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 #include <linux/module.h>
37 #include <linux/moduleparam.h>
38 #include <linux/init.h>
39 #include <linux/pci.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/netdevice.h>
42 #include <linux/etherdevice.h>
43 #include <linux/debugfs.h>
44 #include <linux/ethtool.h>
46 #include "t4vf_common.h"
47 #include "t4vf_defs.h"
49 #include "../cxgb4/t4_regs.h"
50 #include "../cxgb4/t4_msg.h"
53 * Generic information about the driver.
55 #define DRV_VERSION "1.0.0"
56 #define DRV_DESC "Chelsio T4 Virtual Function (VF) Network Driver"
64 * Default ethtool "message level" for adapters.
66 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
67 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
68 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
70 static int dflt_msg_enable = DFLT_MSG_ENABLE;
72 module_param(dflt_msg_enable, int, 0644);
73 MODULE_PARM_DESC(dflt_msg_enable,
74 "default adapter ethtool message level bitmap");
77 * The driver uses the best interrupt scheme available on a platform in the
78 * order MSI-X then MSI. This parameter determines which of these schemes the
79 * driver may consider as follows:
81 * msi = 2: choose from among MSI-X and MSI
82 * msi = 1: only consider MSI interrupts
84 * Note that unlike the Physical Function driver, this Virtual Function driver
85 * does _not_ support legacy INTx interrupts (this limitation is mandated by
86 * the PCI-E SR-IOV standard).
90 #define MSI_DEFAULT MSI_MSIX
92 static int msi = MSI_DEFAULT;
94 module_param(msi, int, 0644);
95 MODULE_PARM_DESC(msi, "whether to use MSI-X or MSI");
98 * Fundamental constants.
99 * ======================
103 MAX_TXQ_ENTRIES = 16384,
104 MAX_RSPQ_ENTRIES = 16384,
105 MAX_RX_BUFFERS = 16384,
107 MIN_TXQ_ENTRIES = 32,
108 MIN_RSPQ_ENTRIES = 128,
112 * For purposes of manipulating the Free List size we need to
113 * recognize that Free Lists are actually Egress Queues (the host
114 * produces free buffers which the hardware consumes), Egress Queues
115 * indices are all in units of Egress Context Units bytes, and free
116 * list entries are 64-bit PCI DMA addresses. And since the state of
117 * the Producer Index == the Consumer Index implies an EMPTY list, we
118 * always have at least one Egress Unit's worth of Free List entries
119 * unused. See sge.c for more details ...
121 EQ_UNIT = SGE_EQ_IDXSIZE,
122 FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64),
123 MIN_FL_RESID = FL_PER_EQ_UNIT,
127 * Global driver state.
128 * ====================
131 static struct dentry *cxgb4vf_debugfs_root;
134 * OS "Callback" functions.
135 * ========================
139 * The link status has changed on the indicated "port" (Virtual Interface).
141 void t4vf_os_link_changed(struct adapter *adapter, int pidx, int link_ok)
143 struct net_device *dev = adapter->port[pidx];
146 * If the port is disabled or the current recorded "link up"
147 * status matches the new status, just return.
149 if (!netif_running(dev) || link_ok == netif_carrier_ok(dev))
153 * Tell the OS that the link status has changed and print a short
154 * informative message on the console about the event.
159 const struct port_info *pi = netdev_priv(dev);
161 netif_carrier_on(dev);
163 switch (pi->link_cfg.speed) {
181 switch (pi->link_cfg.fc) {
190 case PAUSE_RX|PAUSE_TX:
199 printk(KERN_INFO "%s: link up, %s, full-duplex, %s PAUSE\n",
202 netif_carrier_off(dev);
203 printk(KERN_INFO "%s: link down\n", dev->name);
208 * Net device operations.
209 * ======================
213 * Record our new VLAN Group and enable/disable hardware VLAN Tag extraction
214 * based on whether the specified VLAN Group pointer is NULL or not.
216 static void cxgb4vf_vlan_rx_register(struct net_device *dev,
217 struct vlan_group *grp)
219 struct port_info *pi = netdev_priv(dev);
222 t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1, grp != NULL, 0);
226 * Perform the MAC and PHY actions needed to enable a "port" (Virtual
229 static int link_start(struct net_device *dev)
232 struct port_info *pi = netdev_priv(dev);
235 * We do not set address filters and promiscuity here, the stack does
236 * that step explicitly.
238 ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, -1,
241 ret = t4vf_change_mac(pi->adapter, pi->viid,
242 pi->xact_addr_filt, dev->dev_addr, true);
244 pi->xact_addr_filt = ret;
250 * We don't need to actually "start the link" itself since the
251 * firmware will do that for us when the first Virtual Interface
252 * is enabled on a port.
255 ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true);
260 * Name the MSI-X interrupts.
262 static void name_msix_vecs(struct adapter *adapter)
264 int namelen = sizeof(adapter->msix_info[0].desc) - 1;
270 snprintf(adapter->msix_info[MSIX_FW].desc, namelen,
271 "%s-FWeventq", adapter->name);
272 adapter->msix_info[MSIX_FW].desc[namelen] = 0;
277 for_each_port(adapter, pidx) {
278 struct net_device *dev = adapter->port[pidx];
279 const struct port_info *pi = netdev_priv(dev);
282 for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) {
283 snprintf(adapter->msix_info[msi].desc, namelen,
284 "%s-%d", dev->name, qs);
285 adapter->msix_info[msi].desc[namelen] = 0;
291 * Request all of our MSI-X resources.
293 static int request_msix_queue_irqs(struct adapter *adapter)
295 struct sge *s = &adapter->sge;
301 err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix,
302 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq);
310 for_each_ethrxq(s, rxq) {
311 err = request_irq(adapter->msix_info[msi].vec,
312 t4vf_sge_intr_msix, 0,
313 adapter->msix_info[msi].desc,
314 &s->ethrxq[rxq].rspq);
323 free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq);
324 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
329 * Free our MSI-X resources.
331 static void free_msix_queue_irqs(struct adapter *adapter)
333 struct sge *s = &adapter->sge;
336 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
338 for_each_ethrxq(s, rxq)
339 free_irq(adapter->msix_info[msi++].vec,
340 &s->ethrxq[rxq].rspq);
344 * Turn on NAPI and start up interrupts on a response queue.
346 static void qenable(struct sge_rspq *rspq)
348 napi_enable(&rspq->napi);
351 * 0-increment the Going To Sleep register to start the timer and
354 t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
356 SEINTARM(rspq->intr_params) |
357 INGRESSQID(rspq->cntxt_id));
361 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
363 static void enable_rx(struct adapter *adapter)
366 struct sge *s = &adapter->sge;
368 for_each_ethrxq(s, rxq)
369 qenable(&s->ethrxq[rxq].rspq);
370 qenable(&s->fw_evtq);
373 * The interrupt queue doesn't use NAPI so we do the 0-increment of
374 * its Going To Sleep register here to get it started.
376 if (adapter->flags & USING_MSI)
377 t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
379 SEINTARM(s->intrq.intr_params) |
380 INGRESSQID(s->intrq.cntxt_id));
385 * Wait until all NAPI handlers are descheduled.
387 static void quiesce_rx(struct adapter *adapter)
389 struct sge *s = &adapter->sge;
392 for_each_ethrxq(s, rxq)
393 napi_disable(&s->ethrxq[rxq].rspq.napi);
394 napi_disable(&s->fw_evtq.napi);
398 * Response queue handler for the firmware event queue.
400 static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp,
401 const struct pkt_gl *gl)
404 * Extract response opcode and get pointer to CPL message body.
406 struct adapter *adapter = rspq->adapter;
407 u8 opcode = ((const struct rss_header *)rsp)->opcode;
408 void *cpl = (void *)(rsp + 1);
413 * We've received an asynchronous message from the firmware.
415 const struct cpl_fw6_msg *fw_msg = cpl;
416 if (fw_msg->type == FW6_TYPE_CMD_RPL)
417 t4vf_handle_fw_rpl(adapter, fw_msg->data);
421 case CPL_SGE_EGR_UPDATE: {
423 * We've received an Egress Queue Status Update message. We
424 * get these, if the SGE is configured to send these when the
425 * firmware passes certain points in processing our TX
426 * Ethernet Queue or if we make an explicit request for one.
427 * We use these updates to determine when we may need to
428 * restart a TX Ethernet Queue which was stopped for lack of
429 * free TX Queue Descriptors ...
431 const struct cpl_sge_egr_update *p = (void *)cpl;
432 unsigned int qid = EGR_QID(be32_to_cpu(p->opcode_qid));
433 struct sge *s = &adapter->sge;
435 struct sge_eth_txq *txq;
439 * Perform sanity checking on the Queue ID to make sure it
440 * really refers to one of our TX Ethernet Egress Queues which
441 * is active and matches the queue's ID. None of these error
442 * conditions should ever happen so we may want to either make
443 * them fatal and/or conditionalized under DEBUG.
445 eq_idx = EQ_IDX(s, qid);
446 if (unlikely(eq_idx >= MAX_EGRQ)) {
447 dev_err(adapter->pdev_dev,
448 "Egress Update QID %d out of range\n", qid);
451 tq = s->egr_map[eq_idx];
452 if (unlikely(tq == NULL)) {
453 dev_err(adapter->pdev_dev,
454 "Egress Update QID %d TXQ=NULL\n", qid);
457 txq = container_of(tq, struct sge_eth_txq, q);
458 if (unlikely(tq->abs_id != qid)) {
459 dev_err(adapter->pdev_dev,
460 "Egress Update QID %d refers to TXQ %d\n",
466 * Restart a stopped TX Queue which has less than half of its
470 netif_tx_wake_queue(txq->txq);
475 dev_err(adapter->pdev_dev,
476 "unexpected CPL %#x on FW event queue\n", opcode);
483 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
484 * to use and initializes them. We support multiple "Queue Sets" per port if
485 * we have MSI-X, otherwise just one queue set per port.
487 static int setup_sge_queues(struct adapter *adapter)
489 struct sge *s = &adapter->sge;
493 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
496 bitmap_zero(s->starving_fl, MAX_EGRQ);
499 * If we're using MSI interrupt mode we need to set up a "forwarded
500 * interrupt" queue which we'll set up with our MSI vector. The rest
501 * of the ingress queues will be set up to forward their interrupts to
502 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
503 * the intrq's queue ID as the interrupt forwarding queue for the
504 * subsequent calls ...
506 if (adapter->flags & USING_MSI) {
507 err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false,
508 adapter->port[0], 0, NULL, NULL);
510 goto err_free_queues;
514 * Allocate our ingress queue for asynchronous firmware messages.
516 err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0],
517 MSIX_FW, NULL, fwevtq_handler);
519 goto err_free_queues;
522 * Allocate each "port"'s initial Queue Sets. These can be changed
523 * later on ... up to the point where any interface on the adapter is
524 * brought up at which point lots of things get nailed down
528 for_each_port(adapter, pidx) {
529 struct net_device *dev = adapter->port[pidx];
530 struct port_info *pi = netdev_priv(dev);
531 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
532 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
535 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
536 err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false,
538 &rxq->fl, t4vf_ethrx_handler);
540 goto err_free_queues;
542 err = t4vf_sge_alloc_eth_txq(adapter, txq, dev,
543 netdev_get_tx_queue(dev, qs),
544 s->fw_evtq.cntxt_id);
546 goto err_free_queues;
549 memset(&rxq->stats, 0, sizeof(rxq->stats));
554 * Create the reverse mappings for the queues.
556 s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id;
557 s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id;
558 IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq;
559 for_each_port(adapter, pidx) {
560 struct net_device *dev = adapter->port[pidx];
561 struct port_info *pi = netdev_priv(dev);
562 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
563 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
566 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
567 IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq;
568 EQ_MAP(s, txq->q.abs_id) = &txq->q;
571 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
572 * for Free Lists but since all of the Egress Queues
573 * (including Free Lists) have Relative Queue IDs
574 * which are computed as Absolute - Base Queue ID, we
575 * can synthesize the Absolute Queue IDs for the Free
576 * Lists. This is useful for debugging purposes when
577 * we want to dump Queue Contexts via the PF Driver.
579 rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base;
580 EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl;
586 t4vf_free_sge_resources(adapter);
591 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
592 * queues. We configure the RSS CPU lookup table to distribute to the number
593 * of HW receive queues, and the response queue lookup table to narrow that
594 * down to the response queues actually configured for each "port" (Virtual
595 * Interface). We always configure the RSS mapping for all ports since the
596 * mapping table has plenty of entries.
598 static int setup_rss(struct adapter *adapter)
602 for_each_port(adapter, pidx) {
603 struct port_info *pi = adap2pinfo(adapter, pidx);
604 struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
605 u16 rss[MAX_PORT_QSETS];
608 for (qs = 0; qs < pi->nqsets; qs++)
609 rss[qs] = rxq[qs].rspq.abs_id;
611 err = t4vf_config_rss_range(adapter, pi->viid,
612 0, pi->rss_size, rss, pi->nqsets);
617 * Perform Global RSS Mode-specific initialization.
619 switch (adapter->params.rss.mode) {
620 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL:
622 * If Tunnel All Lookup isn't specified in the global
623 * RSS Configuration, then we need to specify a
624 * default Ingress Queue for any ingress packets which
625 * aren't hashed. We'll use our first ingress queue
628 if (!adapter->params.rss.u.basicvirtual.tnlalllookup) {
629 union rss_vi_config config;
630 err = t4vf_read_rss_vi_config(adapter,
635 config.basicvirtual.defaultq =
637 err = t4vf_write_rss_vi_config(adapter,
651 * Bring the adapter up. Called whenever we go from no "ports" open to having
652 * one open. This function performs the actions necessary to make an adapter
653 * operational, such as completing the initialization of HW modules, and
654 * enabling interrupts. Must be called with the rtnl lock held. (Note that
655 * this is called "cxgb_up" in the PF Driver.)
657 static int adapter_up(struct adapter *adapter)
662 * If this is the first time we've been called, perform basic
663 * adapter setup. Once we've done this, many of our adapter
664 * parameters can no longer be changed ...
666 if ((adapter->flags & FULL_INIT_DONE) == 0) {
667 err = setup_sge_queues(adapter);
670 err = setup_rss(adapter);
672 t4vf_free_sge_resources(adapter);
676 if (adapter->flags & USING_MSIX)
677 name_msix_vecs(adapter);
678 adapter->flags |= FULL_INIT_DONE;
682 * Acquire our interrupt resources. We only support MSI-X and MSI.
684 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
685 if (adapter->flags & USING_MSIX)
686 err = request_msix_queue_irqs(adapter);
688 err = request_irq(adapter->pdev->irq,
689 t4vf_intr_handler(adapter), 0,
690 adapter->name, adapter);
692 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
698 * Enable NAPI ingress processing and return success.
701 t4vf_sge_start(adapter);
706 * Bring the adapter down. Called whenever the last "port" (Virtual
707 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
710 static void adapter_down(struct adapter *adapter)
713 * Free interrupt resources.
715 if (adapter->flags & USING_MSIX)
716 free_msix_queue_irqs(adapter);
718 free_irq(adapter->pdev->irq, adapter);
721 * Wait for NAPI handlers to finish.
727 * Start up a net device.
729 static int cxgb4vf_open(struct net_device *dev)
732 struct port_info *pi = netdev_priv(dev);
733 struct adapter *adapter = pi->adapter;
736 * If this is the first interface that we're opening on the "adapter",
737 * bring the "adapter" up now.
739 if (adapter->open_device_map == 0) {
740 err = adapter_up(adapter);
746 * Note that this interface is up and start everything up ...
748 netif_set_real_num_tx_queues(dev, pi->nqsets);
749 err = netif_set_real_num_rx_queues(dev, pi->nqsets);
752 err = link_start(dev);
756 netif_tx_start_all_queues(dev);
757 set_bit(pi->port_id, &adapter->open_device_map);
761 if (adapter->open_device_map == 0)
762 adapter_down(adapter);
767 * Shut down a net device. This routine is called "cxgb_close" in the PF
770 static int cxgb4vf_stop(struct net_device *dev)
772 struct port_info *pi = netdev_priv(dev);
773 struct adapter *adapter = pi->adapter;
775 netif_tx_stop_all_queues(dev);
776 netif_carrier_off(dev);
777 t4vf_enable_vi(adapter, pi->viid, false, false);
778 pi->link_cfg.link_ok = 0;
780 clear_bit(pi->port_id, &adapter->open_device_map);
781 if (adapter->open_device_map == 0)
782 adapter_down(adapter);
787 * Translate our basic statistics into the standard "ifconfig" statistics.
789 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
791 struct t4vf_port_stats stats;
792 struct port_info *pi = netdev2pinfo(dev);
793 struct adapter *adapter = pi->adapter;
794 struct net_device_stats *ns = &dev->stats;
797 spin_lock(&adapter->stats_lock);
798 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
799 spin_unlock(&adapter->stats_lock);
801 memset(ns, 0, sizeof(*ns));
805 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
806 stats.tx_ucast_bytes + stats.tx_offload_bytes);
807 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
808 stats.tx_ucast_frames + stats.tx_offload_frames);
809 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
810 stats.rx_ucast_bytes);
811 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
812 stats.rx_ucast_frames);
813 ns->multicast = stats.rx_mcast_frames;
814 ns->tx_errors = stats.tx_drop_frames;
815 ns->rx_errors = stats.rx_err_frames;
821 * Collect up to maxaddrs worth of a netdevice's unicast addresses, starting
822 * at a specified offset within the list, into an array of addrss pointers and
823 * return the number collected.
825 static inline unsigned int collect_netdev_uc_list_addrs(const struct net_device *dev,
828 unsigned int maxaddrs)
830 unsigned int index = 0;
831 unsigned int naddr = 0;
832 const struct netdev_hw_addr *ha;
834 for_each_dev_addr(dev, ha)
835 if (index++ >= offset) {
836 addr[naddr++] = ha->addr;
837 if (naddr >= maxaddrs)
844 * Collect up to maxaddrs worth of a netdevice's multicast addresses, starting
845 * at a specified offset within the list, into an array of addrss pointers and
846 * return the number collected.
848 static inline unsigned int collect_netdev_mc_list_addrs(const struct net_device *dev,
851 unsigned int maxaddrs)
853 unsigned int index = 0;
854 unsigned int naddr = 0;
855 const struct netdev_hw_addr *ha;
857 netdev_for_each_mc_addr(ha, dev)
858 if (index++ >= offset) {
859 addr[naddr++] = ha->addr;
860 if (naddr >= maxaddrs)
867 * Configure the exact and hash address filters to handle a port's multicast
868 * and secondary unicast MAC addresses.
870 static int set_addr_filters(const struct net_device *dev, bool sleep)
875 unsigned int offset, naddr;
878 const struct port_info *pi = netdev_priv(dev);
880 /* first do the secondary unicast addresses */
881 for (offset = 0; ; offset += naddr) {
882 naddr = collect_netdev_uc_list_addrs(dev, addr, offset,
887 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
888 naddr, addr, NULL, &uhash, sleep);
895 /* next set up the multicast addresses */
896 for (offset = 0; ; offset += naddr) {
897 naddr = collect_netdev_mc_list_addrs(dev, addr, offset,
902 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
903 naddr, addr, NULL, &mhash, sleep);
909 return t4vf_set_addr_hash(pi->adapter, pi->viid, uhash != 0,
910 uhash | mhash, sleep);
914 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
915 * If @mtu is -1 it is left unchanged.
917 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
920 struct port_info *pi = netdev_priv(dev);
922 ret = set_addr_filters(dev, sleep_ok);
924 ret = t4vf_set_rxmode(pi->adapter, pi->viid, -1,
925 (dev->flags & IFF_PROMISC) != 0,
926 (dev->flags & IFF_ALLMULTI) != 0,
932 * Set the current receive modes on the device.
934 static void cxgb4vf_set_rxmode(struct net_device *dev)
936 /* unfortunately we can't return errors to the stack */
937 set_rxmode(dev, -1, false);
941 * Find the entry in the interrupt holdoff timer value array which comes
942 * closest to the specified interrupt holdoff value.
944 static int closest_timer(const struct sge *s, int us)
946 int i, timer_idx = 0, min_delta = INT_MAX;
948 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
949 int delta = us - s->timer_val[i];
952 if (delta < min_delta) {
960 static int closest_thres(const struct sge *s, int thres)
962 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
964 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
965 delta = thres - s->counter_val[i];
968 if (delta < min_delta) {
977 * Return a queue's interrupt hold-off time in us. 0 means no timer.
979 static unsigned int qtimer_val(const struct adapter *adapter,
980 const struct sge_rspq *rspq)
982 unsigned int timer_idx = QINTR_TIMER_IDX_GET(rspq->intr_params);
984 return timer_idx < SGE_NTIMERS
985 ? adapter->sge.timer_val[timer_idx]
990 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
991 * @adapter: the adapter
992 * @rspq: the RX response queue
993 * @us: the hold-off time in us, or 0 to disable timer
994 * @cnt: the hold-off packet count, or 0 to disable counter
996 * Sets an RX response queue's interrupt hold-off time and packet count.
997 * At least one of the two needs to be enabled for the queue to generate
1000 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
1001 unsigned int us, unsigned int cnt)
1003 unsigned int timer_idx;
1006 * If both the interrupt holdoff timer and count are specified as
1007 * zero, default to a holdoff count of 1 ...
1009 if ((us | cnt) == 0)
1013 * If an interrupt holdoff count has been specified, then find the
1014 * closest configured holdoff count and use that. If the response
1015 * queue has already been created, then update its queue context
1022 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1023 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1024 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1026 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1027 FW_PARAMS_PARAM_YZ(rspq->cntxt_id);
1028 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1032 rspq->pktcnt_idx = pktcnt_idx;
1036 * Compute the closest holdoff timer index from the supplied holdoff
1039 timer_idx = (us == 0
1040 ? SGE_TIMER_RSTRT_CNTR
1041 : closest_timer(&adapter->sge, us));
1044 * Update the response queue's interrupt coalescing parameters and
1047 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
1048 (cnt > 0 ? QINTR_CNT_EN : 0));
1053 * Return a version number to identify the type of adapter. The scheme is:
1054 * - bits 0..9: chip version
1055 * - bits 10..15: chip revision
1057 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1060 * Chip version 4, revision 0x3f (cxgb4vf).
1062 return 4 | (0x3f << 10);
1066 * Execute the specified ioctl command.
1068 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1074 * The VF Driver doesn't have access to any of the other
1075 * common Ethernet device ioctl()'s (like reading/writing
1076 * PHY registers, etc.
1087 * Change the device's MTU.
1089 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1092 struct port_info *pi = netdev_priv(dev);
1094 /* accommodate SACK */
1098 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1099 -1, -1, -1, -1, true);
1106 * Change the devices MAC address.
1108 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1111 struct sockaddr *addr = _addr;
1112 struct port_info *pi = netdev_priv(dev);
1114 if (!is_valid_ether_addr(addr->sa_data))
1117 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1118 addr->sa_data, true);
1122 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1123 pi->xact_addr_filt = ret;
1127 #ifdef CONFIG_NET_POLL_CONTROLLER
1129 * Poll all of our receive queues. This is called outside of normal interrupt
1132 static void cxgb4vf_poll_controller(struct net_device *dev)
1134 struct port_info *pi = netdev_priv(dev);
1135 struct adapter *adapter = pi->adapter;
1137 if (adapter->flags & USING_MSIX) {
1138 struct sge_eth_rxq *rxq;
1141 rxq = &adapter->sge.ethrxq[pi->first_qset];
1142 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1143 t4vf_sge_intr_msix(0, &rxq->rspq);
1147 t4vf_intr_handler(adapter)(0, adapter);
1152 * Ethtool operations.
1153 * ===================
1155 * Note that we don't support any ethtool operations which change the physical
1156 * state of the port to which we're linked.
1160 * Return current port link settings.
1162 static int cxgb4vf_get_settings(struct net_device *dev,
1163 struct ethtool_cmd *cmd)
1165 const struct port_info *pi = netdev_priv(dev);
1167 cmd->supported = pi->link_cfg.supported;
1168 cmd->advertising = pi->link_cfg.advertising;
1169 ethtool_cmd_speed_set(cmd,
1170 netif_carrier_ok(dev) ? pi->link_cfg.speed : -1);
1171 cmd->duplex = DUPLEX_FULL;
1173 cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1174 cmd->phy_address = pi->port_id;
1175 cmd->transceiver = XCVR_EXTERNAL;
1176 cmd->autoneg = pi->link_cfg.autoneg;
1183 * Return our driver information.
1185 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1186 struct ethtool_drvinfo *drvinfo)
1188 struct adapter *adapter = netdev2adap(dev);
1190 strcpy(drvinfo->driver, KBUILD_MODNAME);
1191 strcpy(drvinfo->version, DRV_VERSION);
1192 strcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)));
1193 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1194 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1195 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.fwrev),
1196 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.fwrev),
1197 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.fwrev),
1198 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.fwrev),
1199 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.tprev),
1200 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.tprev),
1201 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.tprev),
1202 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.tprev));
1206 * Return current adapter message level.
1208 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1210 return netdev2adap(dev)->msg_enable;
1214 * Set current adapter message level.
1216 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1218 netdev2adap(dev)->msg_enable = msglevel;
1222 * Return the device's current Queue Set ring size parameters along with the
1223 * allowed maximum values. Since ethtool doesn't understand the concept of
1224 * multi-queue devices, we just return the current values associated with the
1227 static void cxgb4vf_get_ringparam(struct net_device *dev,
1228 struct ethtool_ringparam *rp)
1230 const struct port_info *pi = netdev_priv(dev);
1231 const struct sge *s = &pi->adapter->sge;
1233 rp->rx_max_pending = MAX_RX_BUFFERS;
1234 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1235 rp->rx_jumbo_max_pending = 0;
1236 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1238 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1239 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1240 rp->rx_jumbo_pending = 0;
1241 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1245 * Set the Queue Set ring size parameters for the device. Again, since
1246 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1247 * apply these new values across all of the Queue Sets associated with the
1248 * device -- after vetting them of course!
1250 static int cxgb4vf_set_ringparam(struct net_device *dev,
1251 struct ethtool_ringparam *rp)
1253 const struct port_info *pi = netdev_priv(dev);
1254 struct adapter *adapter = pi->adapter;
1255 struct sge *s = &adapter->sge;
1258 if (rp->rx_pending > MAX_RX_BUFFERS ||
1259 rp->rx_jumbo_pending ||
1260 rp->tx_pending > MAX_TXQ_ENTRIES ||
1261 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1262 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1263 rp->rx_pending < MIN_FL_ENTRIES ||
1264 rp->tx_pending < MIN_TXQ_ENTRIES)
1267 if (adapter->flags & FULL_INIT_DONE)
1270 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1271 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1272 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1273 s->ethtxq[qs].q.size = rp->tx_pending;
1279 * Return the interrupt holdoff timer and count for the first Queue Set on the
1280 * device. Our extension ioctl() (the cxgbtool interface) allows the
1281 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1283 static int cxgb4vf_get_coalesce(struct net_device *dev,
1284 struct ethtool_coalesce *coalesce)
1286 const struct port_info *pi = netdev_priv(dev);
1287 const struct adapter *adapter = pi->adapter;
1288 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1290 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1291 coalesce->rx_max_coalesced_frames =
1292 ((rspq->intr_params & QINTR_CNT_EN)
1293 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1299 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1300 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1301 * the interrupt holdoff timer on any of the device's Queue Sets.
1303 static int cxgb4vf_set_coalesce(struct net_device *dev,
1304 struct ethtool_coalesce *coalesce)
1306 const struct port_info *pi = netdev_priv(dev);
1307 struct adapter *adapter = pi->adapter;
1309 return set_rxq_intr_params(adapter,
1310 &adapter->sge.ethrxq[pi->first_qset].rspq,
1311 coalesce->rx_coalesce_usecs,
1312 coalesce->rx_max_coalesced_frames);
1316 * Report current port link pause parameter settings.
1318 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1319 struct ethtool_pauseparam *pauseparam)
1321 struct port_info *pi = netdev_priv(dev);
1323 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1324 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1325 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1329 * Identify the port by blinking the port's LED.
1331 static int cxgb4vf_phys_id(struct net_device *dev,
1332 enum ethtool_phys_id_state state)
1335 struct port_info *pi = netdev_priv(dev);
1337 if (state == ETHTOOL_ID_ACTIVE)
1339 else if (state == ETHTOOL_ID_INACTIVE)
1344 return t4vf_identify_port(pi->adapter, pi->viid, val);
1348 * Port stats maintained per queue of the port.
1350 struct queue_port_stats {
1361 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1362 * these need to match the order of statistics returned by
1363 * t4vf_get_port_stats().
1365 static const char stats_strings[][ETH_GSTRING_LEN] = {
1367 * These must match the layout of the t4vf_port_stats structure.
1369 "TxBroadcastBytes ",
1370 "TxBroadcastFrames ",
1371 "TxMulticastBytes ",
1372 "TxMulticastFrames ",
1378 "RxBroadcastBytes ",
1379 "RxBroadcastFrames ",
1380 "RxMulticastBytes ",
1381 "RxMulticastFrames ",
1387 * These are accumulated per-queue statistics and must match the
1388 * order of the fields in the queue_port_stats structure.
1400 * Return the number of statistics in the specified statistics set.
1402 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1406 return ARRAY_SIZE(stats_strings);
1414 * Return the strings for the specified statistics set.
1416 static void cxgb4vf_get_strings(struct net_device *dev,
1422 memcpy(data, stats_strings, sizeof(stats_strings));
1428 * Small utility routine to accumulate queue statistics across the queues of
1431 static void collect_sge_port_stats(const struct adapter *adapter,
1432 const struct port_info *pi,
1433 struct queue_port_stats *stats)
1435 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1436 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1439 memset(stats, 0, sizeof(*stats));
1440 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1441 stats->tso += txq->tso;
1442 stats->tx_csum += txq->tx_cso;
1443 stats->rx_csum += rxq->stats.rx_cso;
1444 stats->vlan_ex += rxq->stats.vlan_ex;
1445 stats->vlan_ins += txq->vlan_ins;
1446 stats->lro_pkts += rxq->stats.lro_pkts;
1447 stats->lro_merged += rxq->stats.lro_merged;
1452 * Return the ETH_SS_STATS statistics set.
1454 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1455 struct ethtool_stats *stats,
1458 struct port_info *pi = netdev2pinfo(dev);
1459 struct adapter *adapter = pi->adapter;
1460 int err = t4vf_get_port_stats(adapter, pi->pidx,
1461 (struct t4vf_port_stats *)data);
1463 memset(data, 0, sizeof(struct t4vf_port_stats));
1465 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1466 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1470 * Return the size of our register map.
1472 static int cxgb4vf_get_regs_len(struct net_device *dev)
1474 return T4VF_REGMAP_SIZE;
1478 * Dump a block of registers, start to end inclusive, into a buffer.
1480 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1481 unsigned int start, unsigned int end)
1483 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1485 for ( ; start <= end; start += sizeof(u32)) {
1487 * Avoid reading the Mailbox Control register since that
1488 * can trigger a Mailbox Ownership Arbitration cycle and
1489 * interfere with communication with the firmware.
1491 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1494 *bp++ = t4_read_reg(adapter, start);
1499 * Copy our entire register map into the provided buffer.
1501 static void cxgb4vf_get_regs(struct net_device *dev,
1502 struct ethtool_regs *regs,
1505 struct adapter *adapter = netdev2adap(dev);
1507 regs->version = mk_adap_vers(adapter);
1510 * Fill in register buffer with our register map.
1512 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1514 reg_block_dump(adapter, regbuf,
1515 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1516 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1517 reg_block_dump(adapter, regbuf,
1518 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1519 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1520 reg_block_dump(adapter, regbuf,
1521 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1522 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_LAST);
1523 reg_block_dump(adapter, regbuf,
1524 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1525 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1527 reg_block_dump(adapter, regbuf,
1528 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1529 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1533 * Report current Wake On LAN settings.
1535 static void cxgb4vf_get_wol(struct net_device *dev,
1536 struct ethtool_wolinfo *wol)
1540 memset(&wol->sopass, 0, sizeof(wol->sopass));
1544 * TCP Segmentation Offload flags which we support.
1546 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1548 static struct ethtool_ops cxgb4vf_ethtool_ops = {
1549 .get_settings = cxgb4vf_get_settings,
1550 .get_drvinfo = cxgb4vf_get_drvinfo,
1551 .get_msglevel = cxgb4vf_get_msglevel,
1552 .set_msglevel = cxgb4vf_set_msglevel,
1553 .get_ringparam = cxgb4vf_get_ringparam,
1554 .set_ringparam = cxgb4vf_set_ringparam,
1555 .get_coalesce = cxgb4vf_get_coalesce,
1556 .set_coalesce = cxgb4vf_set_coalesce,
1557 .get_pauseparam = cxgb4vf_get_pauseparam,
1558 .get_link = ethtool_op_get_link,
1559 .get_strings = cxgb4vf_get_strings,
1560 .set_phys_id = cxgb4vf_phys_id,
1561 .get_sset_count = cxgb4vf_get_sset_count,
1562 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1563 .get_regs_len = cxgb4vf_get_regs_len,
1564 .get_regs = cxgb4vf_get_regs,
1565 .get_wol = cxgb4vf_get_wol,
1569 * /sys/kernel/debug/cxgb4vf support code and data.
1570 * ================================================
1574 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1578 static int sge_qinfo_show(struct seq_file *seq, void *v)
1580 struct adapter *adapter = seq->private;
1581 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1582 int qs, r = (uintptr_t)v - 1;
1585 seq_putc(seq, '\n');
1587 #define S3(fmt_spec, s, v) \
1589 seq_printf(seq, "%-12s", s); \
1590 for (qs = 0; qs < n; ++qs) \
1591 seq_printf(seq, " %16" fmt_spec, v); \
1592 seq_putc(seq, '\n'); \
1594 #define S(s, v) S3("s", s, v)
1595 #define T(s, v) S3("u", s, txq[qs].v)
1596 #define R(s, v) S3("u", s, rxq[qs].v)
1598 if (r < eth_entries) {
1599 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1600 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1601 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1603 S("QType:", "Ethernet");
1605 (rxq[qs].rspq.netdev
1606 ? rxq[qs].rspq.netdev->name
1609 (rxq[qs].rspq.netdev
1610 ? ((struct port_info *)
1611 netdev_priv(rxq[qs].rspq.netdev))->port_id
1613 T("TxQ ID:", q.abs_id);
1614 T("TxQ size:", q.size);
1615 T("TxQ inuse:", q.in_use);
1616 T("TxQ PIdx:", q.pidx);
1617 T("TxQ CIdx:", q.cidx);
1618 R("RspQ ID:", rspq.abs_id);
1619 R("RspQ size:", rspq.size);
1620 R("RspQE size:", rspq.iqe_len);
1621 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1622 S3("u", "Intr pktcnt:",
1623 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1624 R("RspQ CIdx:", rspq.cidx);
1625 R("RspQ Gen:", rspq.gen);
1626 R("FL ID:", fl.abs_id);
1627 R("FL size:", fl.size - MIN_FL_RESID);
1628 R("FL avail:", fl.avail);
1629 R("FL PIdx:", fl.pidx);
1630 R("FL CIdx:", fl.cidx);
1636 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1638 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1639 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1640 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1641 qtimer_val(adapter, evtq));
1642 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1643 adapter->sge.counter_val[evtq->pktcnt_idx]);
1644 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1645 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1646 } else if (r == 1) {
1647 const struct sge_rspq *intrq = &adapter->sge.intrq;
1649 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1650 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1651 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1652 qtimer_val(adapter, intrq));
1653 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1654 adapter->sge.counter_val[intrq->pktcnt_idx]);
1655 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1656 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1668 * Return the number of "entries" in our "file". We group the multi-Queue
1669 * sections with QPL Queue Sets per "entry". The sections of the output are:
1671 * Ethernet RX/TX Queue Sets
1672 * Firmware Event Queue
1673 * Forwarded Interrupt Queue (if in MSI mode)
1675 static int sge_queue_entries(const struct adapter *adapter)
1677 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1678 ((adapter->flags & USING_MSI) != 0);
1681 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1683 int entries = sge_queue_entries(seq->private);
1685 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1688 static void sge_queue_stop(struct seq_file *seq, void *v)
1692 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
1694 int entries = sge_queue_entries(seq->private);
1697 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1700 static const struct seq_operations sge_qinfo_seq_ops = {
1701 .start = sge_queue_start,
1702 .next = sge_queue_next,
1703 .stop = sge_queue_stop,
1704 .show = sge_qinfo_show
1707 static int sge_qinfo_open(struct inode *inode, struct file *file)
1709 int res = seq_open(file, &sge_qinfo_seq_ops);
1712 struct seq_file *seq = file->private_data;
1713 seq->private = inode->i_private;
1718 static const struct file_operations sge_qinfo_debugfs_fops = {
1719 .owner = THIS_MODULE,
1720 .open = sge_qinfo_open,
1722 .llseek = seq_lseek,
1723 .release = seq_release,
1727 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
1731 static int sge_qstats_show(struct seq_file *seq, void *v)
1733 struct adapter *adapter = seq->private;
1734 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1735 int qs, r = (uintptr_t)v - 1;
1738 seq_putc(seq, '\n');
1740 #define S3(fmt, s, v) \
1742 seq_printf(seq, "%-16s", s); \
1743 for (qs = 0; qs < n; ++qs) \
1744 seq_printf(seq, " %8" fmt, v); \
1745 seq_putc(seq, '\n'); \
1747 #define S(s, v) S3("s", s, v)
1749 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
1750 #define T(s, v) T3("lu", s, v)
1752 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
1753 #define R(s, v) R3("lu", s, v)
1755 if (r < eth_entries) {
1756 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1757 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1758 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1760 S("QType:", "Ethernet");
1762 (rxq[qs].rspq.netdev
1763 ? rxq[qs].rspq.netdev->name
1765 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
1766 R("RxPackets:", stats.pkts);
1767 R("RxCSO:", stats.rx_cso);
1768 R("VLANxtract:", stats.vlan_ex);
1769 R("LROmerged:", stats.lro_merged);
1770 R("LROpackets:", stats.lro_pkts);
1771 R("RxDrops:", stats.rx_drops);
1773 T("TxCSO:", tx_cso);
1774 T("VLANins:", vlan_ins);
1775 T("TxQFull:", q.stops);
1776 T("TxQRestarts:", q.restarts);
1777 T("TxMapErr:", mapping_err);
1778 R("FLAllocErr:", fl.alloc_failed);
1779 R("FLLrgAlcErr:", fl.large_alloc_failed);
1780 R("FLStarving:", fl.starving);
1786 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1788 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
1789 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1790 evtq->unhandled_irqs);
1791 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
1792 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
1793 } else if (r == 1) {
1794 const struct sge_rspq *intrq = &adapter->sge.intrq;
1796 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
1797 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1798 intrq->unhandled_irqs);
1799 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
1800 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
1814 * Return the number of "entries" in our "file". We group the multi-Queue
1815 * sections with QPL Queue Sets per "entry". The sections of the output are:
1817 * Ethernet RX/TX Queue Sets
1818 * Firmware Event Queue
1819 * Forwarded Interrupt Queue (if in MSI mode)
1821 static int sge_qstats_entries(const struct adapter *adapter)
1823 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1824 ((adapter->flags & USING_MSI) != 0);
1827 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
1829 int entries = sge_qstats_entries(seq->private);
1831 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1834 static void sge_qstats_stop(struct seq_file *seq, void *v)
1838 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
1840 int entries = sge_qstats_entries(seq->private);
1843 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1846 static const struct seq_operations sge_qstats_seq_ops = {
1847 .start = sge_qstats_start,
1848 .next = sge_qstats_next,
1849 .stop = sge_qstats_stop,
1850 .show = sge_qstats_show
1853 static int sge_qstats_open(struct inode *inode, struct file *file)
1855 int res = seq_open(file, &sge_qstats_seq_ops);
1858 struct seq_file *seq = file->private_data;
1859 seq->private = inode->i_private;
1864 static const struct file_operations sge_qstats_proc_fops = {
1865 .owner = THIS_MODULE,
1866 .open = sge_qstats_open,
1868 .llseek = seq_lseek,
1869 .release = seq_release,
1873 * Show PCI-E SR-IOV Virtual Function Resource Limits.
1875 static int resources_show(struct seq_file *seq, void *v)
1877 struct adapter *adapter = seq->private;
1878 struct vf_resources *vfres = &adapter->params.vfres;
1880 #define S(desc, fmt, var) \
1881 seq_printf(seq, "%-60s " fmt "\n", \
1882 desc " (" #var "):", vfres->var)
1884 S("Virtual Interfaces", "%d", nvi);
1885 S("Egress Queues", "%d", neq);
1886 S("Ethernet Control", "%d", nethctrl);
1887 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
1888 S("Ingress Queues", "%d", niq);
1889 S("Traffic Class", "%d", tc);
1890 S("Port Access Rights Mask", "%#x", pmask);
1891 S("MAC Address Filters", "%d", nexactf);
1892 S("Firmware Command Read Capabilities", "%#x", r_caps);
1893 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
1900 static int resources_open(struct inode *inode, struct file *file)
1902 return single_open(file, resources_show, inode->i_private);
1905 static const struct file_operations resources_proc_fops = {
1906 .owner = THIS_MODULE,
1907 .open = resources_open,
1909 .llseek = seq_lseek,
1910 .release = single_release,
1914 * Show Virtual Interfaces.
1916 static int interfaces_show(struct seq_file *seq, void *v)
1918 if (v == SEQ_START_TOKEN) {
1919 seq_puts(seq, "Interface Port VIID\n");
1921 struct adapter *adapter = seq->private;
1922 int pidx = (uintptr_t)v - 2;
1923 struct net_device *dev = adapter->port[pidx];
1924 struct port_info *pi = netdev_priv(dev);
1926 seq_printf(seq, "%9s %4d %#5x\n",
1927 dev->name, pi->port_id, pi->viid);
1932 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
1934 return pos <= adapter->params.nports
1935 ? (void *)(uintptr_t)(pos + 1)
1939 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
1942 ? interfaces_get_idx(seq->private, *pos)
1946 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
1949 return interfaces_get_idx(seq->private, *pos);
1952 static void interfaces_stop(struct seq_file *seq, void *v)
1956 static const struct seq_operations interfaces_seq_ops = {
1957 .start = interfaces_start,
1958 .next = interfaces_next,
1959 .stop = interfaces_stop,
1960 .show = interfaces_show
1963 static int interfaces_open(struct inode *inode, struct file *file)
1965 int res = seq_open(file, &interfaces_seq_ops);
1968 struct seq_file *seq = file->private_data;
1969 seq->private = inode->i_private;
1974 static const struct file_operations interfaces_proc_fops = {
1975 .owner = THIS_MODULE,
1976 .open = interfaces_open,
1978 .llseek = seq_lseek,
1979 .release = seq_release,
1983 * /sys/kernel/debugfs/cxgb4vf/ files list.
1985 struct cxgb4vf_debugfs_entry {
1986 const char *name; /* name of debugfs node */
1987 mode_t mode; /* file system mode */
1988 const struct file_operations *fops;
1991 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
1992 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
1993 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
1994 { "resources", S_IRUGO, &resources_proc_fops },
1995 { "interfaces", S_IRUGO, &interfaces_proc_fops },
1999 * Module and device initialization and cleanup code.
2000 * ==================================================
2004 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2005 * directory (debugfs_root) has already been set up.
2007 static int __devinit setup_debugfs(struct adapter *adapter)
2011 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2014 * Debugfs support is best effort.
2016 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2017 (void)debugfs_create_file(debugfs_files[i].name,
2018 debugfs_files[i].mode,
2019 adapter->debugfs_root,
2021 debugfs_files[i].fops);
2027 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2028 * it to our caller to tear down the directory (debugfs_root).
2030 static void cleanup_debugfs(struct adapter *adapter)
2032 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2035 * Unlike our sister routine cleanup_proc(), we don't need to remove
2036 * individual entries because a call will be made to
2037 * debugfs_remove_recursive(). We just need to clean up any ancillary
2044 * Perform early "adapter" initialization. This is where we discover what
2045 * adapter parameters we're going to be using and initialize basic adapter
2048 static int __devinit adap_init0(struct adapter *adapter)
2050 struct vf_resources *vfres = &adapter->params.vfres;
2051 struct sge_params *sge_params = &adapter->params.sge;
2052 struct sge *s = &adapter->sge;
2053 unsigned int ethqsets;
2057 * Wait for the device to become ready before proceeding ...
2059 err = t4vf_wait_dev_ready(adapter);
2061 dev_err(adapter->pdev_dev, "device didn't become ready:"
2067 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2068 * 2.6.31 and later we can't call pci_reset_function() in order to
2069 * issue an FLR because of a self- deadlock on the device semaphore.
2070 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2071 * cases where they're needed -- for instance, some versions of KVM
2072 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2073 * use the firmware based reset in order to reset any per function
2076 err = t4vf_fw_reset(adapter);
2078 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2083 * Grab basic operational parameters. These will predominantly have
2084 * been set up by the Physical Function Driver or will be hard coded
2085 * into the adapter. We just have to live with them ... Note that
2086 * we _must_ get our VPD parameters before our SGE parameters because
2087 * we need to know the adapter's core clock from the VPD in order to
2088 * properly decode the SGE Timer Values.
2090 err = t4vf_get_dev_params(adapter);
2092 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2093 " device parameters: err=%d\n", err);
2096 err = t4vf_get_vpd_params(adapter);
2098 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2099 " VPD parameters: err=%d\n", err);
2102 err = t4vf_get_sge_params(adapter);
2104 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2105 " SGE parameters: err=%d\n", err);
2108 err = t4vf_get_rss_glb_config(adapter);
2110 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2111 " RSS parameters: err=%d\n", err);
2114 if (adapter->params.rss.mode !=
2115 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2116 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2117 " mode %d\n", adapter->params.rss.mode);
2120 err = t4vf_sge_init(adapter);
2122 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2128 * Retrieve our RX interrupt holdoff timer values and counter
2129 * threshold values from the SGE parameters.
2131 s->timer_val[0] = core_ticks_to_us(adapter,
2132 TIMERVALUE0_GET(sge_params->sge_timer_value_0_and_1));
2133 s->timer_val[1] = core_ticks_to_us(adapter,
2134 TIMERVALUE1_GET(sge_params->sge_timer_value_0_and_1));
2135 s->timer_val[2] = core_ticks_to_us(adapter,
2136 TIMERVALUE0_GET(sge_params->sge_timer_value_2_and_3));
2137 s->timer_val[3] = core_ticks_to_us(adapter,
2138 TIMERVALUE1_GET(sge_params->sge_timer_value_2_and_3));
2139 s->timer_val[4] = core_ticks_to_us(adapter,
2140 TIMERVALUE0_GET(sge_params->sge_timer_value_4_and_5));
2141 s->timer_val[5] = core_ticks_to_us(adapter,
2142 TIMERVALUE1_GET(sge_params->sge_timer_value_4_and_5));
2145 THRESHOLD_0_GET(sge_params->sge_ingress_rx_threshold);
2147 THRESHOLD_1_GET(sge_params->sge_ingress_rx_threshold);
2149 THRESHOLD_2_GET(sge_params->sge_ingress_rx_threshold);
2151 THRESHOLD_3_GET(sge_params->sge_ingress_rx_threshold);
2154 * Grab our Virtual Interface resource allocation, extract the
2155 * features that we're interested in and do a bit of sanity testing on
2158 err = t4vf_get_vfres(adapter);
2160 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2161 " resources: err=%d\n", err);
2166 * The number of "ports" which we support is equal to the number of
2167 * Virtual Interfaces with which we've been provisioned.
2169 adapter->params.nports = vfres->nvi;
2170 if (adapter->params.nports > MAX_NPORTS) {
2171 dev_warn(adapter->pdev_dev, "only using %d of %d allowed"
2172 " virtual interfaces\n", MAX_NPORTS,
2173 adapter->params.nports);
2174 adapter->params.nports = MAX_NPORTS;
2178 * We need to reserve a number of the ingress queues with Free List
2179 * and Interrupt capabilities for special interrupt purposes (like
2180 * asynchronous firmware messages, or forwarded interrupts if we're
2181 * using MSI). The rest of the FL/Intr-capable ingress queues will be
2182 * matched up one-for-one with Ethernet/Control egress queues in order
2183 * to form "Queue Sets" which will be aportioned between the "ports".
2184 * For each Queue Set, we'll need the ability to allocate two Egress
2185 * Contexts -- one for the Ingress Queue Free List and one for the TX
2188 ethqsets = vfres->niqflint - INGQ_EXTRAS;
2189 if (vfres->nethctrl != ethqsets) {
2190 dev_warn(adapter->pdev_dev, "unequal number of [available]"
2191 " ingress/egress queues (%d/%d); using minimum for"
2192 " number of Queue Sets\n", ethqsets, vfres->nethctrl);
2193 ethqsets = min(vfres->nethctrl, ethqsets);
2195 if (vfres->neq < ethqsets*2) {
2196 dev_warn(adapter->pdev_dev, "Not enough Egress Contexts (%d)"
2197 " to support Queue Sets (%d); reducing allowed Queue"
2198 " Sets\n", vfres->neq, ethqsets);
2199 ethqsets = vfres->neq/2;
2201 if (ethqsets > MAX_ETH_QSETS) {
2202 dev_warn(adapter->pdev_dev, "only using %d of %d allowed Queue"
2203 " Sets\n", MAX_ETH_QSETS, adapter->sge.max_ethqsets);
2204 ethqsets = MAX_ETH_QSETS;
2206 if (vfres->niq != 0 || vfres->neq > ethqsets*2) {
2207 dev_warn(adapter->pdev_dev, "unused resources niq/neq (%d/%d)"
2208 " ignored\n", vfres->niq, vfres->neq - ethqsets*2);
2210 adapter->sge.max_ethqsets = ethqsets;
2213 * Check for various parameter sanity issues. Most checks simply
2214 * result in us using fewer resources than our provissioning but we
2215 * do need at least one "port" with which to work ...
2217 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2218 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2219 " virtual interfaces (too few Queue Sets)\n",
2220 adapter->sge.max_ethqsets, adapter->params.nports);
2221 adapter->params.nports = adapter->sge.max_ethqsets;
2223 if (adapter->params.nports == 0) {
2224 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2231 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2232 u8 pkt_cnt_idx, unsigned int size,
2233 unsigned int iqe_size)
2235 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
2236 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0));
2237 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2240 rspq->iqe_len = iqe_size;
2245 * Perform default configuration of DMA queues depending on the number and
2246 * type of ports we found and the number of available CPUs. Most settings can
2247 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2248 * being brought up for the first time.
2250 static void __devinit cfg_queues(struct adapter *adapter)
2252 struct sge *s = &adapter->sge;
2253 int q10g, n10g, qidx, pidx, qs;
2257 * We should not be called till we know how many Queue Sets we can
2258 * support. In particular, this means that we need to know what kind
2259 * of interrupts we'll be using ...
2261 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2264 * Count the number of 10GbE Virtual Interfaces that we have.
2267 for_each_port(adapter, pidx)
2268 n10g += is_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2271 * We default to 1 queue per non-10G port and up to # of cores queues
2277 int n1g = (adapter->params.nports - n10g);
2278 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2279 if (q10g > num_online_cpus())
2280 q10g = num_online_cpus();
2284 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2285 * The layout will be established in setup_sge_queues() when the
2286 * adapter is brough up for the first time.
2289 for_each_port(adapter, pidx) {
2290 struct port_info *pi = adap2pinfo(adapter, pidx);
2292 pi->first_qset = qidx;
2293 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
2299 * The Ingress Queue Entry Size for our various Response Queues needs
2300 * to be big enough to accommodate the largest message we can receive
2301 * from the chip/firmware; which is 64 bytes ...
2306 * Set up default Queue Set parameters ... Start off with the
2307 * shortest interrupt holdoff timer.
2309 for (qs = 0; qs < s->max_ethqsets; qs++) {
2310 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2311 struct sge_eth_txq *txq = &s->ethtxq[qs];
2313 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2319 * The firmware event queue is used for link state changes and
2320 * notifications of TX DMA completions.
2322 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2325 * The forwarded interrupt queue is used when we're in MSI interrupt
2326 * mode. In this mode all interrupts associated with RX queues will
2327 * be forwarded to a single queue which we'll associate with our MSI
2328 * interrupt vector. The messages dropped in the forwarded interrupt
2329 * queue will indicate which ingress queue needs servicing ... This
2330 * queue needs to be large enough to accommodate all of the ingress
2331 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2332 * from equalling the CIDX if every ingress queue has an outstanding
2333 * interrupt). The queue doesn't need to be any larger because no
2334 * ingress queue will ever have more than one outstanding interrupt at
2337 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2342 * Reduce the number of Ethernet queues across all ports to at most n.
2343 * n provides at least one queue per port.
2345 static void __devinit reduce_ethqs(struct adapter *adapter, int n)
2348 struct port_info *pi;
2351 * While we have too many active Ether Queue Sets, interate across the
2352 * "ports" and reduce their individual Queue Set allocations.
2354 BUG_ON(n < adapter->params.nports);
2355 while (n < adapter->sge.ethqsets)
2356 for_each_port(adapter, i) {
2357 pi = adap2pinfo(adapter, i);
2358 if (pi->nqsets > 1) {
2360 adapter->sge.ethqsets--;
2361 if (adapter->sge.ethqsets <= n)
2367 * Reassign the starting Queue Sets for each of the "ports" ...
2370 for_each_port(adapter, i) {
2371 pi = adap2pinfo(adapter, i);
2378 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2379 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2380 * need. Minimally we need one for every Virtual Interface plus those needed
2381 * for our "extras". Note that this process may lower the maximum number of
2382 * allowed Queue Sets ...
2384 static int __devinit enable_msix(struct adapter *adapter)
2386 int i, err, want, need;
2387 struct msix_entry entries[MSIX_ENTRIES];
2388 struct sge *s = &adapter->sge;
2390 for (i = 0; i < MSIX_ENTRIES; ++i)
2391 entries[i].entry = i;
2394 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2395 * plus those needed for our "extras" (for example, the firmware
2396 * message queue). We _need_ at least one "Queue Set" per Virtual
2397 * Interface plus those needed for our "extras". So now we get to see
2398 * if the song is right ...
2400 want = s->max_ethqsets + MSIX_EXTRAS;
2401 need = adapter->params.nports + MSIX_EXTRAS;
2402 while ((err = pci_enable_msix(adapter->pdev, entries, want)) >= need)
2406 int nqsets = want - MSIX_EXTRAS;
2407 if (nqsets < s->max_ethqsets) {
2408 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2409 " for %d Queue Sets\n", nqsets);
2410 s->max_ethqsets = nqsets;
2411 if (nqsets < s->ethqsets)
2412 reduce_ethqs(adapter, nqsets);
2414 for (i = 0; i < want; ++i)
2415 adapter->msix_info[i].vec = entries[i].vector;
2416 } else if (err > 0) {
2417 pci_disable_msix(adapter->pdev);
2418 dev_info(adapter->pdev_dev, "only %d MSI-X vectors left,"
2419 " not using MSI-X\n", err);
2424 static const struct net_device_ops cxgb4vf_netdev_ops = {
2425 .ndo_open = cxgb4vf_open,
2426 .ndo_stop = cxgb4vf_stop,
2427 .ndo_start_xmit = t4vf_eth_xmit,
2428 .ndo_get_stats = cxgb4vf_get_stats,
2429 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2430 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2431 .ndo_validate_addr = eth_validate_addr,
2432 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2433 .ndo_change_mtu = cxgb4vf_change_mtu,
2434 .ndo_vlan_rx_register = cxgb4vf_vlan_rx_register,
2435 #ifdef CONFIG_NET_POLL_CONTROLLER
2436 .ndo_poll_controller = cxgb4vf_poll_controller,
2441 * "Probe" a device: initialize a device and construct all kernel and driver
2442 * state needed to manage the device. This routine is called "init_one" in
2445 static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
2446 const struct pci_device_id *ent)
2448 static int version_printed;
2453 struct adapter *adapter;
2454 struct port_info *pi;
2455 struct net_device *netdev;
2458 * Print our driver banner the first time we're called to initialize a
2461 if (version_printed == 0) {
2462 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
2463 version_printed = 1;
2467 * Initialize generic PCI device state.
2469 err = pci_enable_device(pdev);
2471 dev_err(&pdev->dev, "cannot enable PCI device\n");
2476 * Reserve PCI resources for the device. If we can't get them some
2477 * other driver may have already claimed the device ...
2479 err = pci_request_regions(pdev, KBUILD_MODNAME);
2481 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2482 goto err_disable_device;
2486 * Set up our DMA mask: try for 64-bit address masking first and
2487 * fall back to 32-bit if we can't get 64 bits ...
2489 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2491 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2493 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2494 " coherent allocations\n");
2495 goto err_release_regions;
2499 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2501 dev_err(&pdev->dev, "no usable DMA configuration\n");
2502 goto err_release_regions;
2508 * Enable bus mastering for the device ...
2510 pci_set_master(pdev);
2513 * Allocate our adapter data structure and attach it to the device.
2515 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2518 goto err_release_regions;
2520 pci_set_drvdata(pdev, adapter);
2521 adapter->pdev = pdev;
2522 adapter->pdev_dev = &pdev->dev;
2525 * Initialize SMP data synchronization resources.
2527 spin_lock_init(&adapter->stats_lock);
2530 * Map our I/O registers in BAR0.
2532 adapter->regs = pci_ioremap_bar(pdev, 0);
2533 if (!adapter->regs) {
2534 dev_err(&pdev->dev, "cannot map device registers\n");
2536 goto err_free_adapter;
2540 * Initialize adapter level features.
2542 adapter->name = pci_name(pdev);
2543 adapter->msg_enable = dflt_msg_enable;
2544 err = adap_init0(adapter);
2549 * Allocate our "adapter ports" and stitch everything together.
2551 pmask = adapter->params.vfres.pmask;
2552 for_each_port(adapter, pidx) {
2556 * We simplistically allocate our virtual interfaces
2557 * sequentially across the port numbers to which we have
2558 * access rights. This should be configurable in some manner
2563 port_id = ffs(pmask) - 1;
2564 pmask &= ~(1 << port_id);
2565 viid = t4vf_alloc_vi(adapter, port_id);
2567 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2568 " err=%d\n", port_id, viid);
2574 * Allocate our network device and stitch things together.
2576 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2578 if (netdev == NULL) {
2579 dev_err(&pdev->dev, "cannot allocate netdev for"
2580 " port %d\n", port_id);
2581 t4vf_free_vi(adapter, viid);
2585 adapter->port[pidx] = netdev;
2586 SET_NETDEV_DEV(netdev, &pdev->dev);
2587 pi = netdev_priv(netdev);
2588 pi->adapter = adapter;
2590 pi->port_id = port_id;
2594 * Initialize the starting state of our "port" and register
2597 pi->xact_addr_filt = -1;
2598 netif_carrier_off(netdev);
2599 netdev->irq = pdev->irq;
2601 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
2602 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2603 NETIF_F_HW_VLAN_TX | NETIF_F_RXCSUM;
2604 netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
2605 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2607 netdev->features = netdev->hw_features |
2610 netdev->features |= NETIF_F_HIGHDMA;
2612 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2613 SET_ETHTOOL_OPS(netdev, &cxgb4vf_ethtool_ops);
2616 * Initialize the hardware/software state for the port.
2618 err = t4vf_port_init(adapter, pidx);
2620 dev_err(&pdev->dev, "cannot initialize port %d\n",
2627 * The "card" is now ready to go. If any errors occur during device
2628 * registration we do not fail the whole "card" but rather proceed
2629 * only with the ports we manage to register successfully. However we
2630 * must register at least one net device.
2632 for_each_port(adapter, pidx) {
2633 netdev = adapter->port[pidx];
2637 err = register_netdev(netdev);
2639 dev_warn(&pdev->dev, "cannot register net device %s,"
2640 " skipping\n", netdev->name);
2644 set_bit(pidx, &adapter->registered_device_map);
2646 if (adapter->registered_device_map == 0) {
2647 dev_err(&pdev->dev, "could not register any net devices\n");
2652 * Set up our debugfs entries.
2654 if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) {
2655 adapter->debugfs_root =
2656 debugfs_create_dir(pci_name(pdev),
2657 cxgb4vf_debugfs_root);
2658 if (IS_ERR_OR_NULL(adapter->debugfs_root))
2659 dev_warn(&pdev->dev, "could not create debugfs"
2662 setup_debugfs(adapter);
2666 * See what interrupts we'll be using. If we've been configured to
2667 * use MSI-X interrupts, try to enable them but fall back to using
2668 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
2669 * get MSI interrupts we bail with the error.
2671 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
2672 adapter->flags |= USING_MSIX;
2674 err = pci_enable_msi(pdev);
2676 dev_err(&pdev->dev, "Unable to allocate %s interrupts;"
2678 msi == MSI_MSIX ? "MSI-X or MSI" : "MSI", err);
2679 goto err_free_debugfs;
2681 adapter->flags |= USING_MSI;
2685 * Now that we know how many "ports" we have and what their types are,
2686 * and how many Queue Sets we can support, we can configure our queue
2689 cfg_queues(adapter);
2692 * Print a short notice on the existence and configuration of the new
2693 * VF network device ...
2695 for_each_port(adapter, pidx) {
2696 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
2697 adapter->port[pidx]->name,
2698 (adapter->flags & USING_MSIX) ? "MSI-X" :
2699 (adapter->flags & USING_MSI) ? "MSI" : "");
2708 * Error recovery and exit code. Unwind state that's been created
2709 * so far and return the error.
2713 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
2714 cleanup_debugfs(adapter);
2715 debugfs_remove_recursive(adapter->debugfs_root);
2719 for_each_port(adapter, pidx) {
2720 netdev = adapter->port[pidx];
2723 pi = netdev_priv(netdev);
2724 t4vf_free_vi(adapter, pi->viid);
2725 if (test_bit(pidx, &adapter->registered_device_map))
2726 unregister_netdev(netdev);
2727 free_netdev(netdev);
2731 iounmap(adapter->regs);
2735 pci_set_drvdata(pdev, NULL);
2737 err_release_regions:
2738 pci_release_regions(pdev);
2739 pci_set_drvdata(pdev, NULL);
2740 pci_clear_master(pdev);
2743 pci_disable_device(pdev);
2749 * "Remove" a device: tear down all kernel and driver state created in the
2750 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
2751 * that this is called "remove_one" in the PF Driver.)
2753 static void __devexit cxgb4vf_pci_remove(struct pci_dev *pdev)
2755 struct adapter *adapter = pci_get_drvdata(pdev);
2758 * Tear down driver state associated with device.
2764 * Stop all of our activity. Unregister network port,
2765 * disable interrupts, etc.
2767 for_each_port(adapter, pidx)
2768 if (test_bit(pidx, &adapter->registered_device_map))
2769 unregister_netdev(adapter->port[pidx]);
2770 t4vf_sge_stop(adapter);
2771 if (adapter->flags & USING_MSIX) {
2772 pci_disable_msix(adapter->pdev);
2773 adapter->flags &= ~USING_MSIX;
2774 } else if (adapter->flags & USING_MSI) {
2775 pci_disable_msi(adapter->pdev);
2776 adapter->flags &= ~USING_MSI;
2780 * Tear down our debugfs entries.
2782 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
2783 cleanup_debugfs(adapter);
2784 debugfs_remove_recursive(adapter->debugfs_root);
2788 * Free all of the various resources which we've acquired ...
2790 t4vf_free_sge_resources(adapter);
2791 for_each_port(adapter, pidx) {
2792 struct net_device *netdev = adapter->port[pidx];
2793 struct port_info *pi;
2798 pi = netdev_priv(netdev);
2799 t4vf_free_vi(adapter, pi->viid);
2800 free_netdev(netdev);
2802 iounmap(adapter->regs);
2804 pci_set_drvdata(pdev, NULL);
2808 * Disable the device and release its PCI resources.
2810 pci_disable_device(pdev);
2811 pci_clear_master(pdev);
2812 pci_release_regions(pdev);
2816 * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt
2819 static void __devexit cxgb4vf_pci_shutdown(struct pci_dev *pdev)
2821 struct adapter *adapter;
2824 adapter = pci_get_drvdata(pdev);
2829 * Disable all Virtual Interfaces. This will shut down the
2830 * delivery of all ingress packets into the chip for these
2831 * Virtual Interfaces.
2833 for_each_port(adapter, pidx) {
2834 struct net_device *netdev;
2835 struct port_info *pi;
2837 if (!test_bit(pidx, &adapter->registered_device_map))
2840 netdev = adapter->port[pidx];
2844 pi = netdev_priv(netdev);
2845 t4vf_enable_vi(adapter, pi->viid, false, false);
2849 * Free up all Queues which will prevent further DMA and
2850 * Interrupts allowing various internal pathways to drain.
2852 t4vf_free_sge_resources(adapter);
2856 * PCI Device registration data structures.
2858 #define CH_DEVICE(devid, idx) \
2859 { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
2861 static struct pci_device_id cxgb4vf_pci_tbl[] = {
2862 CH_DEVICE(0xb000, 0), /* PE10K FPGA */
2863 CH_DEVICE(0x4800, 0), /* T440-dbg */
2864 CH_DEVICE(0x4801, 0), /* T420-cr */
2865 CH_DEVICE(0x4802, 0), /* T422-cr */
2866 CH_DEVICE(0x4803, 0), /* T440-cr */
2867 CH_DEVICE(0x4804, 0), /* T420-bch */
2868 CH_DEVICE(0x4805, 0), /* T440-bch */
2869 CH_DEVICE(0x4806, 0), /* T460-ch */
2870 CH_DEVICE(0x4807, 0), /* T420-so */
2871 CH_DEVICE(0x4808, 0), /* T420-cx */
2872 CH_DEVICE(0x4809, 0), /* T420-bt */
2873 CH_DEVICE(0x480a, 0), /* T404-bt */
2877 MODULE_DESCRIPTION(DRV_DESC);
2878 MODULE_AUTHOR("Chelsio Communications");
2879 MODULE_LICENSE("Dual BSD/GPL");
2880 MODULE_VERSION(DRV_VERSION);
2881 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
2883 static struct pci_driver cxgb4vf_driver = {
2884 .name = KBUILD_MODNAME,
2885 .id_table = cxgb4vf_pci_tbl,
2886 .probe = cxgb4vf_pci_probe,
2887 .remove = __devexit_p(cxgb4vf_pci_remove),
2888 .shutdown = __devexit_p(cxgb4vf_pci_shutdown),
2892 * Initialize global driver state.
2894 static int __init cxgb4vf_module_init(void)
2899 * Vet our module parameters.
2901 if (msi != MSI_MSIX && msi != MSI_MSI) {
2902 printk(KERN_WARNING KBUILD_MODNAME
2903 ": bad module parameter msi=%d; must be %d"
2904 " (MSI-X or MSI) or %d (MSI)\n",
2905 msi, MSI_MSIX, MSI_MSI);
2909 /* Debugfs support is optional, just warn if this fails */
2910 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
2911 if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
2912 printk(KERN_WARNING KBUILD_MODNAME ": could not create"
2913 " debugfs entry, continuing\n");
2915 ret = pci_register_driver(&cxgb4vf_driver);
2916 if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
2917 debugfs_remove(cxgb4vf_debugfs_root);
2922 * Tear down global driver state.
2924 static void __exit cxgb4vf_module_exit(void)
2926 pci_unregister_driver(&cxgb4vf_driver);
2927 debugfs_remove(cxgb4vf_debugfs_root);
2930 module_init(cxgb4vf_module_init);
2931 module_exit(cxgb4vf_module_exit);