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 * ======================
216 * Perform the MAC and PHY actions needed to enable a "port" (Virtual
219 static int link_start(struct net_device *dev)
222 struct port_info *pi = netdev_priv(dev);
225 * We do not set address filters and promiscuity here, the stack does
226 * that step explicitly. Enable vlan accel.
228 ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, 1,
231 ret = t4vf_change_mac(pi->adapter, pi->viid,
232 pi->xact_addr_filt, dev->dev_addr, true);
234 pi->xact_addr_filt = ret;
240 * We don't need to actually "start the link" itself since the
241 * firmware will do that for us when the first Virtual Interface
242 * is enabled on a port.
245 ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true);
250 * Name the MSI-X interrupts.
252 static void name_msix_vecs(struct adapter *adapter)
254 int namelen = sizeof(adapter->msix_info[0].desc) - 1;
260 snprintf(adapter->msix_info[MSIX_FW].desc, namelen,
261 "%s-FWeventq", adapter->name);
262 adapter->msix_info[MSIX_FW].desc[namelen] = 0;
267 for_each_port(adapter, pidx) {
268 struct net_device *dev = adapter->port[pidx];
269 const struct port_info *pi = netdev_priv(dev);
272 for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) {
273 snprintf(adapter->msix_info[msi].desc, namelen,
274 "%s-%d", dev->name, qs);
275 adapter->msix_info[msi].desc[namelen] = 0;
281 * Request all of our MSI-X resources.
283 static int request_msix_queue_irqs(struct adapter *adapter)
285 struct sge *s = &adapter->sge;
291 err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix,
292 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq);
300 for_each_ethrxq(s, rxq) {
301 err = request_irq(adapter->msix_info[msi].vec,
302 t4vf_sge_intr_msix, 0,
303 adapter->msix_info[msi].desc,
304 &s->ethrxq[rxq].rspq);
313 free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq);
314 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
319 * Free our MSI-X resources.
321 static void free_msix_queue_irqs(struct adapter *adapter)
323 struct sge *s = &adapter->sge;
326 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
328 for_each_ethrxq(s, rxq)
329 free_irq(adapter->msix_info[msi++].vec,
330 &s->ethrxq[rxq].rspq);
334 * Turn on NAPI and start up interrupts on a response queue.
336 static void qenable(struct sge_rspq *rspq)
338 napi_enable(&rspq->napi);
341 * 0-increment the Going To Sleep register to start the timer and
344 t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
346 SEINTARM(rspq->intr_params) |
347 INGRESSQID(rspq->cntxt_id));
351 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
353 static void enable_rx(struct adapter *adapter)
356 struct sge *s = &adapter->sge;
358 for_each_ethrxq(s, rxq)
359 qenable(&s->ethrxq[rxq].rspq);
360 qenable(&s->fw_evtq);
363 * The interrupt queue doesn't use NAPI so we do the 0-increment of
364 * its Going To Sleep register here to get it started.
366 if (adapter->flags & USING_MSI)
367 t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
369 SEINTARM(s->intrq.intr_params) |
370 INGRESSQID(s->intrq.cntxt_id));
375 * Wait until all NAPI handlers are descheduled.
377 static void quiesce_rx(struct adapter *adapter)
379 struct sge *s = &adapter->sge;
382 for_each_ethrxq(s, rxq)
383 napi_disable(&s->ethrxq[rxq].rspq.napi);
384 napi_disable(&s->fw_evtq.napi);
388 * Response queue handler for the firmware event queue.
390 static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp,
391 const struct pkt_gl *gl)
394 * Extract response opcode and get pointer to CPL message body.
396 struct adapter *adapter = rspq->adapter;
397 u8 opcode = ((const struct rss_header *)rsp)->opcode;
398 void *cpl = (void *)(rsp + 1);
403 * We've received an asynchronous message from the firmware.
405 const struct cpl_fw6_msg *fw_msg = cpl;
406 if (fw_msg->type == FW6_TYPE_CMD_RPL)
407 t4vf_handle_fw_rpl(adapter, fw_msg->data);
411 case CPL_SGE_EGR_UPDATE: {
413 * We've received an Egress Queue Status Update message. We
414 * get these, if the SGE is configured to send these when the
415 * firmware passes certain points in processing our TX
416 * Ethernet Queue or if we make an explicit request for one.
417 * We use these updates to determine when we may need to
418 * restart a TX Ethernet Queue which was stopped for lack of
419 * free TX Queue Descriptors ...
421 const struct cpl_sge_egr_update *p = (void *)cpl;
422 unsigned int qid = EGR_QID(be32_to_cpu(p->opcode_qid));
423 struct sge *s = &adapter->sge;
425 struct sge_eth_txq *txq;
429 * Perform sanity checking on the Queue ID to make sure it
430 * really refers to one of our TX Ethernet Egress Queues which
431 * is active and matches the queue's ID. None of these error
432 * conditions should ever happen so we may want to either make
433 * them fatal and/or conditionalized under DEBUG.
435 eq_idx = EQ_IDX(s, qid);
436 if (unlikely(eq_idx >= MAX_EGRQ)) {
437 dev_err(adapter->pdev_dev,
438 "Egress Update QID %d out of range\n", qid);
441 tq = s->egr_map[eq_idx];
442 if (unlikely(tq == NULL)) {
443 dev_err(adapter->pdev_dev,
444 "Egress Update QID %d TXQ=NULL\n", qid);
447 txq = container_of(tq, struct sge_eth_txq, q);
448 if (unlikely(tq->abs_id != qid)) {
449 dev_err(adapter->pdev_dev,
450 "Egress Update QID %d refers to TXQ %d\n",
456 * Restart a stopped TX Queue which has less than half of its
460 netif_tx_wake_queue(txq->txq);
465 dev_err(adapter->pdev_dev,
466 "unexpected CPL %#x on FW event queue\n", opcode);
473 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
474 * to use and initializes them. We support multiple "Queue Sets" per port if
475 * we have MSI-X, otherwise just one queue set per port.
477 static int setup_sge_queues(struct adapter *adapter)
479 struct sge *s = &adapter->sge;
483 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
486 bitmap_zero(s->starving_fl, MAX_EGRQ);
489 * If we're using MSI interrupt mode we need to set up a "forwarded
490 * interrupt" queue which we'll set up with our MSI vector. The rest
491 * of the ingress queues will be set up to forward their interrupts to
492 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
493 * the intrq's queue ID as the interrupt forwarding queue for the
494 * subsequent calls ...
496 if (adapter->flags & USING_MSI) {
497 err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false,
498 adapter->port[0], 0, NULL, NULL);
500 goto err_free_queues;
504 * Allocate our ingress queue for asynchronous firmware messages.
506 err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0],
507 MSIX_FW, NULL, fwevtq_handler);
509 goto err_free_queues;
512 * Allocate each "port"'s initial Queue Sets. These can be changed
513 * later on ... up to the point where any interface on the adapter is
514 * brought up at which point lots of things get nailed down
518 for_each_port(adapter, pidx) {
519 struct net_device *dev = adapter->port[pidx];
520 struct port_info *pi = netdev_priv(dev);
521 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
522 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
525 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
526 err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false,
528 &rxq->fl, t4vf_ethrx_handler);
530 goto err_free_queues;
532 err = t4vf_sge_alloc_eth_txq(adapter, txq, dev,
533 netdev_get_tx_queue(dev, qs),
534 s->fw_evtq.cntxt_id);
536 goto err_free_queues;
539 memset(&rxq->stats, 0, sizeof(rxq->stats));
544 * Create the reverse mappings for the queues.
546 s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id;
547 s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id;
548 IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq;
549 for_each_port(adapter, pidx) {
550 struct net_device *dev = adapter->port[pidx];
551 struct port_info *pi = netdev_priv(dev);
552 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
553 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
556 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
557 IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq;
558 EQ_MAP(s, txq->q.abs_id) = &txq->q;
561 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
562 * for Free Lists but since all of the Egress Queues
563 * (including Free Lists) have Relative Queue IDs
564 * which are computed as Absolute - Base Queue ID, we
565 * can synthesize the Absolute Queue IDs for the Free
566 * Lists. This is useful for debugging purposes when
567 * we want to dump Queue Contexts via the PF Driver.
569 rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base;
570 EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl;
576 t4vf_free_sge_resources(adapter);
581 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
582 * queues. We configure the RSS CPU lookup table to distribute to the number
583 * of HW receive queues, and the response queue lookup table to narrow that
584 * down to the response queues actually configured for each "port" (Virtual
585 * Interface). We always configure the RSS mapping for all ports since the
586 * mapping table has plenty of entries.
588 static int setup_rss(struct adapter *adapter)
592 for_each_port(adapter, pidx) {
593 struct port_info *pi = adap2pinfo(adapter, pidx);
594 struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
595 u16 rss[MAX_PORT_QSETS];
598 for (qs = 0; qs < pi->nqsets; qs++)
599 rss[qs] = rxq[qs].rspq.abs_id;
601 err = t4vf_config_rss_range(adapter, pi->viid,
602 0, pi->rss_size, rss, pi->nqsets);
607 * Perform Global RSS Mode-specific initialization.
609 switch (adapter->params.rss.mode) {
610 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL:
612 * If Tunnel All Lookup isn't specified in the global
613 * RSS Configuration, then we need to specify a
614 * default Ingress Queue for any ingress packets which
615 * aren't hashed. We'll use our first ingress queue
618 if (!adapter->params.rss.u.basicvirtual.tnlalllookup) {
619 union rss_vi_config config;
620 err = t4vf_read_rss_vi_config(adapter,
625 config.basicvirtual.defaultq =
627 err = t4vf_write_rss_vi_config(adapter,
641 * Bring the adapter up. Called whenever we go from no "ports" open to having
642 * one open. This function performs the actions necessary to make an adapter
643 * operational, such as completing the initialization of HW modules, and
644 * enabling interrupts. Must be called with the rtnl lock held. (Note that
645 * this is called "cxgb_up" in the PF Driver.)
647 static int adapter_up(struct adapter *adapter)
652 * If this is the first time we've been called, perform basic
653 * adapter setup. Once we've done this, many of our adapter
654 * parameters can no longer be changed ...
656 if ((adapter->flags & FULL_INIT_DONE) == 0) {
657 err = setup_sge_queues(adapter);
660 err = setup_rss(adapter);
662 t4vf_free_sge_resources(adapter);
666 if (adapter->flags & USING_MSIX)
667 name_msix_vecs(adapter);
668 adapter->flags |= FULL_INIT_DONE;
672 * Acquire our interrupt resources. We only support MSI-X and MSI.
674 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
675 if (adapter->flags & USING_MSIX)
676 err = request_msix_queue_irqs(adapter);
678 err = request_irq(adapter->pdev->irq,
679 t4vf_intr_handler(adapter), 0,
680 adapter->name, adapter);
682 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
688 * Enable NAPI ingress processing and return success.
691 t4vf_sge_start(adapter);
696 * Bring the adapter down. Called whenever the last "port" (Virtual
697 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
700 static void adapter_down(struct adapter *adapter)
703 * Free interrupt resources.
705 if (adapter->flags & USING_MSIX)
706 free_msix_queue_irqs(adapter);
708 free_irq(adapter->pdev->irq, adapter);
711 * Wait for NAPI handlers to finish.
717 * Start up a net device.
719 static int cxgb4vf_open(struct net_device *dev)
722 struct port_info *pi = netdev_priv(dev);
723 struct adapter *adapter = pi->adapter;
726 * If this is the first interface that we're opening on the "adapter",
727 * bring the "adapter" up now.
729 if (adapter->open_device_map == 0) {
730 err = adapter_up(adapter);
736 * Note that this interface is up and start everything up ...
738 netif_set_real_num_tx_queues(dev, pi->nqsets);
739 err = netif_set_real_num_rx_queues(dev, pi->nqsets);
742 err = link_start(dev);
746 netif_tx_start_all_queues(dev);
747 set_bit(pi->port_id, &adapter->open_device_map);
751 if (adapter->open_device_map == 0)
752 adapter_down(adapter);
757 * Shut down a net device. This routine is called "cxgb_close" in the PF
760 static int cxgb4vf_stop(struct net_device *dev)
762 struct port_info *pi = netdev_priv(dev);
763 struct adapter *adapter = pi->adapter;
765 netif_tx_stop_all_queues(dev);
766 netif_carrier_off(dev);
767 t4vf_enable_vi(adapter, pi->viid, false, false);
768 pi->link_cfg.link_ok = 0;
770 clear_bit(pi->port_id, &adapter->open_device_map);
771 if (adapter->open_device_map == 0)
772 adapter_down(adapter);
777 * Translate our basic statistics into the standard "ifconfig" statistics.
779 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
781 struct t4vf_port_stats stats;
782 struct port_info *pi = netdev2pinfo(dev);
783 struct adapter *adapter = pi->adapter;
784 struct net_device_stats *ns = &dev->stats;
787 spin_lock(&adapter->stats_lock);
788 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
789 spin_unlock(&adapter->stats_lock);
791 memset(ns, 0, sizeof(*ns));
795 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
796 stats.tx_ucast_bytes + stats.tx_offload_bytes);
797 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
798 stats.tx_ucast_frames + stats.tx_offload_frames);
799 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
800 stats.rx_ucast_bytes);
801 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
802 stats.rx_ucast_frames);
803 ns->multicast = stats.rx_mcast_frames;
804 ns->tx_errors = stats.tx_drop_frames;
805 ns->rx_errors = stats.rx_err_frames;
811 * Collect up to maxaddrs worth of a netdevice's unicast addresses, starting
812 * at a specified offset within the list, into an array of addrss pointers and
813 * return the number collected.
815 static inline unsigned int collect_netdev_uc_list_addrs(const struct net_device *dev,
818 unsigned int maxaddrs)
820 unsigned int index = 0;
821 unsigned int naddr = 0;
822 const struct netdev_hw_addr *ha;
824 for_each_dev_addr(dev, ha)
825 if (index++ >= offset) {
826 addr[naddr++] = ha->addr;
827 if (naddr >= maxaddrs)
834 * Collect up to maxaddrs worth of a netdevice's multicast addresses, starting
835 * at a specified offset within the list, into an array of addrss pointers and
836 * return the number collected.
838 static inline unsigned int collect_netdev_mc_list_addrs(const struct net_device *dev,
841 unsigned int maxaddrs)
843 unsigned int index = 0;
844 unsigned int naddr = 0;
845 const struct netdev_hw_addr *ha;
847 netdev_for_each_mc_addr(ha, dev)
848 if (index++ >= offset) {
849 addr[naddr++] = ha->addr;
850 if (naddr >= maxaddrs)
857 * Configure the exact and hash address filters to handle a port's multicast
858 * and secondary unicast MAC addresses.
860 static int set_addr_filters(const struct net_device *dev, bool sleep)
865 unsigned int offset, naddr;
868 const struct port_info *pi = netdev_priv(dev);
870 /* first do the secondary unicast addresses */
871 for (offset = 0; ; offset += naddr) {
872 naddr = collect_netdev_uc_list_addrs(dev, addr, offset,
877 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
878 naddr, addr, NULL, &uhash, sleep);
885 /* next set up the multicast addresses */
886 for (offset = 0; ; offset += naddr) {
887 naddr = collect_netdev_mc_list_addrs(dev, addr, offset,
892 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
893 naddr, addr, NULL, &mhash, sleep);
899 return t4vf_set_addr_hash(pi->adapter, pi->viid, uhash != 0,
900 uhash | mhash, sleep);
904 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
905 * If @mtu is -1 it is left unchanged.
907 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
910 struct port_info *pi = netdev_priv(dev);
912 ret = set_addr_filters(dev, sleep_ok);
914 ret = t4vf_set_rxmode(pi->adapter, pi->viid, -1,
915 (dev->flags & IFF_PROMISC) != 0,
916 (dev->flags & IFF_ALLMULTI) != 0,
922 * Set the current receive modes on the device.
924 static void cxgb4vf_set_rxmode(struct net_device *dev)
926 /* unfortunately we can't return errors to the stack */
927 set_rxmode(dev, -1, false);
931 * Find the entry in the interrupt holdoff timer value array which comes
932 * closest to the specified interrupt holdoff value.
934 static int closest_timer(const struct sge *s, int us)
936 int i, timer_idx = 0, min_delta = INT_MAX;
938 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
939 int delta = us - s->timer_val[i];
942 if (delta < min_delta) {
950 static int closest_thres(const struct sge *s, int thres)
952 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
954 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
955 delta = thres - s->counter_val[i];
958 if (delta < min_delta) {
967 * Return a queue's interrupt hold-off time in us. 0 means no timer.
969 static unsigned int qtimer_val(const struct adapter *adapter,
970 const struct sge_rspq *rspq)
972 unsigned int timer_idx = QINTR_TIMER_IDX_GET(rspq->intr_params);
974 return timer_idx < SGE_NTIMERS
975 ? adapter->sge.timer_val[timer_idx]
980 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
981 * @adapter: the adapter
982 * @rspq: the RX response queue
983 * @us: the hold-off time in us, or 0 to disable timer
984 * @cnt: the hold-off packet count, or 0 to disable counter
986 * Sets an RX response queue's interrupt hold-off time and packet count.
987 * At least one of the two needs to be enabled for the queue to generate
990 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
991 unsigned int us, unsigned int cnt)
993 unsigned int timer_idx;
996 * If both the interrupt holdoff timer and count are specified as
997 * zero, default to a holdoff count of 1 ...
1003 * If an interrupt holdoff count has been specified, then find the
1004 * closest configured holdoff count and use that. If the response
1005 * queue has already been created, then update its queue context
1012 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1013 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1014 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1016 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1017 FW_PARAMS_PARAM_YZ(rspq->cntxt_id);
1018 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1022 rspq->pktcnt_idx = pktcnt_idx;
1026 * Compute the closest holdoff timer index from the supplied holdoff
1029 timer_idx = (us == 0
1030 ? SGE_TIMER_RSTRT_CNTR
1031 : closest_timer(&adapter->sge, us));
1034 * Update the response queue's interrupt coalescing parameters and
1037 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
1038 (cnt > 0 ? QINTR_CNT_EN : 0));
1043 * Return a version number to identify the type of adapter. The scheme is:
1044 * - bits 0..9: chip version
1045 * - bits 10..15: chip revision
1047 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1050 * Chip version 4, revision 0x3f (cxgb4vf).
1052 return 4 | (0x3f << 10);
1056 * Execute the specified ioctl command.
1058 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1064 * The VF Driver doesn't have access to any of the other
1065 * common Ethernet device ioctl()'s (like reading/writing
1066 * PHY registers, etc.
1077 * Change the device's MTU.
1079 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1082 struct port_info *pi = netdev_priv(dev);
1084 /* accommodate SACK */
1088 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1089 -1, -1, -1, -1, true);
1095 static u32 cxgb4vf_fix_features(struct net_device *dev, u32 features)
1098 * Since there is no support for separate rx/tx vlan accel
1099 * enable/disable make sure tx flag is always in same state as rx.
1101 if (features & NETIF_F_HW_VLAN_RX)
1102 features |= NETIF_F_HW_VLAN_TX;
1104 features &= ~NETIF_F_HW_VLAN_TX;
1109 static int cxgb4vf_set_features(struct net_device *dev, u32 features)
1111 struct port_info *pi = netdev_priv(dev);
1112 u32 changed = dev->features ^ features;
1114 if (changed & NETIF_F_HW_VLAN_RX)
1115 t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1,
1116 features & NETIF_F_HW_VLAN_TX, 0);
1122 * Change the devices MAC address.
1124 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1127 struct sockaddr *addr = _addr;
1128 struct port_info *pi = netdev_priv(dev);
1130 if (!is_valid_ether_addr(addr->sa_data))
1133 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1134 addr->sa_data, true);
1138 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1139 pi->xact_addr_filt = ret;
1143 #ifdef CONFIG_NET_POLL_CONTROLLER
1145 * Poll all of our receive queues. This is called outside of normal interrupt
1148 static void cxgb4vf_poll_controller(struct net_device *dev)
1150 struct port_info *pi = netdev_priv(dev);
1151 struct adapter *adapter = pi->adapter;
1153 if (adapter->flags & USING_MSIX) {
1154 struct sge_eth_rxq *rxq;
1157 rxq = &adapter->sge.ethrxq[pi->first_qset];
1158 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1159 t4vf_sge_intr_msix(0, &rxq->rspq);
1163 t4vf_intr_handler(adapter)(0, adapter);
1168 * Ethtool operations.
1169 * ===================
1171 * Note that we don't support any ethtool operations which change the physical
1172 * state of the port to which we're linked.
1176 * Return current port link settings.
1178 static int cxgb4vf_get_settings(struct net_device *dev,
1179 struct ethtool_cmd *cmd)
1181 const struct port_info *pi = netdev_priv(dev);
1183 cmd->supported = pi->link_cfg.supported;
1184 cmd->advertising = pi->link_cfg.advertising;
1185 ethtool_cmd_speed_set(cmd,
1186 netif_carrier_ok(dev) ? pi->link_cfg.speed : -1);
1187 cmd->duplex = DUPLEX_FULL;
1189 cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1190 cmd->phy_address = pi->port_id;
1191 cmd->transceiver = XCVR_EXTERNAL;
1192 cmd->autoneg = pi->link_cfg.autoneg;
1199 * Return our driver information.
1201 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1202 struct ethtool_drvinfo *drvinfo)
1204 struct adapter *adapter = netdev2adap(dev);
1206 strcpy(drvinfo->driver, KBUILD_MODNAME);
1207 strcpy(drvinfo->version, DRV_VERSION);
1208 strcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)));
1209 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1210 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1211 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.fwrev),
1212 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.fwrev),
1213 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.fwrev),
1214 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.fwrev),
1215 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.tprev),
1216 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.tprev),
1217 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.tprev),
1218 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.tprev));
1222 * Return current adapter message level.
1224 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1226 return netdev2adap(dev)->msg_enable;
1230 * Set current adapter message level.
1232 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1234 netdev2adap(dev)->msg_enable = msglevel;
1238 * Return the device's current Queue Set ring size parameters along with the
1239 * allowed maximum values. Since ethtool doesn't understand the concept of
1240 * multi-queue devices, we just return the current values associated with the
1243 static void cxgb4vf_get_ringparam(struct net_device *dev,
1244 struct ethtool_ringparam *rp)
1246 const struct port_info *pi = netdev_priv(dev);
1247 const struct sge *s = &pi->adapter->sge;
1249 rp->rx_max_pending = MAX_RX_BUFFERS;
1250 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1251 rp->rx_jumbo_max_pending = 0;
1252 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1254 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1255 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1256 rp->rx_jumbo_pending = 0;
1257 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1261 * Set the Queue Set ring size parameters for the device. Again, since
1262 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1263 * apply these new values across all of the Queue Sets associated with the
1264 * device -- after vetting them of course!
1266 static int cxgb4vf_set_ringparam(struct net_device *dev,
1267 struct ethtool_ringparam *rp)
1269 const struct port_info *pi = netdev_priv(dev);
1270 struct adapter *adapter = pi->adapter;
1271 struct sge *s = &adapter->sge;
1274 if (rp->rx_pending > MAX_RX_BUFFERS ||
1275 rp->rx_jumbo_pending ||
1276 rp->tx_pending > MAX_TXQ_ENTRIES ||
1277 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1278 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1279 rp->rx_pending < MIN_FL_ENTRIES ||
1280 rp->tx_pending < MIN_TXQ_ENTRIES)
1283 if (adapter->flags & FULL_INIT_DONE)
1286 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1287 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1288 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1289 s->ethtxq[qs].q.size = rp->tx_pending;
1295 * Return the interrupt holdoff timer and count for the first Queue Set on the
1296 * device. Our extension ioctl() (the cxgbtool interface) allows the
1297 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1299 static int cxgb4vf_get_coalesce(struct net_device *dev,
1300 struct ethtool_coalesce *coalesce)
1302 const struct port_info *pi = netdev_priv(dev);
1303 const struct adapter *adapter = pi->adapter;
1304 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1306 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1307 coalesce->rx_max_coalesced_frames =
1308 ((rspq->intr_params & QINTR_CNT_EN)
1309 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1315 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1316 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1317 * the interrupt holdoff timer on any of the device's Queue Sets.
1319 static int cxgb4vf_set_coalesce(struct net_device *dev,
1320 struct ethtool_coalesce *coalesce)
1322 const struct port_info *pi = netdev_priv(dev);
1323 struct adapter *adapter = pi->adapter;
1325 return set_rxq_intr_params(adapter,
1326 &adapter->sge.ethrxq[pi->first_qset].rspq,
1327 coalesce->rx_coalesce_usecs,
1328 coalesce->rx_max_coalesced_frames);
1332 * Report current port link pause parameter settings.
1334 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1335 struct ethtool_pauseparam *pauseparam)
1337 struct port_info *pi = netdev_priv(dev);
1339 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1340 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1341 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1345 * Identify the port by blinking the port's LED.
1347 static int cxgb4vf_phys_id(struct net_device *dev,
1348 enum ethtool_phys_id_state state)
1351 struct port_info *pi = netdev_priv(dev);
1353 if (state == ETHTOOL_ID_ACTIVE)
1355 else if (state == ETHTOOL_ID_INACTIVE)
1360 return t4vf_identify_port(pi->adapter, pi->viid, val);
1364 * Port stats maintained per queue of the port.
1366 struct queue_port_stats {
1377 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1378 * these need to match the order of statistics returned by
1379 * t4vf_get_port_stats().
1381 static const char stats_strings[][ETH_GSTRING_LEN] = {
1383 * These must match the layout of the t4vf_port_stats structure.
1385 "TxBroadcastBytes ",
1386 "TxBroadcastFrames ",
1387 "TxMulticastBytes ",
1388 "TxMulticastFrames ",
1394 "RxBroadcastBytes ",
1395 "RxBroadcastFrames ",
1396 "RxMulticastBytes ",
1397 "RxMulticastFrames ",
1403 * These are accumulated per-queue statistics and must match the
1404 * order of the fields in the queue_port_stats structure.
1416 * Return the number of statistics in the specified statistics set.
1418 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1422 return ARRAY_SIZE(stats_strings);
1430 * Return the strings for the specified statistics set.
1432 static void cxgb4vf_get_strings(struct net_device *dev,
1438 memcpy(data, stats_strings, sizeof(stats_strings));
1444 * Small utility routine to accumulate queue statistics across the queues of
1447 static void collect_sge_port_stats(const struct adapter *adapter,
1448 const struct port_info *pi,
1449 struct queue_port_stats *stats)
1451 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1452 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1455 memset(stats, 0, sizeof(*stats));
1456 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1457 stats->tso += txq->tso;
1458 stats->tx_csum += txq->tx_cso;
1459 stats->rx_csum += rxq->stats.rx_cso;
1460 stats->vlan_ex += rxq->stats.vlan_ex;
1461 stats->vlan_ins += txq->vlan_ins;
1462 stats->lro_pkts += rxq->stats.lro_pkts;
1463 stats->lro_merged += rxq->stats.lro_merged;
1468 * Return the ETH_SS_STATS statistics set.
1470 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1471 struct ethtool_stats *stats,
1474 struct port_info *pi = netdev2pinfo(dev);
1475 struct adapter *adapter = pi->adapter;
1476 int err = t4vf_get_port_stats(adapter, pi->pidx,
1477 (struct t4vf_port_stats *)data);
1479 memset(data, 0, sizeof(struct t4vf_port_stats));
1481 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1482 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1486 * Return the size of our register map.
1488 static int cxgb4vf_get_regs_len(struct net_device *dev)
1490 return T4VF_REGMAP_SIZE;
1494 * Dump a block of registers, start to end inclusive, into a buffer.
1496 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1497 unsigned int start, unsigned int end)
1499 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1501 for ( ; start <= end; start += sizeof(u32)) {
1503 * Avoid reading the Mailbox Control register since that
1504 * can trigger a Mailbox Ownership Arbitration cycle and
1505 * interfere with communication with the firmware.
1507 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1510 *bp++ = t4_read_reg(adapter, start);
1515 * Copy our entire register map into the provided buffer.
1517 static void cxgb4vf_get_regs(struct net_device *dev,
1518 struct ethtool_regs *regs,
1521 struct adapter *adapter = netdev2adap(dev);
1523 regs->version = mk_adap_vers(adapter);
1526 * Fill in register buffer with our register map.
1528 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1530 reg_block_dump(adapter, regbuf,
1531 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1532 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1533 reg_block_dump(adapter, regbuf,
1534 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1535 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1536 reg_block_dump(adapter, regbuf,
1537 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1538 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_LAST);
1539 reg_block_dump(adapter, regbuf,
1540 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1541 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1543 reg_block_dump(adapter, regbuf,
1544 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1545 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1549 * Report current Wake On LAN settings.
1551 static void cxgb4vf_get_wol(struct net_device *dev,
1552 struct ethtool_wolinfo *wol)
1556 memset(&wol->sopass, 0, sizeof(wol->sopass));
1560 * TCP Segmentation Offload flags which we support.
1562 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1564 static struct ethtool_ops cxgb4vf_ethtool_ops = {
1565 .get_settings = cxgb4vf_get_settings,
1566 .get_drvinfo = cxgb4vf_get_drvinfo,
1567 .get_msglevel = cxgb4vf_get_msglevel,
1568 .set_msglevel = cxgb4vf_set_msglevel,
1569 .get_ringparam = cxgb4vf_get_ringparam,
1570 .set_ringparam = cxgb4vf_set_ringparam,
1571 .get_coalesce = cxgb4vf_get_coalesce,
1572 .set_coalesce = cxgb4vf_set_coalesce,
1573 .get_pauseparam = cxgb4vf_get_pauseparam,
1574 .get_link = ethtool_op_get_link,
1575 .get_strings = cxgb4vf_get_strings,
1576 .set_phys_id = cxgb4vf_phys_id,
1577 .get_sset_count = cxgb4vf_get_sset_count,
1578 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1579 .get_regs_len = cxgb4vf_get_regs_len,
1580 .get_regs = cxgb4vf_get_regs,
1581 .get_wol = cxgb4vf_get_wol,
1585 * /sys/kernel/debug/cxgb4vf support code and data.
1586 * ================================================
1590 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1594 static int sge_qinfo_show(struct seq_file *seq, void *v)
1596 struct adapter *adapter = seq->private;
1597 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1598 int qs, r = (uintptr_t)v - 1;
1601 seq_putc(seq, '\n');
1603 #define S3(fmt_spec, s, v) \
1605 seq_printf(seq, "%-12s", s); \
1606 for (qs = 0; qs < n; ++qs) \
1607 seq_printf(seq, " %16" fmt_spec, v); \
1608 seq_putc(seq, '\n'); \
1610 #define S(s, v) S3("s", s, v)
1611 #define T(s, v) S3("u", s, txq[qs].v)
1612 #define R(s, v) S3("u", s, rxq[qs].v)
1614 if (r < eth_entries) {
1615 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1616 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1617 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1619 S("QType:", "Ethernet");
1621 (rxq[qs].rspq.netdev
1622 ? rxq[qs].rspq.netdev->name
1625 (rxq[qs].rspq.netdev
1626 ? ((struct port_info *)
1627 netdev_priv(rxq[qs].rspq.netdev))->port_id
1629 T("TxQ ID:", q.abs_id);
1630 T("TxQ size:", q.size);
1631 T("TxQ inuse:", q.in_use);
1632 T("TxQ PIdx:", q.pidx);
1633 T("TxQ CIdx:", q.cidx);
1634 R("RspQ ID:", rspq.abs_id);
1635 R("RspQ size:", rspq.size);
1636 R("RspQE size:", rspq.iqe_len);
1637 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1638 S3("u", "Intr pktcnt:",
1639 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1640 R("RspQ CIdx:", rspq.cidx);
1641 R("RspQ Gen:", rspq.gen);
1642 R("FL ID:", fl.abs_id);
1643 R("FL size:", fl.size - MIN_FL_RESID);
1644 R("FL avail:", fl.avail);
1645 R("FL PIdx:", fl.pidx);
1646 R("FL CIdx:", fl.cidx);
1652 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1654 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1655 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1656 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1657 qtimer_val(adapter, evtq));
1658 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1659 adapter->sge.counter_val[evtq->pktcnt_idx]);
1660 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1661 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1662 } else if (r == 1) {
1663 const struct sge_rspq *intrq = &adapter->sge.intrq;
1665 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1666 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1667 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1668 qtimer_val(adapter, intrq));
1669 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1670 adapter->sge.counter_val[intrq->pktcnt_idx]);
1671 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1672 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1684 * Return the number of "entries" in our "file". We group the multi-Queue
1685 * sections with QPL Queue Sets per "entry". The sections of the output are:
1687 * Ethernet RX/TX Queue Sets
1688 * Firmware Event Queue
1689 * Forwarded Interrupt Queue (if in MSI mode)
1691 static int sge_queue_entries(const struct adapter *adapter)
1693 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1694 ((adapter->flags & USING_MSI) != 0);
1697 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1699 int entries = sge_queue_entries(seq->private);
1701 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1704 static void sge_queue_stop(struct seq_file *seq, void *v)
1708 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
1710 int entries = sge_queue_entries(seq->private);
1713 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1716 static const struct seq_operations sge_qinfo_seq_ops = {
1717 .start = sge_queue_start,
1718 .next = sge_queue_next,
1719 .stop = sge_queue_stop,
1720 .show = sge_qinfo_show
1723 static int sge_qinfo_open(struct inode *inode, struct file *file)
1725 int res = seq_open(file, &sge_qinfo_seq_ops);
1728 struct seq_file *seq = file->private_data;
1729 seq->private = inode->i_private;
1734 static const struct file_operations sge_qinfo_debugfs_fops = {
1735 .owner = THIS_MODULE,
1736 .open = sge_qinfo_open,
1738 .llseek = seq_lseek,
1739 .release = seq_release,
1743 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
1747 static int sge_qstats_show(struct seq_file *seq, void *v)
1749 struct adapter *adapter = seq->private;
1750 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1751 int qs, r = (uintptr_t)v - 1;
1754 seq_putc(seq, '\n');
1756 #define S3(fmt, s, v) \
1758 seq_printf(seq, "%-16s", s); \
1759 for (qs = 0; qs < n; ++qs) \
1760 seq_printf(seq, " %8" fmt, v); \
1761 seq_putc(seq, '\n'); \
1763 #define S(s, v) S3("s", s, v)
1765 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
1766 #define T(s, v) T3("lu", s, v)
1768 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
1769 #define R(s, v) R3("lu", s, v)
1771 if (r < eth_entries) {
1772 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1773 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1774 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1776 S("QType:", "Ethernet");
1778 (rxq[qs].rspq.netdev
1779 ? rxq[qs].rspq.netdev->name
1781 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
1782 R("RxPackets:", stats.pkts);
1783 R("RxCSO:", stats.rx_cso);
1784 R("VLANxtract:", stats.vlan_ex);
1785 R("LROmerged:", stats.lro_merged);
1786 R("LROpackets:", stats.lro_pkts);
1787 R("RxDrops:", stats.rx_drops);
1789 T("TxCSO:", tx_cso);
1790 T("VLANins:", vlan_ins);
1791 T("TxQFull:", q.stops);
1792 T("TxQRestarts:", q.restarts);
1793 T("TxMapErr:", mapping_err);
1794 R("FLAllocErr:", fl.alloc_failed);
1795 R("FLLrgAlcErr:", fl.large_alloc_failed);
1796 R("FLStarving:", fl.starving);
1802 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1804 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
1805 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1806 evtq->unhandled_irqs);
1807 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
1808 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
1809 } else if (r == 1) {
1810 const struct sge_rspq *intrq = &adapter->sge.intrq;
1812 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
1813 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1814 intrq->unhandled_irqs);
1815 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
1816 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
1830 * Return the number of "entries" in our "file". We group the multi-Queue
1831 * sections with QPL Queue Sets per "entry". The sections of the output are:
1833 * Ethernet RX/TX Queue Sets
1834 * Firmware Event Queue
1835 * Forwarded Interrupt Queue (if in MSI mode)
1837 static int sge_qstats_entries(const struct adapter *adapter)
1839 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1840 ((adapter->flags & USING_MSI) != 0);
1843 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
1845 int entries = sge_qstats_entries(seq->private);
1847 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1850 static void sge_qstats_stop(struct seq_file *seq, void *v)
1854 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
1856 int entries = sge_qstats_entries(seq->private);
1859 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1862 static const struct seq_operations sge_qstats_seq_ops = {
1863 .start = sge_qstats_start,
1864 .next = sge_qstats_next,
1865 .stop = sge_qstats_stop,
1866 .show = sge_qstats_show
1869 static int sge_qstats_open(struct inode *inode, struct file *file)
1871 int res = seq_open(file, &sge_qstats_seq_ops);
1874 struct seq_file *seq = file->private_data;
1875 seq->private = inode->i_private;
1880 static const struct file_operations sge_qstats_proc_fops = {
1881 .owner = THIS_MODULE,
1882 .open = sge_qstats_open,
1884 .llseek = seq_lseek,
1885 .release = seq_release,
1889 * Show PCI-E SR-IOV Virtual Function Resource Limits.
1891 static int resources_show(struct seq_file *seq, void *v)
1893 struct adapter *adapter = seq->private;
1894 struct vf_resources *vfres = &adapter->params.vfres;
1896 #define S(desc, fmt, var) \
1897 seq_printf(seq, "%-60s " fmt "\n", \
1898 desc " (" #var "):", vfres->var)
1900 S("Virtual Interfaces", "%d", nvi);
1901 S("Egress Queues", "%d", neq);
1902 S("Ethernet Control", "%d", nethctrl);
1903 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
1904 S("Ingress Queues", "%d", niq);
1905 S("Traffic Class", "%d", tc);
1906 S("Port Access Rights Mask", "%#x", pmask);
1907 S("MAC Address Filters", "%d", nexactf);
1908 S("Firmware Command Read Capabilities", "%#x", r_caps);
1909 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
1916 static int resources_open(struct inode *inode, struct file *file)
1918 return single_open(file, resources_show, inode->i_private);
1921 static const struct file_operations resources_proc_fops = {
1922 .owner = THIS_MODULE,
1923 .open = resources_open,
1925 .llseek = seq_lseek,
1926 .release = single_release,
1930 * Show Virtual Interfaces.
1932 static int interfaces_show(struct seq_file *seq, void *v)
1934 if (v == SEQ_START_TOKEN) {
1935 seq_puts(seq, "Interface Port VIID\n");
1937 struct adapter *adapter = seq->private;
1938 int pidx = (uintptr_t)v - 2;
1939 struct net_device *dev = adapter->port[pidx];
1940 struct port_info *pi = netdev_priv(dev);
1942 seq_printf(seq, "%9s %4d %#5x\n",
1943 dev->name, pi->port_id, pi->viid);
1948 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
1950 return pos <= adapter->params.nports
1951 ? (void *)(uintptr_t)(pos + 1)
1955 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
1958 ? interfaces_get_idx(seq->private, *pos)
1962 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
1965 return interfaces_get_idx(seq->private, *pos);
1968 static void interfaces_stop(struct seq_file *seq, void *v)
1972 static const struct seq_operations interfaces_seq_ops = {
1973 .start = interfaces_start,
1974 .next = interfaces_next,
1975 .stop = interfaces_stop,
1976 .show = interfaces_show
1979 static int interfaces_open(struct inode *inode, struct file *file)
1981 int res = seq_open(file, &interfaces_seq_ops);
1984 struct seq_file *seq = file->private_data;
1985 seq->private = inode->i_private;
1990 static const struct file_operations interfaces_proc_fops = {
1991 .owner = THIS_MODULE,
1992 .open = interfaces_open,
1994 .llseek = seq_lseek,
1995 .release = seq_release,
1999 * /sys/kernel/debugfs/cxgb4vf/ files list.
2001 struct cxgb4vf_debugfs_entry {
2002 const char *name; /* name of debugfs node */
2003 mode_t mode; /* file system mode */
2004 const struct file_operations *fops;
2007 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
2008 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
2009 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
2010 { "resources", S_IRUGO, &resources_proc_fops },
2011 { "interfaces", S_IRUGO, &interfaces_proc_fops },
2015 * Module and device initialization and cleanup code.
2016 * ==================================================
2020 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2021 * directory (debugfs_root) has already been set up.
2023 static int __devinit setup_debugfs(struct adapter *adapter)
2027 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2030 * Debugfs support is best effort.
2032 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2033 (void)debugfs_create_file(debugfs_files[i].name,
2034 debugfs_files[i].mode,
2035 adapter->debugfs_root,
2037 debugfs_files[i].fops);
2043 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2044 * it to our caller to tear down the directory (debugfs_root).
2046 static void cleanup_debugfs(struct adapter *adapter)
2048 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2051 * Unlike our sister routine cleanup_proc(), we don't need to remove
2052 * individual entries because a call will be made to
2053 * debugfs_remove_recursive(). We just need to clean up any ancillary
2060 * Perform early "adapter" initialization. This is where we discover what
2061 * adapter parameters we're going to be using and initialize basic adapter
2064 static int __devinit adap_init0(struct adapter *adapter)
2066 struct vf_resources *vfres = &adapter->params.vfres;
2067 struct sge_params *sge_params = &adapter->params.sge;
2068 struct sge *s = &adapter->sge;
2069 unsigned int ethqsets;
2073 * Wait for the device to become ready before proceeding ...
2075 err = t4vf_wait_dev_ready(adapter);
2077 dev_err(adapter->pdev_dev, "device didn't become ready:"
2083 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2084 * 2.6.31 and later we can't call pci_reset_function() in order to
2085 * issue an FLR because of a self- deadlock on the device semaphore.
2086 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2087 * cases where they're needed -- for instance, some versions of KVM
2088 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2089 * use the firmware based reset in order to reset any per function
2092 err = t4vf_fw_reset(adapter);
2094 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2099 * Grab basic operational parameters. These will predominantly have
2100 * been set up by the Physical Function Driver or will be hard coded
2101 * into the adapter. We just have to live with them ... Note that
2102 * we _must_ get our VPD parameters before our SGE parameters because
2103 * we need to know the adapter's core clock from the VPD in order to
2104 * properly decode the SGE Timer Values.
2106 err = t4vf_get_dev_params(adapter);
2108 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2109 " device parameters: err=%d\n", err);
2112 err = t4vf_get_vpd_params(adapter);
2114 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2115 " VPD parameters: err=%d\n", err);
2118 err = t4vf_get_sge_params(adapter);
2120 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2121 " SGE parameters: err=%d\n", err);
2124 err = t4vf_get_rss_glb_config(adapter);
2126 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2127 " RSS parameters: err=%d\n", err);
2130 if (adapter->params.rss.mode !=
2131 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2132 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2133 " mode %d\n", adapter->params.rss.mode);
2136 err = t4vf_sge_init(adapter);
2138 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2144 * Retrieve our RX interrupt holdoff timer values and counter
2145 * threshold values from the SGE parameters.
2147 s->timer_val[0] = core_ticks_to_us(adapter,
2148 TIMERVALUE0_GET(sge_params->sge_timer_value_0_and_1));
2149 s->timer_val[1] = core_ticks_to_us(adapter,
2150 TIMERVALUE1_GET(sge_params->sge_timer_value_0_and_1));
2151 s->timer_val[2] = core_ticks_to_us(adapter,
2152 TIMERVALUE0_GET(sge_params->sge_timer_value_2_and_3));
2153 s->timer_val[3] = core_ticks_to_us(adapter,
2154 TIMERVALUE1_GET(sge_params->sge_timer_value_2_and_3));
2155 s->timer_val[4] = core_ticks_to_us(adapter,
2156 TIMERVALUE0_GET(sge_params->sge_timer_value_4_and_5));
2157 s->timer_val[5] = core_ticks_to_us(adapter,
2158 TIMERVALUE1_GET(sge_params->sge_timer_value_4_and_5));
2161 THRESHOLD_0_GET(sge_params->sge_ingress_rx_threshold);
2163 THRESHOLD_1_GET(sge_params->sge_ingress_rx_threshold);
2165 THRESHOLD_2_GET(sge_params->sge_ingress_rx_threshold);
2167 THRESHOLD_3_GET(sge_params->sge_ingress_rx_threshold);
2170 * Grab our Virtual Interface resource allocation, extract the
2171 * features that we're interested in and do a bit of sanity testing on
2174 err = t4vf_get_vfres(adapter);
2176 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2177 " resources: err=%d\n", err);
2182 * The number of "ports" which we support is equal to the number of
2183 * Virtual Interfaces with which we've been provisioned.
2185 adapter->params.nports = vfres->nvi;
2186 if (adapter->params.nports > MAX_NPORTS) {
2187 dev_warn(adapter->pdev_dev, "only using %d of %d allowed"
2188 " virtual interfaces\n", MAX_NPORTS,
2189 adapter->params.nports);
2190 adapter->params.nports = MAX_NPORTS;
2194 * We need to reserve a number of the ingress queues with Free List
2195 * and Interrupt capabilities for special interrupt purposes (like
2196 * asynchronous firmware messages, or forwarded interrupts if we're
2197 * using MSI). The rest of the FL/Intr-capable ingress queues will be
2198 * matched up one-for-one with Ethernet/Control egress queues in order
2199 * to form "Queue Sets" which will be aportioned between the "ports".
2200 * For each Queue Set, we'll need the ability to allocate two Egress
2201 * Contexts -- one for the Ingress Queue Free List and one for the TX
2204 ethqsets = vfres->niqflint - INGQ_EXTRAS;
2205 if (vfres->nethctrl != ethqsets) {
2206 dev_warn(adapter->pdev_dev, "unequal number of [available]"
2207 " ingress/egress queues (%d/%d); using minimum for"
2208 " number of Queue Sets\n", ethqsets, vfres->nethctrl);
2209 ethqsets = min(vfres->nethctrl, ethqsets);
2211 if (vfres->neq < ethqsets*2) {
2212 dev_warn(adapter->pdev_dev, "Not enough Egress Contexts (%d)"
2213 " to support Queue Sets (%d); reducing allowed Queue"
2214 " Sets\n", vfres->neq, ethqsets);
2215 ethqsets = vfres->neq/2;
2217 if (ethqsets > MAX_ETH_QSETS) {
2218 dev_warn(adapter->pdev_dev, "only using %d of %d allowed Queue"
2219 " Sets\n", MAX_ETH_QSETS, adapter->sge.max_ethqsets);
2220 ethqsets = MAX_ETH_QSETS;
2222 if (vfres->niq != 0 || vfres->neq > ethqsets*2) {
2223 dev_warn(adapter->pdev_dev, "unused resources niq/neq (%d/%d)"
2224 " ignored\n", vfres->niq, vfres->neq - ethqsets*2);
2226 adapter->sge.max_ethqsets = ethqsets;
2229 * Check for various parameter sanity issues. Most checks simply
2230 * result in us using fewer resources than our provissioning but we
2231 * do need at least one "port" with which to work ...
2233 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2234 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2235 " virtual interfaces (too few Queue Sets)\n",
2236 adapter->sge.max_ethqsets, adapter->params.nports);
2237 adapter->params.nports = adapter->sge.max_ethqsets;
2239 if (adapter->params.nports == 0) {
2240 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2247 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2248 u8 pkt_cnt_idx, unsigned int size,
2249 unsigned int iqe_size)
2251 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
2252 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0));
2253 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2256 rspq->iqe_len = iqe_size;
2261 * Perform default configuration of DMA queues depending on the number and
2262 * type of ports we found and the number of available CPUs. Most settings can
2263 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2264 * being brought up for the first time.
2266 static void __devinit cfg_queues(struct adapter *adapter)
2268 struct sge *s = &adapter->sge;
2269 int q10g, n10g, qidx, pidx, qs;
2273 * We should not be called till we know how many Queue Sets we can
2274 * support. In particular, this means that we need to know what kind
2275 * of interrupts we'll be using ...
2277 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2280 * Count the number of 10GbE Virtual Interfaces that we have.
2283 for_each_port(adapter, pidx)
2284 n10g += is_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2287 * We default to 1 queue per non-10G port and up to # of cores queues
2293 int n1g = (adapter->params.nports - n10g);
2294 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2295 if (q10g > num_online_cpus())
2296 q10g = num_online_cpus();
2300 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2301 * The layout will be established in setup_sge_queues() when the
2302 * adapter is brough up for the first time.
2305 for_each_port(adapter, pidx) {
2306 struct port_info *pi = adap2pinfo(adapter, pidx);
2308 pi->first_qset = qidx;
2309 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
2315 * The Ingress Queue Entry Size for our various Response Queues needs
2316 * to be big enough to accommodate the largest message we can receive
2317 * from the chip/firmware; which is 64 bytes ...
2322 * Set up default Queue Set parameters ... Start off with the
2323 * shortest interrupt holdoff timer.
2325 for (qs = 0; qs < s->max_ethqsets; qs++) {
2326 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2327 struct sge_eth_txq *txq = &s->ethtxq[qs];
2329 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2335 * The firmware event queue is used for link state changes and
2336 * notifications of TX DMA completions.
2338 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2341 * The forwarded interrupt queue is used when we're in MSI interrupt
2342 * mode. In this mode all interrupts associated with RX queues will
2343 * be forwarded to a single queue which we'll associate with our MSI
2344 * interrupt vector. The messages dropped in the forwarded interrupt
2345 * queue will indicate which ingress queue needs servicing ... This
2346 * queue needs to be large enough to accommodate all of the ingress
2347 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2348 * from equalling the CIDX if every ingress queue has an outstanding
2349 * interrupt). The queue doesn't need to be any larger because no
2350 * ingress queue will ever have more than one outstanding interrupt at
2353 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2358 * Reduce the number of Ethernet queues across all ports to at most n.
2359 * n provides at least one queue per port.
2361 static void __devinit reduce_ethqs(struct adapter *adapter, int n)
2364 struct port_info *pi;
2367 * While we have too many active Ether Queue Sets, interate across the
2368 * "ports" and reduce their individual Queue Set allocations.
2370 BUG_ON(n < adapter->params.nports);
2371 while (n < adapter->sge.ethqsets)
2372 for_each_port(adapter, i) {
2373 pi = adap2pinfo(adapter, i);
2374 if (pi->nqsets > 1) {
2376 adapter->sge.ethqsets--;
2377 if (adapter->sge.ethqsets <= n)
2383 * Reassign the starting Queue Sets for each of the "ports" ...
2386 for_each_port(adapter, i) {
2387 pi = adap2pinfo(adapter, i);
2394 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2395 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2396 * need. Minimally we need one for every Virtual Interface plus those needed
2397 * for our "extras". Note that this process may lower the maximum number of
2398 * allowed Queue Sets ...
2400 static int __devinit enable_msix(struct adapter *adapter)
2402 int i, err, want, need;
2403 struct msix_entry entries[MSIX_ENTRIES];
2404 struct sge *s = &adapter->sge;
2406 for (i = 0; i < MSIX_ENTRIES; ++i)
2407 entries[i].entry = i;
2410 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2411 * plus those needed for our "extras" (for example, the firmware
2412 * message queue). We _need_ at least one "Queue Set" per Virtual
2413 * Interface plus those needed for our "extras". So now we get to see
2414 * if the song is right ...
2416 want = s->max_ethqsets + MSIX_EXTRAS;
2417 need = adapter->params.nports + MSIX_EXTRAS;
2418 while ((err = pci_enable_msix(adapter->pdev, entries, want)) >= need)
2422 int nqsets = want - MSIX_EXTRAS;
2423 if (nqsets < s->max_ethqsets) {
2424 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2425 " for %d Queue Sets\n", nqsets);
2426 s->max_ethqsets = nqsets;
2427 if (nqsets < s->ethqsets)
2428 reduce_ethqs(adapter, nqsets);
2430 for (i = 0; i < want; ++i)
2431 adapter->msix_info[i].vec = entries[i].vector;
2432 } else if (err > 0) {
2433 pci_disable_msix(adapter->pdev);
2434 dev_info(adapter->pdev_dev, "only %d MSI-X vectors left,"
2435 " not using MSI-X\n", err);
2440 static const struct net_device_ops cxgb4vf_netdev_ops = {
2441 .ndo_open = cxgb4vf_open,
2442 .ndo_stop = cxgb4vf_stop,
2443 .ndo_start_xmit = t4vf_eth_xmit,
2444 .ndo_get_stats = cxgb4vf_get_stats,
2445 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2446 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2447 .ndo_validate_addr = eth_validate_addr,
2448 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2449 .ndo_change_mtu = cxgb4vf_change_mtu,
2450 .ndo_fix_features = cxgb4vf_fix_features,
2451 .ndo_set_features = cxgb4vf_set_features,
2452 #ifdef CONFIG_NET_POLL_CONTROLLER
2453 .ndo_poll_controller = cxgb4vf_poll_controller,
2458 * "Probe" a device: initialize a device and construct all kernel and driver
2459 * state needed to manage the device. This routine is called "init_one" in
2462 static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
2463 const struct pci_device_id *ent)
2465 static int version_printed;
2470 struct adapter *adapter;
2471 struct port_info *pi;
2472 struct net_device *netdev;
2475 * Print our driver banner the first time we're called to initialize a
2478 if (version_printed == 0) {
2479 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
2480 version_printed = 1;
2484 * Initialize generic PCI device state.
2486 err = pci_enable_device(pdev);
2488 dev_err(&pdev->dev, "cannot enable PCI device\n");
2493 * Reserve PCI resources for the device. If we can't get them some
2494 * other driver may have already claimed the device ...
2496 err = pci_request_regions(pdev, KBUILD_MODNAME);
2498 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2499 goto err_disable_device;
2503 * Set up our DMA mask: try for 64-bit address masking first and
2504 * fall back to 32-bit if we can't get 64 bits ...
2506 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2508 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2510 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2511 " coherent allocations\n");
2512 goto err_release_regions;
2516 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2518 dev_err(&pdev->dev, "no usable DMA configuration\n");
2519 goto err_release_regions;
2525 * Enable bus mastering for the device ...
2527 pci_set_master(pdev);
2530 * Allocate our adapter data structure and attach it to the device.
2532 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2535 goto err_release_regions;
2537 pci_set_drvdata(pdev, adapter);
2538 adapter->pdev = pdev;
2539 adapter->pdev_dev = &pdev->dev;
2542 * Initialize SMP data synchronization resources.
2544 spin_lock_init(&adapter->stats_lock);
2547 * Map our I/O registers in BAR0.
2549 adapter->regs = pci_ioremap_bar(pdev, 0);
2550 if (!adapter->regs) {
2551 dev_err(&pdev->dev, "cannot map device registers\n");
2553 goto err_free_adapter;
2557 * Initialize adapter level features.
2559 adapter->name = pci_name(pdev);
2560 adapter->msg_enable = dflt_msg_enable;
2561 err = adap_init0(adapter);
2566 * Allocate our "adapter ports" and stitch everything together.
2568 pmask = adapter->params.vfres.pmask;
2569 for_each_port(adapter, pidx) {
2573 * We simplistically allocate our virtual interfaces
2574 * sequentially across the port numbers to which we have
2575 * access rights. This should be configurable in some manner
2580 port_id = ffs(pmask) - 1;
2581 pmask &= ~(1 << port_id);
2582 viid = t4vf_alloc_vi(adapter, port_id);
2584 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2585 " err=%d\n", port_id, viid);
2591 * Allocate our network device and stitch things together.
2593 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2595 if (netdev == NULL) {
2596 dev_err(&pdev->dev, "cannot allocate netdev for"
2597 " port %d\n", port_id);
2598 t4vf_free_vi(adapter, viid);
2602 adapter->port[pidx] = netdev;
2603 SET_NETDEV_DEV(netdev, &pdev->dev);
2604 pi = netdev_priv(netdev);
2605 pi->adapter = adapter;
2607 pi->port_id = port_id;
2611 * Initialize the starting state of our "port" and register
2614 pi->xact_addr_filt = -1;
2615 netif_carrier_off(netdev);
2616 netdev->irq = pdev->irq;
2618 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
2619 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2620 NETIF_F_HW_VLAN_RX | NETIF_F_RXCSUM;
2621 netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
2622 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2624 netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_TX;
2626 netdev->features |= NETIF_F_HIGHDMA;
2628 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2629 SET_ETHTOOL_OPS(netdev, &cxgb4vf_ethtool_ops);
2632 * Initialize the hardware/software state for the port.
2634 err = t4vf_port_init(adapter, pidx);
2636 dev_err(&pdev->dev, "cannot initialize port %d\n",
2643 * The "card" is now ready to go. If any errors occur during device
2644 * registration we do not fail the whole "card" but rather proceed
2645 * only with the ports we manage to register successfully. However we
2646 * must register at least one net device.
2648 for_each_port(adapter, pidx) {
2649 netdev = adapter->port[pidx];
2653 err = register_netdev(netdev);
2655 dev_warn(&pdev->dev, "cannot register net device %s,"
2656 " skipping\n", netdev->name);
2660 set_bit(pidx, &adapter->registered_device_map);
2662 if (adapter->registered_device_map == 0) {
2663 dev_err(&pdev->dev, "could not register any net devices\n");
2668 * Set up our debugfs entries.
2670 if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) {
2671 adapter->debugfs_root =
2672 debugfs_create_dir(pci_name(pdev),
2673 cxgb4vf_debugfs_root);
2674 if (IS_ERR_OR_NULL(adapter->debugfs_root))
2675 dev_warn(&pdev->dev, "could not create debugfs"
2678 setup_debugfs(adapter);
2682 * See what interrupts we'll be using. If we've been configured to
2683 * use MSI-X interrupts, try to enable them but fall back to using
2684 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
2685 * get MSI interrupts we bail with the error.
2687 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
2688 adapter->flags |= USING_MSIX;
2690 err = pci_enable_msi(pdev);
2692 dev_err(&pdev->dev, "Unable to allocate %s interrupts;"
2694 msi == MSI_MSIX ? "MSI-X or MSI" : "MSI", err);
2695 goto err_free_debugfs;
2697 adapter->flags |= USING_MSI;
2701 * Now that we know how many "ports" we have and what their types are,
2702 * and how many Queue Sets we can support, we can configure our queue
2705 cfg_queues(adapter);
2708 * Print a short notice on the existence and configuration of the new
2709 * VF network device ...
2711 for_each_port(adapter, pidx) {
2712 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
2713 adapter->port[pidx]->name,
2714 (adapter->flags & USING_MSIX) ? "MSI-X" :
2715 (adapter->flags & USING_MSI) ? "MSI" : "");
2724 * Error recovery and exit code. Unwind state that's been created
2725 * so far and return the error.
2729 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
2730 cleanup_debugfs(adapter);
2731 debugfs_remove_recursive(adapter->debugfs_root);
2735 for_each_port(adapter, pidx) {
2736 netdev = adapter->port[pidx];
2739 pi = netdev_priv(netdev);
2740 t4vf_free_vi(adapter, pi->viid);
2741 if (test_bit(pidx, &adapter->registered_device_map))
2742 unregister_netdev(netdev);
2743 free_netdev(netdev);
2747 iounmap(adapter->regs);
2751 pci_set_drvdata(pdev, NULL);
2753 err_release_regions:
2754 pci_release_regions(pdev);
2755 pci_set_drvdata(pdev, NULL);
2756 pci_clear_master(pdev);
2759 pci_disable_device(pdev);
2765 * "Remove" a device: tear down all kernel and driver state created in the
2766 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
2767 * that this is called "remove_one" in the PF Driver.)
2769 static void __devexit cxgb4vf_pci_remove(struct pci_dev *pdev)
2771 struct adapter *adapter = pci_get_drvdata(pdev);
2774 * Tear down driver state associated with device.
2780 * Stop all of our activity. Unregister network port,
2781 * disable interrupts, etc.
2783 for_each_port(adapter, pidx)
2784 if (test_bit(pidx, &adapter->registered_device_map))
2785 unregister_netdev(adapter->port[pidx]);
2786 t4vf_sge_stop(adapter);
2787 if (adapter->flags & USING_MSIX) {
2788 pci_disable_msix(adapter->pdev);
2789 adapter->flags &= ~USING_MSIX;
2790 } else if (adapter->flags & USING_MSI) {
2791 pci_disable_msi(adapter->pdev);
2792 adapter->flags &= ~USING_MSI;
2796 * Tear down our debugfs entries.
2798 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
2799 cleanup_debugfs(adapter);
2800 debugfs_remove_recursive(adapter->debugfs_root);
2804 * Free all of the various resources which we've acquired ...
2806 t4vf_free_sge_resources(adapter);
2807 for_each_port(adapter, pidx) {
2808 struct net_device *netdev = adapter->port[pidx];
2809 struct port_info *pi;
2814 pi = netdev_priv(netdev);
2815 t4vf_free_vi(adapter, pi->viid);
2816 free_netdev(netdev);
2818 iounmap(adapter->regs);
2820 pci_set_drvdata(pdev, NULL);
2824 * Disable the device and release its PCI resources.
2826 pci_disable_device(pdev);
2827 pci_clear_master(pdev);
2828 pci_release_regions(pdev);
2832 * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt
2835 static void __devexit cxgb4vf_pci_shutdown(struct pci_dev *pdev)
2837 struct adapter *adapter;
2840 adapter = pci_get_drvdata(pdev);
2845 * Disable all Virtual Interfaces. This will shut down the
2846 * delivery of all ingress packets into the chip for these
2847 * Virtual Interfaces.
2849 for_each_port(adapter, pidx) {
2850 struct net_device *netdev;
2851 struct port_info *pi;
2853 if (!test_bit(pidx, &adapter->registered_device_map))
2856 netdev = adapter->port[pidx];
2860 pi = netdev_priv(netdev);
2861 t4vf_enable_vi(adapter, pi->viid, false, false);
2865 * Free up all Queues which will prevent further DMA and
2866 * Interrupts allowing various internal pathways to drain.
2868 t4vf_free_sge_resources(adapter);
2872 * PCI Device registration data structures.
2874 #define CH_DEVICE(devid, idx) \
2875 { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
2877 static struct pci_device_id cxgb4vf_pci_tbl[] = {
2878 CH_DEVICE(0xb000, 0), /* PE10K FPGA */
2879 CH_DEVICE(0x4800, 0), /* T440-dbg */
2880 CH_DEVICE(0x4801, 0), /* T420-cr */
2881 CH_DEVICE(0x4802, 0), /* T422-cr */
2882 CH_DEVICE(0x4803, 0), /* T440-cr */
2883 CH_DEVICE(0x4804, 0), /* T420-bch */
2884 CH_DEVICE(0x4805, 0), /* T440-bch */
2885 CH_DEVICE(0x4806, 0), /* T460-ch */
2886 CH_DEVICE(0x4807, 0), /* T420-so */
2887 CH_DEVICE(0x4808, 0), /* T420-cx */
2888 CH_DEVICE(0x4809, 0), /* T420-bt */
2889 CH_DEVICE(0x480a, 0), /* T404-bt */
2893 MODULE_DESCRIPTION(DRV_DESC);
2894 MODULE_AUTHOR("Chelsio Communications");
2895 MODULE_LICENSE("Dual BSD/GPL");
2896 MODULE_VERSION(DRV_VERSION);
2897 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
2899 static struct pci_driver cxgb4vf_driver = {
2900 .name = KBUILD_MODNAME,
2901 .id_table = cxgb4vf_pci_tbl,
2902 .probe = cxgb4vf_pci_probe,
2903 .remove = __devexit_p(cxgb4vf_pci_remove),
2904 .shutdown = __devexit_p(cxgb4vf_pci_shutdown),
2908 * Initialize global driver state.
2910 static int __init cxgb4vf_module_init(void)
2915 * Vet our module parameters.
2917 if (msi != MSI_MSIX && msi != MSI_MSI) {
2918 printk(KERN_WARNING KBUILD_MODNAME
2919 ": bad module parameter msi=%d; must be %d"
2920 " (MSI-X or MSI) or %d (MSI)\n",
2921 msi, MSI_MSIX, MSI_MSI);
2925 /* Debugfs support is optional, just warn if this fails */
2926 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
2927 if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
2928 printk(KERN_WARNING KBUILD_MODNAME ": could not create"
2929 " debugfs entry, continuing\n");
2931 ret = pci_register_driver(&cxgb4vf_driver);
2932 if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
2933 debugfs_remove(cxgb4vf_debugfs_root);
2938 * Tear down global driver state.
2940 static void __exit cxgb4vf_module_exit(void)
2942 pci_unregister_driver(&cxgb4vf_driver);
2943 debugfs_remove(cxgb4vf_debugfs_root);
2946 module_init(cxgb4vf_module_init);
2947 module_exit(cxgb4vf_module_exit);