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>
47 #include <linux/mdio.h>
49 #include "t4vf_common.h"
50 #include "t4vf_defs.h"
52 #include "../cxgb4/t4_regs.h"
53 #include "../cxgb4/t4_msg.h"
56 * Generic information about the driver.
58 #define DRV_VERSION "2.0.0-ko"
59 #define DRV_DESC "Chelsio T4/T5/T6 Virtual Function (VF) Network Driver"
67 * Default ethtool "message level" for adapters.
69 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
70 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
71 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
73 static int dflt_msg_enable = DFLT_MSG_ENABLE;
75 module_param(dflt_msg_enable, int, 0644);
76 MODULE_PARM_DESC(dflt_msg_enable,
77 "default adapter ethtool message level bitmap, "
78 "deprecated parameter");
81 * The driver uses the best interrupt scheme available on a platform in the
82 * order MSI-X then MSI. This parameter determines which of these schemes the
83 * driver may consider as follows:
85 * msi = 2: choose from among MSI-X and MSI
86 * msi = 1: only consider MSI interrupts
88 * Note that unlike the Physical Function driver, this Virtual Function driver
89 * does _not_ support legacy INTx interrupts (this limitation is mandated by
90 * the PCI-E SR-IOV standard).
94 #define MSI_DEFAULT MSI_MSIX
96 static int msi = MSI_DEFAULT;
98 module_param(msi, int, 0644);
99 MODULE_PARM_DESC(msi, "whether to use MSI-X or MSI");
102 * Fundamental constants.
103 * ======================
107 MAX_TXQ_ENTRIES = 16384,
108 MAX_RSPQ_ENTRIES = 16384,
109 MAX_RX_BUFFERS = 16384,
111 MIN_TXQ_ENTRIES = 32,
112 MIN_RSPQ_ENTRIES = 128,
116 * For purposes of manipulating the Free List size we need to
117 * recognize that Free Lists are actually Egress Queues (the host
118 * produces free buffers which the hardware consumes), Egress Queues
119 * indices are all in units of Egress Context Units bytes, and free
120 * list entries are 64-bit PCI DMA addresses. And since the state of
121 * the Producer Index == the Consumer Index implies an EMPTY list, we
122 * always have at least one Egress Unit's worth of Free List entries
123 * unused. See sge.c for more details ...
125 EQ_UNIT = SGE_EQ_IDXSIZE,
126 FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64),
127 MIN_FL_RESID = FL_PER_EQ_UNIT,
131 * Global driver state.
132 * ====================
135 static struct dentry *cxgb4vf_debugfs_root;
138 * OS "Callback" functions.
139 * ========================
143 * The link status has changed on the indicated "port" (Virtual Interface).
145 void t4vf_os_link_changed(struct adapter *adapter, int pidx, int link_ok)
147 struct net_device *dev = adapter->port[pidx];
150 * If the port is disabled or the current recorded "link up"
151 * status matches the new status, just return.
153 if (!netif_running(dev) || link_ok == netif_carrier_ok(dev))
157 * Tell the OS that the link status has changed and print a short
158 * informative message on the console about the event.
163 const struct port_info *pi = netdev_priv(dev);
165 netif_carrier_on(dev);
167 switch (pi->link_cfg.speed) {
189 switch (pi->link_cfg.fc) {
198 case PAUSE_RX|PAUSE_TX:
207 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s, fc);
209 netif_carrier_off(dev);
210 netdev_info(dev, "link down\n");
215 * THe port module type has changed on the indicated "port" (Virtual
218 void t4vf_os_portmod_changed(struct adapter *adapter, int pidx)
220 static const char * const mod_str[] = {
221 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
223 const struct net_device *dev = adapter->port[pidx];
224 const struct port_info *pi = netdev_priv(dev);
226 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
227 dev_info(adapter->pdev_dev, "%s: port module unplugged\n",
229 else if (pi->mod_type < ARRAY_SIZE(mod_str))
230 dev_info(adapter->pdev_dev, "%s: %s port module inserted\n",
231 dev->name, mod_str[pi->mod_type]);
232 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
233 dev_info(adapter->pdev_dev, "%s: unsupported optical port "
234 "module inserted\n", dev->name);
235 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
236 dev_info(adapter->pdev_dev, "%s: unknown port module inserted,"
237 "forcing TWINAX\n", dev->name);
238 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
239 dev_info(adapter->pdev_dev, "%s: transceiver module error\n",
242 dev_info(adapter->pdev_dev, "%s: unknown module type %d "
243 "inserted\n", dev->name, pi->mod_type);
247 * Net device operations.
248 * ======================
255 * Perform the MAC and PHY actions needed to enable a "port" (Virtual
258 static int link_start(struct net_device *dev)
261 struct port_info *pi = netdev_priv(dev);
264 * We do not set address filters and promiscuity here, the stack does
265 * that step explicitly. Enable vlan accel.
267 ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, 1,
270 ret = t4vf_change_mac(pi->adapter, pi->viid,
271 pi->xact_addr_filt, dev->dev_addr, true);
273 pi->xact_addr_filt = ret;
279 * We don't need to actually "start the link" itself since the
280 * firmware will do that for us when the first Virtual Interface
281 * is enabled on a port.
284 ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true);
289 * Name the MSI-X interrupts.
291 static void name_msix_vecs(struct adapter *adapter)
293 int namelen = sizeof(adapter->msix_info[0].desc) - 1;
299 snprintf(adapter->msix_info[MSIX_FW].desc, namelen,
300 "%s-FWeventq", adapter->name);
301 adapter->msix_info[MSIX_FW].desc[namelen] = 0;
306 for_each_port(adapter, pidx) {
307 struct net_device *dev = adapter->port[pidx];
308 const struct port_info *pi = netdev_priv(dev);
311 for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) {
312 snprintf(adapter->msix_info[msi].desc, namelen,
313 "%s-%d", dev->name, qs);
314 adapter->msix_info[msi].desc[namelen] = 0;
320 * Request all of our MSI-X resources.
322 static int request_msix_queue_irqs(struct adapter *adapter)
324 struct sge *s = &adapter->sge;
330 err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix,
331 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq);
339 for_each_ethrxq(s, rxq) {
340 err = request_irq(adapter->msix_info[msi].vec,
341 t4vf_sge_intr_msix, 0,
342 adapter->msix_info[msi].desc,
343 &s->ethrxq[rxq].rspq);
352 free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq);
353 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
358 * Free our MSI-X resources.
360 static void free_msix_queue_irqs(struct adapter *adapter)
362 struct sge *s = &adapter->sge;
365 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
367 for_each_ethrxq(s, rxq)
368 free_irq(adapter->msix_info[msi++].vec,
369 &s->ethrxq[rxq].rspq);
373 * Turn on NAPI and start up interrupts on a response queue.
375 static void qenable(struct sge_rspq *rspq)
377 napi_enable(&rspq->napi);
380 * 0-increment the Going To Sleep register to start the timer and
383 t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
385 SEINTARM_V(rspq->intr_params) |
386 INGRESSQID_V(rspq->cntxt_id));
390 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
392 static void enable_rx(struct adapter *adapter)
395 struct sge *s = &adapter->sge;
397 for_each_ethrxq(s, rxq)
398 qenable(&s->ethrxq[rxq].rspq);
399 qenable(&s->fw_evtq);
402 * The interrupt queue doesn't use NAPI so we do the 0-increment of
403 * its Going To Sleep register here to get it started.
405 if (adapter->flags & USING_MSI)
406 t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
408 SEINTARM_V(s->intrq.intr_params) |
409 INGRESSQID_V(s->intrq.cntxt_id));
414 * Wait until all NAPI handlers are descheduled.
416 static void quiesce_rx(struct adapter *adapter)
418 struct sge *s = &adapter->sge;
421 for_each_ethrxq(s, rxq)
422 napi_disable(&s->ethrxq[rxq].rspq.napi);
423 napi_disable(&s->fw_evtq.napi);
427 * Response queue handler for the firmware event queue.
429 static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp,
430 const struct pkt_gl *gl)
433 * Extract response opcode and get pointer to CPL message body.
435 struct adapter *adapter = rspq->adapter;
436 u8 opcode = ((const struct rss_header *)rsp)->opcode;
437 void *cpl = (void *)(rsp + 1);
442 * We've received an asynchronous message from the firmware.
444 const struct cpl_fw6_msg *fw_msg = cpl;
445 if (fw_msg->type == FW6_TYPE_CMD_RPL)
446 t4vf_handle_fw_rpl(adapter, fw_msg->data);
451 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
453 const struct cpl_sge_egr_update *p = (void *)(rsp + 3);
454 opcode = CPL_OPCODE_G(ntohl(p->opcode_qid));
455 if (opcode != CPL_SGE_EGR_UPDATE) {
456 dev_err(adapter->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
464 case CPL_SGE_EGR_UPDATE: {
466 * We've received an Egress Queue Status Update message. We
467 * get these, if the SGE is configured to send these when the
468 * firmware passes certain points in processing our TX
469 * Ethernet Queue or if we make an explicit request for one.
470 * We use these updates to determine when we may need to
471 * restart a TX Ethernet Queue which was stopped for lack of
472 * free TX Queue Descriptors ...
474 const struct cpl_sge_egr_update *p = cpl;
475 unsigned int qid = EGR_QID_G(be32_to_cpu(p->opcode_qid));
476 struct sge *s = &adapter->sge;
478 struct sge_eth_txq *txq;
482 * Perform sanity checking on the Queue ID to make sure it
483 * really refers to one of our TX Ethernet Egress Queues which
484 * is active and matches the queue's ID. None of these error
485 * conditions should ever happen so we may want to either make
486 * them fatal and/or conditionalized under DEBUG.
488 eq_idx = EQ_IDX(s, qid);
489 if (unlikely(eq_idx >= MAX_EGRQ)) {
490 dev_err(adapter->pdev_dev,
491 "Egress Update QID %d out of range\n", qid);
494 tq = s->egr_map[eq_idx];
495 if (unlikely(tq == NULL)) {
496 dev_err(adapter->pdev_dev,
497 "Egress Update QID %d TXQ=NULL\n", qid);
500 txq = container_of(tq, struct sge_eth_txq, q);
501 if (unlikely(tq->abs_id != qid)) {
502 dev_err(adapter->pdev_dev,
503 "Egress Update QID %d refers to TXQ %d\n",
509 * Restart a stopped TX Queue which has less than half of its
513 netif_tx_wake_queue(txq->txq);
518 dev_err(adapter->pdev_dev,
519 "unexpected CPL %#x on FW event queue\n", opcode);
526 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
527 * to use and initializes them. We support multiple "Queue Sets" per port if
528 * we have MSI-X, otherwise just one queue set per port.
530 static int setup_sge_queues(struct adapter *adapter)
532 struct sge *s = &adapter->sge;
536 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
539 bitmap_zero(s->starving_fl, MAX_EGRQ);
542 * If we're using MSI interrupt mode we need to set up a "forwarded
543 * interrupt" queue which we'll set up with our MSI vector. The rest
544 * of the ingress queues will be set up to forward their interrupts to
545 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
546 * the intrq's queue ID as the interrupt forwarding queue for the
547 * subsequent calls ...
549 if (adapter->flags & USING_MSI) {
550 err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false,
551 adapter->port[0], 0, NULL, NULL);
553 goto err_free_queues;
557 * Allocate our ingress queue for asynchronous firmware messages.
559 err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0],
560 MSIX_FW, NULL, fwevtq_handler);
562 goto err_free_queues;
565 * Allocate each "port"'s initial Queue Sets. These can be changed
566 * later on ... up to the point where any interface on the adapter is
567 * brought up at which point lots of things get nailed down
571 for_each_port(adapter, pidx) {
572 struct net_device *dev = adapter->port[pidx];
573 struct port_info *pi = netdev_priv(dev);
574 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
575 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
578 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
579 err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false,
581 &rxq->fl, t4vf_ethrx_handler);
583 goto err_free_queues;
585 err = t4vf_sge_alloc_eth_txq(adapter, txq, dev,
586 netdev_get_tx_queue(dev, qs),
587 s->fw_evtq.cntxt_id);
589 goto err_free_queues;
592 memset(&rxq->stats, 0, sizeof(rxq->stats));
597 * Create the reverse mappings for the queues.
599 s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id;
600 s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id;
601 IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq;
602 for_each_port(adapter, pidx) {
603 struct net_device *dev = adapter->port[pidx];
604 struct port_info *pi = netdev_priv(dev);
605 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
606 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
609 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
610 IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq;
611 EQ_MAP(s, txq->q.abs_id) = &txq->q;
614 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
615 * for Free Lists but since all of the Egress Queues
616 * (including Free Lists) have Relative Queue IDs
617 * which are computed as Absolute - Base Queue ID, we
618 * can synthesize the Absolute Queue IDs for the Free
619 * Lists. This is useful for debugging purposes when
620 * we want to dump Queue Contexts via the PF Driver.
622 rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base;
623 EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl;
629 t4vf_free_sge_resources(adapter);
634 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
635 * queues. We configure the RSS CPU lookup table to distribute to the number
636 * of HW receive queues, and the response queue lookup table to narrow that
637 * down to the response queues actually configured for each "port" (Virtual
638 * Interface). We always configure the RSS mapping for all ports since the
639 * mapping table has plenty of entries.
641 static int setup_rss(struct adapter *adapter)
645 for_each_port(adapter, pidx) {
646 struct port_info *pi = adap2pinfo(adapter, pidx);
647 struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
648 u16 rss[MAX_PORT_QSETS];
651 for (qs = 0; qs < pi->nqsets; qs++)
652 rss[qs] = rxq[qs].rspq.abs_id;
654 err = t4vf_config_rss_range(adapter, pi->viid,
655 0, pi->rss_size, rss, pi->nqsets);
660 * Perform Global RSS Mode-specific initialization.
662 switch (adapter->params.rss.mode) {
663 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL:
665 * If Tunnel All Lookup isn't specified in the global
666 * RSS Configuration, then we need to specify a
667 * default Ingress Queue for any ingress packets which
668 * aren't hashed. We'll use our first ingress queue
671 if (!adapter->params.rss.u.basicvirtual.tnlalllookup) {
672 union rss_vi_config config;
673 err = t4vf_read_rss_vi_config(adapter,
678 config.basicvirtual.defaultq =
680 err = t4vf_write_rss_vi_config(adapter,
694 * Bring the adapter up. Called whenever we go from no "ports" open to having
695 * one open. This function performs the actions necessary to make an adapter
696 * operational, such as completing the initialization of HW modules, and
697 * enabling interrupts. Must be called with the rtnl lock held. (Note that
698 * this is called "cxgb_up" in the PF Driver.)
700 static int adapter_up(struct adapter *adapter)
705 * If this is the first time we've been called, perform basic
706 * adapter setup. Once we've done this, many of our adapter
707 * parameters can no longer be changed ...
709 if ((adapter->flags & FULL_INIT_DONE) == 0) {
710 err = setup_sge_queues(adapter);
713 err = setup_rss(adapter);
715 t4vf_free_sge_resources(adapter);
719 if (adapter->flags & USING_MSIX)
720 name_msix_vecs(adapter);
722 adapter->flags |= FULL_INIT_DONE;
726 * Acquire our interrupt resources. We only support MSI-X and MSI.
728 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
729 if (adapter->flags & USING_MSIX)
730 err = request_msix_queue_irqs(adapter);
732 err = request_irq(adapter->pdev->irq,
733 t4vf_intr_handler(adapter), 0,
734 adapter->name, adapter);
736 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
742 * Enable NAPI ingress processing and return success.
745 t4vf_sge_start(adapter);
751 * Bring the adapter down. Called whenever the last "port" (Virtual
752 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
755 static void adapter_down(struct adapter *adapter)
758 * Free interrupt resources.
760 if (adapter->flags & USING_MSIX)
761 free_msix_queue_irqs(adapter);
763 free_irq(adapter->pdev->irq, adapter);
766 * Wait for NAPI handlers to finish.
772 * Start up a net device.
774 static int cxgb4vf_open(struct net_device *dev)
777 struct port_info *pi = netdev_priv(dev);
778 struct adapter *adapter = pi->adapter;
781 * If this is the first interface that we're opening on the "adapter",
782 * bring the "adapter" up now.
784 if (adapter->open_device_map == 0) {
785 err = adapter_up(adapter);
791 * Note that this interface is up and start everything up ...
793 err = link_start(dev);
797 netif_tx_start_all_queues(dev);
798 set_bit(pi->port_id, &adapter->open_device_map);
802 if (adapter->open_device_map == 0)
803 adapter_down(adapter);
808 * Shut down a net device. This routine is called "cxgb_close" in the PF
811 static int cxgb4vf_stop(struct net_device *dev)
813 struct port_info *pi = netdev_priv(dev);
814 struct adapter *adapter = pi->adapter;
816 netif_tx_stop_all_queues(dev);
817 netif_carrier_off(dev);
818 t4vf_enable_vi(adapter, pi->viid, false, false);
819 pi->link_cfg.link_ok = 0;
821 clear_bit(pi->port_id, &adapter->open_device_map);
822 if (adapter->open_device_map == 0)
823 adapter_down(adapter);
828 * Translate our basic statistics into the standard "ifconfig" statistics.
830 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
832 struct t4vf_port_stats stats;
833 struct port_info *pi = netdev2pinfo(dev);
834 struct adapter *adapter = pi->adapter;
835 struct net_device_stats *ns = &dev->stats;
838 spin_lock(&adapter->stats_lock);
839 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
840 spin_unlock(&adapter->stats_lock);
842 memset(ns, 0, sizeof(*ns));
846 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
847 stats.tx_ucast_bytes + stats.tx_offload_bytes);
848 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
849 stats.tx_ucast_frames + stats.tx_offload_frames);
850 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
851 stats.rx_ucast_bytes);
852 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
853 stats.rx_ucast_frames);
854 ns->multicast = stats.rx_mcast_frames;
855 ns->tx_errors = stats.tx_drop_frames;
856 ns->rx_errors = stats.rx_err_frames;
861 static inline int cxgb4vf_set_addr_hash(struct port_info *pi)
863 struct adapter *adapter = pi->adapter;
866 struct hash_mac_addr *entry;
868 /* Calculate the hash vector for the updated list and program it */
869 list_for_each_entry(entry, &adapter->mac_hlist, list) {
870 ucast |= is_unicast_ether_addr(entry->addr);
871 vec |= (1ULL << hash_mac_addr(entry->addr));
873 return t4vf_set_addr_hash(adapter, pi->viid, ucast, vec, false);
876 static int cxgb4vf_mac_sync(struct net_device *netdev, const u8 *mac_addr)
878 struct port_info *pi = netdev_priv(netdev);
879 struct adapter *adapter = pi->adapter;
884 bool ucast = is_unicast_ether_addr(mac_addr);
885 const u8 *maclist[1] = {mac_addr};
886 struct hash_mac_addr *new_entry;
888 ret = t4vf_alloc_mac_filt(adapter, pi->viid, free, 1, maclist,
889 NULL, ucast ? &uhash : &mhash, false);
892 /* if hash != 0, then add the addr to hash addr list
893 * so on the end we will calculate the hash for the
894 * list and program it
896 if (uhash || mhash) {
897 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
900 ether_addr_copy(new_entry->addr, mac_addr);
901 list_add_tail(&new_entry->list, &adapter->mac_hlist);
902 ret = cxgb4vf_set_addr_hash(pi);
905 return ret < 0 ? ret : 0;
908 static int cxgb4vf_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
910 struct port_info *pi = netdev_priv(netdev);
911 struct adapter *adapter = pi->adapter;
913 const u8 *maclist[1] = {mac_addr};
914 struct hash_mac_addr *entry, *tmp;
916 /* If the MAC address to be removed is in the hash addr
917 * list, delete it from the list and update hash vector
919 list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist, list) {
920 if (ether_addr_equal(entry->addr, mac_addr)) {
921 list_del(&entry->list);
923 return cxgb4vf_set_addr_hash(pi);
927 ret = t4vf_free_mac_filt(adapter, pi->viid, 1, maclist, false);
928 return ret < 0 ? -EINVAL : 0;
932 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
933 * If @mtu is -1 it is left unchanged.
935 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
937 struct port_info *pi = netdev_priv(dev);
939 __dev_uc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync);
940 __dev_mc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync);
941 return t4vf_set_rxmode(pi->adapter, pi->viid, -1,
942 (dev->flags & IFF_PROMISC) != 0,
943 (dev->flags & IFF_ALLMULTI) != 0,
948 * Set the current receive modes on the device.
950 static void cxgb4vf_set_rxmode(struct net_device *dev)
952 /* unfortunately we can't return errors to the stack */
953 set_rxmode(dev, -1, false);
957 * Find the entry in the interrupt holdoff timer value array which comes
958 * closest to the specified interrupt holdoff value.
960 static int closest_timer(const struct sge *s, int us)
962 int i, timer_idx = 0, min_delta = INT_MAX;
964 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
965 int delta = us - s->timer_val[i];
968 if (delta < min_delta) {
976 static int closest_thres(const struct sge *s, int thres)
978 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
980 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
981 delta = thres - s->counter_val[i];
984 if (delta < min_delta) {
993 * Return a queue's interrupt hold-off time in us. 0 means no timer.
995 static unsigned int qtimer_val(const struct adapter *adapter,
996 const struct sge_rspq *rspq)
998 unsigned int timer_idx = QINTR_TIMER_IDX_G(rspq->intr_params);
1000 return timer_idx < SGE_NTIMERS
1001 ? adapter->sge.timer_val[timer_idx]
1006 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1007 * @adapter: the adapter
1008 * @rspq: the RX response queue
1009 * @us: the hold-off time in us, or 0 to disable timer
1010 * @cnt: the hold-off packet count, or 0 to disable counter
1012 * Sets an RX response queue's interrupt hold-off time and packet count.
1013 * At least one of the two needs to be enabled for the queue to generate
1016 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
1017 unsigned int us, unsigned int cnt)
1019 unsigned int timer_idx;
1022 * If both the interrupt holdoff timer and count are specified as
1023 * zero, default to a holdoff count of 1 ...
1025 if ((us | cnt) == 0)
1029 * If an interrupt holdoff count has been specified, then find the
1030 * closest configured holdoff count and use that. If the response
1031 * queue has already been created, then update its queue context
1038 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1039 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1040 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1041 FW_PARAMS_PARAM_X_V(
1042 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1043 FW_PARAMS_PARAM_YZ_V(rspq->cntxt_id);
1044 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1048 rspq->pktcnt_idx = pktcnt_idx;
1052 * Compute the closest holdoff timer index from the supplied holdoff
1055 timer_idx = (us == 0
1056 ? SGE_TIMER_RSTRT_CNTR
1057 : closest_timer(&adapter->sge, us));
1060 * Update the response queue's interrupt coalescing parameters and
1063 rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) |
1064 QINTR_CNT_EN_V(cnt > 0));
1069 * Return a version number to identify the type of adapter. The scheme is:
1070 * - bits 0..9: chip version
1071 * - bits 10..15: chip revision
1073 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1076 * Chip version 4, revision 0x3f (cxgb4vf).
1078 return CHELSIO_CHIP_VERSION(adapter->params.chip) | (0x3f << 10);
1082 * Execute the specified ioctl command.
1084 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1090 * The VF Driver doesn't have access to any of the other
1091 * common Ethernet device ioctl()'s (like reading/writing
1092 * PHY registers, etc.
1103 * Change the device's MTU.
1105 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1108 struct port_info *pi = netdev_priv(dev);
1110 /* accommodate SACK */
1114 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1115 -1, -1, -1, -1, true);
1121 static netdev_features_t cxgb4vf_fix_features(struct net_device *dev,
1122 netdev_features_t features)
1125 * Since there is no support for separate rx/tx vlan accel
1126 * enable/disable make sure tx flag is always in same state as rx.
1128 if (features & NETIF_F_HW_VLAN_CTAG_RX)
1129 features |= NETIF_F_HW_VLAN_CTAG_TX;
1131 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
1136 static int cxgb4vf_set_features(struct net_device *dev,
1137 netdev_features_t features)
1139 struct port_info *pi = netdev_priv(dev);
1140 netdev_features_t changed = dev->features ^ features;
1142 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1143 t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1,
1144 features & NETIF_F_HW_VLAN_CTAG_TX, 0);
1150 * Change the devices MAC address.
1152 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1155 struct sockaddr *addr = _addr;
1156 struct port_info *pi = netdev_priv(dev);
1158 if (!is_valid_ether_addr(addr->sa_data))
1159 return -EADDRNOTAVAIL;
1161 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1162 addr->sa_data, true);
1166 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1167 pi->xact_addr_filt = ret;
1171 #ifdef CONFIG_NET_POLL_CONTROLLER
1173 * Poll all of our receive queues. This is called outside of normal interrupt
1176 static void cxgb4vf_poll_controller(struct net_device *dev)
1178 struct port_info *pi = netdev_priv(dev);
1179 struct adapter *adapter = pi->adapter;
1181 if (adapter->flags & USING_MSIX) {
1182 struct sge_eth_rxq *rxq;
1185 rxq = &adapter->sge.ethrxq[pi->first_qset];
1186 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1187 t4vf_sge_intr_msix(0, &rxq->rspq);
1191 t4vf_intr_handler(adapter)(0, adapter);
1196 * Ethtool operations.
1197 * ===================
1199 * Note that we don't support any ethtool operations which change the physical
1200 * state of the port to which we're linked.
1204 * from_fw_port_mod_type - translate Firmware Port/Module type to Ethtool
1205 * @port_type: Firmware Port Type
1206 * @mod_type: Firmware Module Type
1208 * Translate Firmware Port/Module type to Ethtool Port Type.
1210 static int from_fw_port_mod_type(enum fw_port_type port_type,
1211 enum fw_port_module_type mod_type)
1213 if (port_type == FW_PORT_TYPE_BT_SGMII ||
1214 port_type == FW_PORT_TYPE_BT_XFI ||
1215 port_type == FW_PORT_TYPE_BT_XAUI) {
1217 } else if (port_type == FW_PORT_TYPE_FIBER_XFI ||
1218 port_type == FW_PORT_TYPE_FIBER_XAUI) {
1220 } else if (port_type == FW_PORT_TYPE_SFP ||
1221 port_type == FW_PORT_TYPE_QSFP_10G ||
1222 port_type == FW_PORT_TYPE_QSA ||
1223 port_type == FW_PORT_TYPE_QSFP) {
1224 if (mod_type == FW_PORT_MOD_TYPE_LR ||
1225 mod_type == FW_PORT_MOD_TYPE_SR ||
1226 mod_type == FW_PORT_MOD_TYPE_ER ||
1227 mod_type == FW_PORT_MOD_TYPE_LRM)
1229 else if (mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
1230 mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
1240 * fw_caps_to_lmm - translate Firmware to ethtool Link Mode Mask
1241 * @port_type: Firmware Port Type
1242 * @fw_caps: Firmware Port Capabilities
1243 * @link_mode_mask: ethtool Link Mode Mask
1245 * Translate a Firmware Port Capabilities specification to an ethtool
1248 static void fw_caps_to_lmm(enum fw_port_type port_type,
1249 unsigned int fw_caps,
1250 unsigned long *link_mode_mask)
1252 #define SET_LMM(__lmm_name) __set_bit(ETHTOOL_LINK_MODE_ ## __lmm_name\
1253 ## _BIT, link_mode_mask)
1255 #define FW_CAPS_TO_LMM(__fw_name, __lmm_name) \
1257 if (fw_caps & FW_PORT_CAP_ ## __fw_name) \
1258 SET_LMM(__lmm_name); \
1261 switch (port_type) {
1262 case FW_PORT_TYPE_BT_SGMII:
1263 case FW_PORT_TYPE_BT_XFI:
1264 case FW_PORT_TYPE_BT_XAUI:
1266 FW_CAPS_TO_LMM(SPEED_100M, 100baseT_Full);
1267 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1268 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1271 case FW_PORT_TYPE_KX4:
1272 case FW_PORT_TYPE_KX:
1274 FW_CAPS_TO_LMM(SPEED_1G, 1000baseKX_Full);
1275 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKX4_Full);
1278 case FW_PORT_TYPE_KR:
1280 SET_LMM(10000baseKR_Full);
1283 case FW_PORT_TYPE_BP_AP:
1285 SET_LMM(10000baseR_FEC);
1286 SET_LMM(10000baseKR_Full);
1287 SET_LMM(1000baseKX_Full);
1290 case FW_PORT_TYPE_BP4_AP:
1292 SET_LMM(10000baseR_FEC);
1293 SET_LMM(10000baseKR_Full);
1294 SET_LMM(1000baseKX_Full);
1295 SET_LMM(10000baseKX4_Full);
1298 case FW_PORT_TYPE_FIBER_XFI:
1299 case FW_PORT_TYPE_FIBER_XAUI:
1300 case FW_PORT_TYPE_SFP:
1301 case FW_PORT_TYPE_QSFP_10G:
1302 case FW_PORT_TYPE_QSA:
1304 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1305 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1308 case FW_PORT_TYPE_BP40_BA:
1309 case FW_PORT_TYPE_QSFP:
1311 SET_LMM(40000baseSR4_Full);
1314 case FW_PORT_TYPE_CR_QSFP:
1315 case FW_PORT_TYPE_SFP28:
1317 SET_LMM(25000baseCR_Full);
1320 case FW_PORT_TYPE_KR4_100G:
1321 case FW_PORT_TYPE_CR4_QSFP:
1323 SET_LMM(100000baseCR4_Full);
1330 FW_CAPS_TO_LMM(ANEG, Autoneg);
1331 FW_CAPS_TO_LMM(802_3_PAUSE, Pause);
1332 FW_CAPS_TO_LMM(802_3_ASM_DIR, Asym_Pause);
1334 #undef FW_CAPS_TO_LMM
1338 static int cxgb4vf_get_link_ksettings(struct net_device *dev,
1339 struct ethtool_link_ksettings
1342 const struct port_info *pi = netdev_priv(dev);
1343 struct ethtool_link_settings *base = &link_ksettings->base;
1345 ethtool_link_ksettings_zero_link_mode(link_ksettings, supported);
1346 ethtool_link_ksettings_zero_link_mode(link_ksettings, advertising);
1347 ethtool_link_ksettings_zero_link_mode(link_ksettings, lp_advertising);
1349 base->port = from_fw_port_mod_type(pi->port_type, pi->mod_type);
1351 if (pi->mdio_addr >= 0) {
1352 base->phy_address = pi->mdio_addr;
1353 base->mdio_support = (pi->port_type == FW_PORT_TYPE_BT_SGMII
1354 ? ETH_MDIO_SUPPORTS_C22
1355 : ETH_MDIO_SUPPORTS_C45);
1357 base->phy_address = 255;
1358 base->mdio_support = 0;
1361 fw_caps_to_lmm(pi->port_type, pi->link_cfg.supported,
1362 link_ksettings->link_modes.supported);
1363 fw_caps_to_lmm(pi->port_type, pi->link_cfg.advertising,
1364 link_ksettings->link_modes.advertising);
1365 fw_caps_to_lmm(pi->port_type, pi->link_cfg.lp_advertising,
1366 link_ksettings->link_modes.lp_advertising);
1368 if (netif_carrier_ok(dev)) {
1369 base->speed = pi->link_cfg.speed;
1370 base->duplex = DUPLEX_FULL;
1372 base->speed = SPEED_UNKNOWN;
1373 base->duplex = DUPLEX_UNKNOWN;
1376 base->autoneg = pi->link_cfg.autoneg;
1377 if (pi->link_cfg.supported & FW_PORT_CAP_ANEG)
1378 ethtool_link_ksettings_add_link_mode(link_ksettings,
1379 supported, Autoneg);
1380 if (pi->link_cfg.autoneg)
1381 ethtool_link_ksettings_add_link_mode(link_ksettings,
1382 advertising, Autoneg);
1388 * Return our driver information.
1390 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1391 struct ethtool_drvinfo *drvinfo)
1393 struct adapter *adapter = netdev2adap(dev);
1395 strlcpy(drvinfo->driver, KBUILD_MODNAME, sizeof(drvinfo->driver));
1396 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
1397 strlcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)),
1398 sizeof(drvinfo->bus_info));
1399 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1400 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1401 FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.fwrev),
1402 FW_HDR_FW_VER_MINOR_G(adapter->params.dev.fwrev),
1403 FW_HDR_FW_VER_MICRO_G(adapter->params.dev.fwrev),
1404 FW_HDR_FW_VER_BUILD_G(adapter->params.dev.fwrev),
1405 FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.tprev),
1406 FW_HDR_FW_VER_MINOR_G(adapter->params.dev.tprev),
1407 FW_HDR_FW_VER_MICRO_G(adapter->params.dev.tprev),
1408 FW_HDR_FW_VER_BUILD_G(adapter->params.dev.tprev));
1412 * Return current adapter message level.
1414 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1416 return netdev2adap(dev)->msg_enable;
1420 * Set current adapter message level.
1422 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1424 netdev2adap(dev)->msg_enable = msglevel;
1428 * Return the device's current Queue Set ring size parameters along with the
1429 * allowed maximum values. Since ethtool doesn't understand the concept of
1430 * multi-queue devices, we just return the current values associated with the
1433 static void cxgb4vf_get_ringparam(struct net_device *dev,
1434 struct ethtool_ringparam *rp)
1436 const struct port_info *pi = netdev_priv(dev);
1437 const struct sge *s = &pi->adapter->sge;
1439 rp->rx_max_pending = MAX_RX_BUFFERS;
1440 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1441 rp->rx_jumbo_max_pending = 0;
1442 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1444 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1445 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1446 rp->rx_jumbo_pending = 0;
1447 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1451 * Set the Queue Set ring size parameters for the device. Again, since
1452 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1453 * apply these new values across all of the Queue Sets associated with the
1454 * device -- after vetting them of course!
1456 static int cxgb4vf_set_ringparam(struct net_device *dev,
1457 struct ethtool_ringparam *rp)
1459 const struct port_info *pi = netdev_priv(dev);
1460 struct adapter *adapter = pi->adapter;
1461 struct sge *s = &adapter->sge;
1464 if (rp->rx_pending > MAX_RX_BUFFERS ||
1465 rp->rx_jumbo_pending ||
1466 rp->tx_pending > MAX_TXQ_ENTRIES ||
1467 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1468 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1469 rp->rx_pending < MIN_FL_ENTRIES ||
1470 rp->tx_pending < MIN_TXQ_ENTRIES)
1473 if (adapter->flags & FULL_INIT_DONE)
1476 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1477 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1478 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1479 s->ethtxq[qs].q.size = rp->tx_pending;
1485 * Return the interrupt holdoff timer and count for the first Queue Set on the
1486 * device. Our extension ioctl() (the cxgbtool interface) allows the
1487 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1489 static int cxgb4vf_get_coalesce(struct net_device *dev,
1490 struct ethtool_coalesce *coalesce)
1492 const struct port_info *pi = netdev_priv(dev);
1493 const struct adapter *adapter = pi->adapter;
1494 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1496 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1497 coalesce->rx_max_coalesced_frames =
1498 ((rspq->intr_params & QINTR_CNT_EN_F)
1499 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1505 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1506 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1507 * the interrupt holdoff timer on any of the device's Queue Sets.
1509 static int cxgb4vf_set_coalesce(struct net_device *dev,
1510 struct ethtool_coalesce *coalesce)
1512 const struct port_info *pi = netdev_priv(dev);
1513 struct adapter *adapter = pi->adapter;
1515 return set_rxq_intr_params(adapter,
1516 &adapter->sge.ethrxq[pi->first_qset].rspq,
1517 coalesce->rx_coalesce_usecs,
1518 coalesce->rx_max_coalesced_frames);
1522 * Report current port link pause parameter settings.
1524 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1525 struct ethtool_pauseparam *pauseparam)
1527 struct port_info *pi = netdev_priv(dev);
1529 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1530 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1531 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1535 * Identify the port by blinking the port's LED.
1537 static int cxgb4vf_phys_id(struct net_device *dev,
1538 enum ethtool_phys_id_state state)
1541 struct port_info *pi = netdev_priv(dev);
1543 if (state == ETHTOOL_ID_ACTIVE)
1545 else if (state == ETHTOOL_ID_INACTIVE)
1550 return t4vf_identify_port(pi->adapter, pi->viid, val);
1554 * Port stats maintained per queue of the port.
1556 struct queue_port_stats {
1567 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1568 * these need to match the order of statistics returned by
1569 * t4vf_get_port_stats().
1571 static const char stats_strings[][ETH_GSTRING_LEN] = {
1573 * These must match the layout of the t4vf_port_stats structure.
1575 "TxBroadcastBytes ",
1576 "TxBroadcastFrames ",
1577 "TxMulticastBytes ",
1578 "TxMulticastFrames ",
1584 "RxBroadcastBytes ",
1585 "RxBroadcastFrames ",
1586 "RxMulticastBytes ",
1587 "RxMulticastFrames ",
1593 * These are accumulated per-queue statistics and must match the
1594 * order of the fields in the queue_port_stats structure.
1606 * Return the number of statistics in the specified statistics set.
1608 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1612 return ARRAY_SIZE(stats_strings);
1620 * Return the strings for the specified statistics set.
1622 static void cxgb4vf_get_strings(struct net_device *dev,
1628 memcpy(data, stats_strings, sizeof(stats_strings));
1634 * Small utility routine to accumulate queue statistics across the queues of
1637 static void collect_sge_port_stats(const struct adapter *adapter,
1638 const struct port_info *pi,
1639 struct queue_port_stats *stats)
1641 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1642 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1645 memset(stats, 0, sizeof(*stats));
1646 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1647 stats->tso += txq->tso;
1648 stats->tx_csum += txq->tx_cso;
1649 stats->rx_csum += rxq->stats.rx_cso;
1650 stats->vlan_ex += rxq->stats.vlan_ex;
1651 stats->vlan_ins += txq->vlan_ins;
1652 stats->lro_pkts += rxq->stats.lro_pkts;
1653 stats->lro_merged += rxq->stats.lro_merged;
1658 * Return the ETH_SS_STATS statistics set.
1660 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1661 struct ethtool_stats *stats,
1664 struct port_info *pi = netdev2pinfo(dev);
1665 struct adapter *adapter = pi->adapter;
1666 int err = t4vf_get_port_stats(adapter, pi->pidx,
1667 (struct t4vf_port_stats *)data);
1669 memset(data, 0, sizeof(struct t4vf_port_stats));
1671 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1672 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1676 * Return the size of our register map.
1678 static int cxgb4vf_get_regs_len(struct net_device *dev)
1680 return T4VF_REGMAP_SIZE;
1684 * Dump a block of registers, start to end inclusive, into a buffer.
1686 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1687 unsigned int start, unsigned int end)
1689 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1691 for ( ; start <= end; start += sizeof(u32)) {
1693 * Avoid reading the Mailbox Control register since that
1694 * can trigger a Mailbox Ownership Arbitration cycle and
1695 * interfere with communication with the firmware.
1697 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1700 *bp++ = t4_read_reg(adapter, start);
1705 * Copy our entire register map into the provided buffer.
1707 static void cxgb4vf_get_regs(struct net_device *dev,
1708 struct ethtool_regs *regs,
1711 struct adapter *adapter = netdev2adap(dev);
1713 regs->version = mk_adap_vers(adapter);
1716 * Fill in register buffer with our register map.
1718 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1720 reg_block_dump(adapter, regbuf,
1721 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1722 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1723 reg_block_dump(adapter, regbuf,
1724 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1725 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1727 /* T5 adds new registers in the PL Register map.
1729 reg_block_dump(adapter, regbuf,
1730 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1731 T4VF_PL_BASE_ADDR + (is_t4(adapter->params.chip)
1732 ? PL_VF_WHOAMI_A : PL_VF_REVISION_A));
1733 reg_block_dump(adapter, regbuf,
1734 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1735 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1737 reg_block_dump(adapter, regbuf,
1738 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1739 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1743 * Report current Wake On LAN settings.
1745 static void cxgb4vf_get_wol(struct net_device *dev,
1746 struct ethtool_wolinfo *wol)
1750 memset(&wol->sopass, 0, sizeof(wol->sopass));
1754 * TCP Segmentation Offload flags which we support.
1756 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1758 static const struct ethtool_ops cxgb4vf_ethtool_ops = {
1759 .get_link_ksettings = cxgb4vf_get_link_ksettings,
1760 .get_drvinfo = cxgb4vf_get_drvinfo,
1761 .get_msglevel = cxgb4vf_get_msglevel,
1762 .set_msglevel = cxgb4vf_set_msglevel,
1763 .get_ringparam = cxgb4vf_get_ringparam,
1764 .set_ringparam = cxgb4vf_set_ringparam,
1765 .get_coalesce = cxgb4vf_get_coalesce,
1766 .set_coalesce = cxgb4vf_set_coalesce,
1767 .get_pauseparam = cxgb4vf_get_pauseparam,
1768 .get_link = ethtool_op_get_link,
1769 .get_strings = cxgb4vf_get_strings,
1770 .set_phys_id = cxgb4vf_phys_id,
1771 .get_sset_count = cxgb4vf_get_sset_count,
1772 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1773 .get_regs_len = cxgb4vf_get_regs_len,
1774 .get_regs = cxgb4vf_get_regs,
1775 .get_wol = cxgb4vf_get_wol,
1779 * /sys/kernel/debug/cxgb4vf support code and data.
1780 * ================================================
1784 * Show Firmware Mailbox Command/Reply Log
1786 * Note that we don't do any locking when dumping the Firmware Mailbox Log so
1787 * it's possible that we can catch things during a log update and therefore
1788 * see partially corrupted log entries. But i9t's probably Good Enough(tm).
1789 * If we ever decide that we want to make sure that we're dumping a coherent
1790 * log, we'd need to perform locking in the mailbox logging and in
1791 * mboxlog_open() where we'd need to grab the entire mailbox log in one go
1792 * like we do for the Firmware Device Log. But as stated above, meh ...
1794 static int mboxlog_show(struct seq_file *seq, void *v)
1796 struct adapter *adapter = seq->private;
1797 struct mbox_cmd_log *log = adapter->mbox_log;
1798 struct mbox_cmd *entry;
1801 if (v == SEQ_START_TOKEN) {
1803 "%10s %15s %5s %5s %s\n",
1804 "Seq#", "Tstamp", "Atime", "Etime",
1809 entry_idx = log->cursor + ((uintptr_t)v - 2);
1810 if (entry_idx >= log->size)
1811 entry_idx -= log->size;
1812 entry = mbox_cmd_log_entry(log, entry_idx);
1814 /* skip over unused entries */
1815 if (entry->timestamp == 0)
1818 seq_printf(seq, "%10u %15llu %5d %5d",
1819 entry->seqno, entry->timestamp,
1820 entry->access, entry->execute);
1821 for (i = 0; i < MBOX_LEN / 8; i++) {
1822 u64 flit = entry->cmd[i];
1823 u32 hi = (u32)(flit >> 32);
1826 seq_printf(seq, " %08x %08x", hi, lo);
1828 seq_puts(seq, "\n");
1832 static inline void *mboxlog_get_idx(struct seq_file *seq, loff_t pos)
1834 struct adapter *adapter = seq->private;
1835 struct mbox_cmd_log *log = adapter->mbox_log;
1837 return ((pos <= log->size) ? (void *)(uintptr_t)(pos + 1) : NULL);
1840 static void *mboxlog_start(struct seq_file *seq, loff_t *pos)
1842 return *pos ? mboxlog_get_idx(seq, *pos) : SEQ_START_TOKEN;
1845 static void *mboxlog_next(struct seq_file *seq, void *v, loff_t *pos)
1848 return mboxlog_get_idx(seq, *pos);
1851 static void mboxlog_stop(struct seq_file *seq, void *v)
1855 static const struct seq_operations mboxlog_seq_ops = {
1856 .start = mboxlog_start,
1857 .next = mboxlog_next,
1858 .stop = mboxlog_stop,
1859 .show = mboxlog_show
1862 static int mboxlog_open(struct inode *inode, struct file *file)
1864 int res = seq_open(file, &mboxlog_seq_ops);
1867 struct seq_file *seq = file->private_data;
1869 seq->private = inode->i_private;
1874 static const struct file_operations mboxlog_fops = {
1875 .owner = THIS_MODULE,
1876 .open = mboxlog_open,
1878 .llseek = seq_lseek,
1879 .release = seq_release,
1883 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1887 static int sge_qinfo_show(struct seq_file *seq, void *v)
1889 struct adapter *adapter = seq->private;
1890 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1891 int qs, r = (uintptr_t)v - 1;
1894 seq_putc(seq, '\n');
1896 #define S3(fmt_spec, s, v) \
1898 seq_printf(seq, "%-12s", s); \
1899 for (qs = 0; qs < n; ++qs) \
1900 seq_printf(seq, " %16" fmt_spec, v); \
1901 seq_putc(seq, '\n'); \
1903 #define S(s, v) S3("s", s, v)
1904 #define T(s, v) S3("u", s, txq[qs].v)
1905 #define R(s, v) S3("u", s, rxq[qs].v)
1907 if (r < eth_entries) {
1908 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1909 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1910 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1912 S("QType:", "Ethernet");
1914 (rxq[qs].rspq.netdev
1915 ? rxq[qs].rspq.netdev->name
1918 (rxq[qs].rspq.netdev
1919 ? ((struct port_info *)
1920 netdev_priv(rxq[qs].rspq.netdev))->port_id
1922 T("TxQ ID:", q.abs_id);
1923 T("TxQ size:", q.size);
1924 T("TxQ inuse:", q.in_use);
1925 T("TxQ PIdx:", q.pidx);
1926 T("TxQ CIdx:", q.cidx);
1927 R("RspQ ID:", rspq.abs_id);
1928 R("RspQ size:", rspq.size);
1929 R("RspQE size:", rspq.iqe_len);
1930 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1931 S3("u", "Intr pktcnt:",
1932 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1933 R("RspQ CIdx:", rspq.cidx);
1934 R("RspQ Gen:", rspq.gen);
1935 R("FL ID:", fl.abs_id);
1936 R("FL size:", fl.size - MIN_FL_RESID);
1937 R("FL avail:", fl.avail);
1938 R("FL PIdx:", fl.pidx);
1939 R("FL CIdx:", fl.cidx);
1945 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1947 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1948 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1949 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1950 qtimer_val(adapter, evtq));
1951 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1952 adapter->sge.counter_val[evtq->pktcnt_idx]);
1953 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1954 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1955 } else if (r == 1) {
1956 const struct sge_rspq *intrq = &adapter->sge.intrq;
1958 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1959 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1960 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1961 qtimer_val(adapter, intrq));
1962 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1963 adapter->sge.counter_val[intrq->pktcnt_idx]);
1964 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1965 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1977 * Return the number of "entries" in our "file". We group the multi-Queue
1978 * sections with QPL Queue Sets per "entry". The sections of the output are:
1980 * Ethernet RX/TX Queue Sets
1981 * Firmware Event Queue
1982 * Forwarded Interrupt Queue (if in MSI mode)
1984 static int sge_queue_entries(const struct adapter *adapter)
1986 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1987 ((adapter->flags & USING_MSI) != 0);
1990 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1992 int entries = sge_queue_entries(seq->private);
1994 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1997 static void sge_queue_stop(struct seq_file *seq, void *v)
2001 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
2003 int entries = sge_queue_entries(seq->private);
2006 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2009 static const struct seq_operations sge_qinfo_seq_ops = {
2010 .start = sge_queue_start,
2011 .next = sge_queue_next,
2012 .stop = sge_queue_stop,
2013 .show = sge_qinfo_show
2016 static int sge_qinfo_open(struct inode *inode, struct file *file)
2018 int res = seq_open(file, &sge_qinfo_seq_ops);
2021 struct seq_file *seq = file->private_data;
2022 seq->private = inode->i_private;
2027 static const struct file_operations sge_qinfo_debugfs_fops = {
2028 .owner = THIS_MODULE,
2029 .open = sge_qinfo_open,
2031 .llseek = seq_lseek,
2032 .release = seq_release,
2036 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
2040 static int sge_qstats_show(struct seq_file *seq, void *v)
2042 struct adapter *adapter = seq->private;
2043 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
2044 int qs, r = (uintptr_t)v - 1;
2047 seq_putc(seq, '\n');
2049 #define S3(fmt, s, v) \
2051 seq_printf(seq, "%-16s", s); \
2052 for (qs = 0; qs < n; ++qs) \
2053 seq_printf(seq, " %8" fmt, v); \
2054 seq_putc(seq, '\n'); \
2056 #define S(s, v) S3("s", s, v)
2058 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
2059 #define T(s, v) T3("lu", s, v)
2061 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
2062 #define R(s, v) R3("lu", s, v)
2064 if (r < eth_entries) {
2065 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
2066 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
2067 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
2069 S("QType:", "Ethernet");
2071 (rxq[qs].rspq.netdev
2072 ? rxq[qs].rspq.netdev->name
2074 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
2075 R("RxPackets:", stats.pkts);
2076 R("RxCSO:", stats.rx_cso);
2077 R("VLANxtract:", stats.vlan_ex);
2078 R("LROmerged:", stats.lro_merged);
2079 R("LROpackets:", stats.lro_pkts);
2080 R("RxDrops:", stats.rx_drops);
2082 T("TxCSO:", tx_cso);
2083 T("VLANins:", vlan_ins);
2084 T("TxQFull:", q.stops);
2085 T("TxQRestarts:", q.restarts);
2086 T("TxMapErr:", mapping_err);
2087 R("FLAllocErr:", fl.alloc_failed);
2088 R("FLLrgAlcErr:", fl.large_alloc_failed);
2089 R("FLStarving:", fl.starving);
2095 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
2097 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
2098 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
2099 evtq->unhandled_irqs);
2100 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
2101 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
2102 } else if (r == 1) {
2103 const struct sge_rspq *intrq = &adapter->sge.intrq;
2105 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
2106 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
2107 intrq->unhandled_irqs);
2108 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
2109 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
2123 * Return the number of "entries" in our "file". We group the multi-Queue
2124 * sections with QPL Queue Sets per "entry". The sections of the output are:
2126 * Ethernet RX/TX Queue Sets
2127 * Firmware Event Queue
2128 * Forwarded Interrupt Queue (if in MSI mode)
2130 static int sge_qstats_entries(const struct adapter *adapter)
2132 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
2133 ((adapter->flags & USING_MSI) != 0);
2136 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
2138 int entries = sge_qstats_entries(seq->private);
2140 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2143 static void sge_qstats_stop(struct seq_file *seq, void *v)
2147 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
2149 int entries = sge_qstats_entries(seq->private);
2152 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2155 static const struct seq_operations sge_qstats_seq_ops = {
2156 .start = sge_qstats_start,
2157 .next = sge_qstats_next,
2158 .stop = sge_qstats_stop,
2159 .show = sge_qstats_show
2162 static int sge_qstats_open(struct inode *inode, struct file *file)
2164 int res = seq_open(file, &sge_qstats_seq_ops);
2167 struct seq_file *seq = file->private_data;
2168 seq->private = inode->i_private;
2173 static const struct file_operations sge_qstats_proc_fops = {
2174 .owner = THIS_MODULE,
2175 .open = sge_qstats_open,
2177 .llseek = seq_lseek,
2178 .release = seq_release,
2182 * Show PCI-E SR-IOV Virtual Function Resource Limits.
2184 static int resources_show(struct seq_file *seq, void *v)
2186 struct adapter *adapter = seq->private;
2187 struct vf_resources *vfres = &adapter->params.vfres;
2189 #define S(desc, fmt, var) \
2190 seq_printf(seq, "%-60s " fmt "\n", \
2191 desc " (" #var "):", vfres->var)
2193 S("Virtual Interfaces", "%d", nvi);
2194 S("Egress Queues", "%d", neq);
2195 S("Ethernet Control", "%d", nethctrl);
2196 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
2197 S("Ingress Queues", "%d", niq);
2198 S("Traffic Class", "%d", tc);
2199 S("Port Access Rights Mask", "%#x", pmask);
2200 S("MAC Address Filters", "%d", nexactf);
2201 S("Firmware Command Read Capabilities", "%#x", r_caps);
2202 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
2209 static int resources_open(struct inode *inode, struct file *file)
2211 return single_open(file, resources_show, inode->i_private);
2214 static const struct file_operations resources_proc_fops = {
2215 .owner = THIS_MODULE,
2216 .open = resources_open,
2218 .llseek = seq_lseek,
2219 .release = single_release,
2223 * Show Virtual Interfaces.
2225 static int interfaces_show(struct seq_file *seq, void *v)
2227 if (v == SEQ_START_TOKEN) {
2228 seq_puts(seq, "Interface Port VIID\n");
2230 struct adapter *adapter = seq->private;
2231 int pidx = (uintptr_t)v - 2;
2232 struct net_device *dev = adapter->port[pidx];
2233 struct port_info *pi = netdev_priv(dev);
2235 seq_printf(seq, "%9s %4d %#5x\n",
2236 dev->name, pi->port_id, pi->viid);
2241 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
2243 return pos <= adapter->params.nports
2244 ? (void *)(uintptr_t)(pos + 1)
2248 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
2251 ? interfaces_get_idx(seq->private, *pos)
2255 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
2258 return interfaces_get_idx(seq->private, *pos);
2261 static void interfaces_stop(struct seq_file *seq, void *v)
2265 static const struct seq_operations interfaces_seq_ops = {
2266 .start = interfaces_start,
2267 .next = interfaces_next,
2268 .stop = interfaces_stop,
2269 .show = interfaces_show
2272 static int interfaces_open(struct inode *inode, struct file *file)
2274 int res = seq_open(file, &interfaces_seq_ops);
2277 struct seq_file *seq = file->private_data;
2278 seq->private = inode->i_private;
2283 static const struct file_operations interfaces_proc_fops = {
2284 .owner = THIS_MODULE,
2285 .open = interfaces_open,
2287 .llseek = seq_lseek,
2288 .release = seq_release,
2292 * /sys/kernel/debugfs/cxgb4vf/ files list.
2294 struct cxgb4vf_debugfs_entry {
2295 const char *name; /* name of debugfs node */
2296 umode_t mode; /* file system mode */
2297 const struct file_operations *fops;
2300 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
2301 { "mboxlog", S_IRUGO, &mboxlog_fops },
2302 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
2303 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
2304 { "resources", S_IRUGO, &resources_proc_fops },
2305 { "interfaces", S_IRUGO, &interfaces_proc_fops },
2309 * Module and device initialization and cleanup code.
2310 * ==================================================
2314 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2315 * directory (debugfs_root) has already been set up.
2317 static int setup_debugfs(struct adapter *adapter)
2321 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2324 * Debugfs support is best effort.
2326 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2327 (void)debugfs_create_file(debugfs_files[i].name,
2328 debugfs_files[i].mode,
2329 adapter->debugfs_root,
2331 debugfs_files[i].fops);
2337 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2338 * it to our caller to tear down the directory (debugfs_root).
2340 static void cleanup_debugfs(struct adapter *adapter)
2342 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2345 * Unlike our sister routine cleanup_proc(), we don't need to remove
2346 * individual entries because a call will be made to
2347 * debugfs_remove_recursive(). We just need to clean up any ancillary
2353 /* Figure out how many Ports and Queue Sets we can support. This depends on
2354 * knowing our Virtual Function Resources and may be called a second time if
2355 * we fall back from MSI-X to MSI Interrupt Mode.
2357 static void size_nports_qsets(struct adapter *adapter)
2359 struct vf_resources *vfres = &adapter->params.vfres;
2360 unsigned int ethqsets, pmask_nports;
2362 /* The number of "ports" which we support is equal to the number of
2363 * Virtual Interfaces with which we've been provisioned.
2365 adapter->params.nports = vfres->nvi;
2366 if (adapter->params.nports > MAX_NPORTS) {
2367 dev_warn(adapter->pdev_dev, "only using %d of %d maximum"
2368 " allowed virtual interfaces\n", MAX_NPORTS,
2369 adapter->params.nports);
2370 adapter->params.nports = MAX_NPORTS;
2373 /* We may have been provisioned with more VIs than the number of
2374 * ports we're allowed to access (our Port Access Rights Mask).
2375 * This is obviously a configuration conflict but we don't want to
2376 * crash the kernel or anything silly just because of that.
2378 pmask_nports = hweight32(adapter->params.vfres.pmask);
2379 if (pmask_nports < adapter->params.nports) {
2380 dev_warn(adapter->pdev_dev, "only using %d of %d provisioned"
2381 " virtual interfaces; limited by Port Access Rights"
2382 " mask %#x\n", pmask_nports, adapter->params.nports,
2383 adapter->params.vfres.pmask);
2384 adapter->params.nports = pmask_nports;
2387 /* We need to reserve an Ingress Queue for the Asynchronous Firmware
2388 * Event Queue. And if we're using MSI Interrupts, we'll also need to
2389 * reserve an Ingress Queue for a Forwarded Interrupts.
2391 * The rest of the FL/Intr-capable ingress queues will be matched up
2392 * one-for-one with Ethernet/Control egress queues in order to form
2393 * "Queue Sets" which will be aportioned between the "ports". For
2394 * each Queue Set, we'll need the ability to allocate two Egress
2395 * Contexts -- one for the Ingress Queue Free List and one for the TX
2398 * Note that even if we're currently configured to use MSI-X
2399 * Interrupts (module variable msi == MSI_MSIX) we may get downgraded
2400 * to MSI Interrupts if we can't get enough MSI-X Interrupts. If that
2401 * happens we'll need to adjust things later.
2403 ethqsets = vfres->niqflint - 1 - (msi == MSI_MSI);
2404 if (vfres->nethctrl != ethqsets)
2405 ethqsets = min(vfres->nethctrl, ethqsets);
2406 if (vfres->neq < ethqsets*2)
2407 ethqsets = vfres->neq/2;
2408 if (ethqsets > MAX_ETH_QSETS)
2409 ethqsets = MAX_ETH_QSETS;
2410 adapter->sge.max_ethqsets = ethqsets;
2412 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2413 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2414 " virtual interfaces (too few Queue Sets)\n",
2415 adapter->sge.max_ethqsets, adapter->params.nports);
2416 adapter->params.nports = adapter->sge.max_ethqsets;
2421 * Perform early "adapter" initialization. This is where we discover what
2422 * adapter parameters we're going to be using and initialize basic adapter
2425 static int adap_init0(struct adapter *adapter)
2427 struct sge_params *sge_params = &adapter->params.sge;
2428 struct sge *s = &adapter->sge;
2433 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2434 * 2.6.31 and later we can't call pci_reset_function() in order to
2435 * issue an FLR because of a self- deadlock on the device semaphore.
2436 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2437 * cases where they're needed -- for instance, some versions of KVM
2438 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2439 * use the firmware based reset in order to reset any per function
2442 err = t4vf_fw_reset(adapter);
2444 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2449 * Grab basic operational parameters. These will predominantly have
2450 * been set up by the Physical Function Driver or will be hard coded
2451 * into the adapter. We just have to live with them ... Note that
2452 * we _must_ get our VPD parameters before our SGE parameters because
2453 * we need to know the adapter's core clock from the VPD in order to
2454 * properly decode the SGE Timer Values.
2456 err = t4vf_get_dev_params(adapter);
2458 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2459 " device parameters: err=%d\n", err);
2462 err = t4vf_get_vpd_params(adapter);
2464 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2465 " VPD parameters: err=%d\n", err);
2468 err = t4vf_get_sge_params(adapter);
2470 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2471 " SGE parameters: err=%d\n", err);
2474 err = t4vf_get_rss_glb_config(adapter);
2476 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2477 " RSS parameters: err=%d\n", err);
2480 if (adapter->params.rss.mode !=
2481 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2482 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2483 " mode %d\n", adapter->params.rss.mode);
2486 err = t4vf_sge_init(adapter);
2488 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2493 /* If we're running on newer firmware, let it know that we're
2494 * prepared to deal with encapsulated CPL messages. Older
2495 * firmware won't understand this and we'll just get
2496 * unencapsulated messages ...
2498 param = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2499 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_CPLFW4MSG_ENCAP);
2501 (void) t4vf_set_params(adapter, 1, ¶m, &val);
2504 * Retrieve our RX interrupt holdoff timer values and counter
2505 * threshold values from the SGE parameters.
2507 s->timer_val[0] = core_ticks_to_us(adapter,
2508 TIMERVALUE0_G(sge_params->sge_timer_value_0_and_1));
2509 s->timer_val[1] = core_ticks_to_us(adapter,
2510 TIMERVALUE1_G(sge_params->sge_timer_value_0_and_1));
2511 s->timer_val[2] = core_ticks_to_us(adapter,
2512 TIMERVALUE0_G(sge_params->sge_timer_value_2_and_3));
2513 s->timer_val[3] = core_ticks_to_us(adapter,
2514 TIMERVALUE1_G(sge_params->sge_timer_value_2_and_3));
2515 s->timer_val[4] = core_ticks_to_us(adapter,
2516 TIMERVALUE0_G(sge_params->sge_timer_value_4_and_5));
2517 s->timer_val[5] = core_ticks_to_us(adapter,
2518 TIMERVALUE1_G(sge_params->sge_timer_value_4_and_5));
2520 s->counter_val[0] = THRESHOLD_0_G(sge_params->sge_ingress_rx_threshold);
2521 s->counter_val[1] = THRESHOLD_1_G(sge_params->sge_ingress_rx_threshold);
2522 s->counter_val[2] = THRESHOLD_2_G(sge_params->sge_ingress_rx_threshold);
2523 s->counter_val[3] = THRESHOLD_3_G(sge_params->sge_ingress_rx_threshold);
2526 * Grab our Virtual Interface resource allocation, extract the
2527 * features that we're interested in and do a bit of sanity testing on
2530 err = t4vf_get_vfres(adapter);
2532 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2533 " resources: err=%d\n", err);
2537 /* Check for various parameter sanity issues */
2538 if (adapter->params.vfres.pmask == 0) {
2539 dev_err(adapter->pdev_dev, "no port access configured\n"
2543 if (adapter->params.vfres.nvi == 0) {
2544 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2549 /* Initialize nports and max_ethqsets now that we have our Virtual
2550 * Function Resources.
2552 size_nports_qsets(adapter);
2557 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2558 u8 pkt_cnt_idx, unsigned int size,
2559 unsigned int iqe_size)
2561 rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) |
2562 (pkt_cnt_idx < SGE_NCOUNTERS ?
2563 QINTR_CNT_EN_F : 0));
2564 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2567 rspq->iqe_len = iqe_size;
2572 * Perform default configuration of DMA queues depending on the number and
2573 * type of ports we found and the number of available CPUs. Most settings can
2574 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2575 * being brought up for the first time.
2577 static void cfg_queues(struct adapter *adapter)
2579 struct sge *s = &adapter->sge;
2580 int q10g, n10g, qidx, pidx, qs;
2584 * We should not be called till we know how many Queue Sets we can
2585 * support. In particular, this means that we need to know what kind
2586 * of interrupts we'll be using ...
2588 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2591 * Count the number of 10GbE Virtual Interfaces that we have.
2594 for_each_port(adapter, pidx)
2595 n10g += is_x_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2598 * We default to 1 queue per non-10G port and up to # of cores queues
2604 int n1g = (adapter->params.nports - n10g);
2605 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2606 if (q10g > num_online_cpus())
2607 q10g = num_online_cpus();
2611 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2612 * The layout will be established in setup_sge_queues() when the
2613 * adapter is brough up for the first time.
2616 for_each_port(adapter, pidx) {
2617 struct port_info *pi = adap2pinfo(adapter, pidx);
2619 pi->first_qset = qidx;
2620 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
2626 * The Ingress Queue Entry Size for our various Response Queues needs
2627 * to be big enough to accommodate the largest message we can receive
2628 * from the chip/firmware; which is 64 bytes ...
2633 * Set up default Queue Set parameters ... Start off with the
2634 * shortest interrupt holdoff timer.
2636 for (qs = 0; qs < s->max_ethqsets; qs++) {
2637 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2638 struct sge_eth_txq *txq = &s->ethtxq[qs];
2640 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2646 * The firmware event queue is used for link state changes and
2647 * notifications of TX DMA completions.
2649 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2652 * The forwarded interrupt queue is used when we're in MSI interrupt
2653 * mode. In this mode all interrupts associated with RX queues will
2654 * be forwarded to a single queue which we'll associate with our MSI
2655 * interrupt vector. The messages dropped in the forwarded interrupt
2656 * queue will indicate which ingress queue needs servicing ... This
2657 * queue needs to be large enough to accommodate all of the ingress
2658 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2659 * from equalling the CIDX if every ingress queue has an outstanding
2660 * interrupt). The queue doesn't need to be any larger because no
2661 * ingress queue will ever have more than one outstanding interrupt at
2664 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2669 * Reduce the number of Ethernet queues across all ports to at most n.
2670 * n provides at least one queue per port.
2672 static void reduce_ethqs(struct adapter *adapter, int n)
2675 struct port_info *pi;
2678 * While we have too many active Ether Queue Sets, interate across the
2679 * "ports" and reduce their individual Queue Set allocations.
2681 BUG_ON(n < adapter->params.nports);
2682 while (n < adapter->sge.ethqsets)
2683 for_each_port(adapter, i) {
2684 pi = adap2pinfo(adapter, i);
2685 if (pi->nqsets > 1) {
2687 adapter->sge.ethqsets--;
2688 if (adapter->sge.ethqsets <= n)
2694 * Reassign the starting Queue Sets for each of the "ports" ...
2697 for_each_port(adapter, i) {
2698 pi = adap2pinfo(adapter, i);
2705 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2706 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2707 * need. Minimally we need one for every Virtual Interface plus those needed
2708 * for our "extras". Note that this process may lower the maximum number of
2709 * allowed Queue Sets ...
2711 static int enable_msix(struct adapter *adapter)
2713 int i, want, need, nqsets;
2714 struct msix_entry entries[MSIX_ENTRIES];
2715 struct sge *s = &adapter->sge;
2717 for (i = 0; i < MSIX_ENTRIES; ++i)
2718 entries[i].entry = i;
2721 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2722 * plus those needed for our "extras" (for example, the firmware
2723 * message queue). We _need_ at least one "Queue Set" per Virtual
2724 * Interface plus those needed for our "extras". So now we get to see
2725 * if the song is right ...
2727 want = s->max_ethqsets + MSIX_EXTRAS;
2728 need = adapter->params.nports + MSIX_EXTRAS;
2730 want = pci_enable_msix_range(adapter->pdev, entries, need, want);
2734 nqsets = want - MSIX_EXTRAS;
2735 if (nqsets < s->max_ethqsets) {
2736 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2737 " for %d Queue Sets\n", nqsets);
2738 s->max_ethqsets = nqsets;
2739 if (nqsets < s->ethqsets)
2740 reduce_ethqs(adapter, nqsets);
2742 for (i = 0; i < want; ++i)
2743 adapter->msix_info[i].vec = entries[i].vector;
2748 static const struct net_device_ops cxgb4vf_netdev_ops = {
2749 .ndo_open = cxgb4vf_open,
2750 .ndo_stop = cxgb4vf_stop,
2751 .ndo_start_xmit = t4vf_eth_xmit,
2752 .ndo_get_stats = cxgb4vf_get_stats,
2753 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2754 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2755 .ndo_validate_addr = eth_validate_addr,
2756 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2757 .ndo_change_mtu = cxgb4vf_change_mtu,
2758 .ndo_fix_features = cxgb4vf_fix_features,
2759 .ndo_set_features = cxgb4vf_set_features,
2760 #ifdef CONFIG_NET_POLL_CONTROLLER
2761 .ndo_poll_controller = cxgb4vf_poll_controller,
2766 * "Probe" a device: initialize a device and construct all kernel and driver
2767 * state needed to manage the device. This routine is called "init_one" in
2770 static int cxgb4vf_pci_probe(struct pci_dev *pdev,
2771 const struct pci_device_id *ent)
2776 struct adapter *adapter;
2777 struct port_info *pi;
2778 struct net_device *netdev;
2782 * Print our driver banner the first time we're called to initialize a
2785 pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION);
2788 * Initialize generic PCI device state.
2790 err = pci_enable_device(pdev);
2792 dev_err(&pdev->dev, "cannot enable PCI device\n");
2797 * Reserve PCI resources for the device. If we can't get them some
2798 * other driver may have already claimed the device ...
2800 err = pci_request_regions(pdev, KBUILD_MODNAME);
2802 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2803 goto err_disable_device;
2807 * Set up our DMA mask: try for 64-bit address masking first and
2808 * fall back to 32-bit if we can't get 64 bits ...
2810 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2812 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2814 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2815 " coherent allocations\n");
2816 goto err_release_regions;
2820 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2822 dev_err(&pdev->dev, "no usable DMA configuration\n");
2823 goto err_release_regions;
2829 * Enable bus mastering for the device ...
2831 pci_set_master(pdev);
2834 * Allocate our adapter data structure and attach it to the device.
2836 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2839 goto err_release_regions;
2841 pci_set_drvdata(pdev, adapter);
2842 adapter->pdev = pdev;
2843 adapter->pdev_dev = &pdev->dev;
2845 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
2846 (sizeof(struct mbox_cmd) *
2847 T4VF_OS_LOG_MBOX_CMDS),
2849 if (!adapter->mbox_log) {
2851 goto err_free_adapter;
2853 adapter->mbox_log->size = T4VF_OS_LOG_MBOX_CMDS;
2856 * Initialize SMP data synchronization resources.
2858 spin_lock_init(&adapter->stats_lock);
2859 spin_lock_init(&adapter->mbox_lock);
2860 INIT_LIST_HEAD(&adapter->mlist.list);
2863 * Map our I/O registers in BAR0.
2865 adapter->regs = pci_ioremap_bar(pdev, 0);
2866 if (!adapter->regs) {
2867 dev_err(&pdev->dev, "cannot map device registers\n");
2869 goto err_free_adapter;
2872 /* Wait for the device to become ready before proceeding ...
2874 err = t4vf_prep_adapter(adapter);
2876 dev_err(adapter->pdev_dev, "device didn't become ready:"
2878 goto err_unmap_bar0;
2881 /* For T5 and later we want to use the new BAR-based User Doorbells,
2882 * so we need to map BAR2 here ...
2884 if (!is_t4(adapter->params.chip)) {
2885 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
2886 pci_resource_len(pdev, 2));
2887 if (!adapter->bar2) {
2888 dev_err(adapter->pdev_dev, "cannot map BAR2 doorbells\n");
2890 goto err_unmap_bar0;
2894 * Initialize adapter level features.
2896 adapter->name = pci_name(pdev);
2897 adapter->msg_enable = dflt_msg_enable;
2898 err = adap_init0(adapter);
2902 /* Initialize hash mac addr list */
2903 INIT_LIST_HEAD(&adapter->mac_hlist);
2906 * Allocate our "adapter ports" and stitch everything together.
2908 pmask = adapter->params.vfres.pmask;
2909 pf = t4vf_get_pf_from_vf(adapter);
2910 for_each_port(adapter, pidx) {
2913 unsigned int naddr = 1;
2916 * We simplistically allocate our virtual interfaces
2917 * sequentially across the port numbers to which we have
2918 * access rights. This should be configurable in some manner
2923 port_id = ffs(pmask) - 1;
2924 pmask &= ~(1 << port_id);
2925 viid = t4vf_alloc_vi(adapter, port_id);
2927 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2928 " err=%d\n", port_id, viid);
2934 * Allocate our network device and stitch things together.
2936 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2938 if (netdev == NULL) {
2939 t4vf_free_vi(adapter, viid);
2943 adapter->port[pidx] = netdev;
2944 SET_NETDEV_DEV(netdev, &pdev->dev);
2945 pi = netdev_priv(netdev);
2946 pi->adapter = adapter;
2948 pi->port_id = port_id;
2952 * Initialize the starting state of our "port" and register
2955 pi->xact_addr_filt = -1;
2956 netif_carrier_off(netdev);
2957 netdev->irq = pdev->irq;
2959 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
2960 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2961 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXCSUM;
2962 netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
2963 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2965 netdev->features = netdev->hw_features |
2966 NETIF_F_HW_VLAN_CTAG_TX;
2968 netdev->features |= NETIF_F_HIGHDMA;
2970 netdev->priv_flags |= IFF_UNICAST_FLT;
2972 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2973 netdev->ethtool_ops = &cxgb4vf_ethtool_ops;
2974 netdev->dev_port = pi->port_id;
2977 * Initialize the hardware/software state for the port.
2979 err = t4vf_port_init(adapter, pidx);
2981 dev_err(&pdev->dev, "cannot initialize port %d\n",
2986 err = t4vf_get_vf_mac_acl(adapter, pf, &naddr, mac);
2989 "unable to determine MAC ACL address, "
2990 "continuing anyway.. (status %d)\n", err);
2991 } else if (naddr && adapter->params.vfres.nvi == 1) {
2992 struct sockaddr addr;
2994 ether_addr_copy(addr.sa_data, mac);
2995 err = cxgb4vf_set_mac_addr(netdev, &addr);
2998 "unable to set MAC address %pM\n",
3002 dev_info(&pdev->dev,
3003 "Using assigned MAC ACL: %pM\n", mac);
3007 /* See what interrupts we'll be using. If we've been configured to
3008 * use MSI-X interrupts, try to enable them but fall back to using
3009 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
3010 * get MSI interrupts we bail with the error.
3012 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
3013 adapter->flags |= USING_MSIX;
3015 if (msi == MSI_MSIX) {
3016 dev_info(adapter->pdev_dev,
3017 "Unable to use MSI-X Interrupts; falling "
3018 "back to MSI Interrupts\n");
3020 /* We're going to need a Forwarded Interrupt Queue so
3021 * that may cut into how many Queue Sets we can
3025 size_nports_qsets(adapter);
3027 err = pci_enable_msi(pdev);
3029 dev_err(&pdev->dev, "Unable to allocate MSI Interrupts;"
3033 adapter->flags |= USING_MSI;
3036 /* Now that we know how many "ports" we have and what interrupt
3037 * mechanism we're going to use, we can configure our queue resources.
3039 cfg_queues(adapter);
3042 * The "card" is now ready to go. If any errors occur during device
3043 * registration we do not fail the whole "card" but rather proceed
3044 * only with the ports we manage to register successfully. However we
3045 * must register at least one net device.
3047 for_each_port(adapter, pidx) {
3048 struct port_info *pi = netdev_priv(adapter->port[pidx]);
3049 netdev = adapter->port[pidx];
3053 netif_set_real_num_tx_queues(netdev, pi->nqsets);
3054 netif_set_real_num_rx_queues(netdev, pi->nqsets);
3056 err = register_netdev(netdev);
3058 dev_warn(&pdev->dev, "cannot register net device %s,"
3059 " skipping\n", netdev->name);
3063 set_bit(pidx, &adapter->registered_device_map);
3065 if (adapter->registered_device_map == 0) {
3066 dev_err(&pdev->dev, "could not register any net devices\n");
3067 goto err_disable_interrupts;
3071 * Set up our debugfs entries.
3073 if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) {
3074 adapter->debugfs_root =
3075 debugfs_create_dir(pci_name(pdev),
3076 cxgb4vf_debugfs_root);
3077 if (IS_ERR_OR_NULL(adapter->debugfs_root))
3078 dev_warn(&pdev->dev, "could not create debugfs"
3081 setup_debugfs(adapter);
3085 * Print a short notice on the existence and configuration of the new
3086 * VF network device ...
3088 for_each_port(adapter, pidx) {
3089 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
3090 adapter->port[pidx]->name,
3091 (adapter->flags & USING_MSIX) ? "MSI-X" :
3092 (adapter->flags & USING_MSI) ? "MSI" : "");
3101 * Error recovery and exit code. Unwind state that's been created
3102 * so far and return the error.
3104 err_disable_interrupts:
3105 if (adapter->flags & USING_MSIX) {
3106 pci_disable_msix(adapter->pdev);
3107 adapter->flags &= ~USING_MSIX;
3108 } else if (adapter->flags & USING_MSI) {
3109 pci_disable_msi(adapter->pdev);
3110 adapter->flags &= ~USING_MSI;
3114 for_each_port(adapter, pidx) {
3115 netdev = adapter->port[pidx];
3118 pi = netdev_priv(netdev);
3119 t4vf_free_vi(adapter, pi->viid);
3120 if (test_bit(pidx, &adapter->registered_device_map))
3121 unregister_netdev(netdev);
3122 free_netdev(netdev);
3126 if (!is_t4(adapter->params.chip))
3127 iounmap(adapter->bar2);
3130 iounmap(adapter->regs);
3133 kfree(adapter->mbox_log);
3136 err_release_regions:
3137 pci_release_regions(pdev);
3138 pci_clear_master(pdev);
3141 pci_disable_device(pdev);
3147 * "Remove" a device: tear down all kernel and driver state created in the
3148 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
3149 * that this is called "remove_one" in the PF Driver.)
3151 static void cxgb4vf_pci_remove(struct pci_dev *pdev)
3153 struct adapter *adapter = pci_get_drvdata(pdev);
3156 * Tear down driver state associated with device.
3162 * Stop all of our activity. Unregister network port,
3163 * disable interrupts, etc.
3165 for_each_port(adapter, pidx)
3166 if (test_bit(pidx, &adapter->registered_device_map))
3167 unregister_netdev(adapter->port[pidx]);
3168 t4vf_sge_stop(adapter);
3169 if (adapter->flags & USING_MSIX) {
3170 pci_disable_msix(adapter->pdev);
3171 adapter->flags &= ~USING_MSIX;
3172 } else if (adapter->flags & USING_MSI) {
3173 pci_disable_msi(adapter->pdev);
3174 adapter->flags &= ~USING_MSI;
3178 * Tear down our debugfs entries.
3180 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
3181 cleanup_debugfs(adapter);
3182 debugfs_remove_recursive(adapter->debugfs_root);
3186 * Free all of the various resources which we've acquired ...
3188 t4vf_free_sge_resources(adapter);
3189 for_each_port(adapter, pidx) {
3190 struct net_device *netdev = adapter->port[pidx];
3191 struct port_info *pi;
3196 pi = netdev_priv(netdev);
3197 t4vf_free_vi(adapter, pi->viid);
3198 free_netdev(netdev);
3200 iounmap(adapter->regs);
3201 if (!is_t4(adapter->params.chip))
3202 iounmap(adapter->bar2);
3203 kfree(adapter->mbox_log);
3208 * Disable the device and release its PCI resources.
3210 pci_disable_device(pdev);
3211 pci_clear_master(pdev);
3212 pci_release_regions(pdev);
3216 * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt
3219 static void cxgb4vf_pci_shutdown(struct pci_dev *pdev)
3221 struct adapter *adapter;
3224 adapter = pci_get_drvdata(pdev);
3228 /* Disable all Virtual Interfaces. This will shut down the
3229 * delivery of all ingress packets into the chip for these
3230 * Virtual Interfaces.
3232 for_each_port(adapter, pidx)
3233 if (test_bit(pidx, &adapter->registered_device_map))
3234 unregister_netdev(adapter->port[pidx]);
3236 /* Free up all Queues which will prevent further DMA and
3237 * Interrupts allowing various internal pathways to drain.
3239 t4vf_sge_stop(adapter);
3240 if (adapter->flags & USING_MSIX) {
3241 pci_disable_msix(adapter->pdev);
3242 adapter->flags &= ~USING_MSIX;
3243 } else if (adapter->flags & USING_MSI) {
3244 pci_disable_msi(adapter->pdev);
3245 adapter->flags &= ~USING_MSI;
3249 * Free up all Queues which will prevent further DMA and
3250 * Interrupts allowing various internal pathways to drain.
3252 t4vf_free_sge_resources(adapter);
3253 pci_set_drvdata(pdev, NULL);
3256 /* Macros needed to support the PCI Device ID Table ...
3258 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
3259 static const struct pci_device_id cxgb4vf_pci_tbl[] = {
3260 #define CH_PCI_DEVICE_ID_FUNCTION 0x8
3262 #define CH_PCI_ID_TABLE_ENTRY(devid) \
3263 { PCI_VDEVICE(CHELSIO, (devid)), 0 }
3265 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END { 0, } }
3267 #include "../cxgb4/t4_pci_id_tbl.h"
3269 MODULE_DESCRIPTION(DRV_DESC);
3270 MODULE_AUTHOR("Chelsio Communications");
3271 MODULE_LICENSE("Dual BSD/GPL");
3272 MODULE_VERSION(DRV_VERSION);
3273 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
3275 static struct pci_driver cxgb4vf_driver = {
3276 .name = KBUILD_MODNAME,
3277 .id_table = cxgb4vf_pci_tbl,
3278 .probe = cxgb4vf_pci_probe,
3279 .remove = cxgb4vf_pci_remove,
3280 .shutdown = cxgb4vf_pci_shutdown,
3284 * Initialize global driver state.
3286 static int __init cxgb4vf_module_init(void)
3291 * Vet our module parameters.
3293 if (msi != MSI_MSIX && msi != MSI_MSI) {
3294 pr_warn("bad module parameter msi=%d; must be %d (MSI-X or MSI) or %d (MSI)\n",
3295 msi, MSI_MSIX, MSI_MSI);
3299 /* Debugfs support is optional, just warn if this fails */
3300 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3301 if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3302 pr_warn("could not create debugfs entry, continuing\n");
3304 ret = pci_register_driver(&cxgb4vf_driver);
3305 if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3306 debugfs_remove(cxgb4vf_debugfs_root);
3311 * Tear down global driver state.
3313 static void __exit cxgb4vf_module_exit(void)
3315 pci_unregister_driver(&cxgb4vf_driver);
3316 debugfs_remove(cxgb4vf_debugfs_root);
3319 module_init(cxgb4vf_module_init);
3320 module_exit(cxgb4vf_module_exit);