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/version.h>
37 #include <linux/module.h>
38 #include <linux/moduleparam.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/netdevice.h>
43 #include <linux/etherdevice.h>
44 #include <linux/debugfs.h>
45 #include <linux/ethtool.h>
47 #include "t4vf_common.h"
48 #include "t4vf_defs.h"
50 #include "../cxgb4/t4_regs.h"
51 #include "../cxgb4/t4_msg.h"
54 * Generic information about the driver.
56 #define DRV_VERSION "1.0.0"
57 #define DRV_DESC "Chelsio T4 Virtual Function (VF) Network Driver"
65 * Default ethtool "message level" for adapters.
67 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
68 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
69 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
71 static int dflt_msg_enable = DFLT_MSG_ENABLE;
73 module_param(dflt_msg_enable, int, 0644);
74 MODULE_PARM_DESC(dflt_msg_enable,
75 "default adapter ethtool message level bitmap");
78 * The driver uses the best interrupt scheme available on a platform in the
79 * order MSI-X then MSI. This parameter determines which of these schemes the
80 * driver may consider as follows:
82 * msi = 2: choose from among MSI-X and MSI
83 * msi = 1: only consider MSI interrupts
85 * Note that unlike the Physical Function driver, this Virtual Function driver
86 * does _not_ support legacy INTx interrupts (this limitation is mandated by
87 * the PCI-E SR-IOV standard).
91 #define MSI_DEFAULT MSI_MSIX
93 static int msi = MSI_DEFAULT;
95 module_param(msi, int, 0644);
96 MODULE_PARM_DESC(msi, "whether to use MSI-X or MSI");
99 * Fundamental constants.
100 * ======================
104 MAX_TXQ_ENTRIES = 16384,
105 MAX_RSPQ_ENTRIES = 16384,
106 MAX_RX_BUFFERS = 16384,
108 MIN_TXQ_ENTRIES = 32,
109 MIN_RSPQ_ENTRIES = 128,
113 * For purposes of manipulating the Free List size we need to
114 * recognize that Free Lists are actually Egress Queues (the host
115 * produces free buffers which the hardware consumes), Egress Queues
116 * indices are all in units of Egress Context Units bytes, and free
117 * list entries are 64-bit PCI DMA addresses. And since the state of
118 * the Producer Index == the Consumer Index implies an EMPTY list, we
119 * always have at least one Egress Unit's worth of Free List entries
120 * unused. See sge.c for more details ...
122 EQ_UNIT = SGE_EQ_IDXSIZE,
123 FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64),
124 MIN_FL_RESID = FL_PER_EQ_UNIT,
128 * Global driver state.
129 * ====================
132 static struct dentry *cxgb4vf_debugfs_root;
135 * OS "Callback" functions.
136 * ========================
140 * The link status has changed on the indicated "port" (Virtual Interface).
142 void t4vf_os_link_changed(struct adapter *adapter, int pidx, int link_ok)
144 struct net_device *dev = adapter->port[pidx];
147 * If the port is disabled or the current recorded "link up"
148 * status matches the new status, just return.
150 if (!netif_running(dev) || link_ok == netif_carrier_ok(dev))
154 * Tell the OS that the link status has changed and print a short
155 * informative message on the console about the event.
160 const struct port_info *pi = netdev_priv(dev);
162 netif_carrier_on(dev);
164 switch (pi->link_cfg.speed) {
182 switch (pi->link_cfg.fc) {
191 case PAUSE_RX|PAUSE_TX:
200 printk(KERN_INFO "%s: link up, %s, full-duplex, %s PAUSE\n",
203 netif_carrier_off(dev);
204 printk(KERN_INFO "%s: link down\n", dev->name);
209 * Net device operations.
210 * ======================
214 * Record our new VLAN Group and enable/disable hardware VLAN Tag extraction
215 * based on whether the specified VLAN Group pointer is NULL or not.
217 static void cxgb4vf_vlan_rx_register(struct net_device *dev,
218 struct vlan_group *grp)
220 struct port_info *pi = netdev_priv(dev);
223 t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1, grp != NULL, 0);
227 * Perform the MAC and PHY actions needed to enable a "port" (Virtual
230 static int link_start(struct net_device *dev)
233 struct port_info *pi = netdev_priv(dev);
236 * We do not set address filters and promiscuity here, the stack does
237 * that step explicitly.
239 ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, -1,
242 ret = t4vf_change_mac(pi->adapter, pi->viid,
243 pi->xact_addr_filt, dev->dev_addr, true);
245 pi->xact_addr_filt = ret;
251 * We don't need to actually "start the link" itself since the
252 * firmware will do that for us when the first Virtual Interface
253 * is enabled on a port.
256 ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true);
261 * Name the MSI-X interrupts.
263 static void name_msix_vecs(struct adapter *adapter)
265 int namelen = sizeof(adapter->msix_info[0].desc) - 1;
271 snprintf(adapter->msix_info[MSIX_FW].desc, namelen,
272 "%s-FWeventq", adapter->name);
273 adapter->msix_info[MSIX_FW].desc[namelen] = 0;
278 for_each_port(adapter, pidx) {
279 struct net_device *dev = adapter->port[pidx];
280 const struct port_info *pi = netdev_priv(dev);
283 for (qs = 0, msi = MSIX_NIQFLINT;
286 snprintf(adapter->msix_info[msi].desc, namelen,
287 "%s-%d", dev->name, qs);
288 adapter->msix_info[msi].desc[namelen] = 0;
294 * Request all of our MSI-X resources.
296 static int request_msix_queue_irqs(struct adapter *adapter)
298 struct sge *s = &adapter->sge;
304 err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix,
305 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq);
313 for_each_ethrxq(s, rxq) {
314 err = request_irq(adapter->msix_info[msi].vec,
315 t4vf_sge_intr_msix, 0,
316 adapter->msix_info[msi].desc,
317 &s->ethrxq[rxq].rspq);
326 free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq);
327 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
332 * Free our MSI-X resources.
334 static void free_msix_queue_irqs(struct adapter *adapter)
336 struct sge *s = &adapter->sge;
339 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
341 for_each_ethrxq(s, rxq)
342 free_irq(adapter->msix_info[msi++].vec,
343 &s->ethrxq[rxq].rspq);
347 * Turn on NAPI and start up interrupts on a response queue.
349 static void qenable(struct sge_rspq *rspq)
351 napi_enable(&rspq->napi);
354 * 0-increment the Going To Sleep register to start the timer and
357 t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
359 SEINTARM(rspq->intr_params) |
360 INGRESSQID(rspq->cntxt_id));
364 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
366 static void enable_rx(struct adapter *adapter)
369 struct sge *s = &adapter->sge;
371 for_each_ethrxq(s, rxq)
372 qenable(&s->ethrxq[rxq].rspq);
373 qenable(&s->fw_evtq);
376 * The interrupt queue doesn't use NAPI so we do the 0-increment of
377 * its Going To Sleep register here to get it started.
379 if (adapter->flags & USING_MSI)
380 t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
382 SEINTARM(s->intrq.intr_params) |
383 INGRESSQID(s->intrq.cntxt_id));
388 * Wait until all NAPI handlers are descheduled.
390 static void quiesce_rx(struct adapter *adapter)
392 struct sge *s = &adapter->sge;
395 for_each_ethrxq(s, rxq)
396 napi_disable(&s->ethrxq[rxq].rspq.napi);
397 napi_disable(&s->fw_evtq.napi);
401 * Response queue handler for the firmware event queue.
403 static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp,
404 const struct pkt_gl *gl)
407 * Extract response opcode and get pointer to CPL message body.
409 struct adapter *adapter = rspq->adapter;
410 u8 opcode = ((const struct rss_header *)rsp)->opcode;
411 void *cpl = (void *)(rsp + 1);
416 * We've received an asynchronous message from the firmware.
418 const struct cpl_fw6_msg *fw_msg = cpl;
419 if (fw_msg->type == FW6_TYPE_CMD_RPL)
420 t4vf_handle_fw_rpl(adapter, fw_msg->data);
424 case CPL_SGE_EGR_UPDATE: {
426 * We've received an Egress Queue Status Update message. We
427 * get these, if the SGE is configured to send these when the
428 * firmware passes certain points in processing our TX
429 * Ethernet Queue or if we make an explicit request for one.
430 * We use these updates to determine when we may need to
431 * restart a TX Ethernet Queue which was stopped for lack of
432 * free TX Queue Descriptors ...
434 const struct cpl_sge_egr_update *p = (void *)cpl;
435 unsigned int qid = EGR_QID(be32_to_cpu(p->opcode_qid));
436 struct sge *s = &adapter->sge;
438 struct sge_eth_txq *txq;
442 * Perform sanity checking on the Queue ID to make sure it
443 * really refers to one of our TX Ethernet Egress Queues which
444 * is active and matches the queue's ID. None of these error
445 * conditions should ever happen so we may want to either make
446 * them fatal and/or conditionalized under DEBUG.
448 eq_idx = EQ_IDX(s, qid);
449 if (unlikely(eq_idx >= MAX_EGRQ)) {
450 dev_err(adapter->pdev_dev,
451 "Egress Update QID %d out of range\n", qid);
454 tq = s->egr_map[eq_idx];
455 if (unlikely(tq == NULL)) {
456 dev_err(adapter->pdev_dev,
457 "Egress Update QID %d TXQ=NULL\n", qid);
460 txq = container_of(tq, struct sge_eth_txq, q);
461 if (unlikely(tq->abs_id != qid)) {
462 dev_err(adapter->pdev_dev,
463 "Egress Update QID %d refers to TXQ %d\n",
469 * Restart a stopped TX Queue which has less than half of its
473 netif_tx_wake_queue(txq->txq);
478 dev_err(adapter->pdev_dev,
479 "unexpected CPL %#x on FW event queue\n", opcode);
486 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
487 * to use and initializes them. We support multiple "Queue Sets" per port if
488 * we have MSI-X, otherwise just one queue set per port.
490 static int setup_sge_queues(struct adapter *adapter)
492 struct sge *s = &adapter->sge;
496 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
499 bitmap_zero(s->starving_fl, MAX_EGRQ);
502 * If we're using MSI interrupt mode we need to set up a "forwarded
503 * interrupt" queue which we'll set up with our MSI vector. The rest
504 * of the ingress queues will be set up to forward their interrupts to
505 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
506 * the intrq's queue ID as the interrupt forwarding queue for the
507 * subsequent calls ...
509 if (adapter->flags & USING_MSI) {
510 err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false,
511 adapter->port[0], 0, NULL, NULL);
513 goto err_free_queues;
517 * Allocate our ingress queue for asynchronous firmware messages.
519 err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0],
520 MSIX_FW, NULL, fwevtq_handler);
522 goto err_free_queues;
525 * Allocate each "port"'s initial Queue Sets. These can be changed
526 * later on ... up to the point where any interface on the adapter is
527 * brought up at which point lots of things get nailed down
530 msix = MSIX_NIQFLINT;
531 for_each_port(adapter, pidx) {
532 struct net_device *dev = adapter->port[pidx];
533 struct port_info *pi = netdev_priv(dev);
534 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
535 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
538 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
539 err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false,
541 &rxq->fl, t4vf_ethrx_handler);
543 goto err_free_queues;
545 err = t4vf_sge_alloc_eth_txq(adapter, txq, dev,
546 netdev_get_tx_queue(dev, qs),
547 s->fw_evtq.cntxt_id);
549 goto err_free_queues;
552 memset(&rxq->stats, 0, sizeof(rxq->stats));
557 * Create the reverse mappings for the queues.
559 s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id;
560 s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id;
561 IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq;
562 for_each_port(adapter, pidx) {
563 struct net_device *dev = adapter->port[pidx];
564 struct port_info *pi = netdev_priv(dev);
565 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
566 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
569 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
570 IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq;
571 EQ_MAP(s, txq->q.abs_id) = &txq->q;
574 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
575 * for Free Lists but since all of the Egress Queues
576 * (including Free Lists) have Relative Queue IDs
577 * which are computed as Absolute - Base Queue ID, we
578 * can synthesize the Absolute Queue IDs for the Free
579 * Lists. This is useful for debugging purposes when
580 * we want to dump Queue Contexts via the PF Driver.
582 rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base;
583 EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl;
589 t4vf_free_sge_resources(adapter);
594 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
595 * queues. We configure the RSS CPU lookup table to distribute to the number
596 * of HW receive queues, and the response queue lookup table to narrow that
597 * down to the response queues actually configured for each "port" (Virtual
598 * Interface). We always configure the RSS mapping for all ports since the
599 * mapping table has plenty of entries.
601 static int setup_rss(struct adapter *adapter)
605 for_each_port(adapter, pidx) {
606 struct port_info *pi = adap2pinfo(adapter, pidx);
607 struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
608 u16 rss[MAX_PORT_QSETS];
611 for (qs = 0; qs < pi->nqsets; qs++)
612 rss[qs] = rxq[qs].rspq.abs_id;
614 err = t4vf_config_rss_range(adapter, pi->viid,
615 0, pi->rss_size, rss, pi->nqsets);
620 * Perform Global RSS Mode-specific initialization.
622 switch (adapter->params.rss.mode) {
623 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL:
625 * If Tunnel All Lookup isn't specified in the global
626 * RSS Configuration, then we need to specify a
627 * default Ingress Queue for any ingress packets which
628 * aren't hashed. We'll use our first ingress queue
631 if (!adapter->params.rss.u.basicvirtual.tnlalllookup) {
632 union rss_vi_config config;
633 err = t4vf_read_rss_vi_config(adapter,
638 config.basicvirtual.defaultq =
640 err = t4vf_write_rss_vi_config(adapter,
654 * Bring the adapter up. Called whenever we go from no "ports" open to having
655 * one open. This function performs the actions necessary to make an adapter
656 * operational, such as completing the initialization of HW modules, and
657 * enabling interrupts. Must be called with the rtnl lock held. (Note that
658 * this is called "cxgb_up" in the PF Driver.)
660 static int adapter_up(struct adapter *adapter)
665 * If this is the first time we've been called, perform basic
666 * adapter setup. Once we've done this, many of our adapter
667 * parameters can no longer be changed ...
669 if ((adapter->flags & FULL_INIT_DONE) == 0) {
670 err = setup_sge_queues(adapter);
673 err = setup_rss(adapter);
675 t4vf_free_sge_resources(adapter);
679 if (adapter->flags & USING_MSIX)
680 name_msix_vecs(adapter);
681 adapter->flags |= FULL_INIT_DONE;
685 * Acquire our interrupt resources. We only support MSI-X and MSI.
687 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
688 if (adapter->flags & USING_MSIX)
689 err = request_msix_queue_irqs(adapter);
691 err = request_irq(adapter->pdev->irq,
692 t4vf_intr_handler(adapter), 0,
693 adapter->name, adapter);
695 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
701 * Enable NAPI ingress processing and return success.
704 t4vf_sge_start(adapter);
709 * Bring the adapter down. Called whenever the last "port" (Virtual
710 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
713 static void adapter_down(struct adapter *adapter)
716 * Free interrupt resources.
718 if (adapter->flags & USING_MSIX)
719 free_msix_queue_irqs(adapter);
721 free_irq(adapter->pdev->irq, adapter);
724 * Wait for NAPI handlers to finish.
730 * Start up a net device.
732 static int cxgb4vf_open(struct net_device *dev)
735 struct port_info *pi = netdev_priv(dev);
736 struct adapter *adapter = pi->adapter;
739 * If this is the first interface that we're opening on the "adapter",
740 * bring the "adapter" up now.
742 if (adapter->open_device_map == 0) {
743 err = adapter_up(adapter);
749 * Note that this interface is up and start everything up ...
751 netif_set_real_num_tx_queues(dev, pi->nqsets);
752 err = netif_set_real_num_rx_queues(dev, pi->nqsets);
755 set_bit(pi->port_id, &adapter->open_device_map);
756 err = link_start(dev);
759 netif_tx_start_all_queues(dev);
764 * Shut down a net device. This routine is called "cxgb_close" in the PF
767 static int cxgb4vf_stop(struct net_device *dev)
770 struct port_info *pi = netdev_priv(dev);
771 struct adapter *adapter = pi->adapter;
773 netif_tx_stop_all_queues(dev);
774 netif_carrier_off(dev);
775 ret = t4vf_enable_vi(adapter, pi->viid, false, false);
776 pi->link_cfg.link_ok = 0;
778 clear_bit(pi->port_id, &adapter->open_device_map);
779 if (adapter->open_device_map == 0)
780 adapter_down(adapter);
785 * Translate our basic statistics into the standard "ifconfig" statistics.
787 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
789 struct t4vf_port_stats stats;
790 struct port_info *pi = netdev2pinfo(dev);
791 struct adapter *adapter = pi->adapter;
792 struct net_device_stats *ns = &dev->stats;
795 spin_lock(&adapter->stats_lock);
796 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
797 spin_unlock(&adapter->stats_lock);
799 memset(ns, 0, sizeof(*ns));
803 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
804 stats.tx_ucast_bytes + stats.tx_offload_bytes);
805 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
806 stats.tx_ucast_frames + stats.tx_offload_frames);
807 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
808 stats.rx_ucast_bytes);
809 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
810 stats.rx_ucast_frames);
811 ns->multicast = stats.rx_mcast_frames;
812 ns->tx_errors = stats.tx_drop_frames;
813 ns->rx_errors = stats.rx_err_frames;
819 * Collect up to maxaddrs worth of a netdevice's unicast addresses, starting
820 * at a specified offset within the list, into an array of addrss pointers and
821 * return the number collected.
823 static inline unsigned int collect_netdev_uc_list_addrs(const struct net_device *dev,
826 unsigned int maxaddrs)
828 unsigned int index = 0;
829 unsigned int naddr = 0;
830 const struct netdev_hw_addr *ha;
832 for_each_dev_addr(dev, ha)
833 if (index++ >= offset) {
834 addr[naddr++] = ha->addr;
835 if (naddr >= maxaddrs)
842 * Collect up to maxaddrs worth of a netdevice's multicast addresses, starting
843 * at a specified offset within the list, into an array of addrss pointers and
844 * return the number collected.
846 static inline unsigned int collect_netdev_mc_list_addrs(const struct net_device *dev,
849 unsigned int maxaddrs)
851 unsigned int index = 0;
852 unsigned int naddr = 0;
853 const struct netdev_hw_addr *ha;
855 netdev_for_each_mc_addr(ha, dev)
856 if (index++ >= offset) {
857 addr[naddr++] = ha->addr;
858 if (naddr >= maxaddrs)
865 * Configure the exact and hash address filters to handle a port's multicast
866 * and secondary unicast MAC addresses.
868 static int set_addr_filters(const struct net_device *dev, bool sleep)
873 unsigned int offset, naddr;
876 const struct port_info *pi = netdev_priv(dev);
878 /* first do the secondary unicast addresses */
879 for (offset = 0; ; offset += naddr) {
880 naddr = collect_netdev_uc_list_addrs(dev, addr, offset,
885 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
886 naddr, addr, NULL, &uhash, sleep);
893 /* next set up the multicast addresses */
894 for (offset = 0; ; offset += naddr) {
895 naddr = collect_netdev_mc_list_addrs(dev, addr, offset,
900 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
901 naddr, addr, NULL, &mhash, sleep);
907 return t4vf_set_addr_hash(pi->adapter, pi->viid, uhash != 0,
908 uhash | mhash, sleep);
912 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
913 * If @mtu is -1 it is left unchanged.
915 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
918 struct port_info *pi = netdev_priv(dev);
920 ret = set_addr_filters(dev, sleep_ok);
922 ret = t4vf_set_rxmode(pi->adapter, pi->viid, -1,
923 (dev->flags & IFF_PROMISC) != 0,
924 (dev->flags & IFF_ALLMULTI) != 0,
930 * Set the current receive modes on the device.
932 static void cxgb4vf_set_rxmode(struct net_device *dev)
934 /* unfortunately we can't return errors to the stack */
935 set_rxmode(dev, -1, false);
939 * Find the entry in the interrupt holdoff timer value array which comes
940 * closest to the specified interrupt holdoff value.
942 static int closest_timer(const struct sge *s, int us)
944 int i, timer_idx = 0, min_delta = INT_MAX;
946 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
947 int delta = us - s->timer_val[i];
950 if (delta < min_delta) {
958 static int closest_thres(const struct sge *s, int thres)
960 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
962 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
963 delta = thres - s->counter_val[i];
966 if (delta < min_delta) {
975 * Return a queue's interrupt hold-off time in us. 0 means no timer.
977 static unsigned int qtimer_val(const struct adapter *adapter,
978 const struct sge_rspq *rspq)
980 unsigned int timer_idx = QINTR_TIMER_IDX_GET(rspq->intr_params);
982 return timer_idx < SGE_NTIMERS
983 ? adapter->sge.timer_val[timer_idx]
988 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
989 * @adapter: the adapter
990 * @rspq: the RX response queue
991 * @us: the hold-off time in us, or 0 to disable timer
992 * @cnt: the hold-off packet count, or 0 to disable counter
994 * Sets an RX response queue's interrupt hold-off time and packet count.
995 * At least one of the two needs to be enabled for the queue to generate
998 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
999 unsigned int us, unsigned int cnt)
1001 unsigned int timer_idx;
1004 * If both the interrupt holdoff timer and count are specified as
1005 * zero, default to a holdoff count of 1 ...
1007 if ((us | cnt) == 0)
1011 * If an interrupt holdoff count has been specified, then find the
1012 * closest configured holdoff count and use that. If the response
1013 * queue has already been created, then update its queue context
1020 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1021 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1022 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1024 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1025 FW_PARAMS_PARAM_YZ(rspq->cntxt_id);
1026 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1030 rspq->pktcnt_idx = pktcnt_idx;
1034 * Compute the closest holdoff timer index from the supplied holdoff
1037 timer_idx = (us == 0
1038 ? SGE_TIMER_RSTRT_CNTR
1039 : closest_timer(&adapter->sge, us));
1042 * Update the response queue's interrupt coalescing parameters and
1045 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
1046 (cnt > 0 ? QINTR_CNT_EN : 0));
1051 * Return a version number to identify the type of adapter. The scheme is:
1052 * - bits 0..9: chip version
1053 * - bits 10..15: chip revision
1055 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1058 * Chip version 4, revision 0x3f (cxgb4vf).
1060 return 4 | (0x3f << 10);
1064 * Execute the specified ioctl command.
1066 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1072 * The VF Driver doesn't have access to any of the other
1073 * common Ethernet device ioctl()'s (like reading/writing
1074 * PHY registers, etc.
1085 * Change the device's MTU.
1087 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1090 struct port_info *pi = netdev_priv(dev);
1092 /* accommodate SACK */
1096 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1097 -1, -1, -1, -1, true);
1104 * Change the devices MAC address.
1106 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1109 struct sockaddr *addr = _addr;
1110 struct port_info *pi = netdev_priv(dev);
1112 if (!is_valid_ether_addr(addr->sa_data))
1115 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1116 addr->sa_data, true);
1120 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1121 pi->xact_addr_filt = ret;
1125 #ifdef CONFIG_NET_POLL_CONTROLLER
1127 * Poll all of our receive queues. This is called outside of normal interrupt
1130 static void cxgb4vf_poll_controller(struct net_device *dev)
1132 struct port_info *pi = netdev_priv(dev);
1133 struct adapter *adapter = pi->adapter;
1135 if (adapter->flags & USING_MSIX) {
1136 struct sge_eth_rxq *rxq;
1139 rxq = &adapter->sge.ethrxq[pi->first_qset];
1140 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1141 t4vf_sge_intr_msix(0, &rxq->rspq);
1145 t4vf_intr_handler(adapter)(0, adapter);
1150 * Ethtool operations.
1151 * ===================
1153 * Note that we don't support any ethtool operations which change the physical
1154 * state of the port to which we're linked.
1158 * Return current port link settings.
1160 static int cxgb4vf_get_settings(struct net_device *dev,
1161 struct ethtool_cmd *cmd)
1163 const struct port_info *pi = netdev_priv(dev);
1165 cmd->supported = pi->link_cfg.supported;
1166 cmd->advertising = pi->link_cfg.advertising;
1167 cmd->speed = netif_carrier_ok(dev) ? pi->link_cfg.speed : -1;
1168 cmd->duplex = DUPLEX_FULL;
1170 cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1171 cmd->phy_address = pi->port_id;
1172 cmd->transceiver = XCVR_EXTERNAL;
1173 cmd->autoneg = pi->link_cfg.autoneg;
1180 * Return our driver information.
1182 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1183 struct ethtool_drvinfo *drvinfo)
1185 struct adapter *adapter = netdev2adap(dev);
1187 strcpy(drvinfo->driver, KBUILD_MODNAME);
1188 strcpy(drvinfo->version, DRV_VERSION);
1189 strcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)));
1190 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1191 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1192 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.fwrev),
1193 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.fwrev),
1194 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.fwrev),
1195 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.fwrev),
1196 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.tprev),
1197 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.tprev),
1198 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.tprev),
1199 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.tprev));
1203 * Return current adapter message level.
1205 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1207 return netdev2adap(dev)->msg_enable;
1211 * Set current adapter message level.
1213 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1215 netdev2adap(dev)->msg_enable = msglevel;
1219 * Return the device's current Queue Set ring size parameters along with the
1220 * allowed maximum values. Since ethtool doesn't understand the concept of
1221 * multi-queue devices, we just return the current values associated with the
1224 static void cxgb4vf_get_ringparam(struct net_device *dev,
1225 struct ethtool_ringparam *rp)
1227 const struct port_info *pi = netdev_priv(dev);
1228 const struct sge *s = &pi->adapter->sge;
1230 rp->rx_max_pending = MAX_RX_BUFFERS;
1231 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1232 rp->rx_jumbo_max_pending = 0;
1233 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1235 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1236 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1237 rp->rx_jumbo_pending = 0;
1238 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1242 * Set the Queue Set ring size parameters for the device. Again, since
1243 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1244 * apply these new values across all of the Queue Sets associated with the
1245 * device -- after vetting them of course!
1247 static int cxgb4vf_set_ringparam(struct net_device *dev,
1248 struct ethtool_ringparam *rp)
1250 const struct port_info *pi = netdev_priv(dev);
1251 struct adapter *adapter = pi->adapter;
1252 struct sge *s = &adapter->sge;
1255 if (rp->rx_pending > MAX_RX_BUFFERS ||
1256 rp->rx_jumbo_pending ||
1257 rp->tx_pending > MAX_TXQ_ENTRIES ||
1258 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1259 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1260 rp->rx_pending < MIN_FL_ENTRIES ||
1261 rp->tx_pending < MIN_TXQ_ENTRIES)
1264 if (adapter->flags & FULL_INIT_DONE)
1267 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1268 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1269 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1270 s->ethtxq[qs].q.size = rp->tx_pending;
1276 * Return the interrupt holdoff timer and count for the first Queue Set on the
1277 * device. Our extension ioctl() (the cxgbtool interface) allows the
1278 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1280 static int cxgb4vf_get_coalesce(struct net_device *dev,
1281 struct ethtool_coalesce *coalesce)
1283 const struct port_info *pi = netdev_priv(dev);
1284 const struct adapter *adapter = pi->adapter;
1285 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1287 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1288 coalesce->rx_max_coalesced_frames =
1289 ((rspq->intr_params & QINTR_CNT_EN)
1290 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1296 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1297 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1298 * the interrupt holdoff timer on any of the device's Queue Sets.
1300 static int cxgb4vf_set_coalesce(struct net_device *dev,
1301 struct ethtool_coalesce *coalesce)
1303 const struct port_info *pi = netdev_priv(dev);
1304 struct adapter *adapter = pi->adapter;
1306 return set_rxq_intr_params(adapter,
1307 &adapter->sge.ethrxq[pi->first_qset].rspq,
1308 coalesce->rx_coalesce_usecs,
1309 coalesce->rx_max_coalesced_frames);
1313 * Report current port link pause parameter settings.
1315 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1316 struct ethtool_pauseparam *pauseparam)
1318 struct port_info *pi = netdev_priv(dev);
1320 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1321 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1322 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1326 * Return whether RX Checksum Offloading is currently enabled for the device.
1328 static u32 cxgb4vf_get_rx_csum(struct net_device *dev)
1330 struct port_info *pi = netdev_priv(dev);
1332 return (pi->rx_offload & RX_CSO) != 0;
1336 * Turn RX Checksum Offloading on or off for the device.
1338 static int cxgb4vf_set_rx_csum(struct net_device *dev, u32 csum)
1340 struct port_info *pi = netdev_priv(dev);
1343 pi->rx_offload |= RX_CSO;
1345 pi->rx_offload &= ~RX_CSO;
1350 * Identify the port by blinking the port's LED.
1352 static int cxgb4vf_phys_id(struct net_device *dev, u32 id)
1354 struct port_info *pi = netdev_priv(dev);
1356 return t4vf_identify_port(pi->adapter, pi->viid, 5);
1360 * Port stats maintained per queue of the port.
1362 struct queue_port_stats {
1371 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1372 * these need to match the order of statistics returned by
1373 * t4vf_get_port_stats().
1375 static const char stats_strings[][ETH_GSTRING_LEN] = {
1377 * These must match the layout of the t4vf_port_stats structure.
1379 "TxBroadcastBytes ",
1380 "TxBroadcastFrames ",
1381 "TxMulticastBytes ",
1382 "TxMulticastFrames ",
1388 "RxBroadcastBytes ",
1389 "RxBroadcastFrames ",
1390 "RxMulticastBytes ",
1391 "RxMulticastFrames ",
1397 * These are accumulated per-queue statistics and must match the
1398 * order of the fields in the queue_port_stats structure.
1408 * Return the number of statistics in the specified statistics set.
1410 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1414 return ARRAY_SIZE(stats_strings);
1422 * Return the strings for the specified statistics set.
1424 static void cxgb4vf_get_strings(struct net_device *dev,
1430 memcpy(data, stats_strings, sizeof(stats_strings));
1436 * Small utility routine to accumulate queue statistics across the queues of
1439 static void collect_sge_port_stats(const struct adapter *adapter,
1440 const struct port_info *pi,
1441 struct queue_port_stats *stats)
1443 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1444 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1447 memset(stats, 0, sizeof(*stats));
1448 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1449 stats->tso += txq->tso;
1450 stats->tx_csum += txq->tx_cso;
1451 stats->rx_csum += rxq->stats.rx_cso;
1452 stats->vlan_ex += rxq->stats.vlan_ex;
1453 stats->vlan_ins += txq->vlan_ins;
1458 * Return the ETH_SS_STATS statistics set.
1460 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1461 struct ethtool_stats *stats,
1464 struct port_info *pi = netdev2pinfo(dev);
1465 struct adapter *adapter = pi->adapter;
1466 int err = t4vf_get_port_stats(adapter, pi->pidx,
1467 (struct t4vf_port_stats *)data);
1469 memset(data, 0, sizeof(struct t4vf_port_stats));
1471 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1472 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1476 * Return the size of our register map.
1478 static int cxgb4vf_get_regs_len(struct net_device *dev)
1480 return T4VF_REGMAP_SIZE;
1484 * Dump a block of registers, start to end inclusive, into a buffer.
1486 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1487 unsigned int start, unsigned int end)
1489 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1491 for ( ; start <= end; start += sizeof(u32)) {
1493 * Avoid reading the Mailbox Control register since that
1494 * can trigger a Mailbox Ownership Arbitration cycle and
1495 * interfere with communication with the firmware.
1497 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1500 *bp++ = t4_read_reg(adapter, start);
1505 * Copy our entire register map into the provided buffer.
1507 static void cxgb4vf_get_regs(struct net_device *dev,
1508 struct ethtool_regs *regs,
1511 struct adapter *adapter = netdev2adap(dev);
1513 regs->version = mk_adap_vers(adapter);
1516 * Fill in register buffer with our register map.
1518 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1520 reg_block_dump(adapter, regbuf,
1521 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1522 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1523 reg_block_dump(adapter, regbuf,
1524 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1525 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1526 reg_block_dump(adapter, regbuf,
1527 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1528 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_LAST);
1529 reg_block_dump(adapter, regbuf,
1530 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1531 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1533 reg_block_dump(adapter, regbuf,
1534 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1535 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1539 * Report current Wake On LAN settings.
1541 static void cxgb4vf_get_wol(struct net_device *dev,
1542 struct ethtool_wolinfo *wol)
1546 memset(&wol->sopass, 0, sizeof(wol->sopass));
1550 * Set TCP Segmentation Offloading feature capabilities.
1552 static int cxgb4vf_set_tso(struct net_device *dev, u32 tso)
1555 dev->features |= NETIF_F_TSO | NETIF_F_TSO6;
1557 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
1561 static struct ethtool_ops cxgb4vf_ethtool_ops = {
1562 .get_settings = cxgb4vf_get_settings,
1563 .get_drvinfo = cxgb4vf_get_drvinfo,
1564 .get_msglevel = cxgb4vf_get_msglevel,
1565 .set_msglevel = cxgb4vf_set_msglevel,
1566 .get_ringparam = cxgb4vf_get_ringparam,
1567 .set_ringparam = cxgb4vf_set_ringparam,
1568 .get_coalesce = cxgb4vf_get_coalesce,
1569 .set_coalesce = cxgb4vf_set_coalesce,
1570 .get_pauseparam = cxgb4vf_get_pauseparam,
1571 .get_rx_csum = cxgb4vf_get_rx_csum,
1572 .set_rx_csum = cxgb4vf_set_rx_csum,
1573 .set_tx_csum = ethtool_op_set_tx_ipv6_csum,
1574 .set_sg = ethtool_op_set_sg,
1575 .get_link = ethtool_op_get_link,
1576 .get_strings = cxgb4vf_get_strings,
1577 .phys_id = cxgb4vf_phys_id,
1578 .get_sset_count = cxgb4vf_get_sset_count,
1579 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1580 .get_regs_len = cxgb4vf_get_regs_len,
1581 .get_regs = cxgb4vf_get_regs,
1582 .get_wol = cxgb4vf_get_wol,
1583 .set_tso = cxgb4vf_set_tso,
1587 * /sys/kernel/debug/cxgb4vf support code and data.
1588 * ================================================
1592 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1596 static int sge_qinfo_show(struct seq_file *seq, void *v)
1598 struct adapter *adapter = seq->private;
1599 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1600 int qs, r = (uintptr_t)v - 1;
1603 seq_putc(seq, '\n');
1605 #define S3(fmt_spec, s, v) \
1607 seq_printf(seq, "%-12s", s); \
1608 for (qs = 0; qs < n; ++qs) \
1609 seq_printf(seq, " %16" fmt_spec, v); \
1610 seq_putc(seq, '\n'); \
1612 #define S(s, v) S3("s", s, v)
1613 #define T(s, v) S3("u", s, txq[qs].v)
1614 #define R(s, v) S3("u", s, rxq[qs].v)
1616 if (r < eth_entries) {
1617 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1618 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1619 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1621 S("QType:", "Ethernet");
1623 (rxq[qs].rspq.netdev
1624 ? rxq[qs].rspq.netdev->name
1627 (rxq[qs].rspq.netdev
1628 ? ((struct port_info *)
1629 netdev_priv(rxq[qs].rspq.netdev))->port_id
1631 T("TxQ ID:", q.abs_id);
1632 T("TxQ size:", q.size);
1633 T("TxQ inuse:", q.in_use);
1634 T("TxQ PIdx:", q.pidx);
1635 T("TxQ CIdx:", q.cidx);
1636 R("RspQ ID:", rspq.abs_id);
1637 R("RspQ size:", rspq.size);
1638 R("RspQE size:", rspq.iqe_len);
1639 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1640 S3("u", "Intr pktcnt:",
1641 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1642 R("RspQ CIdx:", rspq.cidx);
1643 R("RspQ Gen:", rspq.gen);
1644 R("FL ID:", fl.abs_id);
1645 R("FL size:", fl.size - MIN_FL_RESID);
1646 R("FL avail:", fl.avail);
1647 R("FL PIdx:", fl.pidx);
1648 R("FL CIdx:", fl.cidx);
1654 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1656 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1657 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1658 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1659 qtimer_val(adapter, evtq));
1660 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1661 adapter->sge.counter_val[evtq->pktcnt_idx]);
1662 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1663 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1664 } else if (r == 1) {
1665 const struct sge_rspq *intrq = &adapter->sge.intrq;
1667 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1668 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1669 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1670 qtimer_val(adapter, intrq));
1671 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1672 adapter->sge.counter_val[intrq->pktcnt_idx]);
1673 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1674 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1686 * Return the number of "entries" in our "file". We group the multi-Queue
1687 * sections with QPL Queue Sets per "entry". The sections of the output are:
1689 * Ethernet RX/TX Queue Sets
1690 * Firmware Event Queue
1691 * Forwarded Interrupt Queue (if in MSI mode)
1693 static int sge_queue_entries(const struct adapter *adapter)
1695 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1696 ((adapter->flags & USING_MSI) != 0);
1699 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1701 int entries = sge_queue_entries(seq->private);
1703 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1706 static void sge_queue_stop(struct seq_file *seq, void *v)
1710 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
1712 int entries = sge_queue_entries(seq->private);
1715 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1718 static const struct seq_operations sge_qinfo_seq_ops = {
1719 .start = sge_queue_start,
1720 .next = sge_queue_next,
1721 .stop = sge_queue_stop,
1722 .show = sge_qinfo_show
1725 static int sge_qinfo_open(struct inode *inode, struct file *file)
1727 int res = seq_open(file, &sge_qinfo_seq_ops);
1730 struct seq_file *seq = file->private_data;
1731 seq->private = inode->i_private;
1736 static const struct file_operations sge_qinfo_debugfs_fops = {
1737 .owner = THIS_MODULE,
1738 .open = sge_qinfo_open,
1740 .llseek = seq_lseek,
1741 .release = seq_release,
1745 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
1749 static int sge_qstats_show(struct seq_file *seq, void *v)
1751 struct adapter *adapter = seq->private;
1752 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1753 int qs, r = (uintptr_t)v - 1;
1756 seq_putc(seq, '\n');
1758 #define S3(fmt, s, v) \
1760 seq_printf(seq, "%-16s", s); \
1761 for (qs = 0; qs < n; ++qs) \
1762 seq_printf(seq, " %8" fmt, v); \
1763 seq_putc(seq, '\n'); \
1765 #define S(s, v) S3("s", s, v)
1767 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
1768 #define T(s, v) T3("lu", s, v)
1770 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
1771 #define R(s, v) R3("lu", s, v)
1773 if (r < eth_entries) {
1774 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1775 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1776 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1778 S("QType:", "Ethernet");
1780 (rxq[qs].rspq.netdev
1781 ? rxq[qs].rspq.netdev->name
1783 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
1784 R("RxPackets:", stats.pkts);
1785 R("RxCSO:", stats.rx_cso);
1786 R("VLANxtract:", stats.vlan_ex);
1787 R("LROmerged:", stats.lro_merged);
1788 R("LROpackets:", stats.lro_pkts);
1789 R("RxDrops:", stats.rx_drops);
1791 T("TxCSO:", tx_cso);
1792 T("VLANins:", vlan_ins);
1793 T("TxQFull:", q.stops);
1794 T("TxQRestarts:", q.restarts);
1795 T("TxMapErr:", mapping_err);
1796 R("FLAllocErr:", fl.alloc_failed);
1797 R("FLLrgAlcErr:", fl.large_alloc_failed);
1798 R("FLStarving:", fl.starving);
1804 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1806 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
1807 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1808 evtq->unhandled_irqs);
1809 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
1810 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
1811 } else if (r == 1) {
1812 const struct sge_rspq *intrq = &adapter->sge.intrq;
1814 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
1815 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1816 intrq->unhandled_irqs);
1817 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
1818 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
1832 * Return the number of "entries" in our "file". We group the multi-Queue
1833 * sections with QPL Queue Sets per "entry". The sections of the output are:
1835 * Ethernet RX/TX Queue Sets
1836 * Firmware Event Queue
1837 * Forwarded Interrupt Queue (if in MSI mode)
1839 static int sge_qstats_entries(const struct adapter *adapter)
1841 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1842 ((adapter->flags & USING_MSI) != 0);
1845 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
1847 int entries = sge_qstats_entries(seq->private);
1849 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1852 static void sge_qstats_stop(struct seq_file *seq, void *v)
1856 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
1858 int entries = sge_qstats_entries(seq->private);
1861 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1864 static const struct seq_operations sge_qstats_seq_ops = {
1865 .start = sge_qstats_start,
1866 .next = sge_qstats_next,
1867 .stop = sge_qstats_stop,
1868 .show = sge_qstats_show
1871 static int sge_qstats_open(struct inode *inode, struct file *file)
1873 int res = seq_open(file, &sge_qstats_seq_ops);
1876 struct seq_file *seq = file->private_data;
1877 seq->private = inode->i_private;
1882 static const struct file_operations sge_qstats_proc_fops = {
1883 .owner = THIS_MODULE,
1884 .open = sge_qstats_open,
1886 .llseek = seq_lseek,
1887 .release = seq_release,
1891 * Show PCI-E SR-IOV Virtual Function Resource Limits.
1893 static int resources_show(struct seq_file *seq, void *v)
1895 struct adapter *adapter = seq->private;
1896 struct vf_resources *vfres = &adapter->params.vfres;
1898 #define S(desc, fmt, var) \
1899 seq_printf(seq, "%-60s " fmt "\n", \
1900 desc " (" #var "):", vfres->var)
1902 S("Virtual Interfaces", "%d", nvi);
1903 S("Egress Queues", "%d", neq);
1904 S("Ethernet Control", "%d", nethctrl);
1905 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
1906 S("Ingress Queues", "%d", niq);
1907 S("Traffic Class", "%d", tc);
1908 S("Port Access Rights Mask", "%#x", pmask);
1909 S("MAC Address Filters", "%d", nexactf);
1910 S("Firmware Command Read Capabilities", "%#x", r_caps);
1911 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
1918 static int resources_open(struct inode *inode, struct file *file)
1920 return single_open(file, resources_show, inode->i_private);
1923 static const struct file_operations resources_proc_fops = {
1924 .owner = THIS_MODULE,
1925 .open = resources_open,
1927 .llseek = seq_lseek,
1928 .release = single_release,
1932 * Show Virtual Interfaces.
1934 static int interfaces_show(struct seq_file *seq, void *v)
1936 if (v == SEQ_START_TOKEN) {
1937 seq_puts(seq, "Interface Port VIID\n");
1939 struct adapter *adapter = seq->private;
1940 int pidx = (uintptr_t)v - 2;
1941 struct net_device *dev = adapter->port[pidx];
1942 struct port_info *pi = netdev_priv(dev);
1944 seq_printf(seq, "%9s %4d %#5x\n",
1945 dev->name, pi->port_id, pi->viid);
1950 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
1952 return pos <= adapter->params.nports
1953 ? (void *)(uintptr_t)(pos + 1)
1957 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
1960 ? interfaces_get_idx(seq->private, *pos)
1964 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
1967 return interfaces_get_idx(seq->private, *pos);
1970 static void interfaces_stop(struct seq_file *seq, void *v)
1974 static const struct seq_operations interfaces_seq_ops = {
1975 .start = interfaces_start,
1976 .next = interfaces_next,
1977 .stop = interfaces_stop,
1978 .show = interfaces_show
1981 static int interfaces_open(struct inode *inode, struct file *file)
1983 int res = seq_open(file, &interfaces_seq_ops);
1986 struct seq_file *seq = file->private_data;
1987 seq->private = inode->i_private;
1992 static const struct file_operations interfaces_proc_fops = {
1993 .owner = THIS_MODULE,
1994 .open = interfaces_open,
1996 .llseek = seq_lseek,
1997 .release = seq_release,
2001 * /sys/kernel/debugfs/cxgb4vf/ files list.
2003 struct cxgb4vf_debugfs_entry {
2004 const char *name; /* name of debugfs node */
2005 mode_t mode; /* file system mode */
2006 const struct file_operations *fops;
2009 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
2010 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
2011 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
2012 { "resources", S_IRUGO, &resources_proc_fops },
2013 { "interfaces", S_IRUGO, &interfaces_proc_fops },
2017 * Module and device initialization and cleanup code.
2018 * ==================================================
2022 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2023 * directory (debugfs_root) has already been set up.
2025 static int __devinit setup_debugfs(struct adapter *adapter)
2029 BUG_ON(adapter->debugfs_root == NULL);
2032 * Debugfs support is best effort.
2034 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2035 (void)debugfs_create_file(debugfs_files[i].name,
2036 debugfs_files[i].mode,
2037 adapter->debugfs_root,
2039 debugfs_files[i].fops);
2045 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2046 * it to our caller to tear down the directory (debugfs_root).
2048 static void __devexit cleanup_debugfs(struct adapter *adapter)
2050 BUG_ON(adapter->debugfs_root == NULL);
2053 * Unlike our sister routine cleanup_proc(), we don't need to remove
2054 * individual entries because a call will be made to
2055 * debugfs_remove_recursive(). We just need to clean up any ancillary
2062 * Perform early "adapter" initialization. This is where we discover what
2063 * adapter parameters we're going to be using and initialize basic adapter
2066 static int adap_init0(struct adapter *adapter)
2068 struct vf_resources *vfres = &adapter->params.vfres;
2069 struct sge_params *sge_params = &adapter->params.sge;
2070 struct sge *s = &adapter->sge;
2071 unsigned int ethqsets;
2075 * Wait for the device to become ready before proceeding ...
2077 err = t4vf_wait_dev_ready(adapter);
2079 dev_err(adapter->pdev_dev, "device didn't become ready:"
2085 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2086 * 2.6.31 and later we can't call pci_reset_function() in order to
2087 * issue an FLR because of a self- deadlock on the device semaphore.
2088 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2089 * cases where they're needed -- for instance, some versions of KVM
2090 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2091 * use the firmware based reset in order to reset any per function
2094 err = t4vf_fw_reset(adapter);
2096 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2101 * Grab basic operational parameters. These will predominantly have
2102 * been set up by the Physical Function Driver or will be hard coded
2103 * into the adapter. We just have to live with them ... Note that
2104 * we _must_ get our VPD parameters before our SGE parameters because
2105 * we need to know the adapter's core clock from the VPD in order to
2106 * properly decode the SGE Timer Values.
2108 err = t4vf_get_dev_params(adapter);
2110 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2111 " device parameters: err=%d\n", err);
2114 err = t4vf_get_vpd_params(adapter);
2116 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2117 " VPD parameters: err=%d\n", err);
2120 err = t4vf_get_sge_params(adapter);
2122 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2123 " SGE parameters: err=%d\n", err);
2126 err = t4vf_get_rss_glb_config(adapter);
2128 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2129 " RSS parameters: err=%d\n", err);
2132 if (adapter->params.rss.mode !=
2133 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2134 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2135 " mode %d\n", adapter->params.rss.mode);
2138 err = t4vf_sge_init(adapter);
2140 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2146 * Retrieve our RX interrupt holdoff timer values and counter
2147 * threshold values from the SGE parameters.
2149 s->timer_val[0] = core_ticks_to_us(adapter,
2150 TIMERVALUE0_GET(sge_params->sge_timer_value_0_and_1));
2151 s->timer_val[1] = core_ticks_to_us(adapter,
2152 TIMERVALUE1_GET(sge_params->sge_timer_value_0_and_1));
2153 s->timer_val[2] = core_ticks_to_us(adapter,
2154 TIMERVALUE0_GET(sge_params->sge_timer_value_2_and_3));
2155 s->timer_val[3] = core_ticks_to_us(adapter,
2156 TIMERVALUE1_GET(sge_params->sge_timer_value_2_and_3));
2157 s->timer_val[4] = core_ticks_to_us(adapter,
2158 TIMERVALUE0_GET(sge_params->sge_timer_value_4_and_5));
2159 s->timer_val[5] = core_ticks_to_us(adapter,
2160 TIMERVALUE1_GET(sge_params->sge_timer_value_4_and_5));
2163 THRESHOLD_0_GET(sge_params->sge_ingress_rx_threshold);
2165 THRESHOLD_1_GET(sge_params->sge_ingress_rx_threshold);
2167 THRESHOLD_2_GET(sge_params->sge_ingress_rx_threshold);
2169 THRESHOLD_3_GET(sge_params->sge_ingress_rx_threshold);
2172 * Grab our Virtual Interface resource allocation, extract the
2173 * features that we're interested in and do a bit of sanity testing on
2176 err = t4vf_get_vfres(adapter);
2178 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2179 " resources: err=%d\n", err);
2184 * The number of "ports" which we support is equal to the number of
2185 * Virtual Interfaces with which we've been provisioned.
2187 adapter->params.nports = vfres->nvi;
2188 if (adapter->params.nports > MAX_NPORTS) {
2189 dev_warn(adapter->pdev_dev, "only using %d of %d allowed"
2190 " virtual interfaces\n", MAX_NPORTS,
2191 adapter->params.nports);
2192 adapter->params.nports = MAX_NPORTS;
2196 * We need to reserve a number of the ingress queues with Free List
2197 * and Interrupt capabilities for special interrupt purposes (like
2198 * asynchronous firmware messages, or forwarded interrupts if we're
2199 * using MSI). The rest of the FL/Intr-capable ingress queues will be
2200 * matched up one-for-one with Ethernet/Control egress queues in order
2201 * to form "Queue Sets" which will be aportioned between the "ports".
2202 * For each Queue Set, we'll need the ability to allocate two Egress
2203 * Contexts -- one for the Ingress Queue Free List and one for the TX
2206 ethqsets = vfres->niqflint - INGQ_EXTRAS;
2207 if (vfres->nethctrl != ethqsets) {
2208 dev_warn(adapter->pdev_dev, "unequal number of [available]"
2209 " ingress/egress queues (%d/%d); using minimum for"
2210 " number of Queue Sets\n", ethqsets, vfres->nethctrl);
2211 ethqsets = min(vfres->nethctrl, ethqsets);
2213 if (vfres->neq < ethqsets*2) {
2214 dev_warn(adapter->pdev_dev, "Not enough Egress Contexts (%d)"
2215 " to support Queue Sets (%d); reducing allowed Queue"
2216 " Sets\n", vfres->neq, ethqsets);
2217 ethqsets = vfres->neq/2;
2219 if (ethqsets > MAX_ETH_QSETS) {
2220 dev_warn(adapter->pdev_dev, "only using %d of %d allowed Queue"
2221 " Sets\n", MAX_ETH_QSETS, adapter->sge.max_ethqsets);
2222 ethqsets = MAX_ETH_QSETS;
2224 if (vfres->niq != 0 || vfres->neq > ethqsets*2) {
2225 dev_warn(adapter->pdev_dev, "unused resources niq/neq (%d/%d)"
2226 " ignored\n", vfres->niq, vfres->neq - ethqsets*2);
2228 adapter->sge.max_ethqsets = ethqsets;
2231 * Check for various parameter sanity issues. Most checks simply
2232 * result in us using fewer resources than our provissioning but we
2233 * do need at least one "port" with which to work ...
2235 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2236 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2237 " virtual interfaces (too few Queue Sets)\n",
2238 adapter->sge.max_ethqsets, adapter->params.nports);
2239 adapter->params.nports = adapter->sge.max_ethqsets;
2241 if (adapter->params.nports == 0) {
2242 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2249 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2250 u8 pkt_cnt_idx, unsigned int size,
2251 unsigned int iqe_size)
2253 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
2254 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0));
2255 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2258 rspq->iqe_len = iqe_size;
2263 * Perform default configuration of DMA queues depending on the number and
2264 * type of ports we found and the number of available CPUs. Most settings can
2265 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2266 * being brought up for the first time.
2268 static void __devinit cfg_queues(struct adapter *adapter)
2270 struct sge *s = &adapter->sge;
2271 int q10g, n10g, qidx, pidx, qs;
2275 * We should not be called till we know how many Queue Sets we can
2276 * support. In particular, this means that we need to know what kind
2277 * of interrupts we'll be using ...
2279 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2282 * Count the number of 10GbE Virtual Interfaces that we have.
2285 for_each_port(adapter, pidx)
2286 n10g += is_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2289 * We default to 1 queue per non-10G port and up to # of cores queues
2295 int n1g = (adapter->params.nports - n10g);
2296 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2297 if (q10g > num_online_cpus())
2298 q10g = num_online_cpus();
2302 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2303 * The layout will be established in setup_sge_queues() when the
2304 * adapter is brough up for the first time.
2307 for_each_port(adapter, pidx) {
2308 struct port_info *pi = adap2pinfo(adapter, pidx);
2310 pi->first_qset = qidx;
2311 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
2317 * The Ingress Queue Entry Size for our various Response Queues needs
2318 * to be big enough to accommodate the largest message we can receive
2319 * from the chip/firmware; which is 64 bytes ...
2324 * Set up default Queue Set parameters ... Start off with the
2325 * shortest interrupt holdoff timer.
2327 for (qs = 0; qs < s->max_ethqsets; qs++) {
2328 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2329 struct sge_eth_txq *txq = &s->ethtxq[qs];
2331 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2337 * The firmware event queue is used for link state changes and
2338 * notifications of TX DMA completions.
2340 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2343 * The forwarded interrupt queue is used when we're in MSI interrupt
2344 * mode. In this mode all interrupts associated with RX queues will
2345 * be forwarded to a single queue which we'll associate with our MSI
2346 * interrupt vector. The messages dropped in the forwarded interrupt
2347 * queue will indicate which ingress queue needs servicing ... This
2348 * queue needs to be large enough to accommodate all of the ingress
2349 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2350 * from equalling the CIDX if every ingress queue has an outstanding
2351 * interrupt). The queue doesn't need to be any larger because no
2352 * ingress queue will ever have more than one outstanding interrupt at
2355 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2360 * Reduce the number of Ethernet queues across all ports to at most n.
2361 * n provides at least one queue per port.
2363 static void __devinit reduce_ethqs(struct adapter *adapter, int n)
2366 struct port_info *pi;
2369 * While we have too many active Ether Queue Sets, interate across the
2370 * "ports" and reduce their individual Queue Set allocations.
2372 BUG_ON(n < adapter->params.nports);
2373 while (n < adapter->sge.ethqsets)
2374 for_each_port(adapter, i) {
2375 pi = adap2pinfo(adapter, i);
2376 if (pi->nqsets > 1) {
2378 adapter->sge.ethqsets--;
2379 if (adapter->sge.ethqsets <= n)
2385 * Reassign the starting Queue Sets for each of the "ports" ...
2388 for_each_port(adapter, i) {
2389 pi = adap2pinfo(adapter, i);
2396 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2397 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2398 * need. Minimally we need one for every Virtual Interface plus those needed
2399 * for our "extras". Note that this process may lower the maximum number of
2400 * allowed Queue Sets ...
2402 static int __devinit enable_msix(struct adapter *adapter)
2404 int i, err, want, need;
2405 struct msix_entry entries[MSIX_ENTRIES];
2406 struct sge *s = &adapter->sge;
2408 for (i = 0; i < MSIX_ENTRIES; ++i)
2409 entries[i].entry = i;
2412 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2413 * plus those needed for our "extras" (for example, the firmware
2414 * message queue). We _need_ at least one "Queue Set" per Virtual
2415 * Interface plus those needed for our "extras". So now we get to see
2416 * if the song is right ...
2418 want = s->max_ethqsets + MSIX_EXTRAS;
2419 need = adapter->params.nports + MSIX_EXTRAS;
2420 while ((err = pci_enable_msix(adapter->pdev, entries, want)) >= need)
2424 int nqsets = want - MSIX_EXTRAS;
2425 if (nqsets < s->max_ethqsets) {
2426 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2427 " for %d Queue Sets\n", nqsets);
2428 s->max_ethqsets = nqsets;
2429 if (nqsets < s->ethqsets)
2430 reduce_ethqs(adapter, nqsets);
2432 for (i = 0; i < want; ++i)
2433 adapter->msix_info[i].vec = entries[i].vector;
2434 } else if (err > 0) {
2435 pci_disable_msix(adapter->pdev);
2436 dev_info(adapter->pdev_dev, "only %d MSI-X vectors left,"
2437 " not using MSI-X\n", err);
2442 #ifdef HAVE_NET_DEVICE_OPS
2443 static const struct net_device_ops cxgb4vf_netdev_ops = {
2444 .ndo_open = cxgb4vf_open,
2445 .ndo_stop = cxgb4vf_stop,
2446 .ndo_start_xmit = t4vf_eth_xmit,
2447 .ndo_get_stats = cxgb4vf_get_stats,
2448 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2449 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2450 .ndo_validate_addr = eth_validate_addr,
2451 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2452 .ndo_change_mtu = cxgb4vf_change_mtu,
2453 .ndo_vlan_rx_register = cxgb4vf_vlan_rx_register,
2454 #ifdef CONFIG_NET_POLL_CONTROLLER
2455 .ndo_poll_controller = cxgb4vf_poll_controller,
2461 * "Probe" a device: initialize a device and construct all kernel and driver
2462 * state needed to manage the device. This routine is called "init_one" in
2465 static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
2466 const struct pci_device_id *ent)
2468 static int version_printed;
2473 struct adapter *adapter;
2474 struct port_info *pi;
2475 struct net_device *netdev;
2478 * Vet our module parameters.
2480 if (msi != MSI_MSIX && msi != MSI_MSI) {
2481 dev_err(&pdev->dev, "bad module parameter msi=%d; must be %d"
2482 " (MSI-X or MSI) or %d (MSI)\n", msi, MSI_MSIX,
2489 * Print our driver banner the first time we're called to initialize a
2492 if (version_printed == 0) {
2493 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
2494 version_printed = 1;
2499 * Initialize generic PCI device state.
2501 err = pci_enable_device(pdev);
2503 dev_err(&pdev->dev, "cannot enable PCI device\n");
2508 * Reserve PCI resources for the device. If we can't get them some
2509 * other driver may have already claimed the device ...
2511 err = pci_request_regions(pdev, KBUILD_MODNAME);
2513 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2514 goto err_disable_device;
2518 * Set up our DMA mask: try for 64-bit address masking first and
2519 * fall back to 32-bit if we can't get 64 bits ...
2521 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2523 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2525 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2526 " coherent allocations\n");
2527 goto err_release_regions;
2531 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2533 dev_err(&pdev->dev, "no usable DMA configuration\n");
2534 goto err_release_regions;
2540 * Enable bus mastering for the device ...
2542 pci_set_master(pdev);
2545 * Allocate our adapter data structure and attach it to the device.
2547 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2550 goto err_release_regions;
2552 pci_set_drvdata(pdev, adapter);
2553 adapter->pdev = pdev;
2554 adapter->pdev_dev = &pdev->dev;
2557 * Initialize SMP data synchronization resources.
2559 spin_lock_init(&adapter->stats_lock);
2562 * Map our I/O registers in BAR0.
2564 adapter->regs = pci_ioremap_bar(pdev, 0);
2565 if (!adapter->regs) {
2566 dev_err(&pdev->dev, "cannot map device registers\n");
2568 goto err_free_adapter;
2572 * Initialize adapter level features.
2574 adapter->name = pci_name(pdev);
2575 adapter->msg_enable = dflt_msg_enable;
2576 err = adap_init0(adapter);
2581 * Allocate our "adapter ports" and stitch everything together.
2583 pmask = adapter->params.vfres.pmask;
2584 for_each_port(adapter, pidx) {
2588 * We simplistically allocate our virtual interfaces
2589 * sequentially across the port numbers to which we have
2590 * access rights. This should be configurable in some manner
2595 port_id = ffs(pmask) - 1;
2596 pmask &= ~(1 << port_id);
2597 viid = t4vf_alloc_vi(adapter, port_id);
2599 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2600 " err=%d\n", port_id, viid);
2606 * Allocate our network device and stitch things together.
2608 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2610 if (netdev == NULL) {
2611 dev_err(&pdev->dev, "cannot allocate netdev for"
2612 " port %d\n", port_id);
2613 t4vf_free_vi(adapter, viid);
2617 adapter->port[pidx] = netdev;
2618 SET_NETDEV_DEV(netdev, &pdev->dev);
2619 pi = netdev_priv(netdev);
2620 pi->adapter = adapter;
2622 pi->port_id = port_id;
2626 * Initialize the starting state of our "port" and register
2629 pi->xact_addr_filt = -1;
2630 pi->rx_offload = RX_CSO;
2631 netif_carrier_off(netdev);
2632 netdev->irq = pdev->irq;
2634 netdev->features = (NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
2635 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2636 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
2639 netdev->features |= NETIF_F_HIGHDMA;
2640 netdev->vlan_features =
2642 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX));
2644 #ifdef HAVE_NET_DEVICE_OPS
2645 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2647 netdev->vlan_rx_register = cxgb4vf_vlan_rx_register;
2648 netdev->open = cxgb4vf_open;
2649 netdev->stop = cxgb4vf_stop;
2650 netdev->hard_start_xmit = t4vf_eth_xmit;
2651 netdev->get_stats = cxgb4vf_get_stats;
2652 netdev->set_rx_mode = cxgb4vf_set_rxmode;
2653 netdev->do_ioctl = cxgb4vf_do_ioctl;
2654 netdev->change_mtu = cxgb4vf_change_mtu;
2655 netdev->set_mac_address = cxgb4vf_set_mac_addr;
2656 #ifdef CONFIG_NET_POLL_CONTROLLER
2657 netdev->poll_controller = cxgb4vf_poll_controller;
2660 SET_ETHTOOL_OPS(netdev, &cxgb4vf_ethtool_ops);
2663 * Initialize the hardware/software state for the port.
2665 err = t4vf_port_init(adapter, pidx);
2667 dev_err(&pdev->dev, "cannot initialize port %d\n",
2674 * The "card" is now ready to go. If any errors occur during device
2675 * registration we do not fail the whole "card" but rather proceed
2676 * only with the ports we manage to register successfully. However we
2677 * must register at least one net device.
2679 for_each_port(adapter, pidx) {
2680 netdev = adapter->port[pidx];
2684 err = register_netdev(netdev);
2686 dev_warn(&pdev->dev, "cannot register net device %s,"
2687 " skipping\n", netdev->name);
2691 set_bit(pidx, &adapter->registered_device_map);
2693 if (adapter->registered_device_map == 0) {
2694 dev_err(&pdev->dev, "could not register any net devices\n");
2699 * Set up our debugfs entries.
2701 if (cxgb4vf_debugfs_root) {
2702 adapter->debugfs_root =
2703 debugfs_create_dir(pci_name(pdev),
2704 cxgb4vf_debugfs_root);
2705 if (adapter->debugfs_root == NULL)
2706 dev_warn(&pdev->dev, "could not create debugfs"
2709 setup_debugfs(adapter);
2713 * See what interrupts we'll be using. If we've been configured to
2714 * use MSI-X interrupts, try to enable them but fall back to using
2715 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
2716 * get MSI interrupts we bail with the error.
2718 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
2719 adapter->flags |= USING_MSIX;
2721 err = pci_enable_msi(pdev);
2723 dev_err(&pdev->dev, "Unable to allocate %s interrupts;"
2725 msi == MSI_MSIX ? "MSI-X or MSI" : "MSI", err);
2726 goto err_free_debugfs;
2728 adapter->flags |= USING_MSI;
2732 * Now that we know how many "ports" we have and what their types are,
2733 * and how many Queue Sets we can support, we can configure our queue
2736 cfg_queues(adapter);
2739 * Print a short notice on the existance and configuration of the new
2740 * VF network device ...
2742 for_each_port(adapter, pidx) {
2743 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
2744 adapter->port[pidx]->name,
2745 (adapter->flags & USING_MSIX) ? "MSI-X" :
2746 (adapter->flags & USING_MSI) ? "MSI" : "");
2755 * Error recovery and exit code. Unwind state that's been created
2756 * so far and return the error.
2760 if (adapter->debugfs_root) {
2761 cleanup_debugfs(adapter);
2762 debugfs_remove_recursive(adapter->debugfs_root);
2766 for_each_port(adapter, pidx) {
2767 netdev = adapter->port[pidx];
2770 pi = netdev_priv(netdev);
2771 t4vf_free_vi(adapter, pi->viid);
2772 if (test_bit(pidx, &adapter->registered_device_map))
2773 unregister_netdev(netdev);
2774 free_netdev(netdev);
2778 iounmap(adapter->regs);
2782 pci_set_drvdata(pdev, NULL);
2784 err_release_regions:
2785 pci_release_regions(pdev);
2786 pci_set_drvdata(pdev, NULL);
2787 pci_clear_master(pdev);
2790 pci_disable_device(pdev);
2797 * "Remove" a device: tear down all kernel and driver state created in the
2798 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
2799 * that this is called "remove_one" in the PF Driver.)
2801 static void __devexit cxgb4vf_pci_remove(struct pci_dev *pdev)
2803 struct adapter *adapter = pci_get_drvdata(pdev);
2806 * Tear down driver state associated with device.
2812 * Stop all of our activity. Unregister network port,
2813 * disable interrupts, etc.
2815 for_each_port(adapter, pidx)
2816 if (test_bit(pidx, &adapter->registered_device_map))
2817 unregister_netdev(adapter->port[pidx]);
2818 t4vf_sge_stop(adapter);
2819 if (adapter->flags & USING_MSIX) {
2820 pci_disable_msix(adapter->pdev);
2821 adapter->flags &= ~USING_MSIX;
2822 } else if (adapter->flags & USING_MSI) {
2823 pci_disable_msi(adapter->pdev);
2824 adapter->flags &= ~USING_MSI;
2828 * Tear down our debugfs entries.
2830 if (adapter->debugfs_root) {
2831 cleanup_debugfs(adapter);
2832 debugfs_remove_recursive(adapter->debugfs_root);
2836 * Free all of the various resources which we've acquired ...
2838 t4vf_free_sge_resources(adapter);
2839 for_each_port(adapter, pidx) {
2840 struct net_device *netdev = adapter->port[pidx];
2841 struct port_info *pi;
2846 pi = netdev_priv(netdev);
2847 t4vf_free_vi(adapter, pi->viid);
2848 free_netdev(netdev);
2850 iounmap(adapter->regs);
2852 pci_set_drvdata(pdev, NULL);
2856 * Disable the device and release its PCI resources.
2858 pci_disable_device(pdev);
2859 pci_clear_master(pdev);
2860 pci_release_regions(pdev);
2864 * PCI Device registration data structures.
2866 #define CH_DEVICE(devid, idx) \
2867 { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
2869 static struct pci_device_id cxgb4vf_pci_tbl[] = {
2870 CH_DEVICE(0xb000, 0), /* PE10K FPGA */
2871 CH_DEVICE(0x4800, 0), /* T440-dbg */
2872 CH_DEVICE(0x4801, 0), /* T420-cr */
2873 CH_DEVICE(0x4802, 0), /* T422-cr */
2874 CH_DEVICE(0x4803, 0), /* T440-cr */
2875 CH_DEVICE(0x4804, 0), /* T420-bch */
2876 CH_DEVICE(0x4805, 0), /* T440-bch */
2877 CH_DEVICE(0x4806, 0), /* T460-ch */
2878 CH_DEVICE(0x4807, 0), /* T420-so */
2879 CH_DEVICE(0x4808, 0), /* T420-cx */
2880 CH_DEVICE(0x4809, 0), /* T420-bt */
2881 CH_DEVICE(0x480a, 0), /* T404-bt */
2885 MODULE_DESCRIPTION(DRV_DESC);
2886 MODULE_AUTHOR("Chelsio Communications");
2887 MODULE_LICENSE("Dual BSD/GPL");
2888 MODULE_VERSION(DRV_VERSION);
2889 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
2891 static struct pci_driver cxgb4vf_driver = {
2892 .name = KBUILD_MODNAME,
2893 .id_table = cxgb4vf_pci_tbl,
2894 .probe = cxgb4vf_pci_probe,
2895 .remove = __devexit_p(cxgb4vf_pci_remove),
2899 * Initialize global driver state.
2901 static int __init cxgb4vf_module_init(void)
2905 /* Debugfs support is optional, just warn if this fails */
2906 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
2907 if (!cxgb4vf_debugfs_root)
2908 printk(KERN_WARNING KBUILD_MODNAME ": could not create"
2909 " debugfs entry, continuing\n");
2911 ret = pci_register_driver(&cxgb4vf_driver);
2913 debugfs_remove(cxgb4vf_debugfs_root);
2918 * Tear down global driver state.
2920 static void __exit cxgb4vf_module_exit(void)
2922 pci_unregister_driver(&cxgb4vf_driver);
2923 debugfs_remove(cxgb4vf_debugfs_root);
2926 module_init(cxgb4vf_module_init);
2927 module_exit(cxgb4vf_module_exit);