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)
74 * The driver uses the best interrupt scheme available on a platform in the
75 * order MSI-X then MSI. This parameter determines which of these schemes the
76 * driver may consider as follows:
78 * msi = 2: choose from among MSI-X and MSI
79 * msi = 1: only consider MSI interrupts
81 * Note that unlike the Physical Function driver, this Virtual Function driver
82 * does _not_ support legacy INTx interrupts (this limitation is mandated by
83 * the PCI-E SR-IOV standard).
87 #define MSI_DEFAULT MSI_MSIX
89 static int msi = MSI_DEFAULT;
91 module_param(msi, int, 0644);
92 MODULE_PARM_DESC(msi, "whether to use MSI-X or MSI");
95 * Fundamental constants.
96 * ======================
100 MAX_TXQ_ENTRIES = 16384,
101 MAX_RSPQ_ENTRIES = 16384,
102 MAX_RX_BUFFERS = 16384,
104 MIN_TXQ_ENTRIES = 32,
105 MIN_RSPQ_ENTRIES = 128,
109 * For purposes of manipulating the Free List size we need to
110 * recognize that Free Lists are actually Egress Queues (the host
111 * produces free buffers which the hardware consumes), Egress Queues
112 * indices are all in units of Egress Context Units bytes, and free
113 * list entries are 64-bit PCI DMA addresses. And since the state of
114 * the Producer Index == the Consumer Index implies an EMPTY list, we
115 * always have at least one Egress Unit's worth of Free List entries
116 * unused. See sge.c for more details ...
118 EQ_UNIT = SGE_EQ_IDXSIZE,
119 FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64),
120 MIN_FL_RESID = FL_PER_EQ_UNIT,
124 * Global driver state.
125 * ====================
128 static struct dentry *cxgb4vf_debugfs_root;
131 * OS "Callback" functions.
132 * ========================
136 * The link status has changed on the indicated "port" (Virtual Interface).
138 void t4vf_os_link_changed(struct adapter *adapter, int pidx, int link_ok)
140 struct net_device *dev = adapter->port[pidx];
143 * If the port is disabled or the current recorded "link up"
144 * status matches the new status, just return.
146 if (!netif_running(dev) || link_ok == netif_carrier_ok(dev))
150 * Tell the OS that the link status has changed and print a short
151 * informative message on the console about the event.
156 const struct port_info *pi = netdev_priv(dev);
158 netif_carrier_on(dev);
160 switch (pi->link_cfg.speed) {
185 switch (pi->link_cfg.fc) {
194 case PAUSE_RX|PAUSE_TX:
203 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s, fc);
205 netif_carrier_off(dev);
206 netdev_info(dev, "link down\n");
211 * THe port module type has changed on the indicated "port" (Virtual
214 void t4vf_os_portmod_changed(struct adapter *adapter, int pidx)
216 static const char * const mod_str[] = {
217 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
219 const struct net_device *dev = adapter->port[pidx];
220 const struct port_info *pi = netdev_priv(dev);
222 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
223 dev_info(adapter->pdev_dev, "%s: port module unplugged\n",
225 else if (pi->mod_type < ARRAY_SIZE(mod_str))
226 dev_info(adapter->pdev_dev, "%s: %s port module inserted\n",
227 dev->name, mod_str[pi->mod_type]);
228 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
229 dev_info(adapter->pdev_dev, "%s: unsupported optical port "
230 "module inserted\n", dev->name);
231 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
232 dev_info(adapter->pdev_dev, "%s: unknown port module inserted,"
233 "forcing TWINAX\n", dev->name);
234 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
235 dev_info(adapter->pdev_dev, "%s: transceiver module error\n",
238 dev_info(adapter->pdev_dev, "%s: unknown module type %d "
239 "inserted\n", dev->name, pi->mod_type);
243 * Net device operations.
244 * ======================
251 * Perform the MAC and PHY actions needed to enable a "port" (Virtual
254 static int link_start(struct net_device *dev)
257 struct port_info *pi = netdev_priv(dev);
260 * We do not set address filters and promiscuity here, the stack does
261 * that step explicitly. Enable vlan accel.
263 ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, 1,
266 ret = t4vf_change_mac(pi->adapter, pi->viid,
267 pi->xact_addr_filt, dev->dev_addr, true);
269 pi->xact_addr_filt = ret;
275 * We don't need to actually "start the link" itself since the
276 * firmware will do that for us when the first Virtual Interface
277 * is enabled on a port.
280 ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true);
285 * Name the MSI-X interrupts.
287 static void name_msix_vecs(struct adapter *adapter)
289 int namelen = sizeof(adapter->msix_info[0].desc) - 1;
295 snprintf(adapter->msix_info[MSIX_FW].desc, namelen,
296 "%s-FWeventq", adapter->name);
297 adapter->msix_info[MSIX_FW].desc[namelen] = 0;
302 for_each_port(adapter, pidx) {
303 struct net_device *dev = adapter->port[pidx];
304 const struct port_info *pi = netdev_priv(dev);
307 for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) {
308 snprintf(adapter->msix_info[msi].desc, namelen,
309 "%s-%d", dev->name, qs);
310 adapter->msix_info[msi].desc[namelen] = 0;
316 * Request all of our MSI-X resources.
318 static int request_msix_queue_irqs(struct adapter *adapter)
320 struct sge *s = &adapter->sge;
326 err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix,
327 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq);
335 for_each_ethrxq(s, rxq) {
336 err = request_irq(adapter->msix_info[msi].vec,
337 t4vf_sge_intr_msix, 0,
338 adapter->msix_info[msi].desc,
339 &s->ethrxq[rxq].rspq);
348 free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq);
349 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
354 * Free our MSI-X resources.
356 static void free_msix_queue_irqs(struct adapter *adapter)
358 struct sge *s = &adapter->sge;
361 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
363 for_each_ethrxq(s, rxq)
364 free_irq(adapter->msix_info[msi++].vec,
365 &s->ethrxq[rxq].rspq);
369 * Turn on NAPI and start up interrupts on a response queue.
371 static void qenable(struct sge_rspq *rspq)
373 napi_enable(&rspq->napi);
376 * 0-increment the Going To Sleep register to start the timer and
379 t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
381 SEINTARM_V(rspq->intr_params) |
382 INGRESSQID_V(rspq->cntxt_id));
386 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
388 static void enable_rx(struct adapter *adapter)
391 struct sge *s = &adapter->sge;
393 for_each_ethrxq(s, rxq)
394 qenable(&s->ethrxq[rxq].rspq);
395 qenable(&s->fw_evtq);
398 * The interrupt queue doesn't use NAPI so we do the 0-increment of
399 * its Going To Sleep register here to get it started.
401 if (adapter->flags & USING_MSI)
402 t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
404 SEINTARM_V(s->intrq.intr_params) |
405 INGRESSQID_V(s->intrq.cntxt_id));
410 * Wait until all NAPI handlers are descheduled.
412 static void quiesce_rx(struct adapter *adapter)
414 struct sge *s = &adapter->sge;
417 for_each_ethrxq(s, rxq)
418 napi_disable(&s->ethrxq[rxq].rspq.napi);
419 napi_disable(&s->fw_evtq.napi);
423 * Response queue handler for the firmware event queue.
425 static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp,
426 const struct pkt_gl *gl)
429 * Extract response opcode and get pointer to CPL message body.
431 struct adapter *adapter = rspq->adapter;
432 u8 opcode = ((const struct rss_header *)rsp)->opcode;
433 void *cpl = (void *)(rsp + 1);
438 * We've received an asynchronous message from the firmware.
440 const struct cpl_fw6_msg *fw_msg = cpl;
441 if (fw_msg->type == FW6_TYPE_CMD_RPL)
442 t4vf_handle_fw_rpl(adapter, fw_msg->data);
447 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
449 const struct cpl_sge_egr_update *p = (void *)(rsp + 3);
450 opcode = CPL_OPCODE_G(ntohl(p->opcode_qid));
451 if (opcode != CPL_SGE_EGR_UPDATE) {
452 dev_err(adapter->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
460 case CPL_SGE_EGR_UPDATE: {
462 * We've received an Egress Queue Status Update message. We
463 * get these, if the SGE is configured to send these when the
464 * firmware passes certain points in processing our TX
465 * Ethernet Queue or if we make an explicit request for one.
466 * We use these updates to determine when we may need to
467 * restart a TX Ethernet Queue which was stopped for lack of
468 * free TX Queue Descriptors ...
470 const struct cpl_sge_egr_update *p = cpl;
471 unsigned int qid = EGR_QID_G(be32_to_cpu(p->opcode_qid));
472 struct sge *s = &adapter->sge;
474 struct sge_eth_txq *txq;
478 * Perform sanity checking on the Queue ID to make sure it
479 * really refers to one of our TX Ethernet Egress Queues which
480 * is active and matches the queue's ID. None of these error
481 * conditions should ever happen so we may want to either make
482 * them fatal and/or conditionalized under DEBUG.
484 eq_idx = EQ_IDX(s, qid);
485 if (unlikely(eq_idx >= MAX_EGRQ)) {
486 dev_err(adapter->pdev_dev,
487 "Egress Update QID %d out of range\n", qid);
490 tq = s->egr_map[eq_idx];
491 if (unlikely(tq == NULL)) {
492 dev_err(adapter->pdev_dev,
493 "Egress Update QID %d TXQ=NULL\n", qid);
496 txq = container_of(tq, struct sge_eth_txq, q);
497 if (unlikely(tq->abs_id != qid)) {
498 dev_err(adapter->pdev_dev,
499 "Egress Update QID %d refers to TXQ %d\n",
505 * Restart a stopped TX Queue which has less than half of its
509 netif_tx_wake_queue(txq->txq);
514 dev_err(adapter->pdev_dev,
515 "unexpected CPL %#x on FW event queue\n", opcode);
522 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
523 * to use and initializes them. We support multiple "Queue Sets" per port if
524 * we have MSI-X, otherwise just one queue set per port.
526 static int setup_sge_queues(struct adapter *adapter)
528 struct sge *s = &adapter->sge;
532 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
535 bitmap_zero(s->starving_fl, MAX_EGRQ);
538 * If we're using MSI interrupt mode we need to set up a "forwarded
539 * interrupt" queue which we'll set up with our MSI vector. The rest
540 * of the ingress queues will be set up to forward their interrupts to
541 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
542 * the intrq's queue ID as the interrupt forwarding queue for the
543 * subsequent calls ...
545 if (adapter->flags & USING_MSI) {
546 err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false,
547 adapter->port[0], 0, NULL, NULL);
549 goto err_free_queues;
553 * Allocate our ingress queue for asynchronous firmware messages.
555 err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0],
556 MSIX_FW, NULL, fwevtq_handler);
558 goto err_free_queues;
561 * Allocate each "port"'s initial Queue Sets. These can be changed
562 * later on ... up to the point where any interface on the adapter is
563 * brought up at which point lots of things get nailed down
567 for_each_port(adapter, pidx) {
568 struct net_device *dev = adapter->port[pidx];
569 struct port_info *pi = netdev_priv(dev);
570 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
571 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
574 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
575 err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false,
577 &rxq->fl, t4vf_ethrx_handler);
579 goto err_free_queues;
581 err = t4vf_sge_alloc_eth_txq(adapter, txq, dev,
582 netdev_get_tx_queue(dev, qs),
583 s->fw_evtq.cntxt_id);
585 goto err_free_queues;
588 memset(&rxq->stats, 0, sizeof(rxq->stats));
593 * Create the reverse mappings for the queues.
595 s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id;
596 s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id;
597 IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq;
598 for_each_port(adapter, pidx) {
599 struct net_device *dev = adapter->port[pidx];
600 struct port_info *pi = netdev_priv(dev);
601 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
602 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
605 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
606 IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq;
607 EQ_MAP(s, txq->q.abs_id) = &txq->q;
610 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
611 * for Free Lists but since all of the Egress Queues
612 * (including Free Lists) have Relative Queue IDs
613 * which are computed as Absolute - Base Queue ID, we
614 * can synthesize the Absolute Queue IDs for the Free
615 * Lists. This is useful for debugging purposes when
616 * we want to dump Queue Contexts via the PF Driver.
618 rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base;
619 EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl;
625 t4vf_free_sge_resources(adapter);
630 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
631 * queues. We configure the RSS CPU lookup table to distribute to the number
632 * of HW receive queues, and the response queue lookup table to narrow that
633 * down to the response queues actually configured for each "port" (Virtual
634 * Interface). We always configure the RSS mapping for all ports since the
635 * mapping table has plenty of entries.
637 static int setup_rss(struct adapter *adapter)
641 for_each_port(adapter, pidx) {
642 struct port_info *pi = adap2pinfo(adapter, pidx);
643 struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
644 u16 rss[MAX_PORT_QSETS];
647 for (qs = 0; qs < pi->nqsets; qs++)
648 rss[qs] = rxq[qs].rspq.abs_id;
650 err = t4vf_config_rss_range(adapter, pi->viid,
651 0, pi->rss_size, rss, pi->nqsets);
656 * Perform Global RSS Mode-specific initialization.
658 switch (adapter->params.rss.mode) {
659 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL:
661 * If Tunnel All Lookup isn't specified in the global
662 * RSS Configuration, then we need to specify a
663 * default Ingress Queue for any ingress packets which
664 * aren't hashed. We'll use our first ingress queue
667 if (!adapter->params.rss.u.basicvirtual.tnlalllookup) {
668 union rss_vi_config config;
669 err = t4vf_read_rss_vi_config(adapter,
674 config.basicvirtual.defaultq =
676 err = t4vf_write_rss_vi_config(adapter,
690 * Bring the adapter up. Called whenever we go from no "ports" open to having
691 * one open. This function performs the actions necessary to make an adapter
692 * operational, such as completing the initialization of HW modules, and
693 * enabling interrupts. Must be called with the rtnl lock held. (Note that
694 * this is called "cxgb_up" in the PF Driver.)
696 static int adapter_up(struct adapter *adapter)
701 * If this is the first time we've been called, perform basic
702 * adapter setup. Once we've done this, many of our adapter
703 * parameters can no longer be changed ...
705 if ((adapter->flags & FULL_INIT_DONE) == 0) {
706 err = setup_sge_queues(adapter);
709 err = setup_rss(adapter);
711 t4vf_free_sge_resources(adapter);
715 if (adapter->flags & USING_MSIX)
716 name_msix_vecs(adapter);
717 adapter->flags |= FULL_INIT_DONE;
721 * Acquire our interrupt resources. We only support MSI-X and MSI.
723 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
724 if (adapter->flags & USING_MSIX)
725 err = request_msix_queue_irqs(adapter);
727 err = request_irq(adapter->pdev->irq,
728 t4vf_intr_handler(adapter), 0,
729 adapter->name, adapter);
731 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
737 * Enable NAPI ingress processing and return success.
740 t4vf_sge_start(adapter);
742 /* Initialize hash mac addr list*/
743 INIT_LIST_HEAD(&adapter->mac_hlist);
748 * Bring the adapter down. Called whenever the last "port" (Virtual
749 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
752 static void adapter_down(struct adapter *adapter)
755 * Free interrupt resources.
757 if (adapter->flags & USING_MSIX)
758 free_msix_queue_irqs(adapter);
760 free_irq(adapter->pdev->irq, adapter);
763 * Wait for NAPI handlers to finish.
769 * Start up a net device.
771 static int cxgb4vf_open(struct net_device *dev)
774 struct port_info *pi = netdev_priv(dev);
775 struct adapter *adapter = pi->adapter;
778 * If this is the first interface that we're opening on the "adapter",
779 * bring the "adapter" up now.
781 if (adapter->open_device_map == 0) {
782 err = adapter_up(adapter);
788 * Note that this interface is up and start everything up ...
790 err = link_start(dev);
794 netif_tx_start_all_queues(dev);
795 set_bit(pi->port_id, &adapter->open_device_map);
799 if (adapter->open_device_map == 0)
800 adapter_down(adapter);
805 * Shut down a net device. This routine is called "cxgb_close" in the PF
808 static int cxgb4vf_stop(struct net_device *dev)
810 struct port_info *pi = netdev_priv(dev);
811 struct adapter *adapter = pi->adapter;
813 netif_tx_stop_all_queues(dev);
814 netif_carrier_off(dev);
815 t4vf_enable_vi(adapter, pi->viid, false, false);
816 pi->link_cfg.link_ok = 0;
818 clear_bit(pi->port_id, &adapter->open_device_map);
819 if (adapter->open_device_map == 0)
820 adapter_down(adapter);
825 * Translate our basic statistics into the standard "ifconfig" statistics.
827 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
829 struct t4vf_port_stats stats;
830 struct port_info *pi = netdev2pinfo(dev);
831 struct adapter *adapter = pi->adapter;
832 struct net_device_stats *ns = &dev->stats;
835 spin_lock(&adapter->stats_lock);
836 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
837 spin_unlock(&adapter->stats_lock);
839 memset(ns, 0, sizeof(*ns));
843 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
844 stats.tx_ucast_bytes + stats.tx_offload_bytes);
845 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
846 stats.tx_ucast_frames + stats.tx_offload_frames);
847 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
848 stats.rx_ucast_bytes);
849 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
850 stats.rx_ucast_frames);
851 ns->multicast = stats.rx_mcast_frames;
852 ns->tx_errors = stats.tx_drop_frames;
853 ns->rx_errors = stats.rx_err_frames;
858 static inline int cxgb4vf_set_addr_hash(struct port_info *pi)
860 struct adapter *adapter = pi->adapter;
863 struct hash_mac_addr *entry;
865 /* Calculate the hash vector for the updated list and program it */
866 list_for_each_entry(entry, &adapter->mac_hlist, list) {
867 ucast |= is_unicast_ether_addr(entry->addr);
868 vec |= (1ULL << hash_mac_addr(entry->addr));
870 return t4vf_set_addr_hash(adapter, pi->viid, ucast, vec, false);
873 static int cxgb4vf_mac_sync(struct net_device *netdev, const u8 *mac_addr)
875 struct port_info *pi = netdev_priv(netdev);
876 struct adapter *adapter = pi->adapter;
881 bool ucast = is_unicast_ether_addr(mac_addr);
882 const u8 *maclist[1] = {mac_addr};
883 struct hash_mac_addr *new_entry;
885 ret = t4vf_alloc_mac_filt(adapter, pi->viid, free, 1, maclist,
886 NULL, ucast ? &uhash : &mhash, false);
889 /* if hash != 0, then add the addr to hash addr list
890 * so on the end we will calculate the hash for the
891 * list and program it
893 if (uhash || mhash) {
894 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
897 ether_addr_copy(new_entry->addr, mac_addr);
898 list_add_tail(&new_entry->list, &adapter->mac_hlist);
899 ret = cxgb4vf_set_addr_hash(pi);
902 return ret < 0 ? ret : 0;
905 static int cxgb4vf_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
907 struct port_info *pi = netdev_priv(netdev);
908 struct adapter *adapter = pi->adapter;
910 const u8 *maclist[1] = {mac_addr};
911 struct hash_mac_addr *entry, *tmp;
913 /* If the MAC address to be removed is in the hash addr
914 * list, delete it from the list and update hash vector
916 list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist, list) {
917 if (ether_addr_equal(entry->addr, mac_addr)) {
918 list_del(&entry->list);
920 return cxgb4vf_set_addr_hash(pi);
924 ret = t4vf_free_mac_filt(adapter, pi->viid, 1, maclist, false);
925 return ret < 0 ? -EINVAL : 0;
929 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
930 * If @mtu is -1 it is left unchanged.
932 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
934 struct port_info *pi = netdev_priv(dev);
936 __dev_uc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync);
937 __dev_mc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync);
938 return t4vf_set_rxmode(pi->adapter, pi->viid, -1,
939 (dev->flags & IFF_PROMISC) != 0,
940 (dev->flags & IFF_ALLMULTI) != 0,
945 * Set the current receive modes on the device.
947 static void cxgb4vf_set_rxmode(struct net_device *dev)
949 /* unfortunately we can't return errors to the stack */
950 set_rxmode(dev, -1, false);
954 * Find the entry in the interrupt holdoff timer value array which comes
955 * closest to the specified interrupt holdoff value.
957 static int closest_timer(const struct sge *s, int us)
959 int i, timer_idx = 0, min_delta = INT_MAX;
961 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
962 int delta = us - s->timer_val[i];
965 if (delta < min_delta) {
973 static int closest_thres(const struct sge *s, int thres)
975 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
977 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
978 delta = thres - s->counter_val[i];
981 if (delta < min_delta) {
990 * Return a queue's interrupt hold-off time in us. 0 means no timer.
992 static unsigned int qtimer_val(const struct adapter *adapter,
993 const struct sge_rspq *rspq)
995 unsigned int timer_idx = QINTR_TIMER_IDX_G(rspq->intr_params);
997 return timer_idx < SGE_NTIMERS
998 ? adapter->sge.timer_val[timer_idx]
1003 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1004 * @adapter: the adapter
1005 * @rspq: the RX response queue
1006 * @us: the hold-off time in us, or 0 to disable timer
1007 * @cnt: the hold-off packet count, or 0 to disable counter
1009 * Sets an RX response queue's interrupt hold-off time and packet count.
1010 * At least one of the two needs to be enabled for the queue to generate
1013 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
1014 unsigned int us, unsigned int cnt)
1016 unsigned int timer_idx;
1019 * If both the interrupt holdoff timer and count are specified as
1020 * zero, default to a holdoff count of 1 ...
1022 if ((us | cnt) == 0)
1026 * If an interrupt holdoff count has been specified, then find the
1027 * closest configured holdoff count and use that. If the response
1028 * queue has already been created, then update its queue context
1035 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1036 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1037 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1038 FW_PARAMS_PARAM_X_V(
1039 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1040 FW_PARAMS_PARAM_YZ_V(rspq->cntxt_id);
1041 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1045 rspq->pktcnt_idx = pktcnt_idx;
1049 * Compute the closest holdoff timer index from the supplied holdoff
1052 timer_idx = (us == 0
1053 ? SGE_TIMER_RSTRT_CNTR
1054 : closest_timer(&adapter->sge, us));
1057 * Update the response queue's interrupt coalescing parameters and
1060 rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) |
1061 QINTR_CNT_EN_V(cnt > 0));
1066 * Return a version number to identify the type of adapter. The scheme is:
1067 * - bits 0..9: chip version
1068 * - bits 10..15: chip revision
1070 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1073 * Chip version 4, revision 0x3f (cxgb4vf).
1075 return CHELSIO_CHIP_VERSION(adapter->params.chip) | (0x3f << 10);
1079 * Execute the specified ioctl command.
1081 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1087 * The VF Driver doesn't have access to any of the other
1088 * common Ethernet device ioctl()'s (like reading/writing
1089 * PHY registers, etc.
1100 * Change the device's MTU.
1102 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1105 struct port_info *pi = netdev_priv(dev);
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 * from_fw_port_mod_type - translate Firmware Port/Module type to Ethtool
1198 * @port_type: Firmware Port Type
1199 * @mod_type: Firmware Module Type
1201 * Translate Firmware Port/Module type to Ethtool Port Type.
1203 static int from_fw_port_mod_type(enum fw_port_type port_type,
1204 enum fw_port_module_type mod_type)
1206 if (port_type == FW_PORT_TYPE_BT_SGMII ||
1207 port_type == FW_PORT_TYPE_BT_XFI ||
1208 port_type == FW_PORT_TYPE_BT_XAUI) {
1210 } else if (port_type == FW_PORT_TYPE_FIBER_XFI ||
1211 port_type == FW_PORT_TYPE_FIBER_XAUI) {
1213 } else if (port_type == FW_PORT_TYPE_SFP ||
1214 port_type == FW_PORT_TYPE_QSFP_10G ||
1215 port_type == FW_PORT_TYPE_QSA ||
1216 port_type == FW_PORT_TYPE_QSFP) {
1217 if (mod_type == FW_PORT_MOD_TYPE_LR ||
1218 mod_type == FW_PORT_MOD_TYPE_SR ||
1219 mod_type == FW_PORT_MOD_TYPE_ER ||
1220 mod_type == FW_PORT_MOD_TYPE_LRM)
1222 else if (mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
1223 mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
1233 * fw_caps_to_lmm - translate Firmware to ethtool Link Mode Mask
1234 * @port_type: Firmware Port Type
1235 * @fw_caps: Firmware Port Capabilities
1236 * @link_mode_mask: ethtool Link Mode Mask
1238 * Translate a Firmware Port Capabilities specification to an ethtool
1241 static void fw_caps_to_lmm(enum fw_port_type port_type,
1242 unsigned int fw_caps,
1243 unsigned long *link_mode_mask)
1245 #define SET_LMM(__lmm_name) __set_bit(ETHTOOL_LINK_MODE_ ## __lmm_name\
1246 ## _BIT, link_mode_mask)
1248 #define FW_CAPS_TO_LMM(__fw_name, __lmm_name) \
1250 if (fw_caps & FW_PORT_CAP_ ## __fw_name) \
1251 SET_LMM(__lmm_name); \
1254 switch (port_type) {
1255 case FW_PORT_TYPE_BT_SGMII:
1256 case FW_PORT_TYPE_BT_XFI:
1257 case FW_PORT_TYPE_BT_XAUI:
1259 FW_CAPS_TO_LMM(SPEED_100M, 100baseT_Full);
1260 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1261 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1264 case FW_PORT_TYPE_KX4:
1265 case FW_PORT_TYPE_KX:
1267 FW_CAPS_TO_LMM(SPEED_1G, 1000baseKX_Full);
1268 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKX4_Full);
1271 case FW_PORT_TYPE_KR:
1273 SET_LMM(10000baseKR_Full);
1276 case FW_PORT_TYPE_BP_AP:
1278 SET_LMM(10000baseR_FEC);
1279 SET_LMM(10000baseKR_Full);
1280 SET_LMM(1000baseKX_Full);
1283 case FW_PORT_TYPE_BP4_AP:
1285 SET_LMM(10000baseR_FEC);
1286 SET_LMM(10000baseKR_Full);
1287 SET_LMM(1000baseKX_Full);
1288 SET_LMM(10000baseKX4_Full);
1291 case FW_PORT_TYPE_FIBER_XFI:
1292 case FW_PORT_TYPE_FIBER_XAUI:
1293 case FW_PORT_TYPE_SFP:
1294 case FW_PORT_TYPE_QSFP_10G:
1295 case FW_PORT_TYPE_QSA:
1297 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1298 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1301 case FW_PORT_TYPE_BP40_BA:
1302 case FW_PORT_TYPE_QSFP:
1304 SET_LMM(40000baseSR4_Full);
1307 case FW_PORT_TYPE_CR_QSFP:
1308 case FW_PORT_TYPE_SFP28:
1310 SET_LMM(25000baseCR_Full);
1313 case FW_PORT_TYPE_KR4_100G:
1314 case FW_PORT_TYPE_CR4_QSFP:
1316 SET_LMM(100000baseCR4_Full);
1323 FW_CAPS_TO_LMM(ANEG, Autoneg);
1324 FW_CAPS_TO_LMM(802_3_PAUSE, Pause);
1325 FW_CAPS_TO_LMM(802_3_ASM_DIR, Asym_Pause);
1327 #undef FW_CAPS_TO_LMM
1331 static int cxgb4vf_get_link_ksettings(struct net_device *dev,
1332 struct ethtool_link_ksettings
1335 const struct port_info *pi = netdev_priv(dev);
1336 struct ethtool_link_settings *base = &link_ksettings->base;
1338 ethtool_link_ksettings_zero_link_mode(link_ksettings, supported);
1339 ethtool_link_ksettings_zero_link_mode(link_ksettings, advertising);
1340 ethtool_link_ksettings_zero_link_mode(link_ksettings, lp_advertising);
1342 base->port = from_fw_port_mod_type(pi->port_type, pi->mod_type);
1344 if (pi->mdio_addr >= 0) {
1345 base->phy_address = pi->mdio_addr;
1346 base->mdio_support = (pi->port_type == FW_PORT_TYPE_BT_SGMII
1347 ? ETH_MDIO_SUPPORTS_C22
1348 : ETH_MDIO_SUPPORTS_C45);
1350 base->phy_address = 255;
1351 base->mdio_support = 0;
1354 fw_caps_to_lmm(pi->port_type, pi->link_cfg.supported,
1355 link_ksettings->link_modes.supported);
1356 fw_caps_to_lmm(pi->port_type, pi->link_cfg.advertising,
1357 link_ksettings->link_modes.advertising);
1358 fw_caps_to_lmm(pi->port_type, pi->link_cfg.lp_advertising,
1359 link_ksettings->link_modes.lp_advertising);
1361 if (netif_carrier_ok(dev)) {
1362 base->speed = pi->link_cfg.speed;
1363 base->duplex = DUPLEX_FULL;
1365 base->speed = SPEED_UNKNOWN;
1366 base->duplex = DUPLEX_UNKNOWN;
1369 base->autoneg = pi->link_cfg.autoneg;
1370 if (pi->link_cfg.supported & FW_PORT_CAP_ANEG)
1371 ethtool_link_ksettings_add_link_mode(link_ksettings,
1372 supported, Autoneg);
1373 if (pi->link_cfg.autoneg)
1374 ethtool_link_ksettings_add_link_mode(link_ksettings,
1375 advertising, Autoneg);
1381 * Return our driver information.
1383 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1384 struct ethtool_drvinfo *drvinfo)
1386 struct adapter *adapter = netdev2adap(dev);
1388 strlcpy(drvinfo->driver, KBUILD_MODNAME, sizeof(drvinfo->driver));
1389 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
1390 strlcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)),
1391 sizeof(drvinfo->bus_info));
1392 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1393 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1394 FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.fwrev),
1395 FW_HDR_FW_VER_MINOR_G(adapter->params.dev.fwrev),
1396 FW_HDR_FW_VER_MICRO_G(adapter->params.dev.fwrev),
1397 FW_HDR_FW_VER_BUILD_G(adapter->params.dev.fwrev),
1398 FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.tprev),
1399 FW_HDR_FW_VER_MINOR_G(adapter->params.dev.tprev),
1400 FW_HDR_FW_VER_MICRO_G(adapter->params.dev.tprev),
1401 FW_HDR_FW_VER_BUILD_G(adapter->params.dev.tprev));
1405 * Return current adapter message level.
1407 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1409 return netdev2adap(dev)->msg_enable;
1413 * Set current adapter message level.
1415 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1417 netdev2adap(dev)->msg_enable = msglevel;
1421 * Return the device's current Queue Set ring size parameters along with the
1422 * allowed maximum values. Since ethtool doesn't understand the concept of
1423 * multi-queue devices, we just return the current values associated with the
1426 static void cxgb4vf_get_ringparam(struct net_device *dev,
1427 struct ethtool_ringparam *rp)
1429 const struct port_info *pi = netdev_priv(dev);
1430 const struct sge *s = &pi->adapter->sge;
1432 rp->rx_max_pending = MAX_RX_BUFFERS;
1433 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1434 rp->rx_jumbo_max_pending = 0;
1435 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1437 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1438 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1439 rp->rx_jumbo_pending = 0;
1440 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1444 * Set the Queue Set ring size parameters for the device. Again, since
1445 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1446 * apply these new values across all of the Queue Sets associated with the
1447 * device -- after vetting them of course!
1449 static int cxgb4vf_set_ringparam(struct net_device *dev,
1450 struct ethtool_ringparam *rp)
1452 const struct port_info *pi = netdev_priv(dev);
1453 struct adapter *adapter = pi->adapter;
1454 struct sge *s = &adapter->sge;
1457 if (rp->rx_pending > MAX_RX_BUFFERS ||
1458 rp->rx_jumbo_pending ||
1459 rp->tx_pending > MAX_TXQ_ENTRIES ||
1460 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1461 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1462 rp->rx_pending < MIN_FL_ENTRIES ||
1463 rp->tx_pending < MIN_TXQ_ENTRIES)
1466 if (adapter->flags & FULL_INIT_DONE)
1469 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1470 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1471 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1472 s->ethtxq[qs].q.size = rp->tx_pending;
1478 * Return the interrupt holdoff timer and count for the first Queue Set on the
1479 * device. Our extension ioctl() (the cxgbtool interface) allows the
1480 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1482 static int cxgb4vf_get_coalesce(struct net_device *dev,
1483 struct ethtool_coalesce *coalesce)
1485 const struct port_info *pi = netdev_priv(dev);
1486 const struct adapter *adapter = pi->adapter;
1487 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1489 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1490 coalesce->rx_max_coalesced_frames =
1491 ((rspq->intr_params & QINTR_CNT_EN_F)
1492 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1498 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1499 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1500 * the interrupt holdoff timer on any of the device's Queue Sets.
1502 static int cxgb4vf_set_coalesce(struct net_device *dev,
1503 struct ethtool_coalesce *coalesce)
1505 const struct port_info *pi = netdev_priv(dev);
1506 struct adapter *adapter = pi->adapter;
1508 return set_rxq_intr_params(adapter,
1509 &adapter->sge.ethrxq[pi->first_qset].rspq,
1510 coalesce->rx_coalesce_usecs,
1511 coalesce->rx_max_coalesced_frames);
1515 * Report current port link pause parameter settings.
1517 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1518 struct ethtool_pauseparam *pauseparam)
1520 struct port_info *pi = netdev_priv(dev);
1522 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1523 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1524 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1528 * Identify the port by blinking the port's LED.
1530 static int cxgb4vf_phys_id(struct net_device *dev,
1531 enum ethtool_phys_id_state state)
1534 struct port_info *pi = netdev_priv(dev);
1536 if (state == ETHTOOL_ID_ACTIVE)
1538 else if (state == ETHTOOL_ID_INACTIVE)
1543 return t4vf_identify_port(pi->adapter, pi->viid, val);
1547 * Port stats maintained per queue of the port.
1549 struct queue_port_stats {
1560 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1561 * these need to match the order of statistics returned by
1562 * t4vf_get_port_stats().
1564 static const char stats_strings[][ETH_GSTRING_LEN] = {
1566 * These must match the layout of the t4vf_port_stats structure.
1568 "TxBroadcastBytes ",
1569 "TxBroadcastFrames ",
1570 "TxMulticastBytes ",
1571 "TxMulticastFrames ",
1577 "RxBroadcastBytes ",
1578 "RxBroadcastFrames ",
1579 "RxMulticastBytes ",
1580 "RxMulticastFrames ",
1586 * These are accumulated per-queue statistics and must match the
1587 * order of the fields in the queue_port_stats structure.
1599 * Return the number of statistics in the specified statistics set.
1601 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1605 return ARRAY_SIZE(stats_strings);
1613 * Return the strings for the specified statistics set.
1615 static void cxgb4vf_get_strings(struct net_device *dev,
1621 memcpy(data, stats_strings, sizeof(stats_strings));
1627 * Small utility routine to accumulate queue statistics across the queues of
1630 static void collect_sge_port_stats(const struct adapter *adapter,
1631 const struct port_info *pi,
1632 struct queue_port_stats *stats)
1634 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1635 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1638 memset(stats, 0, sizeof(*stats));
1639 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1640 stats->tso += txq->tso;
1641 stats->tx_csum += txq->tx_cso;
1642 stats->rx_csum += rxq->stats.rx_cso;
1643 stats->vlan_ex += rxq->stats.vlan_ex;
1644 stats->vlan_ins += txq->vlan_ins;
1645 stats->lro_pkts += rxq->stats.lro_pkts;
1646 stats->lro_merged += rxq->stats.lro_merged;
1651 * Return the ETH_SS_STATS statistics set.
1653 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1654 struct ethtool_stats *stats,
1657 struct port_info *pi = netdev2pinfo(dev);
1658 struct adapter *adapter = pi->adapter;
1659 int err = t4vf_get_port_stats(adapter, pi->pidx,
1660 (struct t4vf_port_stats *)data);
1662 memset(data, 0, sizeof(struct t4vf_port_stats));
1664 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1665 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1669 * Return the size of our register map.
1671 static int cxgb4vf_get_regs_len(struct net_device *dev)
1673 return T4VF_REGMAP_SIZE;
1677 * Dump a block of registers, start to end inclusive, into a buffer.
1679 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1680 unsigned int start, unsigned int end)
1682 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1684 for ( ; start <= end; start += sizeof(u32)) {
1686 * Avoid reading the Mailbox Control register since that
1687 * can trigger a Mailbox Ownership Arbitration cycle and
1688 * interfere with communication with the firmware.
1690 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1693 *bp++ = t4_read_reg(adapter, start);
1698 * Copy our entire register map into the provided buffer.
1700 static void cxgb4vf_get_regs(struct net_device *dev,
1701 struct ethtool_regs *regs,
1704 struct adapter *adapter = netdev2adap(dev);
1706 regs->version = mk_adap_vers(adapter);
1709 * Fill in register buffer with our register map.
1711 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1713 reg_block_dump(adapter, regbuf,
1714 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1715 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1716 reg_block_dump(adapter, regbuf,
1717 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1718 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1720 /* T5 adds new registers in the PL Register map.
1722 reg_block_dump(adapter, regbuf,
1723 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1724 T4VF_PL_BASE_ADDR + (is_t4(adapter->params.chip)
1725 ? PL_VF_WHOAMI_A : PL_VF_REVISION_A));
1726 reg_block_dump(adapter, regbuf,
1727 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1728 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1730 reg_block_dump(adapter, regbuf,
1731 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1732 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1736 * Report current Wake On LAN settings.
1738 static void cxgb4vf_get_wol(struct net_device *dev,
1739 struct ethtool_wolinfo *wol)
1743 memset(&wol->sopass, 0, sizeof(wol->sopass));
1747 * TCP Segmentation Offload flags which we support.
1749 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1751 static const struct ethtool_ops cxgb4vf_ethtool_ops = {
1752 .get_link_ksettings = cxgb4vf_get_link_ksettings,
1753 .get_drvinfo = cxgb4vf_get_drvinfo,
1754 .get_msglevel = cxgb4vf_get_msglevel,
1755 .set_msglevel = cxgb4vf_set_msglevel,
1756 .get_ringparam = cxgb4vf_get_ringparam,
1757 .set_ringparam = cxgb4vf_set_ringparam,
1758 .get_coalesce = cxgb4vf_get_coalesce,
1759 .set_coalesce = cxgb4vf_set_coalesce,
1760 .get_pauseparam = cxgb4vf_get_pauseparam,
1761 .get_link = ethtool_op_get_link,
1762 .get_strings = cxgb4vf_get_strings,
1763 .set_phys_id = cxgb4vf_phys_id,
1764 .get_sset_count = cxgb4vf_get_sset_count,
1765 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1766 .get_regs_len = cxgb4vf_get_regs_len,
1767 .get_regs = cxgb4vf_get_regs,
1768 .get_wol = cxgb4vf_get_wol,
1772 * /sys/kernel/debug/cxgb4vf support code and data.
1773 * ================================================
1777 * Show Firmware Mailbox Command/Reply Log
1779 * Note that we don't do any locking when dumping the Firmware Mailbox Log so
1780 * it's possible that we can catch things during a log update and therefore
1781 * see partially corrupted log entries. But i9t's probably Good Enough(tm).
1782 * If we ever decide that we want to make sure that we're dumping a coherent
1783 * log, we'd need to perform locking in the mailbox logging and in
1784 * mboxlog_open() where we'd need to grab the entire mailbox log in one go
1785 * like we do for the Firmware Device Log. But as stated above, meh ...
1787 static int mboxlog_show(struct seq_file *seq, void *v)
1789 struct adapter *adapter = seq->private;
1790 struct mbox_cmd_log *log = adapter->mbox_log;
1791 struct mbox_cmd *entry;
1794 if (v == SEQ_START_TOKEN) {
1796 "%10s %15s %5s %5s %s\n",
1797 "Seq#", "Tstamp", "Atime", "Etime",
1802 entry_idx = log->cursor + ((uintptr_t)v - 2);
1803 if (entry_idx >= log->size)
1804 entry_idx -= log->size;
1805 entry = mbox_cmd_log_entry(log, entry_idx);
1807 /* skip over unused entries */
1808 if (entry->timestamp == 0)
1811 seq_printf(seq, "%10u %15llu %5d %5d",
1812 entry->seqno, entry->timestamp,
1813 entry->access, entry->execute);
1814 for (i = 0; i < MBOX_LEN / 8; i++) {
1815 u64 flit = entry->cmd[i];
1816 u32 hi = (u32)(flit >> 32);
1819 seq_printf(seq, " %08x %08x", hi, lo);
1821 seq_puts(seq, "\n");
1825 static inline void *mboxlog_get_idx(struct seq_file *seq, loff_t pos)
1827 struct adapter *adapter = seq->private;
1828 struct mbox_cmd_log *log = adapter->mbox_log;
1830 return ((pos <= log->size) ? (void *)(uintptr_t)(pos + 1) : NULL);
1833 static void *mboxlog_start(struct seq_file *seq, loff_t *pos)
1835 return *pos ? mboxlog_get_idx(seq, *pos) : SEQ_START_TOKEN;
1838 static void *mboxlog_next(struct seq_file *seq, void *v, loff_t *pos)
1841 return mboxlog_get_idx(seq, *pos);
1844 static void mboxlog_stop(struct seq_file *seq, void *v)
1848 static const struct seq_operations mboxlog_seq_ops = {
1849 .start = mboxlog_start,
1850 .next = mboxlog_next,
1851 .stop = mboxlog_stop,
1852 .show = mboxlog_show
1855 static int mboxlog_open(struct inode *inode, struct file *file)
1857 int res = seq_open(file, &mboxlog_seq_ops);
1860 struct seq_file *seq = file->private_data;
1862 seq->private = inode->i_private;
1867 static const struct file_operations mboxlog_fops = {
1868 .owner = THIS_MODULE,
1869 .open = mboxlog_open,
1871 .llseek = seq_lseek,
1872 .release = seq_release,
1876 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1880 static int sge_qinfo_show(struct seq_file *seq, void *v)
1882 struct adapter *adapter = seq->private;
1883 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1884 int qs, r = (uintptr_t)v - 1;
1887 seq_putc(seq, '\n');
1889 #define S3(fmt_spec, s, v) \
1891 seq_printf(seq, "%-12s", s); \
1892 for (qs = 0; qs < n; ++qs) \
1893 seq_printf(seq, " %16" fmt_spec, v); \
1894 seq_putc(seq, '\n'); \
1896 #define S(s, v) S3("s", s, v)
1897 #define T(s, v) S3("u", s, txq[qs].v)
1898 #define R(s, v) S3("u", s, rxq[qs].v)
1900 if (r < eth_entries) {
1901 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1902 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1903 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1905 S("QType:", "Ethernet");
1907 (rxq[qs].rspq.netdev
1908 ? rxq[qs].rspq.netdev->name
1911 (rxq[qs].rspq.netdev
1912 ? ((struct port_info *)
1913 netdev_priv(rxq[qs].rspq.netdev))->port_id
1915 T("TxQ ID:", q.abs_id);
1916 T("TxQ size:", q.size);
1917 T("TxQ inuse:", q.in_use);
1918 T("TxQ PIdx:", q.pidx);
1919 T("TxQ CIdx:", q.cidx);
1920 R("RspQ ID:", rspq.abs_id);
1921 R("RspQ size:", rspq.size);
1922 R("RspQE size:", rspq.iqe_len);
1923 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1924 S3("u", "Intr pktcnt:",
1925 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1926 R("RspQ CIdx:", rspq.cidx);
1927 R("RspQ Gen:", rspq.gen);
1928 R("FL ID:", fl.abs_id);
1929 R("FL size:", fl.size - MIN_FL_RESID);
1930 R("FL avail:", fl.avail);
1931 R("FL PIdx:", fl.pidx);
1932 R("FL CIdx:", fl.cidx);
1938 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1940 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1941 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1942 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1943 qtimer_val(adapter, evtq));
1944 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1945 adapter->sge.counter_val[evtq->pktcnt_idx]);
1946 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1947 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1948 } else if (r == 1) {
1949 const struct sge_rspq *intrq = &adapter->sge.intrq;
1951 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1952 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1953 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1954 qtimer_val(adapter, intrq));
1955 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1956 adapter->sge.counter_val[intrq->pktcnt_idx]);
1957 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1958 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1970 * Return the number of "entries" in our "file". We group the multi-Queue
1971 * sections with QPL Queue Sets per "entry". The sections of the output are:
1973 * Ethernet RX/TX Queue Sets
1974 * Firmware Event Queue
1975 * Forwarded Interrupt Queue (if in MSI mode)
1977 static int sge_queue_entries(const struct adapter *adapter)
1979 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1980 ((adapter->flags & USING_MSI) != 0);
1983 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1985 int entries = sge_queue_entries(seq->private);
1987 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1990 static void sge_queue_stop(struct seq_file *seq, void *v)
1994 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
1996 int entries = sge_queue_entries(seq->private);
1999 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2002 static const struct seq_operations sge_qinfo_seq_ops = {
2003 .start = sge_queue_start,
2004 .next = sge_queue_next,
2005 .stop = sge_queue_stop,
2006 .show = sge_qinfo_show
2009 static int sge_qinfo_open(struct inode *inode, struct file *file)
2011 int res = seq_open(file, &sge_qinfo_seq_ops);
2014 struct seq_file *seq = file->private_data;
2015 seq->private = inode->i_private;
2020 static const struct file_operations sge_qinfo_debugfs_fops = {
2021 .owner = THIS_MODULE,
2022 .open = sge_qinfo_open,
2024 .llseek = seq_lseek,
2025 .release = seq_release,
2029 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
2033 static int sge_qstats_show(struct seq_file *seq, void *v)
2035 struct adapter *adapter = seq->private;
2036 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
2037 int qs, r = (uintptr_t)v - 1;
2040 seq_putc(seq, '\n');
2042 #define S3(fmt, s, v) \
2044 seq_printf(seq, "%-16s", s); \
2045 for (qs = 0; qs < n; ++qs) \
2046 seq_printf(seq, " %8" fmt, v); \
2047 seq_putc(seq, '\n'); \
2049 #define S(s, v) S3("s", s, v)
2051 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
2052 #define T(s, v) T3("lu", s, v)
2054 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
2055 #define R(s, v) R3("lu", s, v)
2057 if (r < eth_entries) {
2058 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
2059 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
2060 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
2062 S("QType:", "Ethernet");
2064 (rxq[qs].rspq.netdev
2065 ? rxq[qs].rspq.netdev->name
2067 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
2068 R("RxPackets:", stats.pkts);
2069 R("RxCSO:", stats.rx_cso);
2070 R("VLANxtract:", stats.vlan_ex);
2071 R("LROmerged:", stats.lro_merged);
2072 R("LROpackets:", stats.lro_pkts);
2073 R("RxDrops:", stats.rx_drops);
2075 T("TxCSO:", tx_cso);
2076 T("VLANins:", vlan_ins);
2077 T("TxQFull:", q.stops);
2078 T("TxQRestarts:", q.restarts);
2079 T("TxMapErr:", mapping_err);
2080 R("FLAllocErr:", fl.alloc_failed);
2081 R("FLLrgAlcErr:", fl.large_alloc_failed);
2082 R("FLStarving:", fl.starving);
2088 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
2090 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
2091 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
2092 evtq->unhandled_irqs);
2093 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
2094 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
2095 } else if (r == 1) {
2096 const struct sge_rspq *intrq = &adapter->sge.intrq;
2098 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
2099 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
2100 intrq->unhandled_irqs);
2101 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
2102 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
2116 * Return the number of "entries" in our "file". We group the multi-Queue
2117 * sections with QPL Queue Sets per "entry". The sections of the output are:
2119 * Ethernet RX/TX Queue Sets
2120 * Firmware Event Queue
2121 * Forwarded Interrupt Queue (if in MSI mode)
2123 static int sge_qstats_entries(const struct adapter *adapter)
2125 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
2126 ((adapter->flags & USING_MSI) != 0);
2129 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
2131 int entries = sge_qstats_entries(seq->private);
2133 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2136 static void sge_qstats_stop(struct seq_file *seq, void *v)
2140 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
2142 int entries = sge_qstats_entries(seq->private);
2145 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2148 static const struct seq_operations sge_qstats_seq_ops = {
2149 .start = sge_qstats_start,
2150 .next = sge_qstats_next,
2151 .stop = sge_qstats_stop,
2152 .show = sge_qstats_show
2155 static int sge_qstats_open(struct inode *inode, struct file *file)
2157 int res = seq_open(file, &sge_qstats_seq_ops);
2160 struct seq_file *seq = file->private_data;
2161 seq->private = inode->i_private;
2166 static const struct file_operations sge_qstats_proc_fops = {
2167 .owner = THIS_MODULE,
2168 .open = sge_qstats_open,
2170 .llseek = seq_lseek,
2171 .release = seq_release,
2175 * Show PCI-E SR-IOV Virtual Function Resource Limits.
2177 static int resources_show(struct seq_file *seq, void *v)
2179 struct adapter *adapter = seq->private;
2180 struct vf_resources *vfres = &adapter->params.vfres;
2182 #define S(desc, fmt, var) \
2183 seq_printf(seq, "%-60s " fmt "\n", \
2184 desc " (" #var "):", vfres->var)
2186 S("Virtual Interfaces", "%d", nvi);
2187 S("Egress Queues", "%d", neq);
2188 S("Ethernet Control", "%d", nethctrl);
2189 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
2190 S("Ingress Queues", "%d", niq);
2191 S("Traffic Class", "%d", tc);
2192 S("Port Access Rights Mask", "%#x", pmask);
2193 S("MAC Address Filters", "%d", nexactf);
2194 S("Firmware Command Read Capabilities", "%#x", r_caps);
2195 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
2202 static int resources_open(struct inode *inode, struct file *file)
2204 return single_open(file, resources_show, inode->i_private);
2207 static const struct file_operations resources_proc_fops = {
2208 .owner = THIS_MODULE,
2209 .open = resources_open,
2211 .llseek = seq_lseek,
2212 .release = single_release,
2216 * Show Virtual Interfaces.
2218 static int interfaces_show(struct seq_file *seq, void *v)
2220 if (v == SEQ_START_TOKEN) {
2221 seq_puts(seq, "Interface Port VIID\n");
2223 struct adapter *adapter = seq->private;
2224 int pidx = (uintptr_t)v - 2;
2225 struct net_device *dev = adapter->port[pidx];
2226 struct port_info *pi = netdev_priv(dev);
2228 seq_printf(seq, "%9s %4d %#5x\n",
2229 dev->name, pi->port_id, pi->viid);
2234 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
2236 return pos <= adapter->params.nports
2237 ? (void *)(uintptr_t)(pos + 1)
2241 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
2244 ? interfaces_get_idx(seq->private, *pos)
2248 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
2251 return interfaces_get_idx(seq->private, *pos);
2254 static void interfaces_stop(struct seq_file *seq, void *v)
2258 static const struct seq_operations interfaces_seq_ops = {
2259 .start = interfaces_start,
2260 .next = interfaces_next,
2261 .stop = interfaces_stop,
2262 .show = interfaces_show
2265 static int interfaces_open(struct inode *inode, struct file *file)
2267 int res = seq_open(file, &interfaces_seq_ops);
2270 struct seq_file *seq = file->private_data;
2271 seq->private = inode->i_private;
2276 static const struct file_operations interfaces_proc_fops = {
2277 .owner = THIS_MODULE,
2278 .open = interfaces_open,
2280 .llseek = seq_lseek,
2281 .release = seq_release,
2285 * /sys/kernel/debugfs/cxgb4vf/ files list.
2287 struct cxgb4vf_debugfs_entry {
2288 const char *name; /* name of debugfs node */
2289 umode_t mode; /* file system mode */
2290 const struct file_operations *fops;
2293 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
2294 { "mboxlog", S_IRUGO, &mboxlog_fops },
2295 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
2296 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
2297 { "resources", S_IRUGO, &resources_proc_fops },
2298 { "interfaces", S_IRUGO, &interfaces_proc_fops },
2302 * Module and device initialization and cleanup code.
2303 * ==================================================
2307 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2308 * directory (debugfs_root) has already been set up.
2310 static int setup_debugfs(struct adapter *adapter)
2314 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2317 * Debugfs support is best effort.
2319 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2320 (void)debugfs_create_file(debugfs_files[i].name,
2321 debugfs_files[i].mode,
2322 adapter->debugfs_root,
2324 debugfs_files[i].fops);
2330 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2331 * it to our caller to tear down the directory (debugfs_root).
2333 static void cleanup_debugfs(struct adapter *adapter)
2335 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2338 * Unlike our sister routine cleanup_proc(), we don't need to remove
2339 * individual entries because a call will be made to
2340 * debugfs_remove_recursive(). We just need to clean up any ancillary
2346 /* Figure out how many Ports and Queue Sets we can support. This depends on
2347 * knowing our Virtual Function Resources and may be called a second time if
2348 * we fall back from MSI-X to MSI Interrupt Mode.
2350 static void size_nports_qsets(struct adapter *adapter)
2352 struct vf_resources *vfres = &adapter->params.vfres;
2353 unsigned int ethqsets, pmask_nports;
2355 /* The number of "ports" which we support is equal to the number of
2356 * Virtual Interfaces with which we've been provisioned.
2358 adapter->params.nports = vfres->nvi;
2359 if (adapter->params.nports > MAX_NPORTS) {
2360 dev_warn(adapter->pdev_dev, "only using %d of %d maximum"
2361 " allowed virtual interfaces\n", MAX_NPORTS,
2362 adapter->params.nports);
2363 adapter->params.nports = MAX_NPORTS;
2366 /* We may have been provisioned with more VIs than the number of
2367 * ports we're allowed to access (our Port Access Rights Mask).
2368 * This is obviously a configuration conflict but we don't want to
2369 * crash the kernel or anything silly just because of that.
2371 pmask_nports = hweight32(adapter->params.vfres.pmask);
2372 if (pmask_nports < adapter->params.nports) {
2373 dev_warn(adapter->pdev_dev, "only using %d of %d provisioned"
2374 " virtual interfaces; limited by Port Access Rights"
2375 " mask %#x\n", pmask_nports, adapter->params.nports,
2376 adapter->params.vfres.pmask);
2377 adapter->params.nports = pmask_nports;
2380 /* We need to reserve an Ingress Queue for the Asynchronous Firmware
2381 * Event Queue. And if we're using MSI Interrupts, we'll also need to
2382 * reserve an Ingress Queue for a Forwarded Interrupts.
2384 * The rest of the FL/Intr-capable ingress queues will be matched up
2385 * one-for-one with Ethernet/Control egress queues in order to form
2386 * "Queue Sets" which will be aportioned between the "ports". For
2387 * each Queue Set, we'll need the ability to allocate two Egress
2388 * Contexts -- one for the Ingress Queue Free List and one for the TX
2391 * Note that even if we're currently configured to use MSI-X
2392 * Interrupts (module variable msi == MSI_MSIX) we may get downgraded
2393 * to MSI Interrupts if we can't get enough MSI-X Interrupts. If that
2394 * happens we'll need to adjust things later.
2396 ethqsets = vfres->niqflint - 1 - (msi == MSI_MSI);
2397 if (vfres->nethctrl != ethqsets)
2398 ethqsets = min(vfres->nethctrl, ethqsets);
2399 if (vfres->neq < ethqsets*2)
2400 ethqsets = vfres->neq/2;
2401 if (ethqsets > MAX_ETH_QSETS)
2402 ethqsets = MAX_ETH_QSETS;
2403 adapter->sge.max_ethqsets = ethqsets;
2405 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2406 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2407 " virtual interfaces (too few Queue Sets)\n",
2408 adapter->sge.max_ethqsets, adapter->params.nports);
2409 adapter->params.nports = adapter->sge.max_ethqsets;
2414 * Perform early "adapter" initialization. This is where we discover what
2415 * adapter parameters we're going to be using and initialize basic adapter
2418 static int adap_init0(struct adapter *adapter)
2420 struct sge_params *sge_params = &adapter->params.sge;
2421 struct sge *s = &adapter->sge;
2426 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2427 * 2.6.31 and later we can't call pci_reset_function() in order to
2428 * issue an FLR because of a self- deadlock on the device semaphore.
2429 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2430 * cases where they're needed -- for instance, some versions of KVM
2431 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2432 * use the firmware based reset in order to reset any per function
2435 err = t4vf_fw_reset(adapter);
2437 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2442 * Grab basic operational parameters. These will predominantly have
2443 * been set up by the Physical Function Driver or will be hard coded
2444 * into the adapter. We just have to live with them ... Note that
2445 * we _must_ get our VPD parameters before our SGE parameters because
2446 * we need to know the adapter's core clock from the VPD in order to
2447 * properly decode the SGE Timer Values.
2449 err = t4vf_get_dev_params(adapter);
2451 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2452 " device parameters: err=%d\n", err);
2455 err = t4vf_get_vpd_params(adapter);
2457 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2458 " VPD parameters: err=%d\n", err);
2461 err = t4vf_get_sge_params(adapter);
2463 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2464 " SGE parameters: err=%d\n", err);
2467 err = t4vf_get_rss_glb_config(adapter);
2469 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2470 " RSS parameters: err=%d\n", err);
2473 if (adapter->params.rss.mode !=
2474 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2475 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2476 " mode %d\n", adapter->params.rss.mode);
2479 err = t4vf_sge_init(adapter);
2481 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2486 /* If we're running on newer firmware, let it know that we're
2487 * prepared to deal with encapsulated CPL messages. Older
2488 * firmware won't understand this and we'll just get
2489 * unencapsulated messages ...
2491 param = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2492 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_CPLFW4MSG_ENCAP);
2494 (void) t4vf_set_params(adapter, 1, ¶m, &val);
2497 * Retrieve our RX interrupt holdoff timer values and counter
2498 * threshold values from the SGE parameters.
2500 s->timer_val[0] = core_ticks_to_us(adapter,
2501 TIMERVALUE0_G(sge_params->sge_timer_value_0_and_1));
2502 s->timer_val[1] = core_ticks_to_us(adapter,
2503 TIMERVALUE1_G(sge_params->sge_timer_value_0_and_1));
2504 s->timer_val[2] = core_ticks_to_us(adapter,
2505 TIMERVALUE0_G(sge_params->sge_timer_value_2_and_3));
2506 s->timer_val[3] = core_ticks_to_us(adapter,
2507 TIMERVALUE1_G(sge_params->sge_timer_value_2_and_3));
2508 s->timer_val[4] = core_ticks_to_us(adapter,
2509 TIMERVALUE0_G(sge_params->sge_timer_value_4_and_5));
2510 s->timer_val[5] = core_ticks_to_us(adapter,
2511 TIMERVALUE1_G(sge_params->sge_timer_value_4_and_5));
2513 s->counter_val[0] = THRESHOLD_0_G(sge_params->sge_ingress_rx_threshold);
2514 s->counter_val[1] = THRESHOLD_1_G(sge_params->sge_ingress_rx_threshold);
2515 s->counter_val[2] = THRESHOLD_2_G(sge_params->sge_ingress_rx_threshold);
2516 s->counter_val[3] = THRESHOLD_3_G(sge_params->sge_ingress_rx_threshold);
2519 * Grab our Virtual Interface resource allocation, extract the
2520 * features that we're interested in and do a bit of sanity testing on
2523 err = t4vf_get_vfres(adapter);
2525 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2526 " resources: err=%d\n", err);
2530 /* Check for various parameter sanity issues */
2531 if (adapter->params.vfres.pmask == 0) {
2532 dev_err(adapter->pdev_dev, "no port access configured\n"
2536 if (adapter->params.vfres.nvi == 0) {
2537 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2542 /* Initialize nports and max_ethqsets now that we have our Virtual
2543 * Function Resources.
2545 size_nports_qsets(adapter);
2550 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2551 u8 pkt_cnt_idx, unsigned int size,
2552 unsigned int iqe_size)
2554 rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) |
2555 (pkt_cnt_idx < SGE_NCOUNTERS ?
2556 QINTR_CNT_EN_F : 0));
2557 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2560 rspq->iqe_len = iqe_size;
2565 * Perform default configuration of DMA queues depending on the number and
2566 * type of ports we found and the number of available CPUs. Most settings can
2567 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2568 * being brought up for the first time.
2570 static void cfg_queues(struct adapter *adapter)
2572 struct sge *s = &adapter->sge;
2573 int q10g, n10g, qidx, pidx, qs;
2577 * We should not be called till we know how many Queue Sets we can
2578 * support. In particular, this means that we need to know what kind
2579 * of interrupts we'll be using ...
2581 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2584 * Count the number of 10GbE Virtual Interfaces that we have.
2587 for_each_port(adapter, pidx)
2588 n10g += is_x_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2591 * We default to 1 queue per non-10G port and up to # of cores queues
2597 int n1g = (adapter->params.nports - n10g);
2598 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2599 if (q10g > num_online_cpus())
2600 q10g = num_online_cpus();
2604 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2605 * The layout will be established in setup_sge_queues() when the
2606 * adapter is brough up for the first time.
2609 for_each_port(adapter, pidx) {
2610 struct port_info *pi = adap2pinfo(adapter, pidx);
2612 pi->first_qset = qidx;
2613 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
2619 * The Ingress Queue Entry Size for our various Response Queues needs
2620 * to be big enough to accommodate the largest message we can receive
2621 * from the chip/firmware; which is 64 bytes ...
2626 * Set up default Queue Set parameters ... Start off with the
2627 * shortest interrupt holdoff timer.
2629 for (qs = 0; qs < s->max_ethqsets; qs++) {
2630 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2631 struct sge_eth_txq *txq = &s->ethtxq[qs];
2633 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2639 * The firmware event queue is used for link state changes and
2640 * notifications of TX DMA completions.
2642 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2645 * The forwarded interrupt queue is used when we're in MSI interrupt
2646 * mode. In this mode all interrupts associated with RX queues will
2647 * be forwarded to a single queue which we'll associate with our MSI
2648 * interrupt vector. The messages dropped in the forwarded interrupt
2649 * queue will indicate which ingress queue needs servicing ... This
2650 * queue needs to be large enough to accommodate all of the ingress
2651 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2652 * from equalling the CIDX if every ingress queue has an outstanding
2653 * interrupt). The queue doesn't need to be any larger because no
2654 * ingress queue will ever have more than one outstanding interrupt at
2657 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2662 * Reduce the number of Ethernet queues across all ports to at most n.
2663 * n provides at least one queue per port.
2665 static void reduce_ethqs(struct adapter *adapter, int n)
2668 struct port_info *pi;
2671 * While we have too many active Ether Queue Sets, interate across the
2672 * "ports" and reduce their individual Queue Set allocations.
2674 BUG_ON(n < adapter->params.nports);
2675 while (n < adapter->sge.ethqsets)
2676 for_each_port(adapter, i) {
2677 pi = adap2pinfo(adapter, i);
2678 if (pi->nqsets > 1) {
2680 adapter->sge.ethqsets--;
2681 if (adapter->sge.ethqsets <= n)
2687 * Reassign the starting Queue Sets for each of the "ports" ...
2690 for_each_port(adapter, i) {
2691 pi = adap2pinfo(adapter, i);
2698 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2699 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2700 * need. Minimally we need one for every Virtual Interface plus those needed
2701 * for our "extras". Note that this process may lower the maximum number of
2702 * allowed Queue Sets ...
2704 static int enable_msix(struct adapter *adapter)
2706 int i, want, need, nqsets;
2707 struct msix_entry entries[MSIX_ENTRIES];
2708 struct sge *s = &adapter->sge;
2710 for (i = 0; i < MSIX_ENTRIES; ++i)
2711 entries[i].entry = i;
2714 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2715 * plus those needed for our "extras" (for example, the firmware
2716 * message queue). We _need_ at least one "Queue Set" per Virtual
2717 * Interface plus those needed for our "extras". So now we get to see
2718 * if the song is right ...
2720 want = s->max_ethqsets + MSIX_EXTRAS;
2721 need = adapter->params.nports + MSIX_EXTRAS;
2723 want = pci_enable_msix_range(adapter->pdev, entries, need, want);
2727 nqsets = want - MSIX_EXTRAS;
2728 if (nqsets < s->max_ethqsets) {
2729 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2730 " for %d Queue Sets\n", nqsets);
2731 s->max_ethqsets = nqsets;
2732 if (nqsets < s->ethqsets)
2733 reduce_ethqs(adapter, nqsets);
2735 for (i = 0; i < want; ++i)
2736 adapter->msix_info[i].vec = entries[i].vector;
2741 static const struct net_device_ops cxgb4vf_netdev_ops = {
2742 .ndo_open = cxgb4vf_open,
2743 .ndo_stop = cxgb4vf_stop,
2744 .ndo_start_xmit = t4vf_eth_xmit,
2745 .ndo_get_stats = cxgb4vf_get_stats,
2746 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2747 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2748 .ndo_validate_addr = eth_validate_addr,
2749 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2750 .ndo_change_mtu = cxgb4vf_change_mtu,
2751 .ndo_fix_features = cxgb4vf_fix_features,
2752 .ndo_set_features = cxgb4vf_set_features,
2753 #ifdef CONFIG_NET_POLL_CONTROLLER
2754 .ndo_poll_controller = cxgb4vf_poll_controller,
2759 * "Probe" a device: initialize a device and construct all kernel and driver
2760 * state needed to manage the device. This routine is called "init_one" in
2763 static int cxgb4vf_pci_probe(struct pci_dev *pdev,
2764 const struct pci_device_id *ent)
2769 struct adapter *adapter;
2770 struct port_info *pi;
2771 struct net_device *netdev;
2775 * Print our driver banner the first time we're called to initialize a
2778 pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION);
2781 * Initialize generic PCI device state.
2783 err = pci_enable_device(pdev);
2785 dev_err(&pdev->dev, "cannot enable PCI device\n");
2790 * Reserve PCI resources for the device. If we can't get them some
2791 * other driver may have already claimed the device ...
2793 err = pci_request_regions(pdev, KBUILD_MODNAME);
2795 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2796 goto err_disable_device;
2800 * Set up our DMA mask: try for 64-bit address masking first and
2801 * fall back to 32-bit if we can't get 64 bits ...
2803 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2805 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2807 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2808 " coherent allocations\n");
2809 goto err_release_regions;
2813 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2815 dev_err(&pdev->dev, "no usable DMA configuration\n");
2816 goto err_release_regions;
2822 * Enable bus mastering for the device ...
2824 pci_set_master(pdev);
2827 * Allocate our adapter data structure and attach it to the device.
2829 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2832 goto err_release_regions;
2834 pci_set_drvdata(pdev, adapter);
2835 adapter->pdev = pdev;
2836 adapter->pdev_dev = &pdev->dev;
2838 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
2839 (sizeof(struct mbox_cmd) *
2840 T4VF_OS_LOG_MBOX_CMDS),
2842 if (!adapter->mbox_log) {
2844 goto err_free_adapter;
2846 adapter->mbox_log->size = T4VF_OS_LOG_MBOX_CMDS;
2849 * Initialize SMP data synchronization resources.
2851 spin_lock_init(&adapter->stats_lock);
2852 spin_lock_init(&adapter->mbox_lock);
2853 INIT_LIST_HEAD(&adapter->mlist.list);
2856 * Map our I/O registers in BAR0.
2858 adapter->regs = pci_ioremap_bar(pdev, 0);
2859 if (!adapter->regs) {
2860 dev_err(&pdev->dev, "cannot map device registers\n");
2862 goto err_free_adapter;
2865 /* Wait for the device to become ready before proceeding ...
2867 err = t4vf_prep_adapter(adapter);
2869 dev_err(adapter->pdev_dev, "device didn't become ready:"
2871 goto err_unmap_bar0;
2874 /* For T5 and later we want to use the new BAR-based User Doorbells,
2875 * so we need to map BAR2 here ...
2877 if (!is_t4(adapter->params.chip)) {
2878 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
2879 pci_resource_len(pdev, 2));
2880 if (!adapter->bar2) {
2881 dev_err(adapter->pdev_dev, "cannot map BAR2 doorbells\n");
2883 goto err_unmap_bar0;
2887 * Initialize adapter level features.
2889 adapter->name = pci_name(pdev);
2890 adapter->msg_enable = DFLT_MSG_ENABLE;
2891 err = adap_init0(adapter);
2896 * Allocate our "adapter ports" and stitch everything together.
2898 pmask = adapter->params.vfres.pmask;
2899 pf = t4vf_get_pf_from_vf(adapter);
2900 for_each_port(adapter, pidx) {
2903 unsigned int naddr = 1;
2906 * We simplistically allocate our virtual interfaces
2907 * sequentially across the port numbers to which we have
2908 * access rights. This should be configurable in some manner
2913 port_id = ffs(pmask) - 1;
2914 pmask &= ~(1 << port_id);
2915 viid = t4vf_alloc_vi(adapter, port_id);
2917 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2918 " err=%d\n", port_id, viid);
2924 * Allocate our network device and stitch things together.
2926 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2928 if (netdev == NULL) {
2929 t4vf_free_vi(adapter, viid);
2933 adapter->port[pidx] = netdev;
2934 SET_NETDEV_DEV(netdev, &pdev->dev);
2935 pi = netdev_priv(netdev);
2936 pi->adapter = adapter;
2938 pi->port_id = port_id;
2942 * Initialize the starting state of our "port" and register
2945 pi->xact_addr_filt = -1;
2946 netif_carrier_off(netdev);
2947 netdev->irq = pdev->irq;
2949 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
2950 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2951 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXCSUM;
2952 netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
2953 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2955 netdev->features = netdev->hw_features |
2956 NETIF_F_HW_VLAN_CTAG_TX;
2958 netdev->features |= NETIF_F_HIGHDMA;
2960 netdev->priv_flags |= IFF_UNICAST_FLT;
2961 netdev->min_mtu = 81;
2962 netdev->max_mtu = ETH_MAX_MTU;
2964 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2965 netdev->ethtool_ops = &cxgb4vf_ethtool_ops;
2966 netdev->dev_port = pi->port_id;
2969 * Initialize the hardware/software state for the port.
2971 err = t4vf_port_init(adapter, pidx);
2973 dev_err(&pdev->dev, "cannot initialize port %d\n",
2978 err = t4vf_get_vf_mac_acl(adapter, pf, &naddr, mac);
2981 "unable to determine MAC ACL address, "
2982 "continuing anyway.. (status %d)\n", err);
2983 } else if (naddr && adapter->params.vfres.nvi == 1) {
2984 struct sockaddr addr;
2986 ether_addr_copy(addr.sa_data, mac);
2987 err = cxgb4vf_set_mac_addr(netdev, &addr);
2990 "unable to set MAC address %pM\n",
2994 dev_info(&pdev->dev,
2995 "Using assigned MAC ACL: %pM\n", mac);
2999 /* See what interrupts we'll be using. If we've been configured to
3000 * use MSI-X interrupts, try to enable them but fall back to using
3001 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
3002 * get MSI interrupts we bail with the error.
3004 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
3005 adapter->flags |= USING_MSIX;
3007 if (msi == MSI_MSIX) {
3008 dev_info(adapter->pdev_dev,
3009 "Unable to use MSI-X Interrupts; falling "
3010 "back to MSI Interrupts\n");
3012 /* We're going to need a Forwarded Interrupt Queue so
3013 * that may cut into how many Queue Sets we can
3017 size_nports_qsets(adapter);
3019 err = pci_enable_msi(pdev);
3021 dev_err(&pdev->dev, "Unable to allocate MSI Interrupts;"
3025 adapter->flags |= USING_MSI;
3028 /* Now that we know how many "ports" we have and what interrupt
3029 * mechanism we're going to use, we can configure our queue resources.
3031 cfg_queues(adapter);
3034 * The "card" is now ready to go. If any errors occur during device
3035 * registration we do not fail the whole "card" but rather proceed
3036 * only with the ports we manage to register successfully. However we
3037 * must register at least one net device.
3039 for_each_port(adapter, pidx) {
3040 struct port_info *pi = netdev_priv(adapter->port[pidx]);
3041 netdev = adapter->port[pidx];
3045 netif_set_real_num_tx_queues(netdev, pi->nqsets);
3046 netif_set_real_num_rx_queues(netdev, pi->nqsets);
3048 err = register_netdev(netdev);
3050 dev_warn(&pdev->dev, "cannot register net device %s,"
3051 " skipping\n", netdev->name);
3055 set_bit(pidx, &adapter->registered_device_map);
3057 if (adapter->registered_device_map == 0) {
3058 dev_err(&pdev->dev, "could not register any net devices\n");
3059 goto err_disable_interrupts;
3063 * Set up our debugfs entries.
3065 if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) {
3066 adapter->debugfs_root =
3067 debugfs_create_dir(pci_name(pdev),
3068 cxgb4vf_debugfs_root);
3069 if (IS_ERR_OR_NULL(adapter->debugfs_root))
3070 dev_warn(&pdev->dev, "could not create debugfs"
3073 setup_debugfs(adapter);
3077 * Print a short notice on the existence and configuration of the new
3078 * VF network device ...
3080 for_each_port(adapter, pidx) {
3081 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
3082 adapter->port[pidx]->name,
3083 (adapter->flags & USING_MSIX) ? "MSI-X" :
3084 (adapter->flags & USING_MSI) ? "MSI" : "");
3093 * Error recovery and exit code. Unwind state that's been created
3094 * so far and return the error.
3096 err_disable_interrupts:
3097 if (adapter->flags & USING_MSIX) {
3098 pci_disable_msix(adapter->pdev);
3099 adapter->flags &= ~USING_MSIX;
3100 } else if (adapter->flags & USING_MSI) {
3101 pci_disable_msi(adapter->pdev);
3102 adapter->flags &= ~USING_MSI;
3106 for_each_port(adapter, pidx) {
3107 netdev = adapter->port[pidx];
3110 pi = netdev_priv(netdev);
3111 t4vf_free_vi(adapter, pi->viid);
3112 if (test_bit(pidx, &adapter->registered_device_map))
3113 unregister_netdev(netdev);
3114 free_netdev(netdev);
3118 if (!is_t4(adapter->params.chip))
3119 iounmap(adapter->bar2);
3122 iounmap(adapter->regs);
3125 kfree(adapter->mbox_log);
3128 err_release_regions:
3129 pci_release_regions(pdev);
3130 pci_clear_master(pdev);
3133 pci_disable_device(pdev);
3139 * "Remove" a device: tear down all kernel and driver state created in the
3140 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
3141 * that this is called "remove_one" in the PF Driver.)
3143 static void cxgb4vf_pci_remove(struct pci_dev *pdev)
3145 struct adapter *adapter = pci_get_drvdata(pdev);
3148 * Tear down driver state associated with device.
3154 * Stop all of our activity. Unregister network port,
3155 * disable interrupts, etc.
3157 for_each_port(adapter, pidx)
3158 if (test_bit(pidx, &adapter->registered_device_map))
3159 unregister_netdev(adapter->port[pidx]);
3160 t4vf_sge_stop(adapter);
3161 if (adapter->flags & USING_MSIX) {
3162 pci_disable_msix(adapter->pdev);
3163 adapter->flags &= ~USING_MSIX;
3164 } else if (adapter->flags & USING_MSI) {
3165 pci_disable_msi(adapter->pdev);
3166 adapter->flags &= ~USING_MSI;
3170 * Tear down our debugfs entries.
3172 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
3173 cleanup_debugfs(adapter);
3174 debugfs_remove_recursive(adapter->debugfs_root);
3178 * Free all of the various resources which we've acquired ...
3180 t4vf_free_sge_resources(adapter);
3181 for_each_port(adapter, pidx) {
3182 struct net_device *netdev = adapter->port[pidx];
3183 struct port_info *pi;
3188 pi = netdev_priv(netdev);
3189 t4vf_free_vi(adapter, pi->viid);
3190 free_netdev(netdev);
3192 iounmap(adapter->regs);
3193 if (!is_t4(adapter->params.chip))
3194 iounmap(adapter->bar2);
3195 kfree(adapter->mbox_log);
3200 * Disable the device and release its PCI resources.
3202 pci_disable_device(pdev);
3203 pci_clear_master(pdev);
3204 pci_release_regions(pdev);
3208 * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt
3211 static void cxgb4vf_pci_shutdown(struct pci_dev *pdev)
3213 struct adapter *adapter;
3216 adapter = pci_get_drvdata(pdev);
3220 /* Disable all Virtual Interfaces. This will shut down the
3221 * delivery of all ingress packets into the chip for these
3222 * Virtual Interfaces.
3224 for_each_port(adapter, pidx)
3225 if (test_bit(pidx, &adapter->registered_device_map))
3226 unregister_netdev(adapter->port[pidx]);
3228 /* Free up all Queues which will prevent further DMA and
3229 * Interrupts allowing various internal pathways to drain.
3231 t4vf_sge_stop(adapter);
3232 if (adapter->flags & USING_MSIX) {
3233 pci_disable_msix(adapter->pdev);
3234 adapter->flags &= ~USING_MSIX;
3235 } else if (adapter->flags & USING_MSI) {
3236 pci_disable_msi(adapter->pdev);
3237 adapter->flags &= ~USING_MSI;
3241 * Free up all Queues which will prevent further DMA and
3242 * Interrupts allowing various internal pathways to drain.
3244 t4vf_free_sge_resources(adapter);
3245 pci_set_drvdata(pdev, NULL);
3248 /* Macros needed to support the PCI Device ID Table ...
3250 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
3251 static const struct pci_device_id cxgb4vf_pci_tbl[] = {
3252 #define CH_PCI_DEVICE_ID_FUNCTION 0x8
3254 #define CH_PCI_ID_TABLE_ENTRY(devid) \
3255 { PCI_VDEVICE(CHELSIO, (devid)), 0 }
3257 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END { 0, } }
3259 #include "../cxgb4/t4_pci_id_tbl.h"
3261 MODULE_DESCRIPTION(DRV_DESC);
3262 MODULE_AUTHOR("Chelsio Communications");
3263 MODULE_LICENSE("Dual BSD/GPL");
3264 MODULE_VERSION(DRV_VERSION);
3265 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
3267 static struct pci_driver cxgb4vf_driver = {
3268 .name = KBUILD_MODNAME,
3269 .id_table = cxgb4vf_pci_tbl,
3270 .probe = cxgb4vf_pci_probe,
3271 .remove = cxgb4vf_pci_remove,
3272 .shutdown = cxgb4vf_pci_shutdown,
3276 * Initialize global driver state.
3278 static int __init cxgb4vf_module_init(void)
3283 * Vet our module parameters.
3285 if (msi != MSI_MSIX && msi != MSI_MSI) {
3286 pr_warn("bad module parameter msi=%d; must be %d (MSI-X or MSI) or %d (MSI)\n",
3287 msi, MSI_MSIX, MSI_MSI);
3291 /* Debugfs support is optional, just warn if this fails */
3292 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3293 if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3294 pr_warn("could not create debugfs entry, continuing\n");
3296 ret = pci_register_driver(&cxgb4vf_driver);
3297 if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3298 debugfs_remove(cxgb4vf_debugfs_root);
3303 * Tear down global driver state.
3305 static void __exit cxgb4vf_module_exit(void)
3307 pci_unregister_driver(&cxgb4vf_driver);
3308 debugfs_remove(cxgb4vf_debugfs_root);
3311 module_init(cxgb4vf_module_init);
3312 module_exit(cxgb4vf_module_exit);