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 switch (pi->link_cfg.speed) {
183 switch ((int)pi->link_cfg.fc) {
192 case PAUSE_RX | PAUSE_TX:
201 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s, fc);
203 netdev_info(dev, "link down\n");
208 * THe port module type has changed on the indicated "port" (Virtual
211 void t4vf_os_portmod_changed(struct adapter *adapter, int pidx)
213 static const char * const mod_str[] = {
214 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
216 const struct net_device *dev = adapter->port[pidx];
217 const struct port_info *pi = netdev_priv(dev);
219 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
220 dev_info(adapter->pdev_dev, "%s: port module unplugged\n",
222 else if (pi->mod_type < ARRAY_SIZE(mod_str))
223 dev_info(adapter->pdev_dev, "%s: %s port module inserted\n",
224 dev->name, mod_str[pi->mod_type]);
225 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
226 dev_info(adapter->pdev_dev, "%s: unsupported optical port "
227 "module inserted\n", dev->name);
228 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
229 dev_info(adapter->pdev_dev, "%s: unknown port module inserted,"
230 "forcing TWINAX\n", dev->name);
231 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
232 dev_info(adapter->pdev_dev, "%s: transceiver module error\n",
235 dev_info(adapter->pdev_dev, "%s: unknown module type %d "
236 "inserted\n", dev->name, pi->mod_type);
240 * Net device operations.
241 * ======================
248 * Perform the MAC and PHY actions needed to enable a "port" (Virtual
251 static int link_start(struct net_device *dev)
254 struct port_info *pi = netdev_priv(dev);
257 * We do not set address filters and promiscuity here, the stack does
258 * that step explicitly. Enable vlan accel.
260 ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, 1,
263 ret = t4vf_change_mac(pi->adapter, pi->viid,
264 pi->xact_addr_filt, dev->dev_addr, true);
266 pi->xact_addr_filt = ret;
272 * We don't need to actually "start the link" itself since the
273 * firmware will do that for us when the first Virtual Interface
274 * is enabled on a port.
277 ret = t4vf_enable_pi(pi->adapter, pi, true, true);
279 /* The Virtual Interfaces are connected to an internal switch on the
280 * chip which allows VIs attached to the same port to talk to each
281 * other even when the port link is down. As a result, we generally
282 * want to always report a VI's link as being "up", provided there are
283 * no errors in enabling vi.
287 netif_carrier_on(dev);
293 * Name the MSI-X interrupts.
295 static void name_msix_vecs(struct adapter *adapter)
297 int namelen = sizeof(adapter->msix_info[0].desc) - 1;
303 snprintf(adapter->msix_info[MSIX_FW].desc, namelen,
304 "%s-FWeventq", adapter->name);
305 adapter->msix_info[MSIX_FW].desc[namelen] = 0;
310 for_each_port(adapter, pidx) {
311 struct net_device *dev = adapter->port[pidx];
312 const struct port_info *pi = netdev_priv(dev);
315 for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) {
316 snprintf(adapter->msix_info[msi].desc, namelen,
317 "%s-%d", dev->name, qs);
318 adapter->msix_info[msi].desc[namelen] = 0;
324 * Request all of our MSI-X resources.
326 static int request_msix_queue_irqs(struct adapter *adapter)
328 struct sge *s = &adapter->sge;
334 err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix,
335 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq);
343 for_each_ethrxq(s, rxq) {
344 err = request_irq(adapter->msix_info[msi].vec,
345 t4vf_sge_intr_msix, 0,
346 adapter->msix_info[msi].desc,
347 &s->ethrxq[rxq].rspq);
356 free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq);
357 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
362 * Free our MSI-X resources.
364 static void free_msix_queue_irqs(struct adapter *adapter)
366 struct sge *s = &adapter->sge;
369 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
371 for_each_ethrxq(s, rxq)
372 free_irq(adapter->msix_info[msi++].vec,
373 &s->ethrxq[rxq].rspq);
377 * Turn on NAPI and start up interrupts on a response queue.
379 static void qenable(struct sge_rspq *rspq)
381 napi_enable(&rspq->napi);
384 * 0-increment the Going To Sleep register to start the timer and
387 t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
389 SEINTARM_V(rspq->intr_params) |
390 INGRESSQID_V(rspq->cntxt_id));
394 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
396 static void enable_rx(struct adapter *adapter)
399 struct sge *s = &adapter->sge;
401 for_each_ethrxq(s, rxq)
402 qenable(&s->ethrxq[rxq].rspq);
403 qenable(&s->fw_evtq);
406 * The interrupt queue doesn't use NAPI so we do the 0-increment of
407 * its Going To Sleep register here to get it started.
409 if (adapter->flags & USING_MSI)
410 t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
412 SEINTARM_V(s->intrq.intr_params) |
413 INGRESSQID_V(s->intrq.cntxt_id));
418 * Wait until all NAPI handlers are descheduled.
420 static void quiesce_rx(struct adapter *adapter)
422 struct sge *s = &adapter->sge;
425 for_each_ethrxq(s, rxq)
426 napi_disable(&s->ethrxq[rxq].rspq.napi);
427 napi_disable(&s->fw_evtq.napi);
431 * Response queue handler for the firmware event queue.
433 static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp,
434 const struct pkt_gl *gl)
437 * Extract response opcode and get pointer to CPL message body.
439 struct adapter *adapter = rspq->adapter;
440 u8 opcode = ((const struct rss_header *)rsp)->opcode;
441 void *cpl = (void *)(rsp + 1);
446 * We've received an asynchronous message from the firmware.
448 const struct cpl_fw6_msg *fw_msg = cpl;
449 if (fw_msg->type == FW6_TYPE_CMD_RPL)
450 t4vf_handle_fw_rpl(adapter, fw_msg->data);
455 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
457 const struct cpl_sge_egr_update *p = (void *)(rsp + 3);
458 opcode = CPL_OPCODE_G(ntohl(p->opcode_qid));
459 if (opcode != CPL_SGE_EGR_UPDATE) {
460 dev_err(adapter->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
468 case CPL_SGE_EGR_UPDATE: {
470 * We've received an Egress Queue Status Update message. We
471 * get these, if the SGE is configured to send these when the
472 * firmware passes certain points in processing our TX
473 * Ethernet Queue or if we make an explicit request for one.
474 * We use these updates to determine when we may need to
475 * restart a TX Ethernet Queue which was stopped for lack of
476 * free TX Queue Descriptors ...
478 const struct cpl_sge_egr_update *p = cpl;
479 unsigned int qid = EGR_QID_G(be32_to_cpu(p->opcode_qid));
480 struct sge *s = &adapter->sge;
482 struct sge_eth_txq *txq;
486 * Perform sanity checking on the Queue ID to make sure it
487 * really refers to one of our TX Ethernet Egress Queues which
488 * is active and matches the queue's ID. None of these error
489 * conditions should ever happen so we may want to either make
490 * them fatal and/or conditionalized under DEBUG.
492 eq_idx = EQ_IDX(s, qid);
493 if (unlikely(eq_idx >= MAX_EGRQ)) {
494 dev_err(adapter->pdev_dev,
495 "Egress Update QID %d out of range\n", qid);
498 tq = s->egr_map[eq_idx];
499 if (unlikely(tq == NULL)) {
500 dev_err(adapter->pdev_dev,
501 "Egress Update QID %d TXQ=NULL\n", qid);
504 txq = container_of(tq, struct sge_eth_txq, q);
505 if (unlikely(tq->abs_id != qid)) {
506 dev_err(adapter->pdev_dev,
507 "Egress Update QID %d refers to TXQ %d\n",
513 * Restart a stopped TX Queue which has less than half of its
517 netif_tx_wake_queue(txq->txq);
522 dev_err(adapter->pdev_dev,
523 "unexpected CPL %#x on FW event queue\n", opcode);
530 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
531 * to use and initializes them. We support multiple "Queue Sets" per port if
532 * we have MSI-X, otherwise just one queue set per port.
534 static int setup_sge_queues(struct adapter *adapter)
536 struct sge *s = &adapter->sge;
540 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
543 bitmap_zero(s->starving_fl, MAX_EGRQ);
546 * If we're using MSI interrupt mode we need to set up a "forwarded
547 * interrupt" queue which we'll set up with our MSI vector. The rest
548 * of the ingress queues will be set up to forward their interrupts to
549 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
550 * the intrq's queue ID as the interrupt forwarding queue for the
551 * subsequent calls ...
553 if (adapter->flags & USING_MSI) {
554 err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false,
555 adapter->port[0], 0, NULL, NULL);
557 goto err_free_queues;
561 * Allocate our ingress queue for asynchronous firmware messages.
563 err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0],
564 MSIX_FW, NULL, fwevtq_handler);
566 goto err_free_queues;
569 * Allocate each "port"'s initial Queue Sets. These can be changed
570 * later on ... up to the point where any interface on the adapter is
571 * brought up at which point lots of things get nailed down
575 for_each_port(adapter, pidx) {
576 struct net_device *dev = adapter->port[pidx];
577 struct port_info *pi = netdev_priv(dev);
578 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
579 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
582 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
583 err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false,
585 &rxq->fl, t4vf_ethrx_handler);
587 goto err_free_queues;
589 err = t4vf_sge_alloc_eth_txq(adapter, txq, dev,
590 netdev_get_tx_queue(dev, qs),
591 s->fw_evtq.cntxt_id);
593 goto err_free_queues;
596 memset(&rxq->stats, 0, sizeof(rxq->stats));
601 * Create the reverse mappings for the queues.
603 s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id;
604 s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id;
605 IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq;
606 for_each_port(adapter, pidx) {
607 struct net_device *dev = adapter->port[pidx];
608 struct port_info *pi = netdev_priv(dev);
609 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
610 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
613 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
614 IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq;
615 EQ_MAP(s, txq->q.abs_id) = &txq->q;
618 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
619 * for Free Lists but since all of the Egress Queues
620 * (including Free Lists) have Relative Queue IDs
621 * which are computed as Absolute - Base Queue ID, we
622 * can synthesize the Absolute Queue IDs for the Free
623 * Lists. This is useful for debugging purposes when
624 * we want to dump Queue Contexts via the PF Driver.
626 rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base;
627 EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl;
633 t4vf_free_sge_resources(adapter);
638 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
639 * queues. We configure the RSS CPU lookup table to distribute to the number
640 * of HW receive queues, and the response queue lookup table to narrow that
641 * down to the response queues actually configured for each "port" (Virtual
642 * Interface). We always configure the RSS mapping for all ports since the
643 * mapping table has plenty of entries.
645 static int setup_rss(struct adapter *adapter)
649 for_each_port(adapter, pidx) {
650 struct port_info *pi = adap2pinfo(adapter, pidx);
651 struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
652 u16 rss[MAX_PORT_QSETS];
655 for (qs = 0; qs < pi->nqsets; qs++)
656 rss[qs] = rxq[qs].rspq.abs_id;
658 err = t4vf_config_rss_range(adapter, pi->viid,
659 0, pi->rss_size, rss, pi->nqsets);
664 * Perform Global RSS Mode-specific initialization.
666 switch (adapter->params.rss.mode) {
667 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL:
669 * If Tunnel All Lookup isn't specified in the global
670 * RSS Configuration, then we need to specify a
671 * default Ingress Queue for any ingress packets which
672 * aren't hashed. We'll use our first ingress queue
675 if (!adapter->params.rss.u.basicvirtual.tnlalllookup) {
676 union rss_vi_config config;
677 err = t4vf_read_rss_vi_config(adapter,
682 config.basicvirtual.defaultq =
684 err = t4vf_write_rss_vi_config(adapter,
698 * Bring the adapter up. Called whenever we go from no "ports" open to having
699 * one open. This function performs the actions necessary to make an adapter
700 * operational, such as completing the initialization of HW modules, and
701 * enabling interrupts. Must be called with the rtnl lock held. (Note that
702 * this is called "cxgb_up" in the PF Driver.)
704 static int adapter_up(struct adapter *adapter)
709 * If this is the first time we've been called, perform basic
710 * adapter setup. Once we've done this, many of our adapter
711 * parameters can no longer be changed ...
713 if ((adapter->flags & FULL_INIT_DONE) == 0) {
714 err = setup_sge_queues(adapter);
717 err = setup_rss(adapter);
719 t4vf_free_sge_resources(adapter);
723 if (adapter->flags & USING_MSIX)
724 name_msix_vecs(adapter);
726 adapter->flags |= FULL_INIT_DONE;
730 * Acquire our interrupt resources. We only support MSI-X and MSI.
732 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
733 if (adapter->flags & USING_MSIX)
734 err = request_msix_queue_irqs(adapter);
736 err = request_irq(adapter->pdev->irq,
737 t4vf_intr_handler(adapter), 0,
738 adapter->name, adapter);
740 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
746 * Enable NAPI ingress processing and return success.
749 t4vf_sge_start(adapter);
755 * Bring the adapter down. Called whenever the last "port" (Virtual
756 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
759 static void adapter_down(struct adapter *adapter)
762 * Free interrupt resources.
764 if (adapter->flags & USING_MSIX)
765 free_msix_queue_irqs(adapter);
767 free_irq(adapter->pdev->irq, adapter);
770 * Wait for NAPI handlers to finish.
776 * Start up a net device.
778 static int cxgb4vf_open(struct net_device *dev)
781 struct port_info *pi = netdev_priv(dev);
782 struct adapter *adapter = pi->adapter;
785 * If this is the first interface that we're opening on the "adapter",
786 * bring the "adapter" up now.
788 if (adapter->open_device_map == 0) {
789 err = adapter_up(adapter);
795 * Note that this interface is up and start everything up ...
797 err = link_start(dev);
801 pi->vlan_id = t4vf_get_vf_vlan_acl(adapter);
803 netif_tx_start_all_queues(dev);
804 set_bit(pi->port_id, &adapter->open_device_map);
808 if (adapter->open_device_map == 0)
809 adapter_down(adapter);
814 * Shut down a net device. This routine is called "cxgb_close" in the PF
817 static int cxgb4vf_stop(struct net_device *dev)
819 struct port_info *pi = netdev_priv(dev);
820 struct adapter *adapter = pi->adapter;
822 netif_tx_stop_all_queues(dev);
823 netif_carrier_off(dev);
824 t4vf_enable_pi(adapter, pi, false, false);
826 clear_bit(pi->port_id, &adapter->open_device_map);
827 if (adapter->open_device_map == 0)
828 adapter_down(adapter);
833 * Translate our basic statistics into the standard "ifconfig" statistics.
835 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
837 struct t4vf_port_stats stats;
838 struct port_info *pi = netdev2pinfo(dev);
839 struct adapter *adapter = pi->adapter;
840 struct net_device_stats *ns = &dev->stats;
843 spin_lock(&adapter->stats_lock);
844 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
845 spin_unlock(&adapter->stats_lock);
847 memset(ns, 0, sizeof(*ns));
851 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
852 stats.tx_ucast_bytes + stats.tx_offload_bytes);
853 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
854 stats.tx_ucast_frames + stats.tx_offload_frames);
855 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
856 stats.rx_ucast_bytes);
857 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
858 stats.rx_ucast_frames);
859 ns->multicast = stats.rx_mcast_frames;
860 ns->tx_errors = stats.tx_drop_frames;
861 ns->rx_errors = stats.rx_err_frames;
866 static inline int cxgb4vf_set_addr_hash(struct port_info *pi)
868 struct adapter *adapter = pi->adapter;
871 struct hash_mac_addr *entry;
873 /* Calculate the hash vector for the updated list and program it */
874 list_for_each_entry(entry, &adapter->mac_hlist, list) {
875 ucast |= is_unicast_ether_addr(entry->addr);
876 vec |= (1ULL << hash_mac_addr(entry->addr));
878 return t4vf_set_addr_hash(adapter, pi->viid, ucast, vec, false);
881 static int cxgb4vf_mac_sync(struct net_device *netdev, const u8 *mac_addr)
883 struct port_info *pi = netdev_priv(netdev);
884 struct adapter *adapter = pi->adapter;
889 bool ucast = is_unicast_ether_addr(mac_addr);
890 const u8 *maclist[1] = {mac_addr};
891 struct hash_mac_addr *new_entry;
893 ret = t4vf_alloc_mac_filt(adapter, pi->viid, free, 1, maclist,
894 NULL, ucast ? &uhash : &mhash, false);
897 /* if hash != 0, then add the addr to hash addr list
898 * so on the end we will calculate the hash for the
899 * list and program it
901 if (uhash || mhash) {
902 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
905 ether_addr_copy(new_entry->addr, mac_addr);
906 list_add_tail(&new_entry->list, &adapter->mac_hlist);
907 ret = cxgb4vf_set_addr_hash(pi);
910 return ret < 0 ? ret : 0;
913 static int cxgb4vf_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
915 struct port_info *pi = netdev_priv(netdev);
916 struct adapter *adapter = pi->adapter;
918 const u8 *maclist[1] = {mac_addr};
919 struct hash_mac_addr *entry, *tmp;
921 /* If the MAC address to be removed is in the hash addr
922 * list, delete it from the list and update hash vector
924 list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist, list) {
925 if (ether_addr_equal(entry->addr, mac_addr)) {
926 list_del(&entry->list);
928 return cxgb4vf_set_addr_hash(pi);
932 ret = t4vf_free_mac_filt(adapter, pi->viid, 1, maclist, false);
933 return ret < 0 ? -EINVAL : 0;
937 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
938 * If @mtu is -1 it is left unchanged.
940 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
942 struct port_info *pi = netdev_priv(dev);
944 __dev_uc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync);
945 __dev_mc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync);
946 return t4vf_set_rxmode(pi->adapter, pi->viid, -1,
947 (dev->flags & IFF_PROMISC) != 0,
948 (dev->flags & IFF_ALLMULTI) != 0,
953 * Set the current receive modes on the device.
955 static void cxgb4vf_set_rxmode(struct net_device *dev)
957 /* unfortunately we can't return errors to the stack */
958 set_rxmode(dev, -1, false);
962 * Find the entry in the interrupt holdoff timer value array which comes
963 * closest to the specified interrupt holdoff value.
965 static int closest_timer(const struct sge *s, int us)
967 int i, timer_idx = 0, min_delta = INT_MAX;
969 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
970 int delta = us - s->timer_val[i];
973 if (delta < min_delta) {
981 static int closest_thres(const struct sge *s, int thres)
983 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
985 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
986 delta = thres - s->counter_val[i];
989 if (delta < min_delta) {
998 * Return a queue's interrupt hold-off time in us. 0 means no timer.
1000 static unsigned int qtimer_val(const struct adapter *adapter,
1001 const struct sge_rspq *rspq)
1003 unsigned int timer_idx = QINTR_TIMER_IDX_G(rspq->intr_params);
1005 return timer_idx < SGE_NTIMERS
1006 ? adapter->sge.timer_val[timer_idx]
1011 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1012 * @adapter: the adapter
1013 * @rspq: the RX response queue
1014 * @us: the hold-off time in us, or 0 to disable timer
1015 * @cnt: the hold-off packet count, or 0 to disable counter
1017 * Sets an RX response queue's interrupt hold-off time and packet count.
1018 * At least one of the two needs to be enabled for the queue to generate
1021 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
1022 unsigned int us, unsigned int cnt)
1024 unsigned int timer_idx;
1027 * If both the interrupt holdoff timer and count are specified as
1028 * zero, default to a holdoff count of 1 ...
1030 if ((us | cnt) == 0)
1034 * If an interrupt holdoff count has been specified, then find the
1035 * closest configured holdoff count and use that. If the response
1036 * queue has already been created, then update its queue context
1043 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1044 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1045 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1046 FW_PARAMS_PARAM_X_V(
1047 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1048 FW_PARAMS_PARAM_YZ_V(rspq->cntxt_id);
1049 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1053 rspq->pktcnt_idx = pktcnt_idx;
1057 * Compute the closest holdoff timer index from the supplied holdoff
1060 timer_idx = (us == 0
1061 ? SGE_TIMER_RSTRT_CNTR
1062 : closest_timer(&adapter->sge, us));
1065 * Update the response queue's interrupt coalescing parameters and
1068 rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) |
1069 QINTR_CNT_EN_V(cnt > 0));
1074 * Return a version number to identify the type of adapter. The scheme is:
1075 * - bits 0..9: chip version
1076 * - bits 10..15: chip revision
1078 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1081 * Chip version 4, revision 0x3f (cxgb4vf).
1083 return CHELSIO_CHIP_VERSION(adapter->params.chip) | (0x3f << 10);
1087 * Execute the specified ioctl command.
1089 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1095 * The VF Driver doesn't have access to any of the other
1096 * common Ethernet device ioctl()'s (like reading/writing
1097 * PHY registers, etc.
1108 * Change the device's MTU.
1110 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1113 struct port_info *pi = netdev_priv(dev);
1115 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1116 -1, -1, -1, -1, true);
1122 static netdev_features_t cxgb4vf_fix_features(struct net_device *dev,
1123 netdev_features_t features)
1126 * Since there is no support for separate rx/tx vlan accel
1127 * enable/disable make sure tx flag is always in same state as rx.
1129 if (features & NETIF_F_HW_VLAN_CTAG_RX)
1130 features |= NETIF_F_HW_VLAN_CTAG_TX;
1132 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
1137 static int cxgb4vf_set_features(struct net_device *dev,
1138 netdev_features_t features)
1140 struct port_info *pi = netdev_priv(dev);
1141 netdev_features_t changed = dev->features ^ features;
1143 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1144 t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1,
1145 features & NETIF_F_HW_VLAN_CTAG_TX, 0);
1151 * Change the devices MAC address.
1153 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1156 struct sockaddr *addr = _addr;
1157 struct port_info *pi = netdev_priv(dev);
1159 if (!is_valid_ether_addr(addr->sa_data))
1160 return -EADDRNOTAVAIL;
1162 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1163 addr->sa_data, true);
1167 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1168 pi->xact_addr_filt = ret;
1172 #ifdef CONFIG_NET_POLL_CONTROLLER
1174 * Poll all of our receive queues. This is called outside of normal interrupt
1177 static void cxgb4vf_poll_controller(struct net_device *dev)
1179 struct port_info *pi = netdev_priv(dev);
1180 struct adapter *adapter = pi->adapter;
1182 if (adapter->flags & USING_MSIX) {
1183 struct sge_eth_rxq *rxq;
1186 rxq = &adapter->sge.ethrxq[pi->first_qset];
1187 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1188 t4vf_sge_intr_msix(0, &rxq->rspq);
1192 t4vf_intr_handler(adapter)(0, adapter);
1197 * Ethtool operations.
1198 * ===================
1200 * Note that we don't support any ethtool operations which change the physical
1201 * state of the port to which we're linked.
1205 * from_fw_port_mod_type - translate Firmware Port/Module type to Ethtool
1206 * @port_type: Firmware Port Type
1207 * @mod_type: Firmware Module Type
1209 * Translate Firmware Port/Module type to Ethtool Port Type.
1211 static int from_fw_port_mod_type(enum fw_port_type port_type,
1212 enum fw_port_module_type mod_type)
1214 if (port_type == FW_PORT_TYPE_BT_SGMII ||
1215 port_type == FW_PORT_TYPE_BT_XFI ||
1216 port_type == FW_PORT_TYPE_BT_XAUI) {
1218 } else if (port_type == FW_PORT_TYPE_FIBER_XFI ||
1219 port_type == FW_PORT_TYPE_FIBER_XAUI) {
1221 } else if (port_type == FW_PORT_TYPE_SFP ||
1222 port_type == FW_PORT_TYPE_QSFP_10G ||
1223 port_type == FW_PORT_TYPE_QSA ||
1224 port_type == FW_PORT_TYPE_QSFP ||
1225 port_type == FW_PORT_TYPE_CR4_QSFP ||
1226 port_type == FW_PORT_TYPE_CR_QSFP ||
1227 port_type == FW_PORT_TYPE_CR2_QSFP ||
1228 port_type == FW_PORT_TYPE_SFP28) {
1229 if (mod_type == FW_PORT_MOD_TYPE_LR ||
1230 mod_type == FW_PORT_MOD_TYPE_SR ||
1231 mod_type == FW_PORT_MOD_TYPE_ER ||
1232 mod_type == FW_PORT_MOD_TYPE_LRM)
1234 else if (mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
1235 mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
1239 } else if (port_type == FW_PORT_TYPE_KR4_100G ||
1240 port_type == FW_PORT_TYPE_KR_SFP28 ||
1241 port_type == FW_PORT_TYPE_KR_XLAUI) {
1249 * fw_caps_to_lmm - translate Firmware to ethtool Link Mode Mask
1250 * @port_type: Firmware Port Type
1251 * @fw_caps: Firmware Port Capabilities
1252 * @link_mode_mask: ethtool Link Mode Mask
1254 * Translate a Firmware Port Capabilities specification to an ethtool
1257 static void fw_caps_to_lmm(enum fw_port_type port_type,
1258 unsigned int fw_caps,
1259 unsigned long *link_mode_mask)
1261 #define SET_LMM(__lmm_name) \
1262 __set_bit(ETHTOOL_LINK_MODE_ ## __lmm_name ## _BIT, \
1265 #define FW_CAPS_TO_LMM(__fw_name, __lmm_name) \
1267 if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1268 SET_LMM(__lmm_name); \
1271 switch (port_type) {
1272 case FW_PORT_TYPE_BT_SGMII:
1273 case FW_PORT_TYPE_BT_XFI:
1274 case FW_PORT_TYPE_BT_XAUI:
1276 FW_CAPS_TO_LMM(SPEED_100M, 100baseT_Full);
1277 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1278 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1281 case FW_PORT_TYPE_KX4:
1282 case FW_PORT_TYPE_KX:
1284 FW_CAPS_TO_LMM(SPEED_1G, 1000baseKX_Full);
1285 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKX4_Full);
1288 case FW_PORT_TYPE_KR:
1290 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKR_Full);
1293 case FW_PORT_TYPE_BP_AP:
1295 FW_CAPS_TO_LMM(SPEED_1G, 1000baseKX_Full);
1296 FW_CAPS_TO_LMM(SPEED_10G, 10000baseR_FEC);
1297 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKR_Full);
1300 case FW_PORT_TYPE_BP4_AP:
1302 FW_CAPS_TO_LMM(SPEED_1G, 1000baseKX_Full);
1303 FW_CAPS_TO_LMM(SPEED_10G, 10000baseR_FEC);
1304 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKR_Full);
1305 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKX4_Full);
1308 case FW_PORT_TYPE_FIBER_XFI:
1309 case FW_PORT_TYPE_FIBER_XAUI:
1310 case FW_PORT_TYPE_SFP:
1311 case FW_PORT_TYPE_QSFP_10G:
1312 case FW_PORT_TYPE_QSA:
1314 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1315 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1318 case FW_PORT_TYPE_BP40_BA:
1319 case FW_PORT_TYPE_QSFP:
1321 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1322 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1323 FW_CAPS_TO_LMM(SPEED_40G, 40000baseSR4_Full);
1326 case FW_PORT_TYPE_CR_QSFP:
1327 case FW_PORT_TYPE_SFP28:
1329 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1330 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1331 FW_CAPS_TO_LMM(SPEED_25G, 25000baseCR_Full);
1334 case FW_PORT_TYPE_KR_SFP28:
1336 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1337 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKR_Full);
1338 FW_CAPS_TO_LMM(SPEED_25G, 25000baseKR_Full);
1341 case FW_PORT_TYPE_KR_XLAUI:
1343 FW_CAPS_TO_LMM(SPEED_1G, 1000baseKX_Full);
1344 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKR_Full);
1345 FW_CAPS_TO_LMM(SPEED_40G, 40000baseKR4_Full);
1348 case FW_PORT_TYPE_CR2_QSFP:
1350 FW_CAPS_TO_LMM(SPEED_50G, 50000baseSR2_Full);
1353 case FW_PORT_TYPE_KR4_100G:
1354 case FW_PORT_TYPE_CR4_QSFP:
1356 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1357 FW_CAPS_TO_LMM(SPEED_10G, 10000baseSR_Full);
1358 FW_CAPS_TO_LMM(SPEED_40G, 40000baseSR4_Full);
1359 FW_CAPS_TO_LMM(SPEED_25G, 25000baseCR_Full);
1360 FW_CAPS_TO_LMM(SPEED_50G, 50000baseCR2_Full);
1361 FW_CAPS_TO_LMM(SPEED_100G, 100000baseCR4_Full);
1368 FW_CAPS_TO_LMM(ANEG, Autoneg);
1369 FW_CAPS_TO_LMM(802_3_PAUSE, Pause);
1370 FW_CAPS_TO_LMM(802_3_ASM_DIR, Asym_Pause);
1372 #undef FW_CAPS_TO_LMM
1376 static int cxgb4vf_get_link_ksettings(struct net_device *dev,
1377 struct ethtool_link_ksettings *link_ksettings)
1379 struct port_info *pi = netdev_priv(dev);
1380 struct ethtool_link_settings *base = &link_ksettings->base;
1382 /* For the nonce, the Firmware doesn't send up Port State changes
1383 * when the Virtual Interface attached to the Port is down. So
1384 * if it's down, let's grab any changes.
1386 if (!netif_running(dev))
1387 (void)t4vf_update_port_info(pi);
1389 ethtool_link_ksettings_zero_link_mode(link_ksettings, supported);
1390 ethtool_link_ksettings_zero_link_mode(link_ksettings, advertising);
1391 ethtool_link_ksettings_zero_link_mode(link_ksettings, lp_advertising);
1393 base->port = from_fw_port_mod_type(pi->port_type, pi->mod_type);
1395 if (pi->mdio_addr >= 0) {
1396 base->phy_address = pi->mdio_addr;
1397 base->mdio_support = (pi->port_type == FW_PORT_TYPE_BT_SGMII
1398 ? ETH_MDIO_SUPPORTS_C22
1399 : ETH_MDIO_SUPPORTS_C45);
1401 base->phy_address = 255;
1402 base->mdio_support = 0;
1405 fw_caps_to_lmm(pi->port_type, pi->link_cfg.pcaps,
1406 link_ksettings->link_modes.supported);
1407 fw_caps_to_lmm(pi->port_type, pi->link_cfg.acaps,
1408 link_ksettings->link_modes.advertising);
1409 fw_caps_to_lmm(pi->port_type, pi->link_cfg.lpacaps,
1410 link_ksettings->link_modes.lp_advertising);
1412 if (netif_carrier_ok(dev)) {
1413 base->speed = pi->link_cfg.speed;
1414 base->duplex = DUPLEX_FULL;
1416 base->speed = SPEED_UNKNOWN;
1417 base->duplex = DUPLEX_UNKNOWN;
1420 if (pi->link_cfg.fc & PAUSE_RX) {
1421 if (pi->link_cfg.fc & PAUSE_TX) {
1422 ethtool_link_ksettings_add_link_mode(link_ksettings,
1426 ethtool_link_ksettings_add_link_mode(link_ksettings,
1430 } else if (pi->link_cfg.fc & PAUSE_TX) {
1431 ethtool_link_ksettings_add_link_mode(link_ksettings,
1436 base->autoneg = pi->link_cfg.autoneg;
1437 if (pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG)
1438 ethtool_link_ksettings_add_link_mode(link_ksettings,
1439 supported, Autoneg);
1440 if (pi->link_cfg.autoneg)
1441 ethtool_link_ksettings_add_link_mode(link_ksettings,
1442 advertising, Autoneg);
1447 /* Translate the Firmware FEC value into the ethtool value. */
1448 static inline unsigned int fwcap_to_eth_fec(unsigned int fw_fec)
1450 unsigned int eth_fec = 0;
1452 if (fw_fec & FW_PORT_CAP32_FEC_RS)
1453 eth_fec |= ETHTOOL_FEC_RS;
1454 if (fw_fec & FW_PORT_CAP32_FEC_BASER_RS)
1455 eth_fec |= ETHTOOL_FEC_BASER;
1457 /* if nothing is set, then FEC is off */
1459 eth_fec = ETHTOOL_FEC_OFF;
1464 /* Translate Common Code FEC value into ethtool value. */
1465 static inline unsigned int cc_to_eth_fec(unsigned int cc_fec)
1467 unsigned int eth_fec = 0;
1469 if (cc_fec & FEC_AUTO)
1470 eth_fec |= ETHTOOL_FEC_AUTO;
1471 if (cc_fec & FEC_RS)
1472 eth_fec |= ETHTOOL_FEC_RS;
1473 if (cc_fec & FEC_BASER_RS)
1474 eth_fec |= ETHTOOL_FEC_BASER;
1476 /* if nothing is set, then FEC is off */
1478 eth_fec = ETHTOOL_FEC_OFF;
1483 static int cxgb4vf_get_fecparam(struct net_device *dev,
1484 struct ethtool_fecparam *fec)
1486 const struct port_info *pi = netdev_priv(dev);
1487 const struct link_config *lc = &pi->link_cfg;
1489 /* Translate the Firmware FEC Support into the ethtool value. We
1490 * always support IEEE 802.3 "automatic" selection of Link FEC type if
1491 * any FEC is supported.
1493 fec->fec = fwcap_to_eth_fec(lc->pcaps);
1494 if (fec->fec != ETHTOOL_FEC_OFF)
1495 fec->fec |= ETHTOOL_FEC_AUTO;
1497 /* Translate the current internal FEC parameters into the
1500 fec->active_fec = cc_to_eth_fec(lc->fec);
1505 * Return our driver information.
1507 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1508 struct ethtool_drvinfo *drvinfo)
1510 struct adapter *adapter = netdev2adap(dev);
1512 strlcpy(drvinfo->driver, KBUILD_MODNAME, sizeof(drvinfo->driver));
1513 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
1514 strlcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)),
1515 sizeof(drvinfo->bus_info));
1516 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1517 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1518 FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.fwrev),
1519 FW_HDR_FW_VER_MINOR_G(adapter->params.dev.fwrev),
1520 FW_HDR_FW_VER_MICRO_G(adapter->params.dev.fwrev),
1521 FW_HDR_FW_VER_BUILD_G(adapter->params.dev.fwrev),
1522 FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.tprev),
1523 FW_HDR_FW_VER_MINOR_G(adapter->params.dev.tprev),
1524 FW_HDR_FW_VER_MICRO_G(adapter->params.dev.tprev),
1525 FW_HDR_FW_VER_BUILD_G(adapter->params.dev.tprev));
1529 * Return current adapter message level.
1531 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1533 return netdev2adap(dev)->msg_enable;
1537 * Set current adapter message level.
1539 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1541 netdev2adap(dev)->msg_enable = msglevel;
1545 * Return the device's current Queue Set ring size parameters along with the
1546 * allowed maximum values. Since ethtool doesn't understand the concept of
1547 * multi-queue devices, we just return the current values associated with the
1550 static void cxgb4vf_get_ringparam(struct net_device *dev,
1551 struct ethtool_ringparam *rp)
1553 const struct port_info *pi = netdev_priv(dev);
1554 const struct sge *s = &pi->adapter->sge;
1556 rp->rx_max_pending = MAX_RX_BUFFERS;
1557 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1558 rp->rx_jumbo_max_pending = 0;
1559 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1561 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1562 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1563 rp->rx_jumbo_pending = 0;
1564 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1568 * Set the Queue Set ring size parameters for the device. Again, since
1569 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1570 * apply these new values across all of the Queue Sets associated with the
1571 * device -- after vetting them of course!
1573 static int cxgb4vf_set_ringparam(struct net_device *dev,
1574 struct ethtool_ringparam *rp)
1576 const struct port_info *pi = netdev_priv(dev);
1577 struct adapter *adapter = pi->adapter;
1578 struct sge *s = &adapter->sge;
1581 if (rp->rx_pending > MAX_RX_BUFFERS ||
1582 rp->rx_jumbo_pending ||
1583 rp->tx_pending > MAX_TXQ_ENTRIES ||
1584 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1585 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1586 rp->rx_pending < MIN_FL_ENTRIES ||
1587 rp->tx_pending < MIN_TXQ_ENTRIES)
1590 if (adapter->flags & FULL_INIT_DONE)
1593 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1594 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1595 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1596 s->ethtxq[qs].q.size = rp->tx_pending;
1602 * Return the interrupt holdoff timer and count for the first Queue Set on the
1603 * device. Our extension ioctl() (the cxgbtool interface) allows the
1604 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1606 static int cxgb4vf_get_coalesce(struct net_device *dev,
1607 struct ethtool_coalesce *coalesce)
1609 const struct port_info *pi = netdev_priv(dev);
1610 const struct adapter *adapter = pi->adapter;
1611 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1613 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1614 coalesce->rx_max_coalesced_frames =
1615 ((rspq->intr_params & QINTR_CNT_EN_F)
1616 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1622 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1623 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1624 * the interrupt holdoff timer on any of the device's Queue Sets.
1626 static int cxgb4vf_set_coalesce(struct net_device *dev,
1627 struct ethtool_coalesce *coalesce)
1629 const struct port_info *pi = netdev_priv(dev);
1630 struct adapter *adapter = pi->adapter;
1632 return set_rxq_intr_params(adapter,
1633 &adapter->sge.ethrxq[pi->first_qset].rspq,
1634 coalesce->rx_coalesce_usecs,
1635 coalesce->rx_max_coalesced_frames);
1639 * Report current port link pause parameter settings.
1641 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1642 struct ethtool_pauseparam *pauseparam)
1644 struct port_info *pi = netdev_priv(dev);
1646 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1647 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1648 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1652 * Identify the port by blinking the port's LED.
1654 static int cxgb4vf_phys_id(struct net_device *dev,
1655 enum ethtool_phys_id_state state)
1658 struct port_info *pi = netdev_priv(dev);
1660 if (state == ETHTOOL_ID_ACTIVE)
1662 else if (state == ETHTOOL_ID_INACTIVE)
1667 return t4vf_identify_port(pi->adapter, pi->viid, val);
1671 * Port stats maintained per queue of the port.
1673 struct queue_port_stats {
1684 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1685 * these need to match the order of statistics returned by
1686 * t4vf_get_port_stats().
1688 static const char stats_strings[][ETH_GSTRING_LEN] = {
1690 * These must match the layout of the t4vf_port_stats structure.
1692 "TxBroadcastBytes ",
1693 "TxBroadcastFrames ",
1694 "TxMulticastBytes ",
1695 "TxMulticastFrames ",
1701 "RxBroadcastBytes ",
1702 "RxBroadcastFrames ",
1703 "RxMulticastBytes ",
1704 "RxMulticastFrames ",
1710 * These are accumulated per-queue statistics and must match the
1711 * order of the fields in the queue_port_stats structure.
1723 * Return the number of statistics in the specified statistics set.
1725 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1729 return ARRAY_SIZE(stats_strings);
1737 * Return the strings for the specified statistics set.
1739 static void cxgb4vf_get_strings(struct net_device *dev,
1745 memcpy(data, stats_strings, sizeof(stats_strings));
1751 * Small utility routine to accumulate queue statistics across the queues of
1754 static void collect_sge_port_stats(const struct adapter *adapter,
1755 const struct port_info *pi,
1756 struct queue_port_stats *stats)
1758 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1759 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1762 memset(stats, 0, sizeof(*stats));
1763 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1764 stats->tso += txq->tso;
1765 stats->tx_csum += txq->tx_cso;
1766 stats->rx_csum += rxq->stats.rx_cso;
1767 stats->vlan_ex += rxq->stats.vlan_ex;
1768 stats->vlan_ins += txq->vlan_ins;
1769 stats->lro_pkts += rxq->stats.lro_pkts;
1770 stats->lro_merged += rxq->stats.lro_merged;
1775 * Return the ETH_SS_STATS statistics set.
1777 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1778 struct ethtool_stats *stats,
1781 struct port_info *pi = netdev2pinfo(dev);
1782 struct adapter *adapter = pi->adapter;
1783 int err = t4vf_get_port_stats(adapter, pi->pidx,
1784 (struct t4vf_port_stats *)data);
1786 memset(data, 0, sizeof(struct t4vf_port_stats));
1788 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1789 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1793 * Return the size of our register map.
1795 static int cxgb4vf_get_regs_len(struct net_device *dev)
1797 return T4VF_REGMAP_SIZE;
1801 * Dump a block of registers, start to end inclusive, into a buffer.
1803 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1804 unsigned int start, unsigned int end)
1806 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1808 for ( ; start <= end; start += sizeof(u32)) {
1810 * Avoid reading the Mailbox Control register since that
1811 * can trigger a Mailbox Ownership Arbitration cycle and
1812 * interfere with communication with the firmware.
1814 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1817 *bp++ = t4_read_reg(adapter, start);
1822 * Copy our entire register map into the provided buffer.
1824 static void cxgb4vf_get_regs(struct net_device *dev,
1825 struct ethtool_regs *regs,
1828 struct adapter *adapter = netdev2adap(dev);
1830 regs->version = mk_adap_vers(adapter);
1833 * Fill in register buffer with our register map.
1835 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1837 reg_block_dump(adapter, regbuf,
1838 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1839 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1840 reg_block_dump(adapter, regbuf,
1841 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1842 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1844 /* T5 adds new registers in the PL Register map.
1846 reg_block_dump(adapter, regbuf,
1847 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1848 T4VF_PL_BASE_ADDR + (is_t4(adapter->params.chip)
1849 ? PL_VF_WHOAMI_A : PL_VF_REVISION_A));
1850 reg_block_dump(adapter, regbuf,
1851 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1852 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1854 reg_block_dump(adapter, regbuf,
1855 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1856 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1860 * Report current Wake On LAN settings.
1862 static void cxgb4vf_get_wol(struct net_device *dev,
1863 struct ethtool_wolinfo *wol)
1867 memset(&wol->sopass, 0, sizeof(wol->sopass));
1871 * TCP Segmentation Offload flags which we support.
1873 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1875 static const struct ethtool_ops cxgb4vf_ethtool_ops = {
1876 .get_link_ksettings = cxgb4vf_get_link_ksettings,
1877 .get_fecparam = cxgb4vf_get_fecparam,
1878 .get_drvinfo = cxgb4vf_get_drvinfo,
1879 .get_msglevel = cxgb4vf_get_msglevel,
1880 .set_msglevel = cxgb4vf_set_msglevel,
1881 .get_ringparam = cxgb4vf_get_ringparam,
1882 .set_ringparam = cxgb4vf_set_ringparam,
1883 .get_coalesce = cxgb4vf_get_coalesce,
1884 .set_coalesce = cxgb4vf_set_coalesce,
1885 .get_pauseparam = cxgb4vf_get_pauseparam,
1886 .get_link = ethtool_op_get_link,
1887 .get_strings = cxgb4vf_get_strings,
1888 .set_phys_id = cxgb4vf_phys_id,
1889 .get_sset_count = cxgb4vf_get_sset_count,
1890 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1891 .get_regs_len = cxgb4vf_get_regs_len,
1892 .get_regs = cxgb4vf_get_regs,
1893 .get_wol = cxgb4vf_get_wol,
1897 * /sys/kernel/debug/cxgb4vf support code and data.
1898 * ================================================
1902 * Show Firmware Mailbox Command/Reply Log
1904 * Note that we don't do any locking when dumping the Firmware Mailbox Log so
1905 * it's possible that we can catch things during a log update and therefore
1906 * see partially corrupted log entries. But i9t's probably Good Enough(tm).
1907 * If we ever decide that we want to make sure that we're dumping a coherent
1908 * log, we'd need to perform locking in the mailbox logging and in
1909 * mboxlog_open() where we'd need to grab the entire mailbox log in one go
1910 * like we do for the Firmware Device Log. But as stated above, meh ...
1912 static int mboxlog_show(struct seq_file *seq, void *v)
1914 struct adapter *adapter = seq->private;
1915 struct mbox_cmd_log *log = adapter->mbox_log;
1916 struct mbox_cmd *entry;
1919 if (v == SEQ_START_TOKEN) {
1921 "%10s %15s %5s %5s %s\n",
1922 "Seq#", "Tstamp", "Atime", "Etime",
1927 entry_idx = log->cursor + ((uintptr_t)v - 2);
1928 if (entry_idx >= log->size)
1929 entry_idx -= log->size;
1930 entry = mbox_cmd_log_entry(log, entry_idx);
1932 /* skip over unused entries */
1933 if (entry->timestamp == 0)
1936 seq_printf(seq, "%10u %15llu %5d %5d",
1937 entry->seqno, entry->timestamp,
1938 entry->access, entry->execute);
1939 for (i = 0; i < MBOX_LEN / 8; i++) {
1940 u64 flit = entry->cmd[i];
1941 u32 hi = (u32)(flit >> 32);
1944 seq_printf(seq, " %08x %08x", hi, lo);
1946 seq_puts(seq, "\n");
1950 static inline void *mboxlog_get_idx(struct seq_file *seq, loff_t pos)
1952 struct adapter *adapter = seq->private;
1953 struct mbox_cmd_log *log = adapter->mbox_log;
1955 return ((pos <= log->size) ? (void *)(uintptr_t)(pos + 1) : NULL);
1958 static void *mboxlog_start(struct seq_file *seq, loff_t *pos)
1960 return *pos ? mboxlog_get_idx(seq, *pos) : SEQ_START_TOKEN;
1963 static void *mboxlog_next(struct seq_file *seq, void *v, loff_t *pos)
1966 return mboxlog_get_idx(seq, *pos);
1969 static void mboxlog_stop(struct seq_file *seq, void *v)
1973 static const struct seq_operations mboxlog_seq_ops = {
1974 .start = mboxlog_start,
1975 .next = mboxlog_next,
1976 .stop = mboxlog_stop,
1977 .show = mboxlog_show
1980 static int mboxlog_open(struct inode *inode, struct file *file)
1982 int res = seq_open(file, &mboxlog_seq_ops);
1985 struct seq_file *seq = file->private_data;
1987 seq->private = inode->i_private;
1992 static const struct file_operations mboxlog_fops = {
1993 .owner = THIS_MODULE,
1994 .open = mboxlog_open,
1996 .llseek = seq_lseek,
1997 .release = seq_release,
2001 * Show SGE Queue Set information. We display QPL Queues Sets per line.
2005 static int sge_qinfo_show(struct seq_file *seq, void *v)
2007 struct adapter *adapter = seq->private;
2008 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
2009 int qs, r = (uintptr_t)v - 1;
2012 seq_putc(seq, '\n');
2014 #define S3(fmt_spec, s, v) \
2016 seq_printf(seq, "%-12s", s); \
2017 for (qs = 0; qs < n; ++qs) \
2018 seq_printf(seq, " %16" fmt_spec, v); \
2019 seq_putc(seq, '\n'); \
2021 #define S(s, v) S3("s", s, v)
2022 #define T(s, v) S3("u", s, txq[qs].v)
2023 #define R(s, v) S3("u", s, rxq[qs].v)
2025 if (r < eth_entries) {
2026 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
2027 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
2028 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
2030 S("QType:", "Ethernet");
2032 (rxq[qs].rspq.netdev
2033 ? rxq[qs].rspq.netdev->name
2036 (rxq[qs].rspq.netdev
2037 ? ((struct port_info *)
2038 netdev_priv(rxq[qs].rspq.netdev))->port_id
2040 T("TxQ ID:", q.abs_id);
2041 T("TxQ size:", q.size);
2042 T("TxQ inuse:", q.in_use);
2043 T("TxQ PIdx:", q.pidx);
2044 T("TxQ CIdx:", q.cidx);
2045 R("RspQ ID:", rspq.abs_id);
2046 R("RspQ size:", rspq.size);
2047 R("RspQE size:", rspq.iqe_len);
2048 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
2049 S3("u", "Intr pktcnt:",
2050 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
2051 R("RspQ CIdx:", rspq.cidx);
2052 R("RspQ Gen:", rspq.gen);
2053 R("FL ID:", fl.abs_id);
2054 R("FL size:", fl.size - MIN_FL_RESID);
2055 R("FL avail:", fl.avail);
2056 R("FL PIdx:", fl.pidx);
2057 R("FL CIdx:", fl.cidx);
2063 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
2065 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
2066 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
2067 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
2068 qtimer_val(adapter, evtq));
2069 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
2070 adapter->sge.counter_val[evtq->pktcnt_idx]);
2071 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
2072 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
2073 } else if (r == 1) {
2074 const struct sge_rspq *intrq = &adapter->sge.intrq;
2076 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
2077 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
2078 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
2079 qtimer_val(adapter, intrq));
2080 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
2081 adapter->sge.counter_val[intrq->pktcnt_idx]);
2082 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
2083 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
2095 * Return the number of "entries" in our "file". We group the multi-Queue
2096 * sections with QPL Queue Sets per "entry". The sections of the output are:
2098 * Ethernet RX/TX Queue Sets
2099 * Firmware Event Queue
2100 * Forwarded Interrupt Queue (if in MSI mode)
2102 static int sge_queue_entries(const struct adapter *adapter)
2104 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
2105 ((adapter->flags & USING_MSI) != 0);
2108 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
2110 int entries = sge_queue_entries(seq->private);
2112 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2115 static void sge_queue_stop(struct seq_file *seq, void *v)
2119 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
2121 int entries = sge_queue_entries(seq->private);
2124 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2127 static const struct seq_operations sge_qinfo_seq_ops = {
2128 .start = sge_queue_start,
2129 .next = sge_queue_next,
2130 .stop = sge_queue_stop,
2131 .show = sge_qinfo_show
2134 static int sge_qinfo_open(struct inode *inode, struct file *file)
2136 int res = seq_open(file, &sge_qinfo_seq_ops);
2139 struct seq_file *seq = file->private_data;
2140 seq->private = inode->i_private;
2145 static const struct file_operations sge_qinfo_debugfs_fops = {
2146 .owner = THIS_MODULE,
2147 .open = sge_qinfo_open,
2149 .llseek = seq_lseek,
2150 .release = seq_release,
2154 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
2158 static int sge_qstats_show(struct seq_file *seq, void *v)
2160 struct adapter *adapter = seq->private;
2161 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
2162 int qs, r = (uintptr_t)v - 1;
2165 seq_putc(seq, '\n');
2167 #define S3(fmt, s, v) \
2169 seq_printf(seq, "%-16s", s); \
2170 for (qs = 0; qs < n; ++qs) \
2171 seq_printf(seq, " %8" fmt, v); \
2172 seq_putc(seq, '\n'); \
2174 #define S(s, v) S3("s", s, v)
2176 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
2177 #define T(s, v) T3("lu", s, v)
2179 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
2180 #define R(s, v) R3("lu", s, v)
2182 if (r < eth_entries) {
2183 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
2184 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
2185 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
2187 S("QType:", "Ethernet");
2189 (rxq[qs].rspq.netdev
2190 ? rxq[qs].rspq.netdev->name
2192 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
2193 R("RxPackets:", stats.pkts);
2194 R("RxCSO:", stats.rx_cso);
2195 R("VLANxtract:", stats.vlan_ex);
2196 R("LROmerged:", stats.lro_merged);
2197 R("LROpackets:", stats.lro_pkts);
2198 R("RxDrops:", stats.rx_drops);
2200 T("TxCSO:", tx_cso);
2201 T("VLANins:", vlan_ins);
2202 T("TxQFull:", q.stops);
2203 T("TxQRestarts:", q.restarts);
2204 T("TxMapErr:", mapping_err);
2205 R("FLAllocErr:", fl.alloc_failed);
2206 R("FLLrgAlcErr:", fl.large_alloc_failed);
2207 R("FLStarving:", fl.starving);
2213 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
2215 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
2216 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
2217 evtq->unhandled_irqs);
2218 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
2219 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
2220 } else if (r == 1) {
2221 const struct sge_rspq *intrq = &adapter->sge.intrq;
2223 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
2224 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
2225 intrq->unhandled_irqs);
2226 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
2227 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
2241 * Return the number of "entries" in our "file". We group the multi-Queue
2242 * sections with QPL Queue Sets per "entry". The sections of the output are:
2244 * Ethernet RX/TX Queue Sets
2245 * Firmware Event Queue
2246 * Forwarded Interrupt Queue (if in MSI mode)
2248 static int sge_qstats_entries(const struct adapter *adapter)
2250 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
2251 ((adapter->flags & USING_MSI) != 0);
2254 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
2256 int entries = sge_qstats_entries(seq->private);
2258 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2261 static void sge_qstats_stop(struct seq_file *seq, void *v)
2265 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
2267 int entries = sge_qstats_entries(seq->private);
2270 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2273 static const struct seq_operations sge_qstats_seq_ops = {
2274 .start = sge_qstats_start,
2275 .next = sge_qstats_next,
2276 .stop = sge_qstats_stop,
2277 .show = sge_qstats_show
2280 static int sge_qstats_open(struct inode *inode, struct file *file)
2282 int res = seq_open(file, &sge_qstats_seq_ops);
2285 struct seq_file *seq = file->private_data;
2286 seq->private = inode->i_private;
2291 static const struct file_operations sge_qstats_proc_fops = {
2292 .owner = THIS_MODULE,
2293 .open = sge_qstats_open,
2295 .llseek = seq_lseek,
2296 .release = seq_release,
2300 * Show PCI-E SR-IOV Virtual Function Resource Limits.
2302 static int resources_show(struct seq_file *seq, void *v)
2304 struct adapter *adapter = seq->private;
2305 struct vf_resources *vfres = &adapter->params.vfres;
2307 #define S(desc, fmt, var) \
2308 seq_printf(seq, "%-60s " fmt "\n", \
2309 desc " (" #var "):", vfres->var)
2311 S("Virtual Interfaces", "%d", nvi);
2312 S("Egress Queues", "%d", neq);
2313 S("Ethernet Control", "%d", nethctrl);
2314 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
2315 S("Ingress Queues", "%d", niq);
2316 S("Traffic Class", "%d", tc);
2317 S("Port Access Rights Mask", "%#x", pmask);
2318 S("MAC Address Filters", "%d", nexactf);
2319 S("Firmware Command Read Capabilities", "%#x", r_caps);
2320 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
2327 static int resources_open(struct inode *inode, struct file *file)
2329 return single_open(file, resources_show, inode->i_private);
2332 static const struct file_operations resources_proc_fops = {
2333 .owner = THIS_MODULE,
2334 .open = resources_open,
2336 .llseek = seq_lseek,
2337 .release = single_release,
2341 * Show Virtual Interfaces.
2343 static int interfaces_show(struct seq_file *seq, void *v)
2345 if (v == SEQ_START_TOKEN) {
2346 seq_puts(seq, "Interface Port VIID\n");
2348 struct adapter *adapter = seq->private;
2349 int pidx = (uintptr_t)v - 2;
2350 struct net_device *dev = adapter->port[pidx];
2351 struct port_info *pi = netdev_priv(dev);
2353 seq_printf(seq, "%9s %4d %#5x\n",
2354 dev->name, pi->port_id, pi->viid);
2359 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
2361 return pos <= adapter->params.nports
2362 ? (void *)(uintptr_t)(pos + 1)
2366 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
2369 ? interfaces_get_idx(seq->private, *pos)
2373 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
2376 return interfaces_get_idx(seq->private, *pos);
2379 static void interfaces_stop(struct seq_file *seq, void *v)
2383 static const struct seq_operations interfaces_seq_ops = {
2384 .start = interfaces_start,
2385 .next = interfaces_next,
2386 .stop = interfaces_stop,
2387 .show = interfaces_show
2390 static int interfaces_open(struct inode *inode, struct file *file)
2392 int res = seq_open(file, &interfaces_seq_ops);
2395 struct seq_file *seq = file->private_data;
2396 seq->private = inode->i_private;
2401 static const struct file_operations interfaces_proc_fops = {
2402 .owner = THIS_MODULE,
2403 .open = interfaces_open,
2405 .llseek = seq_lseek,
2406 .release = seq_release,
2410 * /sys/kernel/debugfs/cxgb4vf/ files list.
2412 struct cxgb4vf_debugfs_entry {
2413 const char *name; /* name of debugfs node */
2414 umode_t mode; /* file system mode */
2415 const struct file_operations *fops;
2418 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
2419 { "mboxlog", 0444, &mboxlog_fops },
2420 { "sge_qinfo", 0444, &sge_qinfo_debugfs_fops },
2421 { "sge_qstats", 0444, &sge_qstats_proc_fops },
2422 { "resources", 0444, &resources_proc_fops },
2423 { "interfaces", 0444, &interfaces_proc_fops },
2427 * Module and device initialization and cleanup code.
2428 * ==================================================
2432 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2433 * directory (debugfs_root) has already been set up.
2435 static int setup_debugfs(struct adapter *adapter)
2439 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2442 * Debugfs support is best effort.
2444 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2445 (void)debugfs_create_file(debugfs_files[i].name,
2446 debugfs_files[i].mode,
2447 adapter->debugfs_root,
2449 debugfs_files[i].fops);
2455 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2456 * it to our caller to tear down the directory (debugfs_root).
2458 static void cleanup_debugfs(struct adapter *adapter)
2460 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2463 * Unlike our sister routine cleanup_proc(), we don't need to remove
2464 * individual entries because a call will be made to
2465 * debugfs_remove_recursive(). We just need to clean up any ancillary
2471 /* Figure out how many Ports and Queue Sets we can support. This depends on
2472 * knowing our Virtual Function Resources and may be called a second time if
2473 * we fall back from MSI-X to MSI Interrupt Mode.
2475 static void size_nports_qsets(struct adapter *adapter)
2477 struct vf_resources *vfres = &adapter->params.vfres;
2478 unsigned int ethqsets, pmask_nports;
2480 /* The number of "ports" which we support is equal to the number of
2481 * Virtual Interfaces with which we've been provisioned.
2483 adapter->params.nports = vfres->nvi;
2484 if (adapter->params.nports > MAX_NPORTS) {
2485 dev_warn(adapter->pdev_dev, "only using %d of %d maximum"
2486 " allowed virtual interfaces\n", MAX_NPORTS,
2487 adapter->params.nports);
2488 adapter->params.nports = MAX_NPORTS;
2491 /* We may have been provisioned with more VIs than the number of
2492 * ports we're allowed to access (our Port Access Rights Mask).
2493 * This is obviously a configuration conflict but we don't want to
2494 * crash the kernel or anything silly just because of that.
2496 pmask_nports = hweight32(adapter->params.vfres.pmask);
2497 if (pmask_nports < adapter->params.nports) {
2498 dev_warn(adapter->pdev_dev, "only using %d of %d provisioned"
2499 " virtual interfaces; limited by Port Access Rights"
2500 " mask %#x\n", pmask_nports, adapter->params.nports,
2501 adapter->params.vfres.pmask);
2502 adapter->params.nports = pmask_nports;
2505 /* We need to reserve an Ingress Queue for the Asynchronous Firmware
2506 * Event Queue. And if we're using MSI Interrupts, we'll also need to
2507 * reserve an Ingress Queue for a Forwarded Interrupts.
2509 * The rest of the FL/Intr-capable ingress queues will be matched up
2510 * one-for-one with Ethernet/Control egress queues in order to form
2511 * "Queue Sets" which will be aportioned between the "ports". For
2512 * each Queue Set, we'll need the ability to allocate two Egress
2513 * Contexts -- one for the Ingress Queue Free List and one for the TX
2516 * Note that even if we're currently configured to use MSI-X
2517 * Interrupts (module variable msi == MSI_MSIX) we may get downgraded
2518 * to MSI Interrupts if we can't get enough MSI-X Interrupts. If that
2519 * happens we'll need to adjust things later.
2521 ethqsets = vfres->niqflint - 1 - (msi == MSI_MSI);
2522 if (vfres->nethctrl != ethqsets)
2523 ethqsets = min(vfres->nethctrl, ethqsets);
2524 if (vfres->neq < ethqsets*2)
2525 ethqsets = vfres->neq/2;
2526 if (ethqsets > MAX_ETH_QSETS)
2527 ethqsets = MAX_ETH_QSETS;
2528 adapter->sge.max_ethqsets = ethqsets;
2530 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2531 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2532 " virtual interfaces (too few Queue Sets)\n",
2533 adapter->sge.max_ethqsets, adapter->params.nports);
2534 adapter->params.nports = adapter->sge.max_ethqsets;
2539 * Perform early "adapter" initialization. This is where we discover what
2540 * adapter parameters we're going to be using and initialize basic adapter
2543 static int adap_init0(struct adapter *adapter)
2545 struct sge_params *sge_params = &adapter->params.sge;
2546 struct sge *s = &adapter->sge;
2551 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2552 * 2.6.31 and later we can't call pci_reset_function() in order to
2553 * issue an FLR because of a self- deadlock on the device semaphore.
2554 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2555 * cases where they're needed -- for instance, some versions of KVM
2556 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2557 * use the firmware based reset in order to reset any per function
2560 err = t4vf_fw_reset(adapter);
2562 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2567 * Grab basic operational parameters. These will predominantly have
2568 * been set up by the Physical Function Driver or will be hard coded
2569 * into the adapter. We just have to live with them ... Note that
2570 * we _must_ get our VPD parameters before our SGE parameters because
2571 * we need to know the adapter's core clock from the VPD in order to
2572 * properly decode the SGE Timer Values.
2574 err = t4vf_get_dev_params(adapter);
2576 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2577 " device parameters: err=%d\n", err);
2580 err = t4vf_get_vpd_params(adapter);
2582 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2583 " VPD parameters: err=%d\n", err);
2586 err = t4vf_get_sge_params(adapter);
2588 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2589 " SGE parameters: err=%d\n", err);
2592 err = t4vf_get_rss_glb_config(adapter);
2594 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2595 " RSS parameters: err=%d\n", err);
2598 if (adapter->params.rss.mode !=
2599 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2600 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2601 " mode %d\n", adapter->params.rss.mode);
2604 err = t4vf_sge_init(adapter);
2606 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2611 /* If we're running on newer firmware, let it know that we're
2612 * prepared to deal with encapsulated CPL messages. Older
2613 * firmware won't understand this and we'll just get
2614 * unencapsulated messages ...
2616 param = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2617 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_CPLFW4MSG_ENCAP);
2619 (void) t4vf_set_params(adapter, 1, ¶m, &val);
2622 * Retrieve our RX interrupt holdoff timer values and counter
2623 * threshold values from the SGE parameters.
2625 s->timer_val[0] = core_ticks_to_us(adapter,
2626 TIMERVALUE0_G(sge_params->sge_timer_value_0_and_1));
2627 s->timer_val[1] = core_ticks_to_us(adapter,
2628 TIMERVALUE1_G(sge_params->sge_timer_value_0_and_1));
2629 s->timer_val[2] = core_ticks_to_us(adapter,
2630 TIMERVALUE0_G(sge_params->sge_timer_value_2_and_3));
2631 s->timer_val[3] = core_ticks_to_us(adapter,
2632 TIMERVALUE1_G(sge_params->sge_timer_value_2_and_3));
2633 s->timer_val[4] = core_ticks_to_us(adapter,
2634 TIMERVALUE0_G(sge_params->sge_timer_value_4_and_5));
2635 s->timer_val[5] = core_ticks_to_us(adapter,
2636 TIMERVALUE1_G(sge_params->sge_timer_value_4_and_5));
2638 s->counter_val[0] = THRESHOLD_0_G(sge_params->sge_ingress_rx_threshold);
2639 s->counter_val[1] = THRESHOLD_1_G(sge_params->sge_ingress_rx_threshold);
2640 s->counter_val[2] = THRESHOLD_2_G(sge_params->sge_ingress_rx_threshold);
2641 s->counter_val[3] = THRESHOLD_3_G(sge_params->sge_ingress_rx_threshold);
2644 * Grab our Virtual Interface resource allocation, extract the
2645 * features that we're interested in and do a bit of sanity testing on
2648 err = t4vf_get_vfres(adapter);
2650 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2651 " resources: err=%d\n", err);
2655 /* Check for various parameter sanity issues */
2656 if (adapter->params.vfres.pmask == 0) {
2657 dev_err(adapter->pdev_dev, "no port access configured\n"
2661 if (adapter->params.vfres.nvi == 0) {
2662 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2667 /* Initialize nports and max_ethqsets now that we have our Virtual
2668 * Function Resources.
2670 size_nports_qsets(adapter);
2675 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2676 u8 pkt_cnt_idx, unsigned int size,
2677 unsigned int iqe_size)
2679 rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) |
2680 (pkt_cnt_idx < SGE_NCOUNTERS ?
2681 QINTR_CNT_EN_F : 0));
2682 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2685 rspq->iqe_len = iqe_size;
2690 * Perform default configuration of DMA queues depending on the number and
2691 * type of ports we found and the number of available CPUs. Most settings can
2692 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2693 * being brought up for the first time.
2695 static void cfg_queues(struct adapter *adapter)
2697 struct sge *s = &adapter->sge;
2698 int q10g, n10g, qidx, pidx, qs;
2702 * We should not be called till we know how many Queue Sets we can
2703 * support. In particular, this means that we need to know what kind
2704 * of interrupts we'll be using ...
2706 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2709 * Count the number of 10GbE Virtual Interfaces that we have.
2712 for_each_port(adapter, pidx)
2713 n10g += is_x_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2716 * We default to 1 queue per non-10G port and up to # of cores queues
2722 int n1g = (adapter->params.nports - n10g);
2723 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2724 if (q10g > num_online_cpus())
2725 q10g = num_online_cpus();
2729 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2730 * The layout will be established in setup_sge_queues() when the
2731 * adapter is brough up for the first time.
2734 for_each_port(adapter, pidx) {
2735 struct port_info *pi = adap2pinfo(adapter, pidx);
2737 pi->first_qset = qidx;
2738 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
2744 * The Ingress Queue Entry Size for our various Response Queues needs
2745 * to be big enough to accommodate the largest message we can receive
2746 * from the chip/firmware; which is 64 bytes ...
2751 * Set up default Queue Set parameters ... Start off with the
2752 * shortest interrupt holdoff timer.
2754 for (qs = 0; qs < s->max_ethqsets; qs++) {
2755 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2756 struct sge_eth_txq *txq = &s->ethtxq[qs];
2758 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2764 * The firmware event queue is used for link state changes and
2765 * notifications of TX DMA completions.
2767 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2770 * The forwarded interrupt queue is used when we're in MSI interrupt
2771 * mode. In this mode all interrupts associated with RX queues will
2772 * be forwarded to a single queue which we'll associate with our MSI
2773 * interrupt vector. The messages dropped in the forwarded interrupt
2774 * queue will indicate which ingress queue needs servicing ... This
2775 * queue needs to be large enough to accommodate all of the ingress
2776 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2777 * from equalling the CIDX if every ingress queue has an outstanding
2778 * interrupt). The queue doesn't need to be any larger because no
2779 * ingress queue will ever have more than one outstanding interrupt at
2782 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2787 * Reduce the number of Ethernet queues across all ports to at most n.
2788 * n provides at least one queue per port.
2790 static void reduce_ethqs(struct adapter *adapter, int n)
2793 struct port_info *pi;
2796 * While we have too many active Ether Queue Sets, interate across the
2797 * "ports" and reduce their individual Queue Set allocations.
2799 BUG_ON(n < adapter->params.nports);
2800 while (n < adapter->sge.ethqsets)
2801 for_each_port(adapter, i) {
2802 pi = adap2pinfo(adapter, i);
2803 if (pi->nqsets > 1) {
2805 adapter->sge.ethqsets--;
2806 if (adapter->sge.ethqsets <= n)
2812 * Reassign the starting Queue Sets for each of the "ports" ...
2815 for_each_port(adapter, i) {
2816 pi = adap2pinfo(adapter, i);
2823 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2824 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2825 * need. Minimally we need one for every Virtual Interface plus those needed
2826 * for our "extras". Note that this process may lower the maximum number of
2827 * allowed Queue Sets ...
2829 static int enable_msix(struct adapter *adapter)
2831 int i, want, need, nqsets;
2832 struct msix_entry entries[MSIX_ENTRIES];
2833 struct sge *s = &adapter->sge;
2835 for (i = 0; i < MSIX_ENTRIES; ++i)
2836 entries[i].entry = i;
2839 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2840 * plus those needed for our "extras" (for example, the firmware
2841 * message queue). We _need_ at least one "Queue Set" per Virtual
2842 * Interface plus those needed for our "extras". So now we get to see
2843 * if the song is right ...
2845 want = s->max_ethqsets + MSIX_EXTRAS;
2846 need = adapter->params.nports + MSIX_EXTRAS;
2848 want = pci_enable_msix_range(adapter->pdev, entries, need, want);
2852 nqsets = want - MSIX_EXTRAS;
2853 if (nqsets < s->max_ethqsets) {
2854 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2855 " for %d Queue Sets\n", nqsets);
2856 s->max_ethqsets = nqsets;
2857 if (nqsets < s->ethqsets)
2858 reduce_ethqs(adapter, nqsets);
2860 for (i = 0; i < want; ++i)
2861 adapter->msix_info[i].vec = entries[i].vector;
2866 static const struct net_device_ops cxgb4vf_netdev_ops = {
2867 .ndo_open = cxgb4vf_open,
2868 .ndo_stop = cxgb4vf_stop,
2869 .ndo_start_xmit = t4vf_eth_xmit,
2870 .ndo_get_stats = cxgb4vf_get_stats,
2871 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2872 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2873 .ndo_validate_addr = eth_validate_addr,
2874 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2875 .ndo_change_mtu = cxgb4vf_change_mtu,
2876 .ndo_fix_features = cxgb4vf_fix_features,
2877 .ndo_set_features = cxgb4vf_set_features,
2878 #ifdef CONFIG_NET_POLL_CONTROLLER
2879 .ndo_poll_controller = cxgb4vf_poll_controller,
2884 * "Probe" a device: initialize a device and construct all kernel and driver
2885 * state needed to manage the device. This routine is called "init_one" in
2888 static int cxgb4vf_pci_probe(struct pci_dev *pdev,
2889 const struct pci_device_id *ent)
2894 struct adapter *adapter;
2895 struct port_info *pi;
2896 struct net_device *netdev;
2900 * Print our driver banner the first time we're called to initialize a
2903 pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION);
2906 * Initialize generic PCI device state.
2908 err = pci_enable_device(pdev);
2910 dev_err(&pdev->dev, "cannot enable PCI device\n");
2915 * Reserve PCI resources for the device. If we can't get them some
2916 * other driver may have already claimed the device ...
2918 err = pci_request_regions(pdev, KBUILD_MODNAME);
2920 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2921 goto err_disable_device;
2925 * Set up our DMA mask: try for 64-bit address masking first and
2926 * fall back to 32-bit if we can't get 64 bits ...
2928 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2930 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2932 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2933 " coherent allocations\n");
2934 goto err_release_regions;
2938 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2940 dev_err(&pdev->dev, "no usable DMA configuration\n");
2941 goto err_release_regions;
2947 * Enable bus mastering for the device ...
2949 pci_set_master(pdev);
2952 * Allocate our adapter data structure and attach it to the device.
2954 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2957 goto err_release_regions;
2959 pci_set_drvdata(pdev, adapter);
2960 adapter->pdev = pdev;
2961 adapter->pdev_dev = &pdev->dev;
2963 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
2964 (sizeof(struct mbox_cmd) *
2965 T4VF_OS_LOG_MBOX_CMDS),
2967 if (!adapter->mbox_log) {
2969 goto err_free_adapter;
2971 adapter->mbox_log->size = T4VF_OS_LOG_MBOX_CMDS;
2974 * Initialize SMP data synchronization resources.
2976 spin_lock_init(&adapter->stats_lock);
2977 spin_lock_init(&adapter->mbox_lock);
2978 INIT_LIST_HEAD(&adapter->mlist.list);
2981 * Map our I/O registers in BAR0.
2983 adapter->regs = pci_ioremap_bar(pdev, 0);
2984 if (!adapter->regs) {
2985 dev_err(&pdev->dev, "cannot map device registers\n");
2987 goto err_free_adapter;
2990 /* Wait for the device to become ready before proceeding ...
2992 err = t4vf_prep_adapter(adapter);
2994 dev_err(adapter->pdev_dev, "device didn't become ready:"
2996 goto err_unmap_bar0;
2999 /* For T5 and later we want to use the new BAR-based User Doorbells,
3000 * so we need to map BAR2 here ...
3002 if (!is_t4(adapter->params.chip)) {
3003 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
3004 pci_resource_len(pdev, 2));
3005 if (!adapter->bar2) {
3006 dev_err(adapter->pdev_dev, "cannot map BAR2 doorbells\n");
3008 goto err_unmap_bar0;
3012 * Initialize adapter level features.
3014 adapter->name = pci_name(pdev);
3015 adapter->msg_enable = DFLT_MSG_ENABLE;
3017 /* If possible, we use PCIe Relaxed Ordering Attribute to deliver
3018 * Ingress Packet Data to Free List Buffers in order to allow for
3019 * chipset performance optimizations between the Root Complex and
3020 * Memory Controllers. (Messages to the associated Ingress Queue
3021 * notifying new Packet Placement in the Free Lists Buffers will be
3022 * send without the Relaxed Ordering Attribute thus guaranteeing that
3023 * all preceding PCIe Transaction Layer Packets will be processed
3024 * first.) But some Root Complexes have various issues with Upstream
3025 * Transaction Layer Packets with the Relaxed Ordering Attribute set.
3026 * The PCIe devices which under the Root Complexes will be cleared the
3027 * Relaxed Ordering bit in the configuration space, So we check our
3028 * PCIe configuration space to see if it's flagged with advice against
3029 * using Relaxed Ordering.
3031 if (!pcie_relaxed_ordering_enabled(pdev))
3032 adapter->flags |= ROOT_NO_RELAXED_ORDERING;
3034 err = adap_init0(adapter);
3038 /* Initialize hash mac addr list */
3039 INIT_LIST_HEAD(&adapter->mac_hlist);
3042 * Allocate our "adapter ports" and stitch everything together.
3044 pmask = adapter->params.vfres.pmask;
3045 pf = t4vf_get_pf_from_vf(adapter);
3046 for_each_port(adapter, pidx) {
3049 unsigned int naddr = 1;
3052 * We simplistically allocate our virtual interfaces
3053 * sequentially across the port numbers to which we have
3054 * access rights. This should be configurable in some manner
3059 port_id = ffs(pmask) - 1;
3060 pmask &= ~(1 << port_id);
3061 viid = t4vf_alloc_vi(adapter, port_id);
3063 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
3064 " err=%d\n", port_id, viid);
3070 * Allocate our network device and stitch things together.
3072 netdev = alloc_etherdev_mq(sizeof(struct port_info),
3074 if (netdev == NULL) {
3075 t4vf_free_vi(adapter, viid);
3079 adapter->port[pidx] = netdev;
3080 SET_NETDEV_DEV(netdev, &pdev->dev);
3081 pi = netdev_priv(netdev);
3082 pi->adapter = adapter;
3084 pi->port_id = port_id;
3088 * Initialize the starting state of our "port" and register
3091 pi->xact_addr_filt = -1;
3092 netif_carrier_off(netdev);
3093 netdev->irq = pdev->irq;
3095 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
3096 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3097 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXCSUM;
3098 netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
3099 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3101 netdev->features = netdev->hw_features |
3102 NETIF_F_HW_VLAN_CTAG_TX;
3104 netdev->features |= NETIF_F_HIGHDMA;
3106 netdev->priv_flags |= IFF_UNICAST_FLT;
3107 netdev->min_mtu = 81;
3108 netdev->max_mtu = ETH_MAX_MTU;
3110 netdev->netdev_ops = &cxgb4vf_netdev_ops;
3111 netdev->ethtool_ops = &cxgb4vf_ethtool_ops;
3112 netdev->dev_port = pi->port_id;
3115 * Initialize the hardware/software state for the port.
3117 err = t4vf_port_init(adapter, pidx);
3119 dev_err(&pdev->dev, "cannot initialize port %d\n",
3124 err = t4vf_get_vf_mac_acl(adapter, pf, &naddr, mac);
3127 "unable to determine MAC ACL address, "
3128 "continuing anyway.. (status %d)\n", err);
3129 } else if (naddr && adapter->params.vfres.nvi == 1) {
3130 struct sockaddr addr;
3132 ether_addr_copy(addr.sa_data, mac);
3133 err = cxgb4vf_set_mac_addr(netdev, &addr);
3136 "unable to set MAC address %pM\n",
3140 dev_info(&pdev->dev,
3141 "Using assigned MAC ACL: %pM\n", mac);
3145 /* See what interrupts we'll be using. If we've been configured to
3146 * use MSI-X interrupts, try to enable them but fall back to using
3147 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
3148 * get MSI interrupts we bail with the error.
3150 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
3151 adapter->flags |= USING_MSIX;
3153 if (msi == MSI_MSIX) {
3154 dev_info(adapter->pdev_dev,
3155 "Unable to use MSI-X Interrupts; falling "
3156 "back to MSI Interrupts\n");
3158 /* We're going to need a Forwarded Interrupt Queue so
3159 * that may cut into how many Queue Sets we can
3163 size_nports_qsets(adapter);
3165 err = pci_enable_msi(pdev);
3167 dev_err(&pdev->dev, "Unable to allocate MSI Interrupts;"
3171 adapter->flags |= USING_MSI;
3174 /* Now that we know how many "ports" we have and what interrupt
3175 * mechanism we're going to use, we can configure our queue resources.
3177 cfg_queues(adapter);
3180 * The "card" is now ready to go. If any errors occur during device
3181 * registration we do not fail the whole "card" but rather proceed
3182 * only with the ports we manage to register successfully. However we
3183 * must register at least one net device.
3185 for_each_port(adapter, pidx) {
3186 struct port_info *pi = netdev_priv(adapter->port[pidx]);
3187 netdev = adapter->port[pidx];
3191 netif_set_real_num_tx_queues(netdev, pi->nqsets);
3192 netif_set_real_num_rx_queues(netdev, pi->nqsets);
3194 err = register_netdev(netdev);
3196 dev_warn(&pdev->dev, "cannot register net device %s,"
3197 " skipping\n", netdev->name);
3201 set_bit(pidx, &adapter->registered_device_map);
3203 if (adapter->registered_device_map == 0) {
3204 dev_err(&pdev->dev, "could not register any net devices\n");
3205 goto err_disable_interrupts;
3209 * Set up our debugfs entries.
3211 if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) {
3212 adapter->debugfs_root =
3213 debugfs_create_dir(pci_name(pdev),
3214 cxgb4vf_debugfs_root);
3215 if (IS_ERR_OR_NULL(adapter->debugfs_root))
3216 dev_warn(&pdev->dev, "could not create debugfs"
3219 setup_debugfs(adapter);
3223 * Print a short notice on the existence and configuration of the new
3224 * VF network device ...
3226 for_each_port(adapter, pidx) {
3227 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
3228 adapter->port[pidx]->name,
3229 (adapter->flags & USING_MSIX) ? "MSI-X" :
3230 (adapter->flags & USING_MSI) ? "MSI" : "");
3239 * Error recovery and exit code. Unwind state that's been created
3240 * so far and return the error.
3242 err_disable_interrupts:
3243 if (adapter->flags & USING_MSIX) {
3244 pci_disable_msix(adapter->pdev);
3245 adapter->flags &= ~USING_MSIX;
3246 } else if (adapter->flags & USING_MSI) {
3247 pci_disable_msi(adapter->pdev);
3248 adapter->flags &= ~USING_MSI;
3252 for_each_port(adapter, pidx) {
3253 netdev = adapter->port[pidx];
3256 pi = netdev_priv(netdev);
3257 t4vf_free_vi(adapter, pi->viid);
3258 if (test_bit(pidx, &adapter->registered_device_map))
3259 unregister_netdev(netdev);
3260 free_netdev(netdev);
3264 if (!is_t4(adapter->params.chip))
3265 iounmap(adapter->bar2);
3268 iounmap(adapter->regs);
3271 kfree(adapter->mbox_log);
3274 err_release_regions:
3275 pci_release_regions(pdev);
3276 pci_clear_master(pdev);
3279 pci_disable_device(pdev);
3285 * "Remove" a device: tear down all kernel and driver state created in the
3286 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
3287 * that this is called "remove_one" in the PF Driver.)
3289 static void cxgb4vf_pci_remove(struct pci_dev *pdev)
3291 struct adapter *adapter = pci_get_drvdata(pdev);
3294 * Tear down driver state associated with device.
3300 * Stop all of our activity. Unregister network port,
3301 * disable interrupts, etc.
3303 for_each_port(adapter, pidx)
3304 if (test_bit(pidx, &adapter->registered_device_map))
3305 unregister_netdev(adapter->port[pidx]);
3306 t4vf_sge_stop(adapter);
3307 if (adapter->flags & USING_MSIX) {
3308 pci_disable_msix(adapter->pdev);
3309 adapter->flags &= ~USING_MSIX;
3310 } else if (adapter->flags & USING_MSI) {
3311 pci_disable_msi(adapter->pdev);
3312 adapter->flags &= ~USING_MSI;
3316 * Tear down our debugfs entries.
3318 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
3319 cleanup_debugfs(adapter);
3320 debugfs_remove_recursive(adapter->debugfs_root);
3324 * Free all of the various resources which we've acquired ...
3326 t4vf_free_sge_resources(adapter);
3327 for_each_port(adapter, pidx) {
3328 struct net_device *netdev = adapter->port[pidx];
3329 struct port_info *pi;
3334 pi = netdev_priv(netdev);
3335 t4vf_free_vi(adapter, pi->viid);
3336 free_netdev(netdev);
3338 iounmap(adapter->regs);
3339 if (!is_t4(adapter->params.chip))
3340 iounmap(adapter->bar2);
3341 kfree(adapter->mbox_log);
3346 * Disable the device and release its PCI resources.
3348 pci_disable_device(pdev);
3349 pci_clear_master(pdev);
3350 pci_release_regions(pdev);
3354 * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt
3357 static void cxgb4vf_pci_shutdown(struct pci_dev *pdev)
3359 struct adapter *adapter;
3362 adapter = pci_get_drvdata(pdev);
3366 /* Disable all Virtual Interfaces. This will shut down the
3367 * delivery of all ingress packets into the chip for these
3368 * Virtual Interfaces.
3370 for_each_port(adapter, pidx)
3371 if (test_bit(pidx, &adapter->registered_device_map))
3372 unregister_netdev(adapter->port[pidx]);
3374 /* Free up all Queues which will prevent further DMA and
3375 * Interrupts allowing various internal pathways to drain.
3377 t4vf_sge_stop(adapter);
3378 if (adapter->flags & USING_MSIX) {
3379 pci_disable_msix(adapter->pdev);
3380 adapter->flags &= ~USING_MSIX;
3381 } else if (adapter->flags & USING_MSI) {
3382 pci_disable_msi(adapter->pdev);
3383 adapter->flags &= ~USING_MSI;
3387 * Free up all Queues which will prevent further DMA and
3388 * Interrupts allowing various internal pathways to drain.
3390 t4vf_free_sge_resources(adapter);
3391 pci_set_drvdata(pdev, NULL);
3394 /* Macros needed to support the PCI Device ID Table ...
3396 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
3397 static const struct pci_device_id cxgb4vf_pci_tbl[] = {
3398 #define CH_PCI_DEVICE_ID_FUNCTION 0x8
3400 #define CH_PCI_ID_TABLE_ENTRY(devid) \
3401 { PCI_VDEVICE(CHELSIO, (devid)), 0 }
3403 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END { 0, } }
3405 #include "../cxgb4/t4_pci_id_tbl.h"
3407 MODULE_DESCRIPTION(DRV_DESC);
3408 MODULE_AUTHOR("Chelsio Communications");
3409 MODULE_LICENSE("Dual BSD/GPL");
3410 MODULE_VERSION(DRV_VERSION);
3411 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
3413 static struct pci_driver cxgb4vf_driver = {
3414 .name = KBUILD_MODNAME,
3415 .id_table = cxgb4vf_pci_tbl,
3416 .probe = cxgb4vf_pci_probe,
3417 .remove = cxgb4vf_pci_remove,
3418 .shutdown = cxgb4vf_pci_shutdown,
3422 * Initialize global driver state.
3424 static int __init cxgb4vf_module_init(void)
3429 * Vet our module parameters.
3431 if (msi != MSI_MSIX && msi != MSI_MSI) {
3432 pr_warn("bad module parameter msi=%d; must be %d (MSI-X or MSI) or %d (MSI)\n",
3433 msi, MSI_MSIX, MSI_MSI);
3437 /* Debugfs support is optional, just warn if this fails */
3438 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3439 if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3440 pr_warn("could not create debugfs entry, continuing\n");
3442 ret = pci_register_driver(&cxgb4vf_driver);
3443 if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3444 debugfs_remove(cxgb4vf_debugfs_root);
3449 * Tear down global driver state.
3451 static void __exit cxgb4vf_module_exit(void)
3453 pci_unregister_driver(&cxgb4vf_driver);
3454 debugfs_remove(cxgb4vf_debugfs_root);
3457 module_init(cxgb4vf_module_init);
3458 module_exit(cxgb4vf_module_exit);