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 ((int)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);
718 adapter->flags |= FULL_INIT_DONE;
722 * Acquire our interrupt resources. We only support MSI-X and MSI.
724 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
725 if (adapter->flags & USING_MSIX)
726 err = request_msix_queue_irqs(adapter);
728 err = request_irq(adapter->pdev->irq,
729 t4vf_intr_handler(adapter), 0,
730 adapter->name, adapter);
732 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
738 * Enable NAPI ingress processing and return success.
741 t4vf_sge_start(adapter);
747 * Bring the adapter down. Called whenever the last "port" (Virtual
748 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
751 static void adapter_down(struct adapter *adapter)
754 * Free interrupt resources.
756 if (adapter->flags & USING_MSIX)
757 free_msix_queue_irqs(adapter);
759 free_irq(adapter->pdev->irq, adapter);
762 * Wait for NAPI handlers to finish.
768 * Start up a net device.
770 static int cxgb4vf_open(struct net_device *dev)
773 struct port_info *pi = netdev_priv(dev);
774 struct adapter *adapter = pi->adapter;
777 * If this is the first interface that we're opening on the "adapter",
778 * bring the "adapter" up now.
780 if (adapter->open_device_map == 0) {
781 err = adapter_up(adapter);
787 * Note that this interface is up and start everything up ...
789 err = link_start(dev);
793 netif_tx_start_all_queues(dev);
794 set_bit(pi->port_id, &adapter->open_device_map);
798 if (adapter->open_device_map == 0)
799 adapter_down(adapter);
804 * Shut down a net device. This routine is called "cxgb_close" in the PF
807 static int cxgb4vf_stop(struct net_device *dev)
809 struct port_info *pi = netdev_priv(dev);
810 struct adapter *adapter = pi->adapter;
812 netif_tx_stop_all_queues(dev);
813 netif_carrier_off(dev);
814 t4vf_enable_vi(adapter, pi->viid, false, false);
815 pi->link_cfg.link_ok = 0;
817 clear_bit(pi->port_id, &adapter->open_device_map);
818 if (adapter->open_device_map == 0)
819 adapter_down(adapter);
824 * Translate our basic statistics into the standard "ifconfig" statistics.
826 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
828 struct t4vf_port_stats stats;
829 struct port_info *pi = netdev2pinfo(dev);
830 struct adapter *adapter = pi->adapter;
831 struct net_device_stats *ns = &dev->stats;
834 spin_lock(&adapter->stats_lock);
835 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
836 spin_unlock(&adapter->stats_lock);
838 memset(ns, 0, sizeof(*ns));
842 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
843 stats.tx_ucast_bytes + stats.tx_offload_bytes);
844 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
845 stats.tx_ucast_frames + stats.tx_offload_frames);
846 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
847 stats.rx_ucast_bytes);
848 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
849 stats.rx_ucast_frames);
850 ns->multicast = stats.rx_mcast_frames;
851 ns->tx_errors = stats.tx_drop_frames;
852 ns->rx_errors = stats.rx_err_frames;
857 static inline int cxgb4vf_set_addr_hash(struct port_info *pi)
859 struct adapter *adapter = pi->adapter;
862 struct hash_mac_addr *entry;
864 /* Calculate the hash vector for the updated list and program it */
865 list_for_each_entry(entry, &adapter->mac_hlist, list) {
866 ucast |= is_unicast_ether_addr(entry->addr);
867 vec |= (1ULL << hash_mac_addr(entry->addr));
869 return t4vf_set_addr_hash(adapter, pi->viid, ucast, vec, false);
872 static int cxgb4vf_mac_sync(struct net_device *netdev, const u8 *mac_addr)
874 struct port_info *pi = netdev_priv(netdev);
875 struct adapter *adapter = pi->adapter;
880 bool ucast = is_unicast_ether_addr(mac_addr);
881 const u8 *maclist[1] = {mac_addr};
882 struct hash_mac_addr *new_entry;
884 ret = t4vf_alloc_mac_filt(adapter, pi->viid, free, 1, maclist,
885 NULL, ucast ? &uhash : &mhash, false);
888 /* if hash != 0, then add the addr to hash addr list
889 * so on the end we will calculate the hash for the
890 * list and program it
892 if (uhash || mhash) {
893 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
896 ether_addr_copy(new_entry->addr, mac_addr);
897 list_add_tail(&new_entry->list, &adapter->mac_hlist);
898 ret = cxgb4vf_set_addr_hash(pi);
901 return ret < 0 ? ret : 0;
904 static int cxgb4vf_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
906 struct port_info *pi = netdev_priv(netdev);
907 struct adapter *adapter = pi->adapter;
909 const u8 *maclist[1] = {mac_addr};
910 struct hash_mac_addr *entry, *tmp;
912 /* If the MAC address to be removed is in the hash addr
913 * list, delete it from the list and update hash vector
915 list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist, list) {
916 if (ether_addr_equal(entry->addr, mac_addr)) {
917 list_del(&entry->list);
919 return cxgb4vf_set_addr_hash(pi);
923 ret = t4vf_free_mac_filt(adapter, pi->viid, 1, maclist, false);
924 return ret < 0 ? -EINVAL : 0;
928 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
929 * If @mtu is -1 it is left unchanged.
931 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
933 struct port_info *pi = netdev_priv(dev);
935 __dev_uc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync);
936 __dev_mc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync);
937 return t4vf_set_rxmode(pi->adapter, pi->viid, -1,
938 (dev->flags & IFF_PROMISC) != 0,
939 (dev->flags & IFF_ALLMULTI) != 0,
944 * Set the current receive modes on the device.
946 static void cxgb4vf_set_rxmode(struct net_device *dev)
948 /* unfortunately we can't return errors to the stack */
949 set_rxmode(dev, -1, false);
953 * Find the entry in the interrupt holdoff timer value array which comes
954 * closest to the specified interrupt holdoff value.
956 static int closest_timer(const struct sge *s, int us)
958 int i, timer_idx = 0, min_delta = INT_MAX;
960 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
961 int delta = us - s->timer_val[i];
964 if (delta < min_delta) {
972 static int closest_thres(const struct sge *s, int thres)
974 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
976 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
977 delta = thres - s->counter_val[i];
980 if (delta < min_delta) {
989 * Return a queue's interrupt hold-off time in us. 0 means no timer.
991 static unsigned int qtimer_val(const struct adapter *adapter,
992 const struct sge_rspq *rspq)
994 unsigned int timer_idx = QINTR_TIMER_IDX_G(rspq->intr_params);
996 return timer_idx < SGE_NTIMERS
997 ? adapter->sge.timer_val[timer_idx]
1002 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1003 * @adapter: the adapter
1004 * @rspq: the RX response queue
1005 * @us: the hold-off time in us, or 0 to disable timer
1006 * @cnt: the hold-off packet count, or 0 to disable counter
1008 * Sets an RX response queue's interrupt hold-off time and packet count.
1009 * At least one of the two needs to be enabled for the queue to generate
1012 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
1013 unsigned int us, unsigned int cnt)
1015 unsigned int timer_idx;
1018 * If both the interrupt holdoff timer and count are specified as
1019 * zero, default to a holdoff count of 1 ...
1021 if ((us | cnt) == 0)
1025 * If an interrupt holdoff count has been specified, then find the
1026 * closest configured holdoff count and use that. If the response
1027 * queue has already been created, then update its queue context
1034 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1035 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1036 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1037 FW_PARAMS_PARAM_X_V(
1038 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1039 FW_PARAMS_PARAM_YZ_V(rspq->cntxt_id);
1040 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1044 rspq->pktcnt_idx = pktcnt_idx;
1048 * Compute the closest holdoff timer index from the supplied holdoff
1051 timer_idx = (us == 0
1052 ? SGE_TIMER_RSTRT_CNTR
1053 : closest_timer(&adapter->sge, us));
1056 * Update the response queue's interrupt coalescing parameters and
1059 rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) |
1060 QINTR_CNT_EN_V(cnt > 0));
1065 * Return a version number to identify the type of adapter. The scheme is:
1066 * - bits 0..9: chip version
1067 * - bits 10..15: chip revision
1069 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1072 * Chip version 4, revision 0x3f (cxgb4vf).
1074 return CHELSIO_CHIP_VERSION(adapter->params.chip) | (0x3f << 10);
1078 * Execute the specified ioctl command.
1080 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1086 * The VF Driver doesn't have access to any of the other
1087 * common Ethernet device ioctl()'s (like reading/writing
1088 * PHY registers, etc.
1099 * Change the device's MTU.
1101 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1104 struct port_info *pi = netdev_priv(dev);
1106 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1107 -1, -1, -1, -1, true);
1113 static netdev_features_t cxgb4vf_fix_features(struct net_device *dev,
1114 netdev_features_t features)
1117 * Since there is no support for separate rx/tx vlan accel
1118 * enable/disable make sure tx flag is always in same state as rx.
1120 if (features & NETIF_F_HW_VLAN_CTAG_RX)
1121 features |= NETIF_F_HW_VLAN_CTAG_TX;
1123 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
1128 static int cxgb4vf_set_features(struct net_device *dev,
1129 netdev_features_t features)
1131 struct port_info *pi = netdev_priv(dev);
1132 netdev_features_t changed = dev->features ^ features;
1134 if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1135 t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1,
1136 features & NETIF_F_HW_VLAN_CTAG_TX, 0);
1142 * Change the devices MAC address.
1144 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1147 struct sockaddr *addr = _addr;
1148 struct port_info *pi = netdev_priv(dev);
1150 if (!is_valid_ether_addr(addr->sa_data))
1151 return -EADDRNOTAVAIL;
1153 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1154 addr->sa_data, true);
1158 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1159 pi->xact_addr_filt = ret;
1163 #ifdef CONFIG_NET_POLL_CONTROLLER
1165 * Poll all of our receive queues. This is called outside of normal interrupt
1168 static void cxgb4vf_poll_controller(struct net_device *dev)
1170 struct port_info *pi = netdev_priv(dev);
1171 struct adapter *adapter = pi->adapter;
1173 if (adapter->flags & USING_MSIX) {
1174 struct sge_eth_rxq *rxq;
1177 rxq = &adapter->sge.ethrxq[pi->first_qset];
1178 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1179 t4vf_sge_intr_msix(0, &rxq->rspq);
1183 t4vf_intr_handler(adapter)(0, adapter);
1188 * Ethtool operations.
1189 * ===================
1191 * Note that we don't support any ethtool operations which change the physical
1192 * state of the port to which we're linked.
1196 * from_fw_port_mod_type - translate Firmware Port/Module type to Ethtool
1197 * @port_type: Firmware Port Type
1198 * @mod_type: Firmware Module Type
1200 * Translate Firmware Port/Module type to Ethtool Port Type.
1202 static int from_fw_port_mod_type(enum fw_port_type port_type,
1203 enum fw_port_module_type mod_type)
1205 if (port_type == FW_PORT_TYPE_BT_SGMII ||
1206 port_type == FW_PORT_TYPE_BT_XFI ||
1207 port_type == FW_PORT_TYPE_BT_XAUI) {
1209 } else if (port_type == FW_PORT_TYPE_FIBER_XFI ||
1210 port_type == FW_PORT_TYPE_FIBER_XAUI) {
1212 } else if (port_type == FW_PORT_TYPE_SFP ||
1213 port_type == FW_PORT_TYPE_QSFP_10G ||
1214 port_type == FW_PORT_TYPE_QSA ||
1215 port_type == FW_PORT_TYPE_QSFP ||
1216 port_type == FW_PORT_TYPE_CR4_QSFP ||
1217 port_type == FW_PORT_TYPE_CR_QSFP ||
1218 port_type == FW_PORT_TYPE_CR2_QSFP ||
1219 port_type == FW_PORT_TYPE_SFP28) {
1220 if (mod_type == FW_PORT_MOD_TYPE_LR ||
1221 mod_type == FW_PORT_MOD_TYPE_SR ||
1222 mod_type == FW_PORT_MOD_TYPE_ER ||
1223 mod_type == FW_PORT_MOD_TYPE_LRM)
1225 else if (mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
1226 mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
1230 } else if (port_type == FW_PORT_TYPE_KR4_100G ||
1231 port_type == FW_PORT_TYPE_KR_SFP28) {
1239 * fw_caps_to_lmm - translate Firmware to ethtool Link Mode Mask
1240 * @port_type: Firmware Port Type
1241 * @fw_caps: Firmware Port Capabilities
1242 * @link_mode_mask: ethtool Link Mode Mask
1244 * Translate a Firmware Port Capabilities specification to an ethtool
1247 static void fw_caps_to_lmm(enum fw_port_type port_type,
1248 unsigned int fw_caps,
1249 unsigned long *link_mode_mask)
1251 #define SET_LMM(__lmm_name) \
1252 __set_bit(ETHTOOL_LINK_MODE_ ## __lmm_name ## _BIT, \
1255 #define FW_CAPS_TO_LMM(__fw_name, __lmm_name) \
1257 if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1258 SET_LMM(__lmm_name); \
1261 switch (port_type) {
1262 case FW_PORT_TYPE_BT_SGMII:
1263 case FW_PORT_TYPE_BT_XFI:
1264 case FW_PORT_TYPE_BT_XAUI:
1266 FW_CAPS_TO_LMM(SPEED_100M, 100baseT_Full);
1267 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1268 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1271 case FW_PORT_TYPE_KX4:
1272 case FW_PORT_TYPE_KX:
1274 FW_CAPS_TO_LMM(SPEED_1G, 1000baseKX_Full);
1275 FW_CAPS_TO_LMM(SPEED_10G, 10000baseKX4_Full);
1278 case FW_PORT_TYPE_KR:
1280 SET_LMM(10000baseKR_Full);
1283 case FW_PORT_TYPE_BP_AP:
1285 SET_LMM(10000baseR_FEC);
1286 SET_LMM(10000baseKR_Full);
1287 SET_LMM(1000baseKX_Full);
1290 case FW_PORT_TYPE_BP4_AP:
1292 SET_LMM(10000baseR_FEC);
1293 SET_LMM(10000baseKR_Full);
1294 SET_LMM(1000baseKX_Full);
1295 SET_LMM(10000baseKX4_Full);
1298 case FW_PORT_TYPE_FIBER_XFI:
1299 case FW_PORT_TYPE_FIBER_XAUI:
1300 case FW_PORT_TYPE_SFP:
1301 case FW_PORT_TYPE_QSFP_10G:
1302 case FW_PORT_TYPE_QSA:
1304 FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full);
1305 FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full);
1308 case FW_PORT_TYPE_BP40_BA:
1309 case FW_PORT_TYPE_QSFP:
1311 SET_LMM(40000baseSR4_Full);
1314 case FW_PORT_TYPE_CR_QSFP:
1315 case FW_PORT_TYPE_SFP28:
1317 SET_LMM(25000baseCR_Full);
1320 case FW_PORT_TYPE_KR_SFP28:
1322 SET_LMM(25000baseKR_Full);
1325 case FW_PORT_TYPE_CR2_QSFP:
1327 SET_LMM(50000baseSR2_Full);
1330 case FW_PORT_TYPE_KR4_100G:
1331 case FW_PORT_TYPE_CR4_QSFP:
1333 SET_LMM(100000baseCR4_Full);
1340 FW_CAPS_TO_LMM(ANEG, Autoneg);
1341 FW_CAPS_TO_LMM(802_3_PAUSE, Pause);
1342 FW_CAPS_TO_LMM(802_3_ASM_DIR, Asym_Pause);
1344 #undef FW_CAPS_TO_LMM
1348 static int cxgb4vf_get_link_ksettings(struct net_device *dev,
1349 struct ethtool_link_ksettings *link_ksettings)
1351 struct port_info *pi = netdev_priv(dev);
1352 struct ethtool_link_settings *base = &link_ksettings->base;
1354 /* For the nonce, the Firmware doesn't send up Port State changes
1355 * when the Virtual Interface attached to the Port is down. So
1356 * if it's down, let's grab any changes.
1358 if (!netif_running(dev))
1359 (void)t4vf_update_port_info(pi);
1361 ethtool_link_ksettings_zero_link_mode(link_ksettings, supported);
1362 ethtool_link_ksettings_zero_link_mode(link_ksettings, advertising);
1363 ethtool_link_ksettings_zero_link_mode(link_ksettings, lp_advertising);
1365 base->port = from_fw_port_mod_type(pi->port_type, pi->mod_type);
1367 if (pi->mdio_addr >= 0) {
1368 base->phy_address = pi->mdio_addr;
1369 base->mdio_support = (pi->port_type == FW_PORT_TYPE_BT_SGMII
1370 ? ETH_MDIO_SUPPORTS_C22
1371 : ETH_MDIO_SUPPORTS_C45);
1373 base->phy_address = 255;
1374 base->mdio_support = 0;
1377 fw_caps_to_lmm(pi->port_type, pi->link_cfg.pcaps,
1378 link_ksettings->link_modes.supported);
1379 fw_caps_to_lmm(pi->port_type, pi->link_cfg.acaps,
1380 link_ksettings->link_modes.advertising);
1381 fw_caps_to_lmm(pi->port_type, pi->link_cfg.lpacaps,
1382 link_ksettings->link_modes.lp_advertising);
1384 if (netif_carrier_ok(dev)) {
1385 base->speed = pi->link_cfg.speed;
1386 base->duplex = DUPLEX_FULL;
1388 base->speed = SPEED_UNKNOWN;
1389 base->duplex = DUPLEX_UNKNOWN;
1392 base->autoneg = pi->link_cfg.autoneg;
1393 if (pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG)
1394 ethtool_link_ksettings_add_link_mode(link_ksettings,
1395 supported, Autoneg);
1396 if (pi->link_cfg.autoneg)
1397 ethtool_link_ksettings_add_link_mode(link_ksettings,
1398 advertising, Autoneg);
1404 * Return our driver information.
1406 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1407 struct ethtool_drvinfo *drvinfo)
1409 struct adapter *adapter = netdev2adap(dev);
1411 strlcpy(drvinfo->driver, KBUILD_MODNAME, sizeof(drvinfo->driver));
1412 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
1413 strlcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)),
1414 sizeof(drvinfo->bus_info));
1415 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1416 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1417 FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.fwrev),
1418 FW_HDR_FW_VER_MINOR_G(adapter->params.dev.fwrev),
1419 FW_HDR_FW_VER_MICRO_G(adapter->params.dev.fwrev),
1420 FW_HDR_FW_VER_BUILD_G(adapter->params.dev.fwrev),
1421 FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.tprev),
1422 FW_HDR_FW_VER_MINOR_G(adapter->params.dev.tprev),
1423 FW_HDR_FW_VER_MICRO_G(adapter->params.dev.tprev),
1424 FW_HDR_FW_VER_BUILD_G(adapter->params.dev.tprev));
1428 * Return current adapter message level.
1430 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1432 return netdev2adap(dev)->msg_enable;
1436 * Set current adapter message level.
1438 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1440 netdev2adap(dev)->msg_enable = msglevel;
1444 * Return the device's current Queue Set ring size parameters along with the
1445 * allowed maximum values. Since ethtool doesn't understand the concept of
1446 * multi-queue devices, we just return the current values associated with the
1449 static void cxgb4vf_get_ringparam(struct net_device *dev,
1450 struct ethtool_ringparam *rp)
1452 const struct port_info *pi = netdev_priv(dev);
1453 const struct sge *s = &pi->adapter->sge;
1455 rp->rx_max_pending = MAX_RX_BUFFERS;
1456 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1457 rp->rx_jumbo_max_pending = 0;
1458 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1460 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1461 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1462 rp->rx_jumbo_pending = 0;
1463 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1467 * Set the Queue Set ring size parameters for the device. Again, since
1468 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1469 * apply these new values across all of the Queue Sets associated with the
1470 * device -- after vetting them of course!
1472 static int cxgb4vf_set_ringparam(struct net_device *dev,
1473 struct ethtool_ringparam *rp)
1475 const struct port_info *pi = netdev_priv(dev);
1476 struct adapter *adapter = pi->adapter;
1477 struct sge *s = &adapter->sge;
1480 if (rp->rx_pending > MAX_RX_BUFFERS ||
1481 rp->rx_jumbo_pending ||
1482 rp->tx_pending > MAX_TXQ_ENTRIES ||
1483 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1484 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1485 rp->rx_pending < MIN_FL_ENTRIES ||
1486 rp->tx_pending < MIN_TXQ_ENTRIES)
1489 if (adapter->flags & FULL_INIT_DONE)
1492 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1493 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1494 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1495 s->ethtxq[qs].q.size = rp->tx_pending;
1501 * Return the interrupt holdoff timer and count for the first Queue Set on the
1502 * device. Our extension ioctl() (the cxgbtool interface) allows the
1503 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1505 static int cxgb4vf_get_coalesce(struct net_device *dev,
1506 struct ethtool_coalesce *coalesce)
1508 const struct port_info *pi = netdev_priv(dev);
1509 const struct adapter *adapter = pi->adapter;
1510 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1512 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1513 coalesce->rx_max_coalesced_frames =
1514 ((rspq->intr_params & QINTR_CNT_EN_F)
1515 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1521 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1522 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1523 * the interrupt holdoff timer on any of the device's Queue Sets.
1525 static int cxgb4vf_set_coalesce(struct net_device *dev,
1526 struct ethtool_coalesce *coalesce)
1528 const struct port_info *pi = netdev_priv(dev);
1529 struct adapter *adapter = pi->adapter;
1531 return set_rxq_intr_params(adapter,
1532 &adapter->sge.ethrxq[pi->first_qset].rspq,
1533 coalesce->rx_coalesce_usecs,
1534 coalesce->rx_max_coalesced_frames);
1538 * Report current port link pause parameter settings.
1540 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1541 struct ethtool_pauseparam *pauseparam)
1543 struct port_info *pi = netdev_priv(dev);
1545 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1546 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1547 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1551 * Identify the port by blinking the port's LED.
1553 static int cxgb4vf_phys_id(struct net_device *dev,
1554 enum ethtool_phys_id_state state)
1557 struct port_info *pi = netdev_priv(dev);
1559 if (state == ETHTOOL_ID_ACTIVE)
1561 else if (state == ETHTOOL_ID_INACTIVE)
1566 return t4vf_identify_port(pi->adapter, pi->viid, val);
1570 * Port stats maintained per queue of the port.
1572 struct queue_port_stats {
1583 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1584 * these need to match the order of statistics returned by
1585 * t4vf_get_port_stats().
1587 static const char stats_strings[][ETH_GSTRING_LEN] = {
1589 * These must match the layout of the t4vf_port_stats structure.
1591 "TxBroadcastBytes ",
1592 "TxBroadcastFrames ",
1593 "TxMulticastBytes ",
1594 "TxMulticastFrames ",
1600 "RxBroadcastBytes ",
1601 "RxBroadcastFrames ",
1602 "RxMulticastBytes ",
1603 "RxMulticastFrames ",
1609 * These are accumulated per-queue statistics and must match the
1610 * order of the fields in the queue_port_stats structure.
1622 * Return the number of statistics in the specified statistics set.
1624 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1628 return ARRAY_SIZE(stats_strings);
1636 * Return the strings for the specified statistics set.
1638 static void cxgb4vf_get_strings(struct net_device *dev,
1644 memcpy(data, stats_strings, sizeof(stats_strings));
1650 * Small utility routine to accumulate queue statistics across the queues of
1653 static void collect_sge_port_stats(const struct adapter *adapter,
1654 const struct port_info *pi,
1655 struct queue_port_stats *stats)
1657 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1658 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1661 memset(stats, 0, sizeof(*stats));
1662 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1663 stats->tso += txq->tso;
1664 stats->tx_csum += txq->tx_cso;
1665 stats->rx_csum += rxq->stats.rx_cso;
1666 stats->vlan_ex += rxq->stats.vlan_ex;
1667 stats->vlan_ins += txq->vlan_ins;
1668 stats->lro_pkts += rxq->stats.lro_pkts;
1669 stats->lro_merged += rxq->stats.lro_merged;
1674 * Return the ETH_SS_STATS statistics set.
1676 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1677 struct ethtool_stats *stats,
1680 struct port_info *pi = netdev2pinfo(dev);
1681 struct adapter *adapter = pi->adapter;
1682 int err = t4vf_get_port_stats(adapter, pi->pidx,
1683 (struct t4vf_port_stats *)data);
1685 memset(data, 0, sizeof(struct t4vf_port_stats));
1687 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1688 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1692 * Return the size of our register map.
1694 static int cxgb4vf_get_regs_len(struct net_device *dev)
1696 return T4VF_REGMAP_SIZE;
1700 * Dump a block of registers, start to end inclusive, into a buffer.
1702 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1703 unsigned int start, unsigned int end)
1705 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1707 for ( ; start <= end; start += sizeof(u32)) {
1709 * Avoid reading the Mailbox Control register since that
1710 * can trigger a Mailbox Ownership Arbitration cycle and
1711 * interfere with communication with the firmware.
1713 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1716 *bp++ = t4_read_reg(adapter, start);
1721 * Copy our entire register map into the provided buffer.
1723 static void cxgb4vf_get_regs(struct net_device *dev,
1724 struct ethtool_regs *regs,
1727 struct adapter *adapter = netdev2adap(dev);
1729 regs->version = mk_adap_vers(adapter);
1732 * Fill in register buffer with our register map.
1734 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1736 reg_block_dump(adapter, regbuf,
1737 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1738 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1739 reg_block_dump(adapter, regbuf,
1740 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1741 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1743 /* T5 adds new registers in the PL Register map.
1745 reg_block_dump(adapter, regbuf,
1746 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1747 T4VF_PL_BASE_ADDR + (is_t4(adapter->params.chip)
1748 ? PL_VF_WHOAMI_A : PL_VF_REVISION_A));
1749 reg_block_dump(adapter, regbuf,
1750 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1751 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1753 reg_block_dump(adapter, regbuf,
1754 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1755 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1759 * Report current Wake On LAN settings.
1761 static void cxgb4vf_get_wol(struct net_device *dev,
1762 struct ethtool_wolinfo *wol)
1766 memset(&wol->sopass, 0, sizeof(wol->sopass));
1770 * TCP Segmentation Offload flags which we support.
1772 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1774 static const struct ethtool_ops cxgb4vf_ethtool_ops = {
1775 .get_link_ksettings = cxgb4vf_get_link_ksettings,
1776 .get_drvinfo = cxgb4vf_get_drvinfo,
1777 .get_msglevel = cxgb4vf_get_msglevel,
1778 .set_msglevel = cxgb4vf_set_msglevel,
1779 .get_ringparam = cxgb4vf_get_ringparam,
1780 .set_ringparam = cxgb4vf_set_ringparam,
1781 .get_coalesce = cxgb4vf_get_coalesce,
1782 .set_coalesce = cxgb4vf_set_coalesce,
1783 .get_pauseparam = cxgb4vf_get_pauseparam,
1784 .get_link = ethtool_op_get_link,
1785 .get_strings = cxgb4vf_get_strings,
1786 .set_phys_id = cxgb4vf_phys_id,
1787 .get_sset_count = cxgb4vf_get_sset_count,
1788 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1789 .get_regs_len = cxgb4vf_get_regs_len,
1790 .get_regs = cxgb4vf_get_regs,
1791 .get_wol = cxgb4vf_get_wol,
1795 * /sys/kernel/debug/cxgb4vf support code and data.
1796 * ================================================
1800 * Show Firmware Mailbox Command/Reply Log
1802 * Note that we don't do any locking when dumping the Firmware Mailbox Log so
1803 * it's possible that we can catch things during a log update and therefore
1804 * see partially corrupted log entries. But i9t's probably Good Enough(tm).
1805 * If we ever decide that we want to make sure that we're dumping a coherent
1806 * log, we'd need to perform locking in the mailbox logging and in
1807 * mboxlog_open() where we'd need to grab the entire mailbox log in one go
1808 * like we do for the Firmware Device Log. But as stated above, meh ...
1810 static int mboxlog_show(struct seq_file *seq, void *v)
1812 struct adapter *adapter = seq->private;
1813 struct mbox_cmd_log *log = adapter->mbox_log;
1814 struct mbox_cmd *entry;
1817 if (v == SEQ_START_TOKEN) {
1819 "%10s %15s %5s %5s %s\n",
1820 "Seq#", "Tstamp", "Atime", "Etime",
1825 entry_idx = log->cursor + ((uintptr_t)v - 2);
1826 if (entry_idx >= log->size)
1827 entry_idx -= log->size;
1828 entry = mbox_cmd_log_entry(log, entry_idx);
1830 /* skip over unused entries */
1831 if (entry->timestamp == 0)
1834 seq_printf(seq, "%10u %15llu %5d %5d",
1835 entry->seqno, entry->timestamp,
1836 entry->access, entry->execute);
1837 for (i = 0; i < MBOX_LEN / 8; i++) {
1838 u64 flit = entry->cmd[i];
1839 u32 hi = (u32)(flit >> 32);
1842 seq_printf(seq, " %08x %08x", hi, lo);
1844 seq_puts(seq, "\n");
1848 static inline void *mboxlog_get_idx(struct seq_file *seq, loff_t pos)
1850 struct adapter *adapter = seq->private;
1851 struct mbox_cmd_log *log = adapter->mbox_log;
1853 return ((pos <= log->size) ? (void *)(uintptr_t)(pos + 1) : NULL);
1856 static void *mboxlog_start(struct seq_file *seq, loff_t *pos)
1858 return *pos ? mboxlog_get_idx(seq, *pos) : SEQ_START_TOKEN;
1861 static void *mboxlog_next(struct seq_file *seq, void *v, loff_t *pos)
1864 return mboxlog_get_idx(seq, *pos);
1867 static void mboxlog_stop(struct seq_file *seq, void *v)
1871 static const struct seq_operations mboxlog_seq_ops = {
1872 .start = mboxlog_start,
1873 .next = mboxlog_next,
1874 .stop = mboxlog_stop,
1875 .show = mboxlog_show
1878 static int mboxlog_open(struct inode *inode, struct file *file)
1880 int res = seq_open(file, &mboxlog_seq_ops);
1883 struct seq_file *seq = file->private_data;
1885 seq->private = inode->i_private;
1890 static const struct file_operations mboxlog_fops = {
1891 .owner = THIS_MODULE,
1892 .open = mboxlog_open,
1894 .llseek = seq_lseek,
1895 .release = seq_release,
1899 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1903 static int sge_qinfo_show(struct seq_file *seq, void *v)
1905 struct adapter *adapter = seq->private;
1906 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1907 int qs, r = (uintptr_t)v - 1;
1910 seq_putc(seq, '\n');
1912 #define S3(fmt_spec, s, v) \
1914 seq_printf(seq, "%-12s", s); \
1915 for (qs = 0; qs < n; ++qs) \
1916 seq_printf(seq, " %16" fmt_spec, v); \
1917 seq_putc(seq, '\n'); \
1919 #define S(s, v) S3("s", s, v)
1920 #define T(s, v) S3("u", s, txq[qs].v)
1921 #define R(s, v) S3("u", s, rxq[qs].v)
1923 if (r < eth_entries) {
1924 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1925 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1926 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1928 S("QType:", "Ethernet");
1930 (rxq[qs].rspq.netdev
1931 ? rxq[qs].rspq.netdev->name
1934 (rxq[qs].rspq.netdev
1935 ? ((struct port_info *)
1936 netdev_priv(rxq[qs].rspq.netdev))->port_id
1938 T("TxQ ID:", q.abs_id);
1939 T("TxQ size:", q.size);
1940 T("TxQ inuse:", q.in_use);
1941 T("TxQ PIdx:", q.pidx);
1942 T("TxQ CIdx:", q.cidx);
1943 R("RspQ ID:", rspq.abs_id);
1944 R("RspQ size:", rspq.size);
1945 R("RspQE size:", rspq.iqe_len);
1946 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1947 S3("u", "Intr pktcnt:",
1948 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1949 R("RspQ CIdx:", rspq.cidx);
1950 R("RspQ Gen:", rspq.gen);
1951 R("FL ID:", fl.abs_id);
1952 R("FL size:", fl.size - MIN_FL_RESID);
1953 R("FL avail:", fl.avail);
1954 R("FL PIdx:", fl.pidx);
1955 R("FL CIdx:", fl.cidx);
1961 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1963 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1964 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1965 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1966 qtimer_val(adapter, evtq));
1967 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1968 adapter->sge.counter_val[evtq->pktcnt_idx]);
1969 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1970 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1971 } else if (r == 1) {
1972 const struct sge_rspq *intrq = &adapter->sge.intrq;
1974 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1975 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1976 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1977 qtimer_val(adapter, intrq));
1978 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1979 adapter->sge.counter_val[intrq->pktcnt_idx]);
1980 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1981 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1993 * Return the number of "entries" in our "file". We group the multi-Queue
1994 * sections with QPL Queue Sets per "entry". The sections of the output are:
1996 * Ethernet RX/TX Queue Sets
1997 * Firmware Event Queue
1998 * Forwarded Interrupt Queue (if in MSI mode)
2000 static int sge_queue_entries(const struct adapter *adapter)
2002 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
2003 ((adapter->flags & USING_MSI) != 0);
2006 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
2008 int entries = sge_queue_entries(seq->private);
2010 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2013 static void sge_queue_stop(struct seq_file *seq, void *v)
2017 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
2019 int entries = sge_queue_entries(seq->private);
2022 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2025 static const struct seq_operations sge_qinfo_seq_ops = {
2026 .start = sge_queue_start,
2027 .next = sge_queue_next,
2028 .stop = sge_queue_stop,
2029 .show = sge_qinfo_show
2032 static int sge_qinfo_open(struct inode *inode, struct file *file)
2034 int res = seq_open(file, &sge_qinfo_seq_ops);
2037 struct seq_file *seq = file->private_data;
2038 seq->private = inode->i_private;
2043 static const struct file_operations sge_qinfo_debugfs_fops = {
2044 .owner = THIS_MODULE,
2045 .open = sge_qinfo_open,
2047 .llseek = seq_lseek,
2048 .release = seq_release,
2052 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
2056 static int sge_qstats_show(struct seq_file *seq, void *v)
2058 struct adapter *adapter = seq->private;
2059 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
2060 int qs, r = (uintptr_t)v - 1;
2063 seq_putc(seq, '\n');
2065 #define S3(fmt, s, v) \
2067 seq_printf(seq, "%-16s", s); \
2068 for (qs = 0; qs < n; ++qs) \
2069 seq_printf(seq, " %8" fmt, v); \
2070 seq_putc(seq, '\n'); \
2072 #define S(s, v) S3("s", s, v)
2074 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
2075 #define T(s, v) T3("lu", s, v)
2077 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
2078 #define R(s, v) R3("lu", s, v)
2080 if (r < eth_entries) {
2081 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
2082 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
2083 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
2085 S("QType:", "Ethernet");
2087 (rxq[qs].rspq.netdev
2088 ? rxq[qs].rspq.netdev->name
2090 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
2091 R("RxPackets:", stats.pkts);
2092 R("RxCSO:", stats.rx_cso);
2093 R("VLANxtract:", stats.vlan_ex);
2094 R("LROmerged:", stats.lro_merged);
2095 R("LROpackets:", stats.lro_pkts);
2096 R("RxDrops:", stats.rx_drops);
2098 T("TxCSO:", tx_cso);
2099 T("VLANins:", vlan_ins);
2100 T("TxQFull:", q.stops);
2101 T("TxQRestarts:", q.restarts);
2102 T("TxMapErr:", mapping_err);
2103 R("FLAllocErr:", fl.alloc_failed);
2104 R("FLLrgAlcErr:", fl.large_alloc_failed);
2105 R("FLStarving:", fl.starving);
2111 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
2113 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
2114 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
2115 evtq->unhandled_irqs);
2116 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
2117 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
2118 } else if (r == 1) {
2119 const struct sge_rspq *intrq = &adapter->sge.intrq;
2121 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
2122 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
2123 intrq->unhandled_irqs);
2124 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
2125 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
2139 * Return the number of "entries" in our "file". We group the multi-Queue
2140 * sections with QPL Queue Sets per "entry". The sections of the output are:
2142 * Ethernet RX/TX Queue Sets
2143 * Firmware Event Queue
2144 * Forwarded Interrupt Queue (if in MSI mode)
2146 static int sge_qstats_entries(const struct adapter *adapter)
2148 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
2149 ((adapter->flags & USING_MSI) != 0);
2152 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
2154 int entries = sge_qstats_entries(seq->private);
2156 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2159 static void sge_qstats_stop(struct seq_file *seq, void *v)
2163 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
2165 int entries = sge_qstats_entries(seq->private);
2168 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
2171 static const struct seq_operations sge_qstats_seq_ops = {
2172 .start = sge_qstats_start,
2173 .next = sge_qstats_next,
2174 .stop = sge_qstats_stop,
2175 .show = sge_qstats_show
2178 static int sge_qstats_open(struct inode *inode, struct file *file)
2180 int res = seq_open(file, &sge_qstats_seq_ops);
2183 struct seq_file *seq = file->private_data;
2184 seq->private = inode->i_private;
2189 static const struct file_operations sge_qstats_proc_fops = {
2190 .owner = THIS_MODULE,
2191 .open = sge_qstats_open,
2193 .llseek = seq_lseek,
2194 .release = seq_release,
2198 * Show PCI-E SR-IOV Virtual Function Resource Limits.
2200 static int resources_show(struct seq_file *seq, void *v)
2202 struct adapter *adapter = seq->private;
2203 struct vf_resources *vfres = &adapter->params.vfres;
2205 #define S(desc, fmt, var) \
2206 seq_printf(seq, "%-60s " fmt "\n", \
2207 desc " (" #var "):", vfres->var)
2209 S("Virtual Interfaces", "%d", nvi);
2210 S("Egress Queues", "%d", neq);
2211 S("Ethernet Control", "%d", nethctrl);
2212 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
2213 S("Ingress Queues", "%d", niq);
2214 S("Traffic Class", "%d", tc);
2215 S("Port Access Rights Mask", "%#x", pmask);
2216 S("MAC Address Filters", "%d", nexactf);
2217 S("Firmware Command Read Capabilities", "%#x", r_caps);
2218 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
2225 static int resources_open(struct inode *inode, struct file *file)
2227 return single_open(file, resources_show, inode->i_private);
2230 static const struct file_operations resources_proc_fops = {
2231 .owner = THIS_MODULE,
2232 .open = resources_open,
2234 .llseek = seq_lseek,
2235 .release = single_release,
2239 * Show Virtual Interfaces.
2241 static int interfaces_show(struct seq_file *seq, void *v)
2243 if (v == SEQ_START_TOKEN) {
2244 seq_puts(seq, "Interface Port VIID\n");
2246 struct adapter *adapter = seq->private;
2247 int pidx = (uintptr_t)v - 2;
2248 struct net_device *dev = adapter->port[pidx];
2249 struct port_info *pi = netdev_priv(dev);
2251 seq_printf(seq, "%9s %4d %#5x\n",
2252 dev->name, pi->port_id, pi->viid);
2257 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
2259 return pos <= adapter->params.nports
2260 ? (void *)(uintptr_t)(pos + 1)
2264 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
2267 ? interfaces_get_idx(seq->private, *pos)
2271 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
2274 return interfaces_get_idx(seq->private, *pos);
2277 static void interfaces_stop(struct seq_file *seq, void *v)
2281 static const struct seq_operations interfaces_seq_ops = {
2282 .start = interfaces_start,
2283 .next = interfaces_next,
2284 .stop = interfaces_stop,
2285 .show = interfaces_show
2288 static int interfaces_open(struct inode *inode, struct file *file)
2290 int res = seq_open(file, &interfaces_seq_ops);
2293 struct seq_file *seq = file->private_data;
2294 seq->private = inode->i_private;
2299 static const struct file_operations interfaces_proc_fops = {
2300 .owner = THIS_MODULE,
2301 .open = interfaces_open,
2303 .llseek = seq_lseek,
2304 .release = seq_release,
2308 * /sys/kernel/debugfs/cxgb4vf/ files list.
2310 struct cxgb4vf_debugfs_entry {
2311 const char *name; /* name of debugfs node */
2312 umode_t mode; /* file system mode */
2313 const struct file_operations *fops;
2316 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
2317 { "mboxlog", S_IRUGO, &mboxlog_fops },
2318 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
2319 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
2320 { "resources", S_IRUGO, &resources_proc_fops },
2321 { "interfaces", S_IRUGO, &interfaces_proc_fops },
2325 * Module and device initialization and cleanup code.
2326 * ==================================================
2330 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2331 * directory (debugfs_root) has already been set up.
2333 static int setup_debugfs(struct adapter *adapter)
2337 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2340 * Debugfs support is best effort.
2342 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2343 (void)debugfs_create_file(debugfs_files[i].name,
2344 debugfs_files[i].mode,
2345 adapter->debugfs_root,
2347 debugfs_files[i].fops);
2353 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2354 * it to our caller to tear down the directory (debugfs_root).
2356 static void cleanup_debugfs(struct adapter *adapter)
2358 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2361 * Unlike our sister routine cleanup_proc(), we don't need to remove
2362 * individual entries because a call will be made to
2363 * debugfs_remove_recursive(). We just need to clean up any ancillary
2369 /* Figure out how many Ports and Queue Sets we can support. This depends on
2370 * knowing our Virtual Function Resources and may be called a second time if
2371 * we fall back from MSI-X to MSI Interrupt Mode.
2373 static void size_nports_qsets(struct adapter *adapter)
2375 struct vf_resources *vfres = &adapter->params.vfres;
2376 unsigned int ethqsets, pmask_nports;
2378 /* The number of "ports" which we support is equal to the number of
2379 * Virtual Interfaces with which we've been provisioned.
2381 adapter->params.nports = vfres->nvi;
2382 if (adapter->params.nports > MAX_NPORTS) {
2383 dev_warn(adapter->pdev_dev, "only using %d of %d maximum"
2384 " allowed virtual interfaces\n", MAX_NPORTS,
2385 adapter->params.nports);
2386 adapter->params.nports = MAX_NPORTS;
2389 /* We may have been provisioned with more VIs than the number of
2390 * ports we're allowed to access (our Port Access Rights Mask).
2391 * This is obviously a configuration conflict but we don't want to
2392 * crash the kernel or anything silly just because of that.
2394 pmask_nports = hweight32(adapter->params.vfres.pmask);
2395 if (pmask_nports < adapter->params.nports) {
2396 dev_warn(adapter->pdev_dev, "only using %d of %d provisioned"
2397 " virtual interfaces; limited by Port Access Rights"
2398 " mask %#x\n", pmask_nports, adapter->params.nports,
2399 adapter->params.vfres.pmask);
2400 adapter->params.nports = pmask_nports;
2403 /* We need to reserve an Ingress Queue for the Asynchronous Firmware
2404 * Event Queue. And if we're using MSI Interrupts, we'll also need to
2405 * reserve an Ingress Queue for a Forwarded Interrupts.
2407 * The rest of the FL/Intr-capable ingress queues will be matched up
2408 * one-for-one with Ethernet/Control egress queues in order to form
2409 * "Queue Sets" which will be aportioned between the "ports". For
2410 * each Queue Set, we'll need the ability to allocate two Egress
2411 * Contexts -- one for the Ingress Queue Free List and one for the TX
2414 * Note that even if we're currently configured to use MSI-X
2415 * Interrupts (module variable msi == MSI_MSIX) we may get downgraded
2416 * to MSI Interrupts if we can't get enough MSI-X Interrupts. If that
2417 * happens we'll need to adjust things later.
2419 ethqsets = vfres->niqflint - 1 - (msi == MSI_MSI);
2420 if (vfres->nethctrl != ethqsets)
2421 ethqsets = min(vfres->nethctrl, ethqsets);
2422 if (vfres->neq < ethqsets*2)
2423 ethqsets = vfres->neq/2;
2424 if (ethqsets > MAX_ETH_QSETS)
2425 ethqsets = MAX_ETH_QSETS;
2426 adapter->sge.max_ethqsets = ethqsets;
2428 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2429 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2430 " virtual interfaces (too few Queue Sets)\n",
2431 adapter->sge.max_ethqsets, adapter->params.nports);
2432 adapter->params.nports = adapter->sge.max_ethqsets;
2437 * Perform early "adapter" initialization. This is where we discover what
2438 * adapter parameters we're going to be using and initialize basic adapter
2441 static int adap_init0(struct adapter *adapter)
2443 struct sge_params *sge_params = &adapter->params.sge;
2444 struct sge *s = &adapter->sge;
2449 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2450 * 2.6.31 and later we can't call pci_reset_function() in order to
2451 * issue an FLR because of a self- deadlock on the device semaphore.
2452 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2453 * cases where they're needed -- for instance, some versions of KVM
2454 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2455 * use the firmware based reset in order to reset any per function
2458 err = t4vf_fw_reset(adapter);
2460 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2465 * Grab basic operational parameters. These will predominantly have
2466 * been set up by the Physical Function Driver or will be hard coded
2467 * into the adapter. We just have to live with them ... Note that
2468 * we _must_ get our VPD parameters before our SGE parameters because
2469 * we need to know the adapter's core clock from the VPD in order to
2470 * properly decode the SGE Timer Values.
2472 err = t4vf_get_dev_params(adapter);
2474 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2475 " device parameters: err=%d\n", err);
2478 err = t4vf_get_vpd_params(adapter);
2480 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2481 " VPD parameters: err=%d\n", err);
2484 err = t4vf_get_sge_params(adapter);
2486 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2487 " SGE parameters: err=%d\n", err);
2490 err = t4vf_get_rss_glb_config(adapter);
2492 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2493 " RSS parameters: err=%d\n", err);
2496 if (adapter->params.rss.mode !=
2497 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2498 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2499 " mode %d\n", adapter->params.rss.mode);
2502 err = t4vf_sge_init(adapter);
2504 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2509 /* If we're running on newer firmware, let it know that we're
2510 * prepared to deal with encapsulated CPL messages. Older
2511 * firmware won't understand this and we'll just get
2512 * unencapsulated messages ...
2514 param = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2515 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_CPLFW4MSG_ENCAP);
2517 (void) t4vf_set_params(adapter, 1, ¶m, &val);
2520 * Retrieve our RX interrupt holdoff timer values and counter
2521 * threshold values from the SGE parameters.
2523 s->timer_val[0] = core_ticks_to_us(adapter,
2524 TIMERVALUE0_G(sge_params->sge_timer_value_0_and_1));
2525 s->timer_val[1] = core_ticks_to_us(adapter,
2526 TIMERVALUE1_G(sge_params->sge_timer_value_0_and_1));
2527 s->timer_val[2] = core_ticks_to_us(adapter,
2528 TIMERVALUE0_G(sge_params->sge_timer_value_2_and_3));
2529 s->timer_val[3] = core_ticks_to_us(adapter,
2530 TIMERVALUE1_G(sge_params->sge_timer_value_2_and_3));
2531 s->timer_val[4] = core_ticks_to_us(adapter,
2532 TIMERVALUE0_G(sge_params->sge_timer_value_4_and_5));
2533 s->timer_val[5] = core_ticks_to_us(adapter,
2534 TIMERVALUE1_G(sge_params->sge_timer_value_4_and_5));
2536 s->counter_val[0] = THRESHOLD_0_G(sge_params->sge_ingress_rx_threshold);
2537 s->counter_val[1] = THRESHOLD_1_G(sge_params->sge_ingress_rx_threshold);
2538 s->counter_val[2] = THRESHOLD_2_G(sge_params->sge_ingress_rx_threshold);
2539 s->counter_val[3] = THRESHOLD_3_G(sge_params->sge_ingress_rx_threshold);
2542 * Grab our Virtual Interface resource allocation, extract the
2543 * features that we're interested in and do a bit of sanity testing on
2546 err = t4vf_get_vfres(adapter);
2548 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2549 " resources: err=%d\n", err);
2553 /* Check for various parameter sanity issues */
2554 if (adapter->params.vfres.pmask == 0) {
2555 dev_err(adapter->pdev_dev, "no port access configured\n"
2559 if (adapter->params.vfres.nvi == 0) {
2560 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2565 /* Initialize nports and max_ethqsets now that we have our Virtual
2566 * Function Resources.
2568 size_nports_qsets(adapter);
2573 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2574 u8 pkt_cnt_idx, unsigned int size,
2575 unsigned int iqe_size)
2577 rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) |
2578 (pkt_cnt_idx < SGE_NCOUNTERS ?
2579 QINTR_CNT_EN_F : 0));
2580 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2583 rspq->iqe_len = iqe_size;
2588 * Perform default configuration of DMA queues depending on the number and
2589 * type of ports we found and the number of available CPUs. Most settings can
2590 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2591 * being brought up for the first time.
2593 static void cfg_queues(struct adapter *adapter)
2595 struct sge *s = &adapter->sge;
2596 int q10g, n10g, qidx, pidx, qs;
2600 * We should not be called till we know how many Queue Sets we can
2601 * support. In particular, this means that we need to know what kind
2602 * of interrupts we'll be using ...
2604 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2607 * Count the number of 10GbE Virtual Interfaces that we have.
2610 for_each_port(adapter, pidx)
2611 n10g += is_x_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2614 * We default to 1 queue per non-10G port and up to # of cores queues
2620 int n1g = (adapter->params.nports - n10g);
2621 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2622 if (q10g > num_online_cpus())
2623 q10g = num_online_cpus();
2627 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2628 * The layout will be established in setup_sge_queues() when the
2629 * adapter is brough up for the first time.
2632 for_each_port(adapter, pidx) {
2633 struct port_info *pi = adap2pinfo(adapter, pidx);
2635 pi->first_qset = qidx;
2636 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
2642 * The Ingress Queue Entry Size for our various Response Queues needs
2643 * to be big enough to accommodate the largest message we can receive
2644 * from the chip/firmware; which is 64 bytes ...
2649 * Set up default Queue Set parameters ... Start off with the
2650 * shortest interrupt holdoff timer.
2652 for (qs = 0; qs < s->max_ethqsets; qs++) {
2653 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2654 struct sge_eth_txq *txq = &s->ethtxq[qs];
2656 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2662 * The firmware event queue is used for link state changes and
2663 * notifications of TX DMA completions.
2665 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2668 * The forwarded interrupt queue is used when we're in MSI interrupt
2669 * mode. In this mode all interrupts associated with RX queues will
2670 * be forwarded to a single queue which we'll associate with our MSI
2671 * interrupt vector. The messages dropped in the forwarded interrupt
2672 * queue will indicate which ingress queue needs servicing ... This
2673 * queue needs to be large enough to accommodate all of the ingress
2674 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2675 * from equalling the CIDX if every ingress queue has an outstanding
2676 * interrupt). The queue doesn't need to be any larger because no
2677 * ingress queue will ever have more than one outstanding interrupt at
2680 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2685 * Reduce the number of Ethernet queues across all ports to at most n.
2686 * n provides at least one queue per port.
2688 static void reduce_ethqs(struct adapter *adapter, int n)
2691 struct port_info *pi;
2694 * While we have too many active Ether Queue Sets, interate across the
2695 * "ports" and reduce their individual Queue Set allocations.
2697 BUG_ON(n < adapter->params.nports);
2698 while (n < adapter->sge.ethqsets)
2699 for_each_port(adapter, i) {
2700 pi = adap2pinfo(adapter, i);
2701 if (pi->nqsets > 1) {
2703 adapter->sge.ethqsets--;
2704 if (adapter->sge.ethqsets <= n)
2710 * Reassign the starting Queue Sets for each of the "ports" ...
2713 for_each_port(adapter, i) {
2714 pi = adap2pinfo(adapter, i);
2721 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2722 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2723 * need. Minimally we need one for every Virtual Interface plus those needed
2724 * for our "extras". Note that this process may lower the maximum number of
2725 * allowed Queue Sets ...
2727 static int enable_msix(struct adapter *adapter)
2729 int i, want, need, nqsets;
2730 struct msix_entry entries[MSIX_ENTRIES];
2731 struct sge *s = &adapter->sge;
2733 for (i = 0; i < MSIX_ENTRIES; ++i)
2734 entries[i].entry = i;
2737 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2738 * plus those needed for our "extras" (for example, the firmware
2739 * message queue). We _need_ at least one "Queue Set" per Virtual
2740 * Interface plus those needed for our "extras". So now we get to see
2741 * if the song is right ...
2743 want = s->max_ethqsets + MSIX_EXTRAS;
2744 need = adapter->params.nports + MSIX_EXTRAS;
2746 want = pci_enable_msix_range(adapter->pdev, entries, need, want);
2750 nqsets = want - MSIX_EXTRAS;
2751 if (nqsets < s->max_ethqsets) {
2752 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2753 " for %d Queue Sets\n", nqsets);
2754 s->max_ethqsets = nqsets;
2755 if (nqsets < s->ethqsets)
2756 reduce_ethqs(adapter, nqsets);
2758 for (i = 0; i < want; ++i)
2759 adapter->msix_info[i].vec = entries[i].vector;
2764 static const struct net_device_ops cxgb4vf_netdev_ops = {
2765 .ndo_open = cxgb4vf_open,
2766 .ndo_stop = cxgb4vf_stop,
2767 .ndo_start_xmit = t4vf_eth_xmit,
2768 .ndo_get_stats = cxgb4vf_get_stats,
2769 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2770 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2771 .ndo_validate_addr = eth_validate_addr,
2772 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2773 .ndo_change_mtu = cxgb4vf_change_mtu,
2774 .ndo_fix_features = cxgb4vf_fix_features,
2775 .ndo_set_features = cxgb4vf_set_features,
2776 #ifdef CONFIG_NET_POLL_CONTROLLER
2777 .ndo_poll_controller = cxgb4vf_poll_controller,
2782 * "Probe" a device: initialize a device and construct all kernel and driver
2783 * state needed to manage the device. This routine is called "init_one" in
2786 static int cxgb4vf_pci_probe(struct pci_dev *pdev,
2787 const struct pci_device_id *ent)
2792 struct adapter *adapter;
2793 struct port_info *pi;
2794 struct net_device *netdev;
2798 * Print our driver banner the first time we're called to initialize a
2801 pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION);
2804 * Initialize generic PCI device state.
2806 err = pci_enable_device(pdev);
2808 dev_err(&pdev->dev, "cannot enable PCI device\n");
2813 * Reserve PCI resources for the device. If we can't get them some
2814 * other driver may have already claimed the device ...
2816 err = pci_request_regions(pdev, KBUILD_MODNAME);
2818 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2819 goto err_disable_device;
2823 * Set up our DMA mask: try for 64-bit address masking first and
2824 * fall back to 32-bit if we can't get 64 bits ...
2826 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2828 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2830 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2831 " coherent allocations\n");
2832 goto err_release_regions;
2836 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2838 dev_err(&pdev->dev, "no usable DMA configuration\n");
2839 goto err_release_regions;
2845 * Enable bus mastering for the device ...
2847 pci_set_master(pdev);
2850 * Allocate our adapter data structure and attach it to the device.
2852 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2855 goto err_release_regions;
2857 pci_set_drvdata(pdev, adapter);
2858 adapter->pdev = pdev;
2859 adapter->pdev_dev = &pdev->dev;
2861 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
2862 (sizeof(struct mbox_cmd) *
2863 T4VF_OS_LOG_MBOX_CMDS),
2865 if (!adapter->mbox_log) {
2867 goto err_free_adapter;
2869 adapter->mbox_log->size = T4VF_OS_LOG_MBOX_CMDS;
2872 * Initialize SMP data synchronization resources.
2874 spin_lock_init(&adapter->stats_lock);
2875 spin_lock_init(&adapter->mbox_lock);
2876 INIT_LIST_HEAD(&adapter->mlist.list);
2879 * Map our I/O registers in BAR0.
2881 adapter->regs = pci_ioremap_bar(pdev, 0);
2882 if (!adapter->regs) {
2883 dev_err(&pdev->dev, "cannot map device registers\n");
2885 goto err_free_adapter;
2888 /* Wait for the device to become ready before proceeding ...
2890 err = t4vf_prep_adapter(adapter);
2892 dev_err(adapter->pdev_dev, "device didn't become ready:"
2894 goto err_unmap_bar0;
2897 /* For T5 and later we want to use the new BAR-based User Doorbells,
2898 * so we need to map BAR2 here ...
2900 if (!is_t4(adapter->params.chip)) {
2901 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
2902 pci_resource_len(pdev, 2));
2903 if (!adapter->bar2) {
2904 dev_err(adapter->pdev_dev, "cannot map BAR2 doorbells\n");
2906 goto err_unmap_bar0;
2910 * Initialize adapter level features.
2912 adapter->name = pci_name(pdev);
2913 adapter->msg_enable = DFLT_MSG_ENABLE;
2915 /* If possible, we use PCIe Relaxed Ordering Attribute to deliver
2916 * Ingress Packet Data to Free List Buffers in order to allow for
2917 * chipset performance optimizations between the Root Complex and
2918 * Memory Controllers. (Messages to the associated Ingress Queue
2919 * notifying new Packet Placement in the Free Lists Buffers will be
2920 * send without the Relaxed Ordering Attribute thus guaranteeing that
2921 * all preceding PCIe Transaction Layer Packets will be processed
2922 * first.) But some Root Complexes have various issues with Upstream
2923 * Transaction Layer Packets with the Relaxed Ordering Attribute set.
2924 * The PCIe devices which under the Root Complexes will be cleared the
2925 * Relaxed Ordering bit in the configuration space, So we check our
2926 * PCIe configuration space to see if it's flagged with advice against
2927 * using Relaxed Ordering.
2929 if (!pcie_relaxed_ordering_enabled(pdev))
2930 adapter->flags |= ROOT_NO_RELAXED_ORDERING;
2932 err = adap_init0(adapter);
2936 /* Initialize hash mac addr list */
2937 INIT_LIST_HEAD(&adapter->mac_hlist);
2940 * Allocate our "adapter ports" and stitch everything together.
2942 pmask = adapter->params.vfres.pmask;
2943 pf = t4vf_get_pf_from_vf(adapter);
2944 for_each_port(adapter, pidx) {
2947 unsigned int naddr = 1;
2950 * We simplistically allocate our virtual interfaces
2951 * sequentially across the port numbers to which we have
2952 * access rights. This should be configurable in some manner
2957 port_id = ffs(pmask) - 1;
2958 pmask &= ~(1 << port_id);
2959 viid = t4vf_alloc_vi(adapter, port_id);
2961 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2962 " err=%d\n", port_id, viid);
2968 * Allocate our network device and stitch things together.
2970 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2972 if (netdev == NULL) {
2973 t4vf_free_vi(adapter, viid);
2977 adapter->port[pidx] = netdev;
2978 SET_NETDEV_DEV(netdev, &pdev->dev);
2979 pi = netdev_priv(netdev);
2980 pi->adapter = adapter;
2982 pi->port_id = port_id;
2986 * Initialize the starting state of our "port" and register
2989 pi->xact_addr_filt = -1;
2990 netif_carrier_off(netdev);
2991 netdev->irq = pdev->irq;
2993 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
2994 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2995 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXCSUM;
2996 netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
2997 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2999 netdev->features = netdev->hw_features |
3000 NETIF_F_HW_VLAN_CTAG_TX;
3002 netdev->features |= NETIF_F_HIGHDMA;
3004 netdev->priv_flags |= IFF_UNICAST_FLT;
3005 netdev->min_mtu = 81;
3006 netdev->max_mtu = ETH_MAX_MTU;
3008 netdev->netdev_ops = &cxgb4vf_netdev_ops;
3009 netdev->ethtool_ops = &cxgb4vf_ethtool_ops;
3010 netdev->dev_port = pi->port_id;
3013 * Initialize the hardware/software state for the port.
3015 err = t4vf_port_init(adapter, pidx);
3017 dev_err(&pdev->dev, "cannot initialize port %d\n",
3022 err = t4vf_get_vf_mac_acl(adapter, pf, &naddr, mac);
3025 "unable to determine MAC ACL address, "
3026 "continuing anyway.. (status %d)\n", err);
3027 } else if (naddr && adapter->params.vfres.nvi == 1) {
3028 struct sockaddr addr;
3030 ether_addr_copy(addr.sa_data, mac);
3031 err = cxgb4vf_set_mac_addr(netdev, &addr);
3034 "unable to set MAC address %pM\n",
3038 dev_info(&pdev->dev,
3039 "Using assigned MAC ACL: %pM\n", mac);
3043 /* See what interrupts we'll be using. If we've been configured to
3044 * use MSI-X interrupts, try to enable them but fall back to using
3045 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
3046 * get MSI interrupts we bail with the error.
3048 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
3049 adapter->flags |= USING_MSIX;
3051 if (msi == MSI_MSIX) {
3052 dev_info(adapter->pdev_dev,
3053 "Unable to use MSI-X Interrupts; falling "
3054 "back to MSI Interrupts\n");
3056 /* We're going to need a Forwarded Interrupt Queue so
3057 * that may cut into how many Queue Sets we can
3061 size_nports_qsets(adapter);
3063 err = pci_enable_msi(pdev);
3065 dev_err(&pdev->dev, "Unable to allocate MSI Interrupts;"
3069 adapter->flags |= USING_MSI;
3072 /* Now that we know how many "ports" we have and what interrupt
3073 * mechanism we're going to use, we can configure our queue resources.
3075 cfg_queues(adapter);
3078 * The "card" is now ready to go. If any errors occur during device
3079 * registration we do not fail the whole "card" but rather proceed
3080 * only with the ports we manage to register successfully. However we
3081 * must register at least one net device.
3083 for_each_port(adapter, pidx) {
3084 struct port_info *pi = netdev_priv(adapter->port[pidx]);
3085 netdev = adapter->port[pidx];
3089 netif_set_real_num_tx_queues(netdev, pi->nqsets);
3090 netif_set_real_num_rx_queues(netdev, pi->nqsets);
3092 err = register_netdev(netdev);
3094 dev_warn(&pdev->dev, "cannot register net device %s,"
3095 " skipping\n", netdev->name);
3099 set_bit(pidx, &adapter->registered_device_map);
3101 if (adapter->registered_device_map == 0) {
3102 dev_err(&pdev->dev, "could not register any net devices\n");
3103 goto err_disable_interrupts;
3107 * Set up our debugfs entries.
3109 if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) {
3110 adapter->debugfs_root =
3111 debugfs_create_dir(pci_name(pdev),
3112 cxgb4vf_debugfs_root);
3113 if (IS_ERR_OR_NULL(adapter->debugfs_root))
3114 dev_warn(&pdev->dev, "could not create debugfs"
3117 setup_debugfs(adapter);
3121 * Print a short notice on the existence and configuration of the new
3122 * VF network device ...
3124 for_each_port(adapter, pidx) {
3125 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
3126 adapter->port[pidx]->name,
3127 (adapter->flags & USING_MSIX) ? "MSI-X" :
3128 (adapter->flags & USING_MSI) ? "MSI" : "");
3137 * Error recovery and exit code. Unwind state that's been created
3138 * so far and return the error.
3140 err_disable_interrupts:
3141 if (adapter->flags & USING_MSIX) {
3142 pci_disable_msix(adapter->pdev);
3143 adapter->flags &= ~USING_MSIX;
3144 } else if (adapter->flags & USING_MSI) {
3145 pci_disable_msi(adapter->pdev);
3146 adapter->flags &= ~USING_MSI;
3150 for_each_port(adapter, pidx) {
3151 netdev = adapter->port[pidx];
3154 pi = netdev_priv(netdev);
3155 t4vf_free_vi(adapter, pi->viid);
3156 if (test_bit(pidx, &adapter->registered_device_map))
3157 unregister_netdev(netdev);
3158 free_netdev(netdev);
3162 if (!is_t4(adapter->params.chip))
3163 iounmap(adapter->bar2);
3166 iounmap(adapter->regs);
3169 kfree(adapter->mbox_log);
3172 err_release_regions:
3173 pci_release_regions(pdev);
3174 pci_clear_master(pdev);
3177 pci_disable_device(pdev);
3183 * "Remove" a device: tear down all kernel and driver state created in the
3184 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
3185 * that this is called "remove_one" in the PF Driver.)
3187 static void cxgb4vf_pci_remove(struct pci_dev *pdev)
3189 struct adapter *adapter = pci_get_drvdata(pdev);
3192 * Tear down driver state associated with device.
3198 * Stop all of our activity. Unregister network port,
3199 * disable interrupts, etc.
3201 for_each_port(adapter, pidx)
3202 if (test_bit(pidx, &adapter->registered_device_map))
3203 unregister_netdev(adapter->port[pidx]);
3204 t4vf_sge_stop(adapter);
3205 if (adapter->flags & USING_MSIX) {
3206 pci_disable_msix(adapter->pdev);
3207 adapter->flags &= ~USING_MSIX;
3208 } else if (adapter->flags & USING_MSI) {
3209 pci_disable_msi(adapter->pdev);
3210 adapter->flags &= ~USING_MSI;
3214 * Tear down our debugfs entries.
3216 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
3217 cleanup_debugfs(adapter);
3218 debugfs_remove_recursive(adapter->debugfs_root);
3222 * Free all of the various resources which we've acquired ...
3224 t4vf_free_sge_resources(adapter);
3225 for_each_port(adapter, pidx) {
3226 struct net_device *netdev = adapter->port[pidx];
3227 struct port_info *pi;
3232 pi = netdev_priv(netdev);
3233 t4vf_free_vi(adapter, pi->viid);
3234 free_netdev(netdev);
3236 iounmap(adapter->regs);
3237 if (!is_t4(adapter->params.chip))
3238 iounmap(adapter->bar2);
3239 kfree(adapter->mbox_log);
3244 * Disable the device and release its PCI resources.
3246 pci_disable_device(pdev);
3247 pci_clear_master(pdev);
3248 pci_release_regions(pdev);
3252 * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt
3255 static void cxgb4vf_pci_shutdown(struct pci_dev *pdev)
3257 struct adapter *adapter;
3260 adapter = pci_get_drvdata(pdev);
3264 /* Disable all Virtual Interfaces. This will shut down the
3265 * delivery of all ingress packets into the chip for these
3266 * Virtual Interfaces.
3268 for_each_port(adapter, pidx)
3269 if (test_bit(pidx, &adapter->registered_device_map))
3270 unregister_netdev(adapter->port[pidx]);
3272 /* Free up all Queues which will prevent further DMA and
3273 * Interrupts allowing various internal pathways to drain.
3275 t4vf_sge_stop(adapter);
3276 if (adapter->flags & USING_MSIX) {
3277 pci_disable_msix(adapter->pdev);
3278 adapter->flags &= ~USING_MSIX;
3279 } else if (adapter->flags & USING_MSI) {
3280 pci_disable_msi(adapter->pdev);
3281 adapter->flags &= ~USING_MSI;
3285 * Free up all Queues which will prevent further DMA and
3286 * Interrupts allowing various internal pathways to drain.
3288 t4vf_free_sge_resources(adapter);
3289 pci_set_drvdata(pdev, NULL);
3292 /* Macros needed to support the PCI Device ID Table ...
3294 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
3295 static const struct pci_device_id cxgb4vf_pci_tbl[] = {
3296 #define CH_PCI_DEVICE_ID_FUNCTION 0x8
3298 #define CH_PCI_ID_TABLE_ENTRY(devid) \
3299 { PCI_VDEVICE(CHELSIO, (devid)), 0 }
3301 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END { 0, } }
3303 #include "../cxgb4/t4_pci_id_tbl.h"
3305 MODULE_DESCRIPTION(DRV_DESC);
3306 MODULE_AUTHOR("Chelsio Communications");
3307 MODULE_LICENSE("Dual BSD/GPL");
3308 MODULE_VERSION(DRV_VERSION);
3309 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
3311 static struct pci_driver cxgb4vf_driver = {
3312 .name = KBUILD_MODNAME,
3313 .id_table = cxgb4vf_pci_tbl,
3314 .probe = cxgb4vf_pci_probe,
3315 .remove = cxgb4vf_pci_remove,
3316 .shutdown = cxgb4vf_pci_shutdown,
3320 * Initialize global driver state.
3322 static int __init cxgb4vf_module_init(void)
3327 * Vet our module parameters.
3329 if (msi != MSI_MSIX && msi != MSI_MSI) {
3330 pr_warn("bad module parameter msi=%d; must be %d (MSI-X or MSI) or %d (MSI)\n",
3331 msi, MSI_MSIX, MSI_MSI);
3335 /* Debugfs support is optional, just warn if this fails */
3336 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3337 if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3338 pr_warn("could not create debugfs entry, continuing\n");
3340 ret = pci_register_driver(&cxgb4vf_driver);
3341 if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
3342 debugfs_remove(cxgb4vf_debugfs_root);
3347 * Tear down global driver state.
3349 static void __exit cxgb4vf_module_exit(void)
3351 pci_unregister_driver(&cxgb4vf_driver);
3352 debugfs_remove(cxgb4vf_debugfs_root);
3355 module_init(cxgb4vf_module_init);
3356 module_exit(cxgb4vf_module_exit);