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_IQFLINT; qs < pi->nqsets; qs++, msi++) {
284 snprintf(adapter->msix_info[msi].desc, namelen,
285 "%s-%d", dev->name, qs);
286 adapter->msix_info[msi].desc[namelen] = 0;
292 * Request all of our MSI-X resources.
294 static int request_msix_queue_irqs(struct adapter *adapter)
296 struct sge *s = &adapter->sge;
302 err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix,
303 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq);
311 for_each_ethrxq(s, rxq) {
312 err = request_irq(adapter->msix_info[msi].vec,
313 t4vf_sge_intr_msix, 0,
314 adapter->msix_info[msi].desc,
315 &s->ethrxq[rxq].rspq);
324 free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq);
325 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
330 * Free our MSI-X resources.
332 static void free_msix_queue_irqs(struct adapter *adapter)
334 struct sge *s = &adapter->sge;
337 free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq);
339 for_each_ethrxq(s, rxq)
340 free_irq(adapter->msix_info[msi++].vec,
341 &s->ethrxq[rxq].rspq);
345 * Turn on NAPI and start up interrupts on a response queue.
347 static void qenable(struct sge_rspq *rspq)
349 napi_enable(&rspq->napi);
352 * 0-increment the Going To Sleep register to start the timer and
355 t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
357 SEINTARM(rspq->intr_params) |
358 INGRESSQID(rspq->cntxt_id));
362 * Enable NAPI scheduling and interrupt generation for all Receive Queues.
364 static void enable_rx(struct adapter *adapter)
367 struct sge *s = &adapter->sge;
369 for_each_ethrxq(s, rxq)
370 qenable(&s->ethrxq[rxq].rspq);
371 qenable(&s->fw_evtq);
374 * The interrupt queue doesn't use NAPI so we do the 0-increment of
375 * its Going To Sleep register here to get it started.
377 if (adapter->flags & USING_MSI)
378 t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
380 SEINTARM(s->intrq.intr_params) |
381 INGRESSQID(s->intrq.cntxt_id));
386 * Wait until all NAPI handlers are descheduled.
388 static void quiesce_rx(struct adapter *adapter)
390 struct sge *s = &adapter->sge;
393 for_each_ethrxq(s, rxq)
394 napi_disable(&s->ethrxq[rxq].rspq.napi);
395 napi_disable(&s->fw_evtq.napi);
399 * Response queue handler for the firmware event queue.
401 static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp,
402 const struct pkt_gl *gl)
405 * Extract response opcode and get pointer to CPL message body.
407 struct adapter *adapter = rspq->adapter;
408 u8 opcode = ((const struct rss_header *)rsp)->opcode;
409 void *cpl = (void *)(rsp + 1);
414 * We've received an asynchronous message from the firmware.
416 const struct cpl_fw6_msg *fw_msg = cpl;
417 if (fw_msg->type == FW6_TYPE_CMD_RPL)
418 t4vf_handle_fw_rpl(adapter, fw_msg->data);
422 case CPL_SGE_EGR_UPDATE: {
424 * We've received an Egress Queue Status Update message. We
425 * get these, if the SGE is configured to send these when the
426 * firmware passes certain points in processing our TX
427 * Ethernet Queue or if we make an explicit request for one.
428 * We use these updates to determine when we may need to
429 * restart a TX Ethernet Queue which was stopped for lack of
430 * free TX Queue Descriptors ...
432 const struct cpl_sge_egr_update *p = (void *)cpl;
433 unsigned int qid = EGR_QID(be32_to_cpu(p->opcode_qid));
434 struct sge *s = &adapter->sge;
436 struct sge_eth_txq *txq;
440 * Perform sanity checking on the Queue ID to make sure it
441 * really refers to one of our TX Ethernet Egress Queues which
442 * is active and matches the queue's ID. None of these error
443 * conditions should ever happen so we may want to either make
444 * them fatal and/or conditionalized under DEBUG.
446 eq_idx = EQ_IDX(s, qid);
447 if (unlikely(eq_idx >= MAX_EGRQ)) {
448 dev_err(adapter->pdev_dev,
449 "Egress Update QID %d out of range\n", qid);
452 tq = s->egr_map[eq_idx];
453 if (unlikely(tq == NULL)) {
454 dev_err(adapter->pdev_dev,
455 "Egress Update QID %d TXQ=NULL\n", qid);
458 txq = container_of(tq, struct sge_eth_txq, q);
459 if (unlikely(tq->abs_id != qid)) {
460 dev_err(adapter->pdev_dev,
461 "Egress Update QID %d refers to TXQ %d\n",
467 * Restart a stopped TX Queue which has less than half of its
471 netif_tx_wake_queue(txq->txq);
476 dev_err(adapter->pdev_dev,
477 "unexpected CPL %#x on FW event queue\n", opcode);
484 * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues
485 * to use and initializes them. We support multiple "Queue Sets" per port if
486 * we have MSI-X, otherwise just one queue set per port.
488 static int setup_sge_queues(struct adapter *adapter)
490 struct sge *s = &adapter->sge;
494 * Clear "Queue Set" Free List Starving and TX Queue Mapping Error
497 bitmap_zero(s->starving_fl, MAX_EGRQ);
500 * If we're using MSI interrupt mode we need to set up a "forwarded
501 * interrupt" queue which we'll set up with our MSI vector. The rest
502 * of the ingress queues will be set up to forward their interrupts to
503 * this queue ... This must be first since t4vf_sge_alloc_rxq() uses
504 * the intrq's queue ID as the interrupt forwarding queue for the
505 * subsequent calls ...
507 if (adapter->flags & USING_MSI) {
508 err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false,
509 adapter->port[0], 0, NULL, NULL);
511 goto err_free_queues;
515 * Allocate our ingress queue for asynchronous firmware messages.
517 err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0],
518 MSIX_FW, NULL, fwevtq_handler);
520 goto err_free_queues;
523 * Allocate each "port"'s initial Queue Sets. These can be changed
524 * later on ... up to the point where any interface on the adapter is
525 * brought up at which point lots of things get nailed down
529 for_each_port(adapter, pidx) {
530 struct net_device *dev = adapter->port[pidx];
531 struct port_info *pi = netdev_priv(dev);
532 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
533 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
536 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
537 err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false,
539 &rxq->fl, t4vf_ethrx_handler);
541 goto err_free_queues;
543 err = t4vf_sge_alloc_eth_txq(adapter, txq, dev,
544 netdev_get_tx_queue(dev, qs),
545 s->fw_evtq.cntxt_id);
547 goto err_free_queues;
550 memset(&rxq->stats, 0, sizeof(rxq->stats));
555 * Create the reverse mappings for the queues.
557 s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id;
558 s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id;
559 IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq;
560 for_each_port(adapter, pidx) {
561 struct net_device *dev = adapter->port[pidx];
562 struct port_info *pi = netdev_priv(dev);
563 struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset];
564 struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset];
567 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
568 IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq;
569 EQ_MAP(s, txq->q.abs_id) = &txq->q;
572 * The FW_IQ_CMD doesn't return the Absolute Queue IDs
573 * for Free Lists but since all of the Egress Queues
574 * (including Free Lists) have Relative Queue IDs
575 * which are computed as Absolute - Base Queue ID, we
576 * can synthesize the Absolute Queue IDs for the Free
577 * Lists. This is useful for debugging purposes when
578 * we want to dump Queue Contexts via the PF Driver.
580 rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base;
581 EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl;
587 t4vf_free_sge_resources(adapter);
592 * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive
593 * queues. We configure the RSS CPU lookup table to distribute to the number
594 * of HW receive queues, and the response queue lookup table to narrow that
595 * down to the response queues actually configured for each "port" (Virtual
596 * Interface). We always configure the RSS mapping for all ports since the
597 * mapping table has plenty of entries.
599 static int setup_rss(struct adapter *adapter)
603 for_each_port(adapter, pidx) {
604 struct port_info *pi = adap2pinfo(adapter, pidx);
605 struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
606 u16 rss[MAX_PORT_QSETS];
609 for (qs = 0; qs < pi->nqsets; qs++)
610 rss[qs] = rxq[qs].rspq.abs_id;
612 err = t4vf_config_rss_range(adapter, pi->viid,
613 0, pi->rss_size, rss, pi->nqsets);
618 * Perform Global RSS Mode-specific initialization.
620 switch (adapter->params.rss.mode) {
621 case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL:
623 * If Tunnel All Lookup isn't specified in the global
624 * RSS Configuration, then we need to specify a
625 * default Ingress Queue for any ingress packets which
626 * aren't hashed. We'll use our first ingress queue
629 if (!adapter->params.rss.u.basicvirtual.tnlalllookup) {
630 union rss_vi_config config;
631 err = t4vf_read_rss_vi_config(adapter,
636 config.basicvirtual.defaultq =
638 err = t4vf_write_rss_vi_config(adapter,
652 * Bring the adapter up. Called whenever we go from no "ports" open to having
653 * one open. This function performs the actions necessary to make an adapter
654 * operational, such as completing the initialization of HW modules, and
655 * enabling interrupts. Must be called with the rtnl lock held. (Note that
656 * this is called "cxgb_up" in the PF Driver.)
658 static int adapter_up(struct adapter *adapter)
663 * If this is the first time we've been called, perform basic
664 * adapter setup. Once we've done this, many of our adapter
665 * parameters can no longer be changed ...
667 if ((adapter->flags & FULL_INIT_DONE) == 0) {
668 err = setup_sge_queues(adapter);
671 err = setup_rss(adapter);
673 t4vf_free_sge_resources(adapter);
677 if (adapter->flags & USING_MSIX)
678 name_msix_vecs(adapter);
679 adapter->flags |= FULL_INIT_DONE;
683 * Acquire our interrupt resources. We only support MSI-X and MSI.
685 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
686 if (adapter->flags & USING_MSIX)
687 err = request_msix_queue_irqs(adapter);
689 err = request_irq(adapter->pdev->irq,
690 t4vf_intr_handler(adapter), 0,
691 adapter->name, adapter);
693 dev_err(adapter->pdev_dev, "request_irq failed, err %d\n",
699 * Enable NAPI ingress processing and return success.
702 t4vf_sge_start(adapter);
707 * Bring the adapter down. Called whenever the last "port" (Virtual
708 * Interface) closed. (Note that this routine is called "cxgb_down" in the PF
711 static void adapter_down(struct adapter *adapter)
714 * Free interrupt resources.
716 if (adapter->flags & USING_MSIX)
717 free_msix_queue_irqs(adapter);
719 free_irq(adapter->pdev->irq, adapter);
722 * Wait for NAPI handlers to finish.
728 * Start up a net device.
730 static int cxgb4vf_open(struct net_device *dev)
733 struct port_info *pi = netdev_priv(dev);
734 struct adapter *adapter = pi->adapter;
737 * If this is the first interface that we're opening on the "adapter",
738 * bring the "adapter" up now.
740 if (adapter->open_device_map == 0) {
741 err = adapter_up(adapter);
747 * Note that this interface is up and start everything up ...
749 netif_set_real_num_tx_queues(dev, pi->nqsets);
750 err = netif_set_real_num_rx_queues(dev, pi->nqsets);
753 set_bit(pi->port_id, &adapter->open_device_map);
754 err = link_start(dev);
757 netif_tx_start_all_queues(dev);
762 * Shut down a net device. This routine is called "cxgb_close" in the PF
765 static int cxgb4vf_stop(struct net_device *dev)
768 struct port_info *pi = netdev_priv(dev);
769 struct adapter *adapter = pi->adapter;
771 netif_tx_stop_all_queues(dev);
772 netif_carrier_off(dev);
773 ret = t4vf_enable_vi(adapter, pi->viid, false, false);
774 pi->link_cfg.link_ok = 0;
776 clear_bit(pi->port_id, &adapter->open_device_map);
777 if (adapter->open_device_map == 0)
778 adapter_down(adapter);
783 * Translate our basic statistics into the standard "ifconfig" statistics.
785 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
787 struct t4vf_port_stats stats;
788 struct port_info *pi = netdev2pinfo(dev);
789 struct adapter *adapter = pi->adapter;
790 struct net_device_stats *ns = &dev->stats;
793 spin_lock(&adapter->stats_lock);
794 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
795 spin_unlock(&adapter->stats_lock);
797 memset(ns, 0, sizeof(*ns));
801 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
802 stats.tx_ucast_bytes + stats.tx_offload_bytes);
803 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
804 stats.tx_ucast_frames + stats.tx_offload_frames);
805 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
806 stats.rx_ucast_bytes);
807 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
808 stats.rx_ucast_frames);
809 ns->multicast = stats.rx_mcast_frames;
810 ns->tx_errors = stats.tx_drop_frames;
811 ns->rx_errors = stats.rx_err_frames;
817 * Collect up to maxaddrs worth of a netdevice's unicast addresses into an
818 * array of addrss pointers and return the number collected.
820 static inline int collect_netdev_uc_list_addrs(const struct net_device *dev,
822 unsigned int maxaddrs)
824 unsigned int naddr = 0;
825 const struct netdev_hw_addr *ha;
827 for_each_dev_addr(dev, ha) {
828 addr[naddr++] = ha->addr;
829 if (naddr >= maxaddrs)
836 * Collect up to maxaddrs worth of a netdevice's multicast addresses into an
837 * array of addrss pointers and return the number collected.
839 static inline int collect_netdev_mc_list_addrs(const struct net_device *dev,
841 unsigned int maxaddrs)
843 unsigned int naddr = 0;
844 const struct netdev_hw_addr *ha;
846 netdev_for_each_mc_addr(ha, dev) {
847 addr[naddr++] = ha->addr;
848 if (naddr >= maxaddrs)
855 * Configure the exact and hash address filters to handle a port's multicast
856 * and secondary unicast MAC addresses.
858 static int set_addr_filters(const struct net_device *dev, bool sleep)
866 const struct port_info *pi = netdev_priv(dev);
868 /* first do the secondary unicast addresses */
869 naddr = collect_netdev_uc_list_addrs(dev, addr, ARRAY_SIZE(addr));
871 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
872 naddr, addr, filt_idx, &uhash, sleep);
879 /* next set up the multicast addresses */
880 naddr = collect_netdev_mc_list_addrs(dev, addr, ARRAY_SIZE(addr));
882 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
883 naddr, addr, filt_idx, &mhash, sleep);
888 return t4vf_set_addr_hash(pi->adapter, pi->viid, uhash != 0,
889 uhash | mhash, sleep);
893 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
894 * If @mtu is -1 it is left unchanged.
896 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
899 struct port_info *pi = netdev_priv(dev);
901 ret = set_addr_filters(dev, sleep_ok);
903 ret = t4vf_set_rxmode(pi->adapter, pi->viid, -1,
904 (dev->flags & IFF_PROMISC) != 0,
905 (dev->flags & IFF_ALLMULTI) != 0,
911 * Set the current receive modes on the device.
913 static void cxgb4vf_set_rxmode(struct net_device *dev)
915 /* unfortunately we can't return errors to the stack */
916 set_rxmode(dev, -1, false);
920 * Find the entry in the interrupt holdoff timer value array which comes
921 * closest to the specified interrupt holdoff value.
923 static int closest_timer(const struct sge *s, int us)
925 int i, timer_idx = 0, min_delta = INT_MAX;
927 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
928 int delta = us - s->timer_val[i];
931 if (delta < min_delta) {
939 static int closest_thres(const struct sge *s, int thres)
941 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
943 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
944 delta = thres - s->counter_val[i];
947 if (delta < min_delta) {
956 * Return a queue's interrupt hold-off time in us. 0 means no timer.
958 static unsigned int qtimer_val(const struct adapter *adapter,
959 const struct sge_rspq *rspq)
961 unsigned int timer_idx = QINTR_TIMER_IDX_GET(rspq->intr_params);
963 return timer_idx < SGE_NTIMERS
964 ? adapter->sge.timer_val[timer_idx]
969 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
970 * @adapter: the adapter
971 * @rspq: the RX response queue
972 * @us: the hold-off time in us, or 0 to disable timer
973 * @cnt: the hold-off packet count, or 0 to disable counter
975 * Sets an RX response queue's interrupt hold-off time and packet count.
976 * At least one of the two needs to be enabled for the queue to generate
979 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
980 unsigned int us, unsigned int cnt)
982 unsigned int timer_idx;
985 * If both the interrupt holdoff timer and count are specified as
986 * zero, default to a holdoff count of 1 ...
992 * If an interrupt holdoff count has been specified, then find the
993 * closest configured holdoff count and use that. If the response
994 * queue has already been created, then update its queue context
1001 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1002 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1003 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1005 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1006 FW_PARAMS_PARAM_YZ(rspq->cntxt_id);
1007 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1011 rspq->pktcnt_idx = pktcnt_idx;
1015 * Compute the closest holdoff timer index from the supplied holdoff
1018 timer_idx = (us == 0
1019 ? SGE_TIMER_RSTRT_CNTR
1020 : closest_timer(&adapter->sge, us));
1023 * Update the response queue's interrupt coalescing parameters and
1026 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
1027 (cnt > 0 ? QINTR_CNT_EN : 0));
1032 * Return a version number to identify the type of adapter. The scheme is:
1033 * - bits 0..9: chip version
1034 * - bits 10..15: chip revision
1036 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1039 * Chip version 4, revision 0x3f (cxgb4vf).
1041 return 4 | (0x3f << 10);
1045 * Execute the specified ioctl command.
1047 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1053 * The VF Driver doesn't have access to any of the other
1054 * common Ethernet device ioctl()'s (like reading/writing
1055 * PHY registers, etc.
1066 * Change the device's MTU.
1068 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1071 struct port_info *pi = netdev_priv(dev);
1073 /* accommodate SACK */
1077 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1078 -1, -1, -1, -1, true);
1085 * Change the devices MAC address.
1087 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1090 struct sockaddr *addr = _addr;
1091 struct port_info *pi = netdev_priv(dev);
1093 if (!is_valid_ether_addr(addr->sa_data))
1096 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1097 addr->sa_data, true);
1101 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1102 pi->xact_addr_filt = ret;
1106 #ifdef CONFIG_NET_POLL_CONTROLLER
1108 * Poll all of our receive queues. This is called outside of normal interrupt
1111 static void cxgb4vf_poll_controller(struct net_device *dev)
1113 struct port_info *pi = netdev_priv(dev);
1114 struct adapter *adapter = pi->adapter;
1116 if (adapter->flags & USING_MSIX) {
1117 struct sge_eth_rxq *rxq;
1120 rxq = &adapter->sge.ethrxq[pi->first_qset];
1121 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1122 t4vf_sge_intr_msix(0, &rxq->rspq);
1126 t4vf_intr_handler(adapter)(0, adapter);
1131 * Ethtool operations.
1132 * ===================
1134 * Note that we don't support any ethtool operations which change the physical
1135 * state of the port to which we're linked.
1139 * Return current port link settings.
1141 static int cxgb4vf_get_settings(struct net_device *dev,
1142 struct ethtool_cmd *cmd)
1144 const struct port_info *pi = netdev_priv(dev);
1146 cmd->supported = pi->link_cfg.supported;
1147 cmd->advertising = pi->link_cfg.advertising;
1148 cmd->speed = netif_carrier_ok(dev) ? pi->link_cfg.speed : -1;
1149 cmd->duplex = DUPLEX_FULL;
1151 cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1152 cmd->phy_address = pi->port_id;
1153 cmd->transceiver = XCVR_EXTERNAL;
1154 cmd->autoneg = pi->link_cfg.autoneg;
1161 * Return our driver information.
1163 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1164 struct ethtool_drvinfo *drvinfo)
1166 struct adapter *adapter = netdev2adap(dev);
1168 strcpy(drvinfo->driver, KBUILD_MODNAME);
1169 strcpy(drvinfo->version, DRV_VERSION);
1170 strcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)));
1171 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1172 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1173 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.fwrev),
1174 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.fwrev),
1175 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.fwrev),
1176 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.fwrev),
1177 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.tprev),
1178 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.tprev),
1179 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.tprev),
1180 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.tprev));
1184 * Return current adapter message level.
1186 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1188 return netdev2adap(dev)->msg_enable;
1192 * Set current adapter message level.
1194 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1196 netdev2adap(dev)->msg_enable = msglevel;
1200 * Return the device's current Queue Set ring size parameters along with the
1201 * allowed maximum values. Since ethtool doesn't understand the concept of
1202 * multi-queue devices, we just return the current values associated with the
1205 static void cxgb4vf_get_ringparam(struct net_device *dev,
1206 struct ethtool_ringparam *rp)
1208 const struct port_info *pi = netdev_priv(dev);
1209 const struct sge *s = &pi->adapter->sge;
1211 rp->rx_max_pending = MAX_RX_BUFFERS;
1212 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1213 rp->rx_jumbo_max_pending = 0;
1214 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1216 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1217 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1218 rp->rx_jumbo_pending = 0;
1219 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1223 * Set the Queue Set ring size parameters for the device. Again, since
1224 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1225 * apply these new values across all of the Queue Sets associated with the
1226 * device -- after vetting them of course!
1228 static int cxgb4vf_set_ringparam(struct net_device *dev,
1229 struct ethtool_ringparam *rp)
1231 const struct port_info *pi = netdev_priv(dev);
1232 struct adapter *adapter = pi->adapter;
1233 struct sge *s = &adapter->sge;
1236 if (rp->rx_pending > MAX_RX_BUFFERS ||
1237 rp->rx_jumbo_pending ||
1238 rp->tx_pending > MAX_TXQ_ENTRIES ||
1239 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1240 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1241 rp->rx_pending < MIN_FL_ENTRIES ||
1242 rp->tx_pending < MIN_TXQ_ENTRIES)
1245 if (adapter->flags & FULL_INIT_DONE)
1248 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1249 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1250 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1251 s->ethtxq[qs].q.size = rp->tx_pending;
1257 * Return the interrupt holdoff timer and count for the first Queue Set on the
1258 * device. Our extension ioctl() (the cxgbtool interface) allows the
1259 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1261 static int cxgb4vf_get_coalesce(struct net_device *dev,
1262 struct ethtool_coalesce *coalesce)
1264 const struct port_info *pi = netdev_priv(dev);
1265 const struct adapter *adapter = pi->adapter;
1266 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1268 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1269 coalesce->rx_max_coalesced_frames =
1270 ((rspq->intr_params & QINTR_CNT_EN)
1271 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1277 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1278 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1279 * the interrupt holdoff timer on any of the device's Queue Sets.
1281 static int cxgb4vf_set_coalesce(struct net_device *dev,
1282 struct ethtool_coalesce *coalesce)
1284 const struct port_info *pi = netdev_priv(dev);
1285 struct adapter *adapter = pi->adapter;
1287 return set_rxq_intr_params(adapter,
1288 &adapter->sge.ethrxq[pi->first_qset].rspq,
1289 coalesce->rx_coalesce_usecs,
1290 coalesce->rx_max_coalesced_frames);
1294 * Report current port link pause parameter settings.
1296 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1297 struct ethtool_pauseparam *pauseparam)
1299 struct port_info *pi = netdev_priv(dev);
1301 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1302 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1303 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1307 * Return whether RX Checksum Offloading is currently enabled for the device.
1309 static u32 cxgb4vf_get_rx_csum(struct net_device *dev)
1311 struct port_info *pi = netdev_priv(dev);
1313 return (pi->rx_offload & RX_CSO) != 0;
1317 * Turn RX Checksum Offloading on or off for the device.
1319 static int cxgb4vf_set_rx_csum(struct net_device *dev, u32 csum)
1321 struct port_info *pi = netdev_priv(dev);
1324 pi->rx_offload |= RX_CSO;
1326 pi->rx_offload &= ~RX_CSO;
1331 * Identify the port by blinking the port's LED.
1333 static int cxgb4vf_phys_id(struct net_device *dev, u32 id)
1335 struct port_info *pi = netdev_priv(dev);
1337 return t4vf_identify_port(pi->adapter, pi->viid, 5);
1341 * Port stats maintained per queue of the port.
1343 struct queue_port_stats {
1352 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1353 * these need to match the order of statistics returned by
1354 * t4vf_get_port_stats().
1356 static const char stats_strings[][ETH_GSTRING_LEN] = {
1358 * These must match the layout of the t4vf_port_stats structure.
1360 "TxBroadcastBytes ",
1361 "TxBroadcastFrames ",
1362 "TxMulticastBytes ",
1363 "TxMulticastFrames ",
1369 "RxBroadcastBytes ",
1370 "RxBroadcastFrames ",
1371 "RxMulticastBytes ",
1372 "RxMulticastFrames ",
1378 * These are accumulated per-queue statistics and must match the
1379 * order of the fields in the queue_port_stats structure.
1389 * Return the number of statistics in the specified statistics set.
1391 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1395 return ARRAY_SIZE(stats_strings);
1403 * Return the strings for the specified statistics set.
1405 static void cxgb4vf_get_strings(struct net_device *dev,
1411 memcpy(data, stats_strings, sizeof(stats_strings));
1417 * Small utility routine to accumulate queue statistics across the queues of
1420 static void collect_sge_port_stats(const struct adapter *adapter,
1421 const struct port_info *pi,
1422 struct queue_port_stats *stats)
1424 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1425 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1428 memset(stats, 0, sizeof(*stats));
1429 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1430 stats->tso += txq->tso;
1431 stats->tx_csum += txq->tx_cso;
1432 stats->rx_csum += rxq->stats.rx_cso;
1433 stats->vlan_ex += rxq->stats.vlan_ex;
1434 stats->vlan_ins += txq->vlan_ins;
1439 * Return the ETH_SS_STATS statistics set.
1441 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1442 struct ethtool_stats *stats,
1445 struct port_info *pi = netdev2pinfo(dev);
1446 struct adapter *adapter = pi->adapter;
1447 int err = t4vf_get_port_stats(adapter, pi->pidx,
1448 (struct t4vf_port_stats *)data);
1450 memset(data, 0, sizeof(struct t4vf_port_stats));
1452 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1453 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1457 * Return the size of our register map.
1459 static int cxgb4vf_get_regs_len(struct net_device *dev)
1461 return T4VF_REGMAP_SIZE;
1465 * Dump a block of registers, start to end inclusive, into a buffer.
1467 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1468 unsigned int start, unsigned int end)
1470 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1472 for ( ; start <= end; start += sizeof(u32)) {
1474 * Avoid reading the Mailbox Control register since that
1475 * can trigger a Mailbox Ownership Arbitration cycle and
1476 * interfere with communication with the firmware.
1478 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1481 *bp++ = t4_read_reg(adapter, start);
1486 * Copy our entire register map into the provided buffer.
1488 static void cxgb4vf_get_regs(struct net_device *dev,
1489 struct ethtool_regs *regs,
1492 struct adapter *adapter = netdev2adap(dev);
1494 regs->version = mk_adap_vers(adapter);
1497 * Fill in register buffer with our register map.
1499 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1501 reg_block_dump(adapter, regbuf,
1502 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1503 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1504 reg_block_dump(adapter, regbuf,
1505 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1506 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1507 reg_block_dump(adapter, regbuf,
1508 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1509 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_LAST);
1510 reg_block_dump(adapter, regbuf,
1511 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1512 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1514 reg_block_dump(adapter, regbuf,
1515 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1516 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1520 * Report current Wake On LAN settings.
1522 static void cxgb4vf_get_wol(struct net_device *dev,
1523 struct ethtool_wolinfo *wol)
1527 memset(&wol->sopass, 0, sizeof(wol->sopass));
1531 * Set TCP Segmentation Offloading feature capabilities.
1533 static int cxgb4vf_set_tso(struct net_device *dev, u32 tso)
1536 dev->features |= NETIF_F_TSO | NETIF_F_TSO6;
1538 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
1542 static struct ethtool_ops cxgb4vf_ethtool_ops = {
1543 .get_settings = cxgb4vf_get_settings,
1544 .get_drvinfo = cxgb4vf_get_drvinfo,
1545 .get_msglevel = cxgb4vf_get_msglevel,
1546 .set_msglevel = cxgb4vf_set_msglevel,
1547 .get_ringparam = cxgb4vf_get_ringparam,
1548 .set_ringparam = cxgb4vf_set_ringparam,
1549 .get_coalesce = cxgb4vf_get_coalesce,
1550 .set_coalesce = cxgb4vf_set_coalesce,
1551 .get_pauseparam = cxgb4vf_get_pauseparam,
1552 .get_rx_csum = cxgb4vf_get_rx_csum,
1553 .set_rx_csum = cxgb4vf_set_rx_csum,
1554 .set_tx_csum = ethtool_op_set_tx_ipv6_csum,
1555 .set_sg = ethtool_op_set_sg,
1556 .get_link = ethtool_op_get_link,
1557 .get_strings = cxgb4vf_get_strings,
1558 .phys_id = cxgb4vf_phys_id,
1559 .get_sset_count = cxgb4vf_get_sset_count,
1560 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1561 .get_regs_len = cxgb4vf_get_regs_len,
1562 .get_regs = cxgb4vf_get_regs,
1563 .get_wol = cxgb4vf_get_wol,
1564 .set_tso = cxgb4vf_set_tso,
1568 * /sys/kernel/debug/cxgb4vf support code and data.
1569 * ================================================
1573 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1577 static int sge_qinfo_show(struct seq_file *seq, void *v)
1579 struct adapter *adapter = seq->private;
1580 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1581 int qs, r = (uintptr_t)v - 1;
1584 seq_putc(seq, '\n');
1586 #define S3(fmt_spec, s, v) \
1588 seq_printf(seq, "%-12s", s); \
1589 for (qs = 0; qs < n; ++qs) \
1590 seq_printf(seq, " %16" fmt_spec, v); \
1591 seq_putc(seq, '\n'); \
1593 #define S(s, v) S3("s", s, v)
1594 #define T(s, v) S3("u", s, txq[qs].v)
1595 #define R(s, v) S3("u", s, rxq[qs].v)
1597 if (r < eth_entries) {
1598 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1599 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1600 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1602 S("QType:", "Ethernet");
1604 (rxq[qs].rspq.netdev
1605 ? rxq[qs].rspq.netdev->name
1608 (rxq[qs].rspq.netdev
1609 ? ((struct port_info *)
1610 netdev_priv(rxq[qs].rspq.netdev))->port_id
1612 T("TxQ ID:", q.abs_id);
1613 T("TxQ size:", q.size);
1614 T("TxQ inuse:", q.in_use);
1615 T("TxQ PIdx:", q.pidx);
1616 T("TxQ CIdx:", q.cidx);
1617 R("RspQ ID:", rspq.abs_id);
1618 R("RspQ size:", rspq.size);
1619 R("RspQE size:", rspq.iqe_len);
1620 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1621 S3("u", "Intr pktcnt:",
1622 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1623 R("RspQ CIdx:", rspq.cidx);
1624 R("RspQ Gen:", rspq.gen);
1625 R("FL ID:", fl.abs_id);
1626 R("FL size:", fl.size - MIN_FL_RESID);
1627 R("FL avail:", fl.avail);
1628 R("FL PIdx:", fl.pidx);
1629 R("FL CIdx:", fl.cidx);
1635 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1637 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1638 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1639 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1640 qtimer_val(adapter, evtq));
1641 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1642 adapter->sge.counter_val[evtq->pktcnt_idx]);
1643 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1644 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1645 } else if (r == 1) {
1646 const struct sge_rspq *intrq = &adapter->sge.intrq;
1648 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1649 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1650 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1651 qtimer_val(adapter, intrq));
1652 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1653 adapter->sge.counter_val[intrq->pktcnt_idx]);
1654 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1655 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1667 * Return the number of "entries" in our "file". We group the multi-Queue
1668 * sections with QPL Queue Sets per "entry". The sections of the output are:
1670 * Ethernet RX/TX Queue Sets
1671 * Firmware Event Queue
1672 * Forwarded Interrupt Queue (if in MSI mode)
1674 static int sge_queue_entries(const struct adapter *adapter)
1676 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1677 ((adapter->flags & USING_MSI) != 0);
1680 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1682 int entries = sge_queue_entries(seq->private);
1684 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1687 static void sge_queue_stop(struct seq_file *seq, void *v)
1691 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
1693 int entries = sge_queue_entries(seq->private);
1696 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1699 static const struct seq_operations sge_qinfo_seq_ops = {
1700 .start = sge_queue_start,
1701 .next = sge_queue_next,
1702 .stop = sge_queue_stop,
1703 .show = sge_qinfo_show
1706 static int sge_qinfo_open(struct inode *inode, struct file *file)
1708 int res = seq_open(file, &sge_qinfo_seq_ops);
1711 struct seq_file *seq = file->private_data;
1712 seq->private = inode->i_private;
1717 static const struct file_operations sge_qinfo_debugfs_fops = {
1718 .owner = THIS_MODULE,
1719 .open = sge_qinfo_open,
1721 .llseek = seq_lseek,
1722 .release = seq_release,
1726 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
1730 static int sge_qstats_show(struct seq_file *seq, void *v)
1732 struct adapter *adapter = seq->private;
1733 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1734 int qs, r = (uintptr_t)v - 1;
1737 seq_putc(seq, '\n');
1739 #define S3(fmt, s, v) \
1741 seq_printf(seq, "%-16s", s); \
1742 for (qs = 0; qs < n; ++qs) \
1743 seq_printf(seq, " %8" fmt, v); \
1744 seq_putc(seq, '\n'); \
1746 #define S(s, v) S3("s", s, v)
1748 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
1749 #define T(s, v) T3("lu", s, v)
1751 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
1752 #define R(s, v) R3("lu", s, v)
1754 if (r < eth_entries) {
1755 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1756 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1757 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1759 S("QType:", "Ethernet");
1761 (rxq[qs].rspq.netdev
1762 ? rxq[qs].rspq.netdev->name
1764 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
1765 R("RxPackets:", stats.pkts);
1766 R("RxCSO:", stats.rx_cso);
1767 R("VLANxtract:", stats.vlan_ex);
1768 R("LROmerged:", stats.lro_merged);
1769 R("LROpackets:", stats.lro_pkts);
1770 R("RxDrops:", stats.rx_drops);
1772 T("TxCSO:", tx_cso);
1773 T("VLANins:", vlan_ins);
1774 T("TxQFull:", q.stops);
1775 T("TxQRestarts:", q.restarts);
1776 T("TxMapErr:", mapping_err);
1777 R("FLAllocErr:", fl.alloc_failed);
1778 R("FLLrgAlcErr:", fl.large_alloc_failed);
1779 R("FLStarving:", fl.starving);
1785 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1787 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
1788 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1789 evtq->unhandled_irqs);
1790 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
1791 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
1792 } else if (r == 1) {
1793 const struct sge_rspq *intrq = &adapter->sge.intrq;
1795 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
1796 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1797 intrq->unhandled_irqs);
1798 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
1799 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
1813 * Return the number of "entries" in our "file". We group the multi-Queue
1814 * sections with QPL Queue Sets per "entry". The sections of the output are:
1816 * Ethernet RX/TX Queue Sets
1817 * Firmware Event Queue
1818 * Forwarded Interrupt Queue (if in MSI mode)
1820 static int sge_qstats_entries(const struct adapter *adapter)
1822 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1823 ((adapter->flags & USING_MSI) != 0);
1826 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
1828 int entries = sge_qstats_entries(seq->private);
1830 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1833 static void sge_qstats_stop(struct seq_file *seq, void *v)
1837 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
1839 int entries = sge_qstats_entries(seq->private);
1842 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1845 static const struct seq_operations sge_qstats_seq_ops = {
1846 .start = sge_qstats_start,
1847 .next = sge_qstats_next,
1848 .stop = sge_qstats_stop,
1849 .show = sge_qstats_show
1852 static int sge_qstats_open(struct inode *inode, struct file *file)
1854 int res = seq_open(file, &sge_qstats_seq_ops);
1857 struct seq_file *seq = file->private_data;
1858 seq->private = inode->i_private;
1863 static const struct file_operations sge_qstats_proc_fops = {
1864 .owner = THIS_MODULE,
1865 .open = sge_qstats_open,
1867 .llseek = seq_lseek,
1868 .release = seq_release,
1872 * Show PCI-E SR-IOV Virtual Function Resource Limits.
1874 static int resources_show(struct seq_file *seq, void *v)
1876 struct adapter *adapter = seq->private;
1877 struct vf_resources *vfres = &adapter->params.vfres;
1879 #define S(desc, fmt, var) \
1880 seq_printf(seq, "%-60s " fmt "\n", \
1881 desc " (" #var "):", vfres->var)
1883 S("Virtual Interfaces", "%d", nvi);
1884 S("Egress Queues", "%d", neq);
1885 S("Ethernet Control", "%d", nethctrl);
1886 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
1887 S("Ingress Queues", "%d", niq);
1888 S("Traffic Class", "%d", tc);
1889 S("Port Access Rights Mask", "%#x", pmask);
1890 S("MAC Address Filters", "%d", nexactf);
1891 S("Firmware Command Read Capabilities", "%#x", r_caps);
1892 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
1899 static int resources_open(struct inode *inode, struct file *file)
1901 return single_open(file, resources_show, inode->i_private);
1904 static const struct file_operations resources_proc_fops = {
1905 .owner = THIS_MODULE,
1906 .open = resources_open,
1908 .llseek = seq_lseek,
1909 .release = single_release,
1913 * Show Virtual Interfaces.
1915 static int interfaces_show(struct seq_file *seq, void *v)
1917 if (v == SEQ_START_TOKEN) {
1918 seq_puts(seq, "Interface Port VIID\n");
1920 struct adapter *adapter = seq->private;
1921 int pidx = (uintptr_t)v - 2;
1922 struct net_device *dev = adapter->port[pidx];
1923 struct port_info *pi = netdev_priv(dev);
1925 seq_printf(seq, "%9s %4d %#5x\n",
1926 dev->name, pi->port_id, pi->viid);
1931 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
1933 return pos <= adapter->params.nports
1934 ? (void *)(uintptr_t)(pos + 1)
1938 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
1941 ? interfaces_get_idx(seq->private, *pos)
1945 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
1948 return interfaces_get_idx(seq->private, *pos);
1951 static void interfaces_stop(struct seq_file *seq, void *v)
1955 static const struct seq_operations interfaces_seq_ops = {
1956 .start = interfaces_start,
1957 .next = interfaces_next,
1958 .stop = interfaces_stop,
1959 .show = interfaces_show
1962 static int interfaces_open(struct inode *inode, struct file *file)
1964 int res = seq_open(file, &interfaces_seq_ops);
1967 struct seq_file *seq = file->private_data;
1968 seq->private = inode->i_private;
1973 static const struct file_operations interfaces_proc_fops = {
1974 .owner = THIS_MODULE,
1975 .open = interfaces_open,
1977 .llseek = seq_lseek,
1978 .release = seq_release,
1982 * /sys/kernel/debugfs/cxgb4vf/ files list.
1984 struct cxgb4vf_debugfs_entry {
1985 const char *name; /* name of debugfs node */
1986 mode_t mode; /* file system mode */
1987 const struct file_operations *fops;
1990 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
1991 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
1992 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
1993 { "resources", S_IRUGO, &resources_proc_fops },
1994 { "interfaces", S_IRUGO, &interfaces_proc_fops },
1998 * Module and device initialization and cleanup code.
1999 * ==================================================
2003 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2004 * directory (debugfs_root) has already been set up.
2006 static int __devinit setup_debugfs(struct adapter *adapter)
2010 BUG_ON(adapter->debugfs_root == NULL);
2013 * Debugfs support is best effort.
2015 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2016 (void)debugfs_create_file(debugfs_files[i].name,
2017 debugfs_files[i].mode,
2018 adapter->debugfs_root,
2020 debugfs_files[i].fops);
2026 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2027 * it to our caller to tear down the directory (debugfs_root).
2029 static void __devexit cleanup_debugfs(struct adapter *adapter)
2031 BUG_ON(adapter->debugfs_root == NULL);
2034 * Unlike our sister routine cleanup_proc(), we don't need to remove
2035 * individual entries because a call will be made to
2036 * debugfs_remove_recursive(). We just need to clean up any ancillary
2043 * Perform early "adapter" initialization. This is where we discover what
2044 * adapter parameters we're going to be using and initialize basic adapter
2047 static int adap_init0(struct adapter *adapter)
2049 struct vf_resources *vfres = &adapter->params.vfres;
2050 struct sge_params *sge_params = &adapter->params.sge;
2051 struct sge *s = &adapter->sge;
2052 unsigned int ethqsets;
2056 * Wait for the device to become ready before proceeding ...
2058 err = t4vf_wait_dev_ready(adapter);
2060 dev_err(adapter->pdev_dev, "device didn't become ready:"
2066 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2067 * 2.6.31 and later we can't call pci_reset_function() in order to
2068 * issue an FLR because of a self- deadlock on the device semaphore.
2069 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2070 * cases where they're needed -- for instance, some versions of KVM
2071 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2072 * use the firmware based reset in order to reset any per function
2075 err = t4vf_fw_reset(adapter);
2077 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2082 * Grab basic operational parameters. These will predominantly have
2083 * been set up by the Physical Function Driver or will be hard coded
2084 * into the adapter. We just have to live with them ... Note that
2085 * we _must_ get our VPD parameters before our SGE parameters because
2086 * we need to know the adapter's core clock from the VPD in order to
2087 * properly decode the SGE Timer Values.
2089 err = t4vf_get_dev_params(adapter);
2091 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2092 " device parameters: err=%d\n", err);
2095 err = t4vf_get_vpd_params(adapter);
2097 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2098 " VPD parameters: err=%d\n", err);
2101 err = t4vf_get_sge_params(adapter);
2103 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2104 " SGE parameters: err=%d\n", err);
2107 err = t4vf_get_rss_glb_config(adapter);
2109 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2110 " RSS parameters: err=%d\n", err);
2113 if (adapter->params.rss.mode !=
2114 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2115 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2116 " mode %d\n", adapter->params.rss.mode);
2119 err = t4vf_sge_init(adapter);
2121 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2127 * Retrieve our RX interrupt holdoff timer values and counter
2128 * threshold values from the SGE parameters.
2130 s->timer_val[0] = core_ticks_to_us(adapter,
2131 TIMERVALUE0_GET(sge_params->sge_timer_value_0_and_1));
2132 s->timer_val[1] = core_ticks_to_us(adapter,
2133 TIMERVALUE1_GET(sge_params->sge_timer_value_0_and_1));
2134 s->timer_val[2] = core_ticks_to_us(adapter,
2135 TIMERVALUE0_GET(sge_params->sge_timer_value_2_and_3));
2136 s->timer_val[3] = core_ticks_to_us(adapter,
2137 TIMERVALUE1_GET(sge_params->sge_timer_value_2_and_3));
2138 s->timer_val[4] = core_ticks_to_us(adapter,
2139 TIMERVALUE0_GET(sge_params->sge_timer_value_4_and_5));
2140 s->timer_val[5] = core_ticks_to_us(adapter,
2141 TIMERVALUE1_GET(sge_params->sge_timer_value_4_and_5));
2144 THRESHOLD_0_GET(sge_params->sge_ingress_rx_threshold);
2146 THRESHOLD_1_GET(sge_params->sge_ingress_rx_threshold);
2148 THRESHOLD_2_GET(sge_params->sge_ingress_rx_threshold);
2150 THRESHOLD_3_GET(sge_params->sge_ingress_rx_threshold);
2153 * Grab our Virtual Interface resource allocation, extract the
2154 * features that we're interested in and do a bit of sanity testing on
2157 err = t4vf_get_vfres(adapter);
2159 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2160 " resources: err=%d\n", err);
2165 * The number of "ports" which we support is equal to the number of
2166 * Virtual Interfaces with which we've been provisioned.
2168 adapter->params.nports = vfres->nvi;
2169 if (adapter->params.nports > MAX_NPORTS) {
2170 dev_warn(adapter->pdev_dev, "only using %d of %d allowed"
2171 " virtual interfaces\n", MAX_NPORTS,
2172 adapter->params.nports);
2173 adapter->params.nports = MAX_NPORTS;
2177 * We need to reserve a number of the ingress queues with Free List
2178 * and Interrupt capabilities for special interrupt purposes (like
2179 * asynchronous firmware messages, or forwarded interrupts if we're
2180 * using MSI). The rest of the FL/Intr-capable ingress queues will be
2181 * matched up one-for-one with Ethernet/Control egress queues in order
2182 * to form "Queue Sets" which will be aportioned between the "ports".
2183 * For each Queue Set, we'll need the ability to allocate two Egress
2184 * Contexts -- one for the Ingress Queue Free List and one for the TX
2187 ethqsets = vfres->niqflint - INGQ_EXTRAS;
2188 if (vfres->nethctrl != ethqsets) {
2189 dev_warn(adapter->pdev_dev, "unequal number of [available]"
2190 " ingress/egress queues (%d/%d); using minimum for"
2191 " number of Queue Sets\n", ethqsets, vfres->nethctrl);
2192 ethqsets = min(vfres->nethctrl, ethqsets);
2194 if (vfres->neq < ethqsets*2) {
2195 dev_warn(adapter->pdev_dev, "Not enough Egress Contexts (%d)"
2196 " to support Queue Sets (%d); reducing allowed Queue"
2197 " Sets\n", vfres->neq, ethqsets);
2198 ethqsets = vfres->neq/2;
2200 if (ethqsets > MAX_ETH_QSETS) {
2201 dev_warn(adapter->pdev_dev, "only using %d of %d allowed Queue"
2202 " Sets\n", MAX_ETH_QSETS, adapter->sge.max_ethqsets);
2203 ethqsets = MAX_ETH_QSETS;
2205 if (vfres->niq != 0 || vfres->neq > ethqsets*2) {
2206 dev_warn(adapter->pdev_dev, "unused resources niq/neq (%d/%d)"
2207 " ignored\n", vfres->niq, vfres->neq - ethqsets*2);
2209 adapter->sge.max_ethqsets = ethqsets;
2212 * Check for various parameter sanity issues. Most checks simply
2213 * result in us using fewer resources than our provissioning but we
2214 * do need at least one "port" with which to work ...
2216 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2217 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2218 " virtual interfaces (too few Queue Sets)\n",
2219 adapter->sge.max_ethqsets, adapter->params.nports);
2220 adapter->params.nports = adapter->sge.max_ethqsets;
2222 if (adapter->params.nports == 0) {
2223 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2230 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2231 u8 pkt_cnt_idx, unsigned int size,
2232 unsigned int iqe_size)
2234 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
2235 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0));
2236 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2239 rspq->iqe_len = iqe_size;
2244 * Perform default configuration of DMA queues depending on the number and
2245 * type of ports we found and the number of available CPUs. Most settings can
2246 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2247 * being brought up for the first time.
2249 static void __devinit cfg_queues(struct adapter *adapter)
2251 struct sge *s = &adapter->sge;
2252 int q10g, n10g, qidx, pidx, qs;
2255 * We should not be called till we know how many Queue Sets we can
2256 * support. In particular, this means that we need to know what kind
2257 * of interrupts we'll be using ...
2259 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2262 * Count the number of 10GbE Virtual Interfaces that we have.
2265 for_each_port(adapter, pidx)
2266 n10g += is_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2269 * We default to 1 queue per non-10G port and up to # of cores queues
2275 int n1g = (adapter->params.nports - n10g);
2276 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2277 if (q10g > num_online_cpus())
2278 q10g = num_online_cpus();
2282 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2283 * The layout will be established in setup_sge_queues() when the
2284 * adapter is brough up for the first time.
2287 for_each_port(adapter, pidx) {
2288 struct port_info *pi = adap2pinfo(adapter, pidx);
2290 pi->first_qset = qidx;
2291 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
2297 * Set up default Queue Set parameters ... Start off with the
2298 * shortest interrupt holdoff timer.
2300 for (qs = 0; qs < s->max_ethqsets; qs++) {
2301 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2302 struct sge_eth_txq *txq = &s->ethtxq[qs];
2304 init_rspq(&rxq->rspq, 0, 0, 1024, L1_CACHE_BYTES);
2310 * The firmware event queue is used for link state changes and
2311 * notifications of TX DMA completions.
2313 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512,
2317 * The forwarded interrupt queue is used when we're in MSI interrupt
2318 * mode. In this mode all interrupts associated with RX queues will
2319 * be forwarded to a single queue which we'll associate with our MSI
2320 * interrupt vector. The messages dropped in the forwarded interrupt
2321 * queue will indicate which ingress queue needs servicing ... This
2322 * queue needs to be large enough to accommodate all of the ingress
2323 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2324 * from equalling the CIDX if every ingress queue has an outstanding
2325 * interrupt). The queue doesn't need to be any larger because no
2326 * ingress queue will ever have more than one outstanding interrupt at
2329 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2334 * Reduce the number of Ethernet queues across all ports to at most n.
2335 * n provides at least one queue per port.
2337 static void __devinit reduce_ethqs(struct adapter *adapter, int n)
2340 struct port_info *pi;
2343 * While we have too many active Ether Queue Sets, interate across the
2344 * "ports" and reduce their individual Queue Set allocations.
2346 BUG_ON(n < adapter->params.nports);
2347 while (n < adapter->sge.ethqsets)
2348 for_each_port(adapter, i) {
2349 pi = adap2pinfo(adapter, i);
2350 if (pi->nqsets > 1) {
2352 adapter->sge.ethqsets--;
2353 if (adapter->sge.ethqsets <= n)
2359 * Reassign the starting Queue Sets for each of the "ports" ...
2362 for_each_port(adapter, i) {
2363 pi = adap2pinfo(adapter, i);
2370 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2371 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2372 * need. Minimally we need one for every Virtual Interface plus those needed
2373 * for our "extras". Note that this process may lower the maximum number of
2374 * allowed Queue Sets ...
2376 static int __devinit enable_msix(struct adapter *adapter)
2378 int i, err, want, need;
2379 struct msix_entry entries[MSIX_ENTRIES];
2380 struct sge *s = &adapter->sge;
2382 for (i = 0; i < MSIX_ENTRIES; ++i)
2383 entries[i].entry = i;
2386 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2387 * plus those needed for our "extras" (for example, the firmware
2388 * message queue). We _need_ at least one "Queue Set" per Virtual
2389 * Interface plus those needed for our "extras". So now we get to see
2390 * if the song is right ...
2392 want = s->max_ethqsets + MSIX_EXTRAS;
2393 need = adapter->params.nports + MSIX_EXTRAS;
2394 while ((err = pci_enable_msix(adapter->pdev, entries, want)) >= need)
2398 int nqsets = want - MSIX_EXTRAS;
2399 if (nqsets < s->max_ethqsets) {
2400 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2401 " for %d Queue Sets\n", nqsets);
2402 s->max_ethqsets = nqsets;
2403 if (nqsets < s->ethqsets)
2404 reduce_ethqs(adapter, nqsets);
2406 for (i = 0; i < want; ++i)
2407 adapter->msix_info[i].vec = entries[i].vector;
2408 } else if (err > 0) {
2409 pci_disable_msix(adapter->pdev);
2410 dev_info(adapter->pdev_dev, "only %d MSI-X vectors left,"
2411 " not using MSI-X\n", err);
2416 #ifdef HAVE_NET_DEVICE_OPS
2417 static const struct net_device_ops cxgb4vf_netdev_ops = {
2418 .ndo_open = cxgb4vf_open,
2419 .ndo_stop = cxgb4vf_stop,
2420 .ndo_start_xmit = t4vf_eth_xmit,
2421 .ndo_get_stats = cxgb4vf_get_stats,
2422 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2423 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2424 .ndo_validate_addr = eth_validate_addr,
2425 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2426 .ndo_change_mtu = cxgb4vf_change_mtu,
2427 .ndo_vlan_rx_register = cxgb4vf_vlan_rx_register,
2428 #ifdef CONFIG_NET_POLL_CONTROLLER
2429 .ndo_poll_controller = cxgb4vf_poll_controller,
2435 * "Probe" a device: initialize a device and construct all kernel and driver
2436 * state needed to manage the device. This routine is called "init_one" in
2439 static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
2440 const struct pci_device_id *ent)
2442 static int version_printed;
2447 struct adapter *adapter;
2448 struct port_info *pi;
2449 struct net_device *netdev;
2452 * Vet our module parameters.
2454 if (msi != MSI_MSIX && msi != MSI_MSI) {
2455 dev_err(&pdev->dev, "bad module parameter msi=%d; must be %d"
2456 " (MSI-X or MSI) or %d (MSI)\n", msi, MSI_MSIX,
2463 * Print our driver banner the first time we're called to initialize a
2466 if (version_printed == 0) {
2467 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
2468 version_printed = 1;
2472 * Initialize generic PCI device state.
2474 err = pci_enable_device(pdev);
2476 dev_err(&pdev->dev, "cannot enable PCI device\n");
2481 * Reserve PCI resources for the device. If we can't get them some
2482 * other driver may have already claimed the device ...
2484 err = pci_request_regions(pdev, KBUILD_MODNAME);
2486 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2487 goto err_disable_device;
2491 * Set up our DMA mask: try for 64-bit address masking first and
2492 * fall back to 32-bit if we can't get 64 bits ...
2494 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2496 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2498 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2499 " coherent allocations\n");
2500 goto err_release_regions;
2504 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2506 dev_err(&pdev->dev, "no usable DMA configuration\n");
2507 goto err_release_regions;
2513 * Enable bus mastering for the device ...
2515 pci_set_master(pdev);
2518 * Allocate our adapter data structure and attach it to the device.
2520 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2523 goto err_release_regions;
2525 pci_set_drvdata(pdev, adapter);
2526 adapter->pdev = pdev;
2527 adapter->pdev_dev = &pdev->dev;
2530 * Initialize SMP data synchronization resources.
2532 spin_lock_init(&adapter->stats_lock);
2535 * Map our I/O registers in BAR0.
2537 adapter->regs = pci_ioremap_bar(pdev, 0);
2538 if (!adapter->regs) {
2539 dev_err(&pdev->dev, "cannot map device registers\n");
2541 goto err_free_adapter;
2545 * Initialize adapter level features.
2547 adapter->name = pci_name(pdev);
2548 adapter->msg_enable = dflt_msg_enable;
2549 err = adap_init0(adapter);
2554 * Allocate our "adapter ports" and stitch everything together.
2556 pmask = adapter->params.vfres.pmask;
2557 for_each_port(adapter, pidx) {
2561 * We simplistically allocate our virtual interfaces
2562 * sequentially across the port numbers to which we have
2563 * access rights. This should be configurable in some manner
2568 port_id = ffs(pmask) - 1;
2569 pmask &= ~(1 << port_id);
2570 viid = t4vf_alloc_vi(adapter, port_id);
2572 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2573 " err=%d\n", port_id, viid);
2579 * Allocate our network device and stitch things together.
2581 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2583 if (netdev == NULL) {
2584 dev_err(&pdev->dev, "cannot allocate netdev for"
2585 " port %d\n", port_id);
2586 t4vf_free_vi(adapter, viid);
2590 adapter->port[pidx] = netdev;
2591 SET_NETDEV_DEV(netdev, &pdev->dev);
2592 pi = netdev_priv(netdev);
2593 pi->adapter = adapter;
2595 pi->port_id = port_id;
2599 * Initialize the starting state of our "port" and register
2602 pi->xact_addr_filt = -1;
2603 pi->rx_offload = RX_CSO;
2604 netif_carrier_off(netdev);
2605 netdev->irq = pdev->irq;
2607 netdev->features = (NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
2608 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2609 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
2612 netdev->features |= NETIF_F_HIGHDMA;
2613 netdev->vlan_features =
2615 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX));
2617 #ifdef HAVE_NET_DEVICE_OPS
2618 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2620 netdev->vlan_rx_register = cxgb4vf_vlan_rx_register;
2621 netdev->open = cxgb4vf_open;
2622 netdev->stop = cxgb4vf_stop;
2623 netdev->hard_start_xmit = t4vf_eth_xmit;
2624 netdev->get_stats = cxgb4vf_get_stats;
2625 netdev->set_rx_mode = cxgb4vf_set_rxmode;
2626 netdev->do_ioctl = cxgb4vf_do_ioctl;
2627 netdev->change_mtu = cxgb4vf_change_mtu;
2628 netdev->set_mac_address = cxgb4vf_set_mac_addr;
2629 #ifdef CONFIG_NET_POLL_CONTROLLER
2630 netdev->poll_controller = cxgb4vf_poll_controller;
2633 SET_ETHTOOL_OPS(netdev, &cxgb4vf_ethtool_ops);
2636 * Initialize the hardware/software state for the port.
2638 err = t4vf_port_init(adapter, pidx);
2640 dev_err(&pdev->dev, "cannot initialize port %d\n",
2647 * The "card" is now ready to go. If any errors occur during device
2648 * registration we do not fail the whole "card" but rather proceed
2649 * only with the ports we manage to register successfully. However we
2650 * must register at least one net device.
2652 for_each_port(adapter, pidx) {
2653 netdev = adapter->port[pidx];
2657 err = register_netdev(netdev);
2659 dev_warn(&pdev->dev, "cannot register net device %s,"
2660 " skipping\n", netdev->name);
2664 set_bit(pidx, &adapter->registered_device_map);
2666 if (adapter->registered_device_map == 0) {
2667 dev_err(&pdev->dev, "could not register any net devices\n");
2672 * Set up our debugfs entries.
2674 if (cxgb4vf_debugfs_root) {
2675 adapter->debugfs_root =
2676 debugfs_create_dir(pci_name(pdev),
2677 cxgb4vf_debugfs_root);
2678 if (adapter->debugfs_root == NULL)
2679 dev_warn(&pdev->dev, "could not create debugfs"
2682 setup_debugfs(adapter);
2686 * See what interrupts we'll be using. If we've been configured to
2687 * use MSI-X interrupts, try to enable them but fall back to using
2688 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
2689 * get MSI interrupts we bail with the error.
2691 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
2692 adapter->flags |= USING_MSIX;
2694 err = pci_enable_msi(pdev);
2696 dev_err(&pdev->dev, "Unable to allocate %s interrupts;"
2698 msi == MSI_MSIX ? "MSI-X or MSI" : "MSI", err);
2699 goto err_free_debugfs;
2701 adapter->flags |= USING_MSI;
2705 * Now that we know how many "ports" we have and what their types are,
2706 * and how many Queue Sets we can support, we can configure our queue
2709 cfg_queues(adapter);
2712 * Print a short notice on the existance and configuration of the new
2713 * VF network device ...
2715 for_each_port(adapter, pidx) {
2716 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
2717 adapter->port[pidx]->name,
2718 (adapter->flags & USING_MSIX) ? "MSI-X" :
2719 (adapter->flags & USING_MSI) ? "MSI" : "");
2728 * Error recovery and exit code. Unwind state that's been created
2729 * so far and return the error.
2733 if (adapter->debugfs_root) {
2734 cleanup_debugfs(adapter);
2735 debugfs_remove_recursive(adapter->debugfs_root);
2739 for_each_port(adapter, pidx) {
2740 netdev = adapter->port[pidx];
2743 pi = netdev_priv(netdev);
2744 t4vf_free_vi(adapter, pi->viid);
2745 if (test_bit(pidx, &adapter->registered_device_map))
2746 unregister_netdev(netdev);
2747 free_netdev(netdev);
2751 iounmap(adapter->regs);
2755 pci_set_drvdata(pdev, NULL);
2757 err_release_regions:
2758 pci_release_regions(pdev);
2759 pci_set_drvdata(pdev, NULL);
2760 pci_clear_master(pdev);
2763 pci_disable_device(pdev);
2770 * "Remove" a device: tear down all kernel and driver state created in the
2771 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
2772 * that this is called "remove_one" in the PF Driver.)
2774 static void __devexit cxgb4vf_pci_remove(struct pci_dev *pdev)
2776 struct adapter *adapter = pci_get_drvdata(pdev);
2779 * Tear down driver state associated with device.
2785 * Stop all of our activity. Unregister network port,
2786 * disable interrupts, etc.
2788 for_each_port(adapter, pidx)
2789 if (test_bit(pidx, &adapter->registered_device_map))
2790 unregister_netdev(adapter->port[pidx]);
2791 t4vf_sge_stop(adapter);
2792 if (adapter->flags & USING_MSIX) {
2793 pci_disable_msix(adapter->pdev);
2794 adapter->flags &= ~USING_MSIX;
2795 } else if (adapter->flags & USING_MSI) {
2796 pci_disable_msi(adapter->pdev);
2797 adapter->flags &= ~USING_MSI;
2801 * Tear down our debugfs entries.
2803 if (adapter->debugfs_root) {
2804 cleanup_debugfs(adapter);
2805 debugfs_remove_recursive(adapter->debugfs_root);
2809 * Free all of the various resources which we've acquired ...
2811 t4vf_free_sge_resources(adapter);
2812 for_each_port(adapter, pidx) {
2813 struct net_device *netdev = adapter->port[pidx];
2814 struct port_info *pi;
2819 pi = netdev_priv(netdev);
2820 t4vf_free_vi(adapter, pi->viid);
2821 free_netdev(netdev);
2823 iounmap(adapter->regs);
2825 pci_set_drvdata(pdev, NULL);
2829 * Disable the device and release its PCI resources.
2831 pci_disable_device(pdev);
2832 pci_clear_master(pdev);
2833 pci_release_regions(pdev);
2837 * PCI Device registration data structures.
2839 #define CH_DEVICE(devid, idx) \
2840 { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
2842 static struct pci_device_id cxgb4vf_pci_tbl[] = {
2843 CH_DEVICE(0xb000, 0), /* PE10K FPGA */
2844 CH_DEVICE(0x4800, 0), /* T440-dbg */
2845 CH_DEVICE(0x4801, 0), /* T420-cr */
2846 CH_DEVICE(0x4802, 0), /* T422-cr */
2847 CH_DEVICE(0x4803, 0), /* T440-cr */
2848 CH_DEVICE(0x4804, 0), /* T420-bch */
2849 CH_DEVICE(0x4805, 0), /* T440-bch */
2850 CH_DEVICE(0x4806, 0), /* T460-ch */
2851 CH_DEVICE(0x4807, 0), /* T420-so */
2852 CH_DEVICE(0x4808, 0), /* T420-cx */
2853 CH_DEVICE(0x4809, 0), /* T420-bt */
2854 CH_DEVICE(0x480a, 0), /* T404-bt */
2858 MODULE_DESCRIPTION(DRV_DESC);
2859 MODULE_AUTHOR("Chelsio Communications");
2860 MODULE_LICENSE("Dual BSD/GPL");
2861 MODULE_VERSION(DRV_VERSION);
2862 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
2864 static struct pci_driver cxgb4vf_driver = {
2865 .name = KBUILD_MODNAME,
2866 .id_table = cxgb4vf_pci_tbl,
2867 .probe = cxgb4vf_pci_probe,
2868 .remove = __devexit_p(cxgb4vf_pci_remove),
2872 * Initialize global driver state.
2874 static int __init cxgb4vf_module_init(void)
2878 /* Debugfs support is optional, just warn if this fails */
2879 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
2880 if (!cxgb4vf_debugfs_root)
2881 printk(KERN_WARNING KBUILD_MODNAME ": could not create"
2882 " debugfs entry, continuing\n");
2884 ret = pci_register_driver(&cxgb4vf_driver);
2886 debugfs_remove(cxgb4vf_debugfs_root);
2891 * Tear down global driver state.
2893 static void __exit cxgb4vf_module_exit(void)
2895 pci_unregister_driver(&cxgb4vf_driver);
2896 debugfs_remove(cxgb4vf_debugfs_root);
2899 module_init(cxgb4vf_module_init);
2900 module_exit(cxgb4vf_module_exit);