n');
1782 #define S3(fmt, s, v) \
1784 seq_printf(seq, "%-16s", s); \
1785 for (qs = 0; qs < n; ++qs) \
1786 seq_printf(seq, " %8" fmt, v); \
1787 seq_putc(seq, '\n'); \
1789 #define S(s, v) S3("s", s, v)
1791 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
1792 #define T(s, v) T3("lu", s, v)
1794 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
1795 #define R(s, v) R3("lu", s, v)
1797 if (r < eth_entries) {
1798 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1799 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1800 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1802 S("QType:", "Ethernet");
1804 (rxq[qs].rspq.netdev
1805 ? rxq[qs].rspq.netdev->name
1807 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
1808 R("RxPackets:", stats.pkts);
1809 R("RxCSO:", stats.rx_cso);
1810 R("VLANxtract:", stats.vlan_ex);
1811 R("LROmerged:", stats.lro_merged);
1812 R("LROpackets:", stats.lro_pkts);
1813 R("RxDrops:", stats.rx_drops);
1815 T("TxCSO:", tx_cso);
1816 T("VLANins:", vlan_ins);
1817 T("TxQFull:", q.stops);
1818 T("TxQRestarts:", q.restarts);
1819 T("TxMapErr:", mapping_err);
1820 R("FLAllocErr:", fl.alloc_failed);
1821 R("FLLrgAlcErr:", fl.large_alloc_failed);
1822 R("FLStarving:", fl.starving);
1828 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1830 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
1831 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1832 evtq->unhandled_irqs);
1833 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
1834 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
1835 } else if (r == 1) {
1836 const struct sge_rspq *intrq = &adapter->sge.intrq;
1838 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
1839 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1840 intrq->unhandled_irqs);
1841 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
1842 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
1856 * Return the number of "entries" in our "file". We group the multi-Queue
1857 * sections with QPL Queue Sets per "entry". The sections of the output are:
1859 * Ethernet RX/TX Queue Sets
1860 * Firmware Event Queue
1861 * Forwarded Interrupt Queue (if in MSI mode)
1863 static int sge_qstats_entries(const struct adapter *adapter)
1865 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1866 ((adapter->flags & USING_MSI) != 0);
1869 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
1871 int entries = sge_qstats_entries(seq->private);
1873 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1876 static void sge_qstats_stop(struct seq_file *seq, void *v)
1880 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
1882 int entries = sge_qstats_entries(seq->private);
1885 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1888 static const struct seq_operations sge_qstats_seq_ops = {
1889 .start = sge_qstats_start,
1890 .next = sge_qstats_next,
1891 .stop = sge_qstats_stop,
1892 .show = sge_qstats_show
1895 static int sge_qstats_open(struct inode *inode, struct file *file)
1897 int res = seq_open(file, &sge_qstats_seq_ops);
1900 struct seq_file *seq = file->private_data;
1901 seq->private = inode->i_private;
1906 static const struct file_operations sge_qstats_proc_fops = {
1907 .owner = THIS_MODULE,
1908 .open = sge_qstats_open,
1910 .llseek = seq_lseek,
1911 .release = seq_release,
1915 * Show PCI-E SR-IOV Virtual Function Resource Limits.
1917 static int resources_show(struct seq_file *seq, void *v)
1919 struct adapter *adapter = seq->private;
1920 struct vf_resources *vfres = &adapter->params.vfres;
1922 #define S(desc, fmt, var) \
1923 seq_printf(seq, "%-60s " fmt "\n", \
1924 desc " (" #var "):", vfres->var)
1926 S("Virtual Interfaces", "%d", nvi);
1927 S("Egress Queues", "%d", neq);
1928 S("Ethernet Control", "%d", nethctrl);
1929 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
1930 S("Ingress Queues", "%d", niq);
1931 S("Traffic Class", "%d", tc);
1932 S("Port Access Rights Mask", "%#x", pmask);
1933 S("MAC Address Filters", "%d", nexactf);
1934 S("Firmware Command Read Capabilities", "%#x", r_caps);
1935 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
1942 static int resources_open(struct inode *inode, struct file *file)
1944 return single_open(file, resources_show, inode->i_private);
1947 static const struct file_operations resources_proc_fops = {
1948 .owner = THIS_MODULE,
1949 .open = resources_open,
1951 .llseek = seq_lseek,
1952 .release = single_release,
1956 * Show Virtual Interfaces.
1958 static int interfaces_show(struct seq_file *seq, void *v)
1960 if (v == SEQ_START_TOKEN) {
1961 seq_puts(seq, "Interface Port VIID\n");
1963 struct adapter *adapter = seq->private;
1964 int pidx = (uintptr_t)v - 2;
1965 struct net_device *dev = adapter->port[pidx];
1966 struct port_info *pi = netdev_priv(dev);
1968 seq_printf(seq, "%9s %4d %#5x\n",
1969 dev->name, pi->port_id, pi->viid);
1974 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
1976 return pos <= adapter->params.nports
1977 ? (void *)(uintptr_t)(pos + 1)
1981 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
1984 ? interfaces_get_idx(seq->private, *pos)
1988 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
1991 return interfaces_get_idx(seq->private, *pos);
1994 static void interfaces_stop(struct seq_file *seq, void *v)
1998 static const struct seq_operations interfaces_seq_ops = {
1999 .start = interfaces_start,
2000 .next = interfaces_next,
2001 .stop = interfaces_stop,
2002 .show = interfaces_show
2005 static int interfaces_open(struct inode *inode, struct file *file)
2007 int res = seq_open(file, &interfaces_seq_ops);
2010 struct seq_file *seq = file->private_data;
2011 seq->private = inode->i_private;
2016 static const struct file_operations interfaces_proc_fops = {
2017 .owner = THIS_MODULE,
2018 .open = interfaces_open,
2020 .llseek = seq_lseek,
2021 .release = seq_release,
2025 * /sys/kernel/debugfs/cxgb4vf/ files list.
2027 struct cxgb4vf_debugfs_entry {
2028 const char *name; /* name of debugfs node */
2029 umode_t mode; /* file system mode */
2030 const struct file_operations *fops;
2033 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
2034 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
2035 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
2036 { "resources", S_IRUGO, &resources_proc_fops },
2037 { "interfaces", S_IRUGO, &interfaces_proc_fops },
2041 * Module and device initialization and cleanup code.
2042 * ==================================================
2046 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2047 * directory (debugfs_root) has already been set up.
2049 static int setup_debugfs(struct adapter *adapter)
2053 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2056 * Debugfs support is best effort.
2058 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2059 (void)debugfs_create_file(debugfs_files[i].name,
2060 debugfs_files[i].mode,
2061 adapter->debugfs_root,
2063 debugfs_files[i].fops);
2069 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2070 * it to our caller to tear down the directory (debugfs_root).
2072 static void cleanup_debugfs(struct adapter *adapter)
2074 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2077 * Unlike our sister routine cleanup_proc(), we don't need to remove
2078 * individual entries because a call will be made to
2079 * debugfs_remove_recursive(). We just need to clean up any ancillary
2086 * Perform early "adapter" initialization. This is where we discover what
2087 * adapter parameters we're going to be using and initialize basic adapter
2090 static int adap_init0(struct adapter *adapter)
2092 struct vf_resources *vfres = &adapter->params.vfres;
2093 struct sge_params *sge_params = &adapter->params.sge;
2094 struct sge *s = &adapter->sge;
2095 unsigned int ethqsets;
2098 unsigned int chipid;
2101 * Wait for the device to become ready before proceeding ...
2103 err = t4vf_wait_dev_ready(adapter);
2105 dev_err(adapter->pdev_dev, "device didn't become ready:"
2111 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2112 * 2.6.31 and later we can't call pci_reset_function() in order to
2113 * issue an FLR because of a self- deadlock on the device semaphore.
2114 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2115 * cases where they're needed -- for instance, some versions of KVM
2116 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2117 * use the firmware based reset in order to reset any per function
2120 err = t4vf_fw_reset(adapter);
2122 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2126 adapter->params.chip = 0;
2127 switch (adapter->pdev->device >> 12) {
2129 adapter->params.chip = CHELSIO_CHIP_CODE(CHELSIO_T4, 0);
2132 chipid = G_REV(t4_read_reg(adapter, A_PL_VF_REV));
2133 adapter->params.chip |= CHELSIO_CHIP_CODE(CHELSIO_T5, chipid);
2138 * Grab basic operational parameters. These will predominantly have
2139 * been set up by the Physical Function Driver or will be hard coded
2140 * into the adapter. We just have to live with them ... Note that
2141 * we _must_ get our VPD parameters before our SGE parameters because
2142 * we need to know the adapter's core clock from the VPD in order to
2143 * properly decode the SGE Timer Values.
2145 err = t4vf_get_dev_params(adapter);
2147 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2148 " device parameters: err=%d\n", err);
2151 err = t4vf_get_vpd_params(adapter);
2153 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2154 " VPD parameters: err=%d\n", err);
2157 err = t4vf_get_sge_params(adapter);
2159 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2160 " SGE parameters: err=%d\n", err);
2163 err = t4vf_get_rss_glb_config(adapter);
2165 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2166 " RSS parameters: err=%d\n", err);
2169 if (adapter->params.rss.mode !=
2170 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2171 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2172 " mode %d\n", adapter->params.rss.mode);
2175 err = t4vf_sge_init(adapter);
2177 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2182 /* If we're running on newer firmware, let it know that we're
2183 * prepared to deal with encapsulated CPL messages. Older
2184 * firmware won't understand this and we'll just get
2185 * unencapsulated messages ...
2187 param = FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) |
2188 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_CPLFW4MSG_ENCAP);
2190 (void) t4vf_set_params(adapter, 1, ¶m, &val);
2193 * Retrieve our RX interrupt holdoff timer values and counter
2194 * threshold values from the SGE parameters.
2196 s->timer_val[0] = core_ticks_to_us(adapter,
2197 TIMERVALUE0_GET(sge_params->sge_timer_value_0_and_1));
2198 s->timer_val[1] = core_ticks_to_us(adapter,
2199 TIMERVALUE1_GET(sge_params->sge_timer_value_0_and_1));
2200 s->timer_val[2] = core_ticks_to_us(adapter,
2201 TIMERVALUE0_GET(sge_params->sge_timer_value_2_and_3));
2202 s->timer_val[3] = core_ticks_to_us(adapter,
2203 TIMERVALUE1_GET(sge_params->sge_timer_value_2_and_3));
2204 s->timer_val[4] = core_ticks_to_us(adapter,
2205 TIMERVALUE0_GET(sge_params->sge_timer_value_4_and_5));
2206 s->timer_val[5] = core_ticks_to_us(adapter,
2207 TIMERVALUE1_GET(sge_params->sge_timer_value_4_and_5));
2210 THRESHOLD_0_GET(sge_params->sge_ingress_rx_threshold);
2212 THRESHOLD_1_GET(sge_params->sge_ingress_rx_threshold);
2214 THRESHOLD_2_GET(sge_params->sge_ingress_rx_threshold);
2216 THRESHOLD_3_GET(sge_params->sge_ingress_rx_threshold);
2219 * Grab our Virtual Interface resource allocation, extract the
2220 * features that we're interested in and do a bit of sanity testing on
2223 err = t4vf_get_vfres(adapter);
2225 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2226 " resources: err=%d\n", err);
2231 * The number of "ports" which we support is equal to the number of
2232 * Virtual Interfaces with which we've been provisioned.
2234 adapter->params.nports = vfres->nvi;
2235 if (adapter->params.nports > MAX_NPORTS) {
2236 dev_warn(adapter->pdev_dev, "only using %d of %d allowed"
2237 " virtual interfaces\n", MAX_NPORTS,
2238 adapter->params.nports);
2239 adapter->params.nports = MAX_NPORTS;
2243 * We need to reserve a number of the ingress queues with Free List
2244 * and Interrupt capabilities for special interrupt purposes (like
2245 * asynchronous firmware messages, or forwarded interrupts if we're
2246 * using MSI). The rest of the FL/Intr-capable ingress queues will be
2247 * matched up one-for-one with Ethernet/Control egress queues in order
2248 * to form "Queue Sets" which will be aportioned between the "ports".
2249 * For each Queue Set, we'll need the ability to allocate two Egress
2250 * Contexts -- one for the Ingress Queue Free List and one for the TX
2253 ethqsets = vfres->niqflint - INGQ_EXTRAS;
2254 if (vfres->nethctrl != ethqsets) {
2255 dev_warn(adapter->pdev_dev, "unequal number of [available]"
2256 " ingress/egress queues (%d/%d); using minimum for"
2257 " number of Queue Sets\n", ethqsets, vfres->nethctrl);
2258 ethqsets = min(vfres->nethctrl, ethqsets);
2260 if (vfres->neq < ethqsets*2) {
2261 dev_warn(adapter->pdev_dev, "Not enough Egress Contexts (%d)"
2262 " to support Queue Sets (%d); reducing allowed Queue"
2263 " Sets\n", vfres->neq, ethqsets);
2264 ethqsets = vfres->neq/2;
2266 if (ethqsets > MAX_ETH_QSETS) {
2267 dev_warn(adapter->pdev_dev, "only using %d of %d allowed Queue"
2268 " Sets\n", MAX_ETH_QSETS, adapter->sge.max_ethqsets);
2269 ethqsets = MAX_ETH_QSETS;
2271 if (vfres->niq != 0 || vfres->neq > ethqsets*2) {
2272 dev_warn(adapter->pdev_dev, "unused resources niq/neq (%d/%d)"
2273 " ignored\n", vfres->niq, vfres->neq - ethqsets*2);
2275 adapter->sge.max_ethqsets = ethqsets;
2278 * Check for various parameter sanity issues. Most checks simply
2279 * result in us using fewer resources than our provissioning but we
2280 * do need at least one "port" with which to work ...
2282 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2283 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2284 " virtual interfaces (too few Queue Sets)\n",
2285 adapter->sge.max_ethqsets, adapter->params.nports);
2286 adapter->params.nports = adapter->sge.max_ethqsets;
2288 if (adapter->params.nports == 0) {
2289 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2296 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2297 u8 pkt_cnt_idx, unsigned int size,
2298 unsigned int iqe_size)
2300 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
2301 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0));
2302 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2305 rspq->iqe_len = iqe_size;
2310 * Perform default configuration of DMA queues depending on the number and
2311 * type of ports we found and the number of available CPUs. Most settings can
2312 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2313 * being brought up for the first time.
2315 static void cfg_queues(struct adapter *adapter)
2317 struct sge *s = &adapter->sge;
2318 int q10g, n10g, qidx, pidx, qs;
2322 * We should not be called till we know how many Queue Sets we can
2323 * support. In particular, this means that we need to know what kind
2324 * of interrupts we'll be using ...
2326 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2329 * Count the number of 10GbE Virtual Interfaces that we have.
2332 for_each_port(adapter, pidx)
2333 n10g += is_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2336 * We default to 1 queue per non-10G port and up to # of cores queues
2342 int n1g = (adapter->params.nports - n10g);
2343 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2344 if (q10g > num_online_cpus())
2345 q10g = num_online_cpus();
2349 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2350 * The layout will be established in setup_sge_queues() when the
2351 * adapter is brough up for the first time.
2354 for_each_port(adapter, pidx) {
2355 struct port_info *pi = adap2pinfo(adapter, pidx);
2357 pi->first_qset = qidx;
2358 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
2364 * The Ingress Queue Entry Size for our various Response Queues needs
2365 * to be big enough to accommodate the largest message we can receive
2366 * from the chip/firmware; which is 64 bytes ...
2371 * Set up default Queue Set parameters ... Start off with the
2372 * shortest interrupt holdoff timer.
2374 for (qs = 0; qs < s->max_ethqsets; qs++) {
2375 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2376 struct sge_eth_txq *txq = &s->ethtxq[qs];
2378 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2384 * The firmware event queue is used for link state changes and
2385 * notifications of TX DMA completions.
2387 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2390 * The forwarded interrupt queue is used when we're in MSI interrupt
2391 * mode. In this mode all interrupts associated with RX queues will
2392 * be forwarded to a single queue which we'll associate with our MSI
2393 * interrupt vector. The messages dropped in the forwarded interrupt
2394 * queue will indicate which ingress queue needs servicing ... This
2395 * queue needs to be large enough to accommodate all of the ingress
2396 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2397 * from equalling the CIDX if every ingress queue has an outstanding
2398 * interrupt). The queue doesn't need to be any larger because no
2399 * ingress queue will ever have more than one outstanding interrupt at
2402 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2407 * Reduce the number of Ethernet queues across all ports to at most n.
2408 * n provides at least one queue per port.
2410 static void reduce_ethqs(struct adapter *adapter, int n)
2413 struct port_info *pi;
2416 * While we have too many active Ether Queue Sets, interate across the
2417 * "ports" and reduce their individual Queue Set allocations.
2419 BUG_ON(n < adapter->params.nports);
2420 while (n < adapter->sge.ethqsets)
2421 for_each_port(adapter, i) {
2422 pi = adap2pinfo(adapter, i);
2423 if (pi->nqsets > 1) {
2425 adapter->sge.ethqsets--;
2426 if (adapter->sge.ethqsets <= n)
2432 * Reassign the starting Queue Sets for each of the "ports" ...
2435 for_each_port(adapter, i) {
2436 pi = adap2pinfo(adapter, i);
2443 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2444 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2445 * need. Minimally we need one for every Virtual Interface plus those needed
2446 * for our "extras". Note that this process may lower the maximum number of
2447 * allowed Queue Sets ...
2449 static int enable_msix(struct adapter *adapter)
2451 int i, want, need, nqsets;
2452 struct msix_entry entries[MSIX_ENTRIES];
2453 struct sge *s = &adapter->sge;
2455 for (i = 0; i < MSIX_ENTRIES; ++i)
2456 entries[i].entry = i;
2459 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2460 * plus those needed for our "extras" (for example, the firmware
2461 * message queue). We _need_ at least one "Queue Set" per Virtual
2462 * Interface plus those needed for our "extras". So now we get to see
2463 * if the song is right ...
2465 want = s->max_ethqsets + MSIX_EXTRAS;
2466 need = adapter->params.nports + MSIX_EXTRAS;
2468 want = pci_enable_msix_range(adapter->pdev, entries, need, want);
2472 nqsets = want - MSIX_EXTRAS;
2473 if (nqsets < s->max_ethqsets) {
2474 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2475 " for %d Queue Sets\n", nqsets);
2476 s->max_ethqsets = nqsets;
2477 if (nqsets < s->ethqsets)
2478 reduce_ethqs(adapter, nqsets);
2480 for (i = 0; i < want; ++i)
2481 adapter->msix_info[i].vec = entries[i].vector;
2486 static const struct net_device_ops cxgb4vf_netdev_ops = {
2487 .ndo_open = cxgb4vf_open,
2488 .ndo_stop = cxgb4vf_stop,
2489 .ndo_start_xmit = t4vf_eth_xmit,
2490 .ndo_get_stats = cxgb4vf_get_stats,
2491 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2492 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2493 .ndo_validate_addr = eth_validate_addr,
2494 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2495 .ndo_change_mtu = cxgb4vf_change_mtu,
2496 .ndo_fix_features = cxgb4vf_fix_features,
2497 .ndo_set_features = cxgb4vf_set_features,
2498 #ifdef CONFIG_NET_POLL_CONTROLLER
2499 .ndo_poll_controller = cxgb4vf_poll_controller,
2504 * "Probe" a device: initialize a device and construct all kernel and driver
2505 * state needed to manage the device. This routine is called "init_one" in
2508 static int cxgb4vf_pci_probe(struct pci_dev *pdev,
2509 const struct pci_device_id *ent)
2514 struct adapter *adapter;
2515 struct port_info *pi;
2516 struct net_device *netdev;
2519 * Print our driver banner the first time we're called to initialize a
2522 pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION);
2525 * Initialize generic PCI device state.
2527 err = pci_enable_device(pdev);
2529 dev_err(&pdev->dev, "cannot enable PCI device\n");
2534 * Reserve PCI resources for the device. If we can't get them some
2535 * other driver may have already claimed the device ...
2537 err = pci_request_regions(pdev, KBUILD_MODNAME);
2539 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2540 goto err_disable_device;
2544 * Set up our DMA mask: try for 64-bit address masking first and
2545 * fall back to 32-bit if we can't get 64 bits ...
2547 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2549 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2551 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2552 " coherent allocations\n");
2553 goto err_release_regions;
2557 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2559 dev_err(&pdev->dev, "no usable DMA configuration\n");
2560 goto err_release_regions;
2566 * Enable bus mastering for the device ...
2568 pci_set_master(pdev);
2571 * Allocate our adapter data structure and attach it to the device.
2573 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2576 goto err_release_regions;
2578 pci_set_drvdata(pdev, adapter);
2579 adapter->pdev = pdev;
2580 adapter->pdev_dev = &pdev->dev;
2583 * Initialize SMP data synchronization resources.
2585 spin_lock_init(&adapter->stats_lock);
2588 * Map our I/O registers in BAR0.
2590 adapter->regs = pci_ioremap_bar(pdev, 0);
2591 if (!adapter->regs) {
2592 dev_err(&pdev->dev, "cannot map device registers\n");
2594 goto err_free_adapter;
2598 * Initialize adapter level features.
2600 adapter->name = pci_name(pdev);
2601 adapter->msg_enable = dflt_msg_enable;
2602 err = adap_init0(adapter);
2607 * Allocate our "adapter ports" and stitch everything together.
2609 pmask = adapter->params.vfres.pmask;
2610 for_each_port(adapter, pidx) {
2614 * We simplistically allocate our virtual interfaces
2615 * sequentially across the port numbers to which we have
2616 * access rights. This should be configurable in some manner
2621 port_id = ffs(pmask) - 1;
2622 pmask &= ~(1 << port_id);
2623 viid = t4vf_alloc_vi(adapter, port_id);
2625 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2626 " err=%d\n", port_id, viid);
2632 * Allocate our network device and stitch things together.
2634 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2636 if (netdev == NULL) {
2637 t4vf_free_vi(adapter, viid);
2641 adapter->port[pidx] = netdev;
2642 SET_NETDEV_DEV(netdev, &pdev->dev);
2643 pi = netdev_priv(netdev);
2644 pi->adapter = adapter;
2646 pi->port_id = port_id;
2650 * Initialize the starting state of our "port" and register
2653 pi->xact_addr_filt = -1;
2654 netif_carrier_off(netdev);
2655 netdev->irq = pdev->irq;
2657 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
2658 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2659 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXCSUM;
2660 netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
2661 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2663 netdev->features = netdev->hw_features |
2664 NETIF_F_HW_VLAN_CTAG_TX;
2666 netdev->features |= NETIF_F_HIGHDMA;
2668 netdev->priv_flags |= IFF_UNICAST_FLT;
2670 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2671 netdev->ethtool_ops = &cxgb4vf_ethtool_ops;
2674 * Initialize the hardware/software state for the port.
2676 err = t4vf_port_init(adapter, pidx);
2678 dev_err(&pdev->dev, "cannot initialize port %d\n",
2685 * The "card" is now ready to go. If any errors occur during device
2686 * registration we do not fail the whole "card" but rather proceed
2687 * only with the ports we manage to register successfully. However we
2688 * must register at least one net device.
2690 for_each_port(adapter, pidx) {
2691 netdev = adapter->port[pidx];
2695 err = register_netdev(netdev);
2697 dev_warn(&pdev->dev, "cannot register net device %s,"
2698 " skipping\n", netdev->name);
2702 set_bit(pidx, &adapter->registered_device_map);
2704 if (adapter->registered_device_map == 0) {
2705 dev_err(&pdev->dev, "could not register any net devices\n");
2710 * Set up our debugfs entries.
2712 if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) {
2713 adapter->debugfs_root =
2714 debugfs_create_dir(pci_name(pdev),
2715 cxgb4vf_debugfs_root);
2716 if (IS_ERR_OR_NULL(adapter->debugfs_root))
2717 dev_warn(&pdev->dev, "could not create debugfs"
2720 setup_debugfs(adapter);
2724 * See what interrupts we'll be using. If we've been configured to
2725 * use MSI-X interrupts, try to enable them but fall back to using
2726 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
2727 * get MSI interrupts we bail with the error.
2729 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
2730 adapter->flags |= USING_MSIX;
2732 err = pci_enable_msi(pdev);
2734 dev_err(&pdev->dev, "Unable to allocate %s interrupts;"
2736 msi == MSI_MSIX ? "MSI-X or MSI" : "MSI", err);
2737 goto err_free_debugfs;
2739 adapter->flags |= USING_MSI;
2743 * Now that we know how many "ports" we have and what their types are,
2744 * and how many Queue Sets we can support, we can configure our queue
2747 cfg_queues(adapter);
2750 * Print a short notice on the existence and configuration of the new
2751 * VF network device ...
2753 for_each_port(adapter, pidx) {
2754 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
2755 adapter->port[pidx]->name,
2756 (adapter->flags & USING_MSIX) ? "MSI-X" :
2757 (adapter->flags & USING_MSI) ? "MSI" : "");
2766 * Error recovery and exit code. Unwind state that's been created
2767 * so far and return the error.
2771 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
2772 cleanup_debugfs(adapter);
2773 debugfs_remove_recursive(adapter->debugfs_root);
2777 for_each_port(adapter, pidx) {
2778 netdev = adapter->port[pidx];
2781 pi = netdev_priv(netdev);
2782 t4vf_free_vi(adapter, pi->viid);
2783 if (test_bit(pidx, &adapter->registered_device_map))
2784 unregister_netdev(netdev);
2785 free_netdev(netdev);
2789 iounmap(adapter->regs);
2794 err_release_regions:
2795 pci_release_regions(pdev);
2796 pci_clear_master(pdev);
2799 pci_disable_device(pdev);
2805 * "Remove" a device: tear down all kernel and driver state created in the
2806 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
2807 * that this is called "remove_one" in the PF Driver.)
2809 static void cxgb4vf_pci_remove(struct pci_dev *pdev)
2811 struct adapter *adapter = pci_get_drvdata(pdev);
2814 * Tear down driver state associated with device.
2820 * Stop all of our activity. Unregister network port,
2821 * disable interrupts, etc.
2823 for_each_port(adapter, pidx)
2824 if (test_bit(pidx, &adapter->registered_device_map))
2825 unregister_netdev(adapter->port[pidx]);
2826 t4vf_sge_stop(adapter);
2827 if (adapter->flags & USING_MSIX) {
2828 pci_disable_msix(adapter->pdev);
2829 adapter->flags &= ~USING_MSIX;
2830 } else if (adapter->flags & USING_MSI) {
2831 pci_disable_msi(adapter->pdev);
2832 adapter->flags &= ~USING_MSI;
2836 * Tear down our debugfs entries.
2838 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
2839 cleanup_debugfs(adapter);
2840 debugfs_remove_recursive(adapter->debugfs_root);
2844 * Free all of the various resources which we've acquired ...
2846 t4vf_free_sge_resources(adapter);
2847 for_each_port(adapter, pidx) {
2848 struct net_device *netdev = adapter->port[pidx];
2849 struct port_info *pi;
2854 pi = netdev_priv(netdev);
2855 t4vf_free_vi(adapter, pi->viid);
2856 free_netdev(netdev);
2858 iounmap(adapter->regs);
2863 * Disable the device and release its PCI resources.
2865 pci_disable_device(pdev);
2866 pci_clear_master(pdev);
2867 pci_release_regions(pdev);
2871 * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt
2874 static void cxgb4vf_pci_shutdown(struct pci_dev *pdev)
2876 struct adapter *adapter;
2879 adapter = pci_get_drvdata(pdev);
2883 /* Disable all Virtual Interfaces. This will shut down the
2884 * delivery of all ingress packets into the chip for these
2885 * Virtual Interfaces.
2887 for_each_port(adapter, pidx)
2888 if (test_bit(pidx, &adapter->registered_device_map))
2889 unregister_netdev(adapter->port[pidx]);
2891 /* Free up all Queues which will prevent further DMA and
2892 * Interrupts allowing various internal pathways to drain.
2894 t4vf_sge_stop(adapter);
2895 if (adapter->flags & USING_MSIX) {
2896 pci_disable_msix(adapter->pdev);
2897 adapter->flags &= ~USING_MSIX;
2898 } else if (adapter->flags & USING_MSI) {
2899 pci_disable_msi(adapter->pdev);
2900 adapter->flags &= ~USING_MSI;
2904 * Free up all Queues which will prevent further DMA and
2905 * Interrupts allowing various internal pathways to drain.
2907 t4vf_free_sge_resources(adapter);
2908 pci_set_drvdata(pdev, NULL);
2912 * PCI Device registration data structures.
2914 #define CH_DEVICE(devid) \
2915 { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }
2917 static const struct pci_device_id cxgb4vf_pci_tbl[] = {
2918 CH_DEVICE(0xb000), /* PE10K FPGA */
2919 CH_DEVICE(0x4801), /* T420-cr */
2920 CH_DEVICE(0x4802), /* T422-cr */
2921 CH_DEVICE(0x4803), /* T440-cr */
2922 CH_DEVICE(0x4804), /* T420-bch */
2923 CH_DEVICE(0x4805), /* T440-bch */
2924 CH_DEVICE(0x4806), /* T460-ch */
2925 CH_DEVICE(0x4807), /* T420-so */
2926 CH_DEVICE(0x4808), /* T420-cx */
2927 CH_DEVICE(0x4809), /* T420-bt */
2928 CH_DEVICE(0x480a), /* T404-bt */
2929 CH_DEVICE(0x480d), /* T480-cr */
2930 CH_DEVICE(0x480e), /* T440-lp-cr */
2940 CH_DEVICE(0x5801), /* T520-cr */
2941 CH_DEVICE(0x5802), /* T522-cr */
2942 CH_DEVICE(0x5803), /* T540-cr */
2943 CH_DEVICE(0x5804), /* T520-bch */
2944 CH_DEVICE(0x5805), /* T540-bch */
2945 CH_DEVICE(0x5806), /* T540-ch */
2946 CH_DEVICE(0x5807), /* T520-so */
2947 CH_DEVICE(0x5808), /* T520-cx */
2948 CH_DEVICE(0x5809), /* T520-bt */
2949 CH_DEVICE(0x580a), /* T504-bt */
2950 CH_DEVICE(0x580b), /* T520-sr */
2951 CH_DEVICE(0x580c), /* T504-bt */
2952 CH_DEVICE(0x580d), /* T580-cr */
2953 CH_DEVICE(0x580e), /* T540-lp-cr */
2954 CH_DEVICE(0x580f), /* Amsterdam */
2955 CH_DEVICE(0x5810), /* T580-lp-cr */
2956 CH_DEVICE(0x5811), /* T520-lp-cr */
2957 CH_DEVICE(0x5812), /* T560-cr */
2958 CH_DEVICE(0x5813), /* T580-cr */
2959 CH_DEVICE(0x5814), /* T580-so-cr */
2960 CH_DEVICE(0x5815), /* T502-bt */
2973 MODULE_DESCRIPTION(DRV_DESC);
2974 MODULE_AUTHOR("Chelsio Communications");
2975 MODULE_LICENSE("Dual BSD/GPL");
2976 MODULE_VERSION(DRV_VERSION);
2977 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
2979 static struct pci_driver cxgb4vf_driver = {
2980 .name = KBUILD_MODNAME,
2981 .id_table = cxgb4vf_pci_tbl,
2982 .probe = cxgb4vf_pci_probe,
2983 .remove = cxgb4vf_pci_remove,
2984 .shutdown = cxgb4vf_pci_shutdown,
2988 * Initialize global driver state.
2990 static int __init cxgb4vf_module_init(void)
2995 * Vet our module parameters.
2997 if (msi != MSI_MSIX && msi != MSI_MSI) {
2998 pr_warn("bad module parameter msi=%d; must be %d (MSI-X or MSI) or %d (MSI)\n",
2999 msi, MSI_MSIX, MSI_MSI);
3003 /* Debugfs support is optional, just warn if this fails */
3004 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3005 if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3006 pr_warn("could not create debugfs entry, continuing\n");
3008 ret = pci_register_driver(&cxgb4vf_driver);
3009 if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3010 debugfs_remove(cxgb4vf_debugfs_root);
3015 * Tear down global driver state.
3017 static void __exit cxgb4vf_module_exit(void)
3019 pci_unregister_driver(&cxgb4vf_driver);
3020 debugfs_remove(cxgb4vf_debugfs_root);
3023 module_init(cxgb4vf_module_init);
3024 module_exit(cxgb4vf_module_exit);