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 err = link_start(dev);
757 netif_tx_start_all_queues(dev);
758 set_bit(pi->port_id, &adapter->open_device_map);
762 if (adapter->open_device_map == 0)
763 adapter_down(adapter);
768 * Shut down a net device. This routine is called "cxgb_close" in the PF
771 static int cxgb4vf_stop(struct net_device *dev)
773 struct port_info *pi = netdev_priv(dev);
774 struct adapter *adapter = pi->adapter;
776 netif_tx_stop_all_queues(dev);
777 netif_carrier_off(dev);
778 t4vf_enable_vi(adapter, pi->viid, false, false);
779 pi->link_cfg.link_ok = 0;
781 clear_bit(pi->port_id, &adapter->open_device_map);
782 if (adapter->open_device_map == 0)
783 adapter_down(adapter);
788 * Translate our basic statistics into the standard "ifconfig" statistics.
790 static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev)
792 struct t4vf_port_stats stats;
793 struct port_info *pi = netdev2pinfo(dev);
794 struct adapter *adapter = pi->adapter;
795 struct net_device_stats *ns = &dev->stats;
798 spin_lock(&adapter->stats_lock);
799 err = t4vf_get_port_stats(adapter, pi->pidx, &stats);
800 spin_unlock(&adapter->stats_lock);
802 memset(ns, 0, sizeof(*ns));
806 ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes +
807 stats.tx_ucast_bytes + stats.tx_offload_bytes);
808 ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames +
809 stats.tx_ucast_frames + stats.tx_offload_frames);
810 ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes +
811 stats.rx_ucast_bytes);
812 ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames +
813 stats.rx_ucast_frames);
814 ns->multicast = stats.rx_mcast_frames;
815 ns->tx_errors = stats.tx_drop_frames;
816 ns->rx_errors = stats.rx_err_frames;
822 * Collect up to maxaddrs worth of a netdevice's unicast addresses, starting
823 * at a specified offset within the list, into an array of addrss pointers and
824 * return the number collected.
826 static inline unsigned int collect_netdev_uc_list_addrs(const struct net_device *dev,
829 unsigned int maxaddrs)
831 unsigned int index = 0;
832 unsigned int naddr = 0;
833 const struct netdev_hw_addr *ha;
835 for_each_dev_addr(dev, ha)
836 if (index++ >= offset) {
837 addr[naddr++] = ha->addr;
838 if (naddr >= maxaddrs)
845 * Collect up to maxaddrs worth of a netdevice's multicast addresses, starting
846 * at a specified offset within the list, into an array of addrss pointers and
847 * return the number collected.
849 static inline unsigned int collect_netdev_mc_list_addrs(const struct net_device *dev,
852 unsigned int maxaddrs)
854 unsigned int index = 0;
855 unsigned int naddr = 0;
856 const struct netdev_hw_addr *ha;
858 netdev_for_each_mc_addr(ha, dev)
859 if (index++ >= offset) {
860 addr[naddr++] = ha->addr;
861 if (naddr >= maxaddrs)
868 * Configure the exact and hash address filters to handle a port's multicast
869 * and secondary unicast MAC addresses.
871 static int set_addr_filters(const struct net_device *dev, bool sleep)
876 unsigned int offset, naddr;
879 const struct port_info *pi = netdev_priv(dev);
881 /* first do the secondary unicast addresses */
882 for (offset = 0; ; offset += naddr) {
883 naddr = collect_netdev_uc_list_addrs(dev, addr, offset,
888 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
889 naddr, addr, NULL, &uhash, sleep);
896 /* next set up the multicast addresses */
897 for (offset = 0; ; offset += naddr) {
898 naddr = collect_netdev_mc_list_addrs(dev, addr, offset,
903 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
904 naddr, addr, NULL, &mhash, sleep);
910 return t4vf_set_addr_hash(pi->adapter, pi->viid, uhash != 0,
911 uhash | mhash, sleep);
915 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
916 * If @mtu is -1 it is left unchanged.
918 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
921 struct port_info *pi = netdev_priv(dev);
923 ret = set_addr_filters(dev, sleep_ok);
925 ret = t4vf_set_rxmode(pi->adapter, pi->viid, -1,
926 (dev->flags & IFF_PROMISC) != 0,
927 (dev->flags & IFF_ALLMULTI) != 0,
933 * Set the current receive modes on the device.
935 static void cxgb4vf_set_rxmode(struct net_device *dev)
937 /* unfortunately we can't return errors to the stack */
938 set_rxmode(dev, -1, false);
942 * Find the entry in the interrupt holdoff timer value array which comes
943 * closest to the specified interrupt holdoff value.
945 static int closest_timer(const struct sge *s, int us)
947 int i, timer_idx = 0, min_delta = INT_MAX;
949 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
950 int delta = us - s->timer_val[i];
953 if (delta < min_delta) {
961 static int closest_thres(const struct sge *s, int thres)
963 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
965 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
966 delta = thres - s->counter_val[i];
969 if (delta < min_delta) {
978 * Return a queue's interrupt hold-off time in us. 0 means no timer.
980 static unsigned int qtimer_val(const struct adapter *adapter,
981 const struct sge_rspq *rspq)
983 unsigned int timer_idx = QINTR_TIMER_IDX_GET(rspq->intr_params);
985 return timer_idx < SGE_NTIMERS
986 ? adapter->sge.timer_val[timer_idx]
991 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
992 * @adapter: the adapter
993 * @rspq: the RX response queue
994 * @us: the hold-off time in us, or 0 to disable timer
995 * @cnt: the hold-off packet count, or 0 to disable counter
997 * Sets an RX response queue's interrupt hold-off time and packet count.
998 * At least one of the two needs to be enabled for the queue to generate
1001 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
1002 unsigned int us, unsigned int cnt)
1004 unsigned int timer_idx;
1007 * If both the interrupt holdoff timer and count are specified as
1008 * zero, default to a holdoff count of 1 ...
1010 if ((us | cnt) == 0)
1014 * If an interrupt holdoff count has been specified, then find the
1015 * closest configured holdoff count and use that. If the response
1016 * queue has already been created, then update its queue context
1023 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1024 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1025 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1027 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1028 FW_PARAMS_PARAM_YZ(rspq->cntxt_id);
1029 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1033 rspq->pktcnt_idx = pktcnt_idx;
1037 * Compute the closest holdoff timer index from the supplied holdoff
1040 timer_idx = (us == 0
1041 ? SGE_TIMER_RSTRT_CNTR
1042 : closest_timer(&adapter->sge, us));
1045 * Update the response queue's interrupt coalescing parameters and
1048 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
1049 (cnt > 0 ? QINTR_CNT_EN : 0));
1054 * Return a version number to identify the type of adapter. The scheme is:
1055 * - bits 0..9: chip version
1056 * - bits 10..15: chip revision
1058 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1061 * Chip version 4, revision 0x3f (cxgb4vf).
1063 return 4 | (0x3f << 10);
1067 * Execute the specified ioctl command.
1069 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1075 * The VF Driver doesn't have access to any of the other
1076 * common Ethernet device ioctl()'s (like reading/writing
1077 * PHY registers, etc.
1088 * Change the device's MTU.
1090 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1093 struct port_info *pi = netdev_priv(dev);
1095 /* accommodate SACK */
1099 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1100 -1, -1, -1, -1, true);
1107 * Change the devices MAC address.
1109 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1112 struct sockaddr *addr = _addr;
1113 struct port_info *pi = netdev_priv(dev);
1115 if (!is_valid_ether_addr(addr->sa_data))
1118 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1119 addr->sa_data, true);
1123 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1124 pi->xact_addr_filt = ret;
1128 #ifdef CONFIG_NET_POLL_CONTROLLER
1130 * Poll all of our receive queues. This is called outside of normal interrupt
1133 static void cxgb4vf_poll_controller(struct net_device *dev)
1135 struct port_info *pi = netdev_priv(dev);
1136 struct adapter *adapter = pi->adapter;
1138 if (adapter->flags & USING_MSIX) {
1139 struct sge_eth_rxq *rxq;
1142 rxq = &adapter->sge.ethrxq[pi->first_qset];
1143 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1144 t4vf_sge_intr_msix(0, &rxq->rspq);
1148 t4vf_intr_handler(adapter)(0, adapter);
1153 * Ethtool operations.
1154 * ===================
1156 * Note that we don't support any ethtool operations which change the physical
1157 * state of the port to which we're linked.
1161 * Return current port link settings.
1163 static int cxgb4vf_get_settings(struct net_device *dev,
1164 struct ethtool_cmd *cmd)
1166 const struct port_info *pi = netdev_priv(dev);
1168 cmd->supported = pi->link_cfg.supported;
1169 cmd->advertising = pi->link_cfg.advertising;
1170 ethtool_cmd_speed_set(cmd,
1171 netif_carrier_ok(dev) ? pi->link_cfg.speed : -1);
1172 cmd->duplex = DUPLEX_FULL;
1174 cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1175 cmd->phy_address = pi->port_id;
1176 cmd->transceiver = XCVR_EXTERNAL;
1177 cmd->autoneg = pi->link_cfg.autoneg;
1184 * Return our driver information.
1186 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1187 struct ethtool_drvinfo *drvinfo)
1189 struct adapter *adapter = netdev2adap(dev);
1191 strcpy(drvinfo->driver, KBUILD_MODNAME);
1192 strcpy(drvinfo->version, DRV_VERSION);
1193 strcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)));
1194 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1195 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1196 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.fwrev),
1197 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.fwrev),
1198 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.fwrev),
1199 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.fwrev),
1200 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.tprev),
1201 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.tprev),
1202 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.tprev),
1203 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.tprev));
1207 * Return current adapter message level.
1209 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1211 return netdev2adap(dev)->msg_enable;
1215 * Set current adapter message level.
1217 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1219 netdev2adap(dev)->msg_enable = msglevel;
1223 * Return the device's current Queue Set ring size parameters along with the
1224 * allowed maximum values. Since ethtool doesn't understand the concept of
1225 * multi-queue devices, we just return the current values associated with the
1228 static void cxgb4vf_get_ringparam(struct net_device *dev,
1229 struct ethtool_ringparam *rp)
1231 const struct port_info *pi = netdev_priv(dev);
1232 const struct sge *s = &pi->adapter->sge;
1234 rp->rx_max_pending = MAX_RX_BUFFERS;
1235 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1236 rp->rx_jumbo_max_pending = 0;
1237 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1239 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1240 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1241 rp->rx_jumbo_pending = 0;
1242 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1246 * Set the Queue Set ring size parameters for the device. Again, since
1247 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1248 * apply these new values across all of the Queue Sets associated with the
1249 * device -- after vetting them of course!
1251 static int cxgb4vf_set_ringparam(struct net_device *dev,
1252 struct ethtool_ringparam *rp)
1254 const struct port_info *pi = netdev_priv(dev);
1255 struct adapter *adapter = pi->adapter;
1256 struct sge *s = &adapter->sge;
1259 if (rp->rx_pending > MAX_RX_BUFFERS ||
1260 rp->rx_jumbo_pending ||
1261 rp->tx_pending > MAX_TXQ_ENTRIES ||
1262 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1263 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1264 rp->rx_pending < MIN_FL_ENTRIES ||
1265 rp->tx_pending < MIN_TXQ_ENTRIES)
1268 if (adapter->flags & FULL_INIT_DONE)
1271 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1272 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1273 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1274 s->ethtxq[qs].q.size = rp->tx_pending;
1280 * Return the interrupt holdoff timer and count for the first Queue Set on the
1281 * device. Our extension ioctl() (the cxgbtool interface) allows the
1282 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1284 static int cxgb4vf_get_coalesce(struct net_device *dev,
1285 struct ethtool_coalesce *coalesce)
1287 const struct port_info *pi = netdev_priv(dev);
1288 const struct adapter *adapter = pi->adapter;
1289 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1291 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1292 coalesce->rx_max_coalesced_frames =
1293 ((rspq->intr_params & QINTR_CNT_EN)
1294 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1300 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1301 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1302 * the interrupt holdoff timer on any of the device's Queue Sets.
1304 static int cxgb4vf_set_coalesce(struct net_device *dev,
1305 struct ethtool_coalesce *coalesce)
1307 const struct port_info *pi = netdev_priv(dev);
1308 struct adapter *adapter = pi->adapter;
1310 return set_rxq_intr_params(adapter,
1311 &adapter->sge.ethrxq[pi->first_qset].rspq,
1312 coalesce->rx_coalesce_usecs,
1313 coalesce->rx_max_coalesced_frames);
1317 * Report current port link pause parameter settings.
1319 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1320 struct ethtool_pauseparam *pauseparam)
1322 struct port_info *pi = netdev_priv(dev);
1324 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1325 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1326 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1330 * Identify the port by blinking the port's LED.
1332 static int cxgb4vf_phys_id(struct net_device *dev,
1333 enum ethtool_phys_id_state state)
1336 struct port_info *pi = netdev_priv(dev);
1338 if (state == ETHTOOL_ID_ACTIVE)
1340 else if (state == ETHTOOL_ID_INACTIVE)
1345 return t4vf_identify_port(pi->adapter, pi->viid, val);
1349 * Port stats maintained per queue of the port.
1351 struct queue_port_stats {
1362 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1363 * these need to match the order of statistics returned by
1364 * t4vf_get_port_stats().
1366 static const char stats_strings[][ETH_GSTRING_LEN] = {
1368 * These must match the layout of the t4vf_port_stats structure.
1370 "TxBroadcastBytes ",
1371 "TxBroadcastFrames ",
1372 "TxMulticastBytes ",
1373 "TxMulticastFrames ",
1379 "RxBroadcastBytes ",
1380 "RxBroadcastFrames ",
1381 "RxMulticastBytes ",
1382 "RxMulticastFrames ",
1388 * These are accumulated per-queue statistics and must match the
1389 * order of the fields in the queue_port_stats structure.
1401 * Return the number of statistics in the specified statistics set.
1403 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1407 return ARRAY_SIZE(stats_strings);
1415 * Return the strings for the specified statistics set.
1417 static void cxgb4vf_get_strings(struct net_device *dev,
1423 memcpy(data, stats_strings, sizeof(stats_strings));
1429 * Small utility routine to accumulate queue statistics across the queues of
1432 static void collect_sge_port_stats(const struct adapter *adapter,
1433 const struct port_info *pi,
1434 struct queue_port_stats *stats)
1436 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1437 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1440 memset(stats, 0, sizeof(*stats));
1441 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1442 stats->tso += txq->tso;
1443 stats->tx_csum += txq->tx_cso;
1444 stats->rx_csum += rxq->stats.rx_cso;
1445 stats->vlan_ex += rxq->stats.vlan_ex;
1446 stats->vlan_ins += txq->vlan_ins;
1447 stats->lro_pkts += rxq->stats.lro_pkts;
1448 stats->lro_merged += rxq->stats.lro_merged;
1453 * Return the ETH_SS_STATS statistics set.
1455 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1456 struct ethtool_stats *stats,
1459 struct port_info *pi = netdev2pinfo(dev);
1460 struct adapter *adapter = pi->adapter;
1461 int err = t4vf_get_port_stats(adapter, pi->pidx,
1462 (struct t4vf_port_stats *)data);
1464 memset(data, 0, sizeof(struct t4vf_port_stats));
1466 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1467 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1471 * Return the size of our register map.
1473 static int cxgb4vf_get_regs_len(struct net_device *dev)
1475 return T4VF_REGMAP_SIZE;
1479 * Dump a block of registers, start to end inclusive, into a buffer.
1481 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1482 unsigned int start, unsigned int end)
1484 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1486 for ( ; start <= end; start += sizeof(u32)) {
1488 * Avoid reading the Mailbox Control register since that
1489 * can trigger a Mailbox Ownership Arbitration cycle and
1490 * interfere with communication with the firmware.
1492 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1495 *bp++ = t4_read_reg(adapter, start);
1500 * Copy our entire register map into the provided buffer.
1502 static void cxgb4vf_get_regs(struct net_device *dev,
1503 struct ethtool_regs *regs,
1506 struct adapter *adapter = netdev2adap(dev);
1508 regs->version = mk_adap_vers(adapter);
1511 * Fill in register buffer with our register map.
1513 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1515 reg_block_dump(adapter, regbuf,
1516 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1517 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1518 reg_block_dump(adapter, regbuf,
1519 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1520 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1521 reg_block_dump(adapter, regbuf,
1522 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1523 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_LAST);
1524 reg_block_dump(adapter, regbuf,
1525 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1526 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1528 reg_block_dump(adapter, regbuf,
1529 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1530 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1534 * Report current Wake On LAN settings.
1536 static void cxgb4vf_get_wol(struct net_device *dev,
1537 struct ethtool_wolinfo *wol)
1541 memset(&wol->sopass, 0, sizeof(wol->sopass));
1545 * TCP Segmentation Offload flags which we support.
1547 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1549 static struct ethtool_ops cxgb4vf_ethtool_ops = {
1550 .get_settings = cxgb4vf_get_settings,
1551 .get_drvinfo = cxgb4vf_get_drvinfo,
1552 .get_msglevel = cxgb4vf_get_msglevel,
1553 .set_msglevel = cxgb4vf_set_msglevel,
1554 .get_ringparam = cxgb4vf_get_ringparam,
1555 .set_ringparam = cxgb4vf_set_ringparam,
1556 .get_coalesce = cxgb4vf_get_coalesce,
1557 .set_coalesce = cxgb4vf_set_coalesce,
1558 .get_pauseparam = cxgb4vf_get_pauseparam,
1559 .get_link = ethtool_op_get_link,
1560 .get_strings = cxgb4vf_get_strings,
1561 .set_phys_id = cxgb4vf_phys_id,
1562 .get_sset_count = cxgb4vf_get_sset_count,
1563 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1564 .get_regs_len = cxgb4vf_get_regs_len,
1565 .get_regs = cxgb4vf_get_regs,
1566 .get_wol = cxgb4vf_get_wol,
1570 * /sys/kernel/debug/cxgb4vf support code and data.
1571 * ================================================
1575 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1579 static int sge_qinfo_show(struct seq_file *seq, void *v)
1581 struct adapter *adapter = seq->private;
1582 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1583 int qs, r = (uintptr_t)v - 1;
1586 seq_putc(seq, '\n');
1588 #define S3(fmt_spec, s, v) \
1590 seq_printf(seq, "%-12s", s); \
1591 for (qs = 0; qs < n; ++qs) \
1592 seq_printf(seq, " %16" fmt_spec, v); \
1593 seq_putc(seq, '\n'); \
1595 #define S(s, v) S3("s", s, v)
1596 #define T(s, v) S3("u", s, txq[qs].v)
1597 #define R(s, v) S3("u", s, rxq[qs].v)
1599 if (r < eth_entries) {
1600 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1601 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1602 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1604 S("QType:", "Ethernet");
1606 (rxq[qs].rspq.netdev
1607 ? rxq[qs].rspq.netdev->name
1610 (rxq[qs].rspq.netdev
1611 ? ((struct port_info *)
1612 netdev_priv(rxq[qs].rspq.netdev))->port_id
1614 T("TxQ ID:", q.abs_id);
1615 T("TxQ size:", q.size);
1616 T("TxQ inuse:", q.in_use);
1617 T("TxQ PIdx:", q.pidx);
1618 T("TxQ CIdx:", q.cidx);
1619 R("RspQ ID:", rspq.abs_id);
1620 R("RspQ size:", rspq.size);
1621 R("RspQE size:", rspq.iqe_len);
1622 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1623 S3("u", "Intr pktcnt:",
1624 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1625 R("RspQ CIdx:", rspq.cidx);
1626 R("RspQ Gen:", rspq.gen);
1627 R("FL ID:", fl.abs_id);
1628 R("FL size:", fl.size - MIN_FL_RESID);
1629 R("FL avail:", fl.avail);
1630 R("FL PIdx:", fl.pidx);
1631 R("FL CIdx:", fl.cidx);
1637 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1639 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1640 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1641 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1642 qtimer_val(adapter, evtq));
1643 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1644 adapter->sge.counter_val[evtq->pktcnt_idx]);
1645 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1646 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1647 } else if (r == 1) {
1648 const struct sge_rspq *intrq = &adapter->sge.intrq;
1650 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1651 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1652 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1653 qtimer_val(adapter, intrq));
1654 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1655 adapter->sge.counter_val[intrq->pktcnt_idx]);
1656 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1657 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1669 * Return the number of "entries" in our "file". We group the multi-Queue
1670 * sections with QPL Queue Sets per "entry". The sections of the output are:
1672 * Ethernet RX/TX Queue Sets
1673 * Firmware Event Queue
1674 * Forwarded Interrupt Queue (if in MSI mode)
1676 static int sge_queue_entries(const struct adapter *adapter)
1678 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1679 ((adapter->flags & USING_MSI) != 0);
1682 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1684 int entries = sge_queue_entries(seq->private);
1686 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1689 static void sge_queue_stop(struct seq_file *seq, void *v)
1693 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
1695 int entries = sge_queue_entries(seq->private);
1698 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1701 static const struct seq_operations sge_qinfo_seq_ops = {
1702 .start = sge_queue_start,
1703 .next = sge_queue_next,
1704 .stop = sge_queue_stop,
1705 .show = sge_qinfo_show
1708 static int sge_qinfo_open(struct inode *inode, struct file *file)
1710 int res = seq_open(file, &sge_qinfo_seq_ops);
1713 struct seq_file *seq = file->private_data;
1714 seq->private = inode->i_private;
1719 static const struct file_operations sge_qinfo_debugfs_fops = {
1720 .owner = THIS_MODULE,
1721 .open = sge_qinfo_open,
1723 .llseek = seq_lseek,
1724 .release = seq_release,
1728 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
1732 static int sge_qstats_show(struct seq_file *seq, void *v)
1734 struct adapter *adapter = seq->private;
1735 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1736 int qs, r = (uintptr_t)v - 1;
1739 seq_putc(seq, '\n');
1741 #define S3(fmt, s, v) \
1743 seq_printf(seq, "%-16s", s); \
1744 for (qs = 0; qs < n; ++qs) \
1745 seq_printf(seq, " %8" fmt, v); \
1746 seq_putc(seq, '\n'); \
1748 #define S(s, v) S3("s", s, v)
1750 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
1751 #define T(s, v) T3("lu", s, v)
1753 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
1754 #define R(s, v) R3("lu", s, v)
1756 if (r < eth_entries) {
1757 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1758 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1759 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1761 S("QType:", "Ethernet");
1763 (rxq[qs].rspq.netdev
1764 ? rxq[qs].rspq.netdev->name
1766 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
1767 R("RxPackets:", stats.pkts);
1768 R("RxCSO:", stats.rx_cso);
1769 R("VLANxtract:", stats.vlan_ex);
1770 R("LROmerged:", stats.lro_merged);
1771 R("LROpackets:", stats.lro_pkts);
1772 R("RxDrops:", stats.rx_drops);
1774 T("TxCSO:", tx_cso);
1775 T("VLANins:", vlan_ins);
1776 T("TxQFull:", q.stops);
1777 T("TxQRestarts:", q.restarts);
1778 T("TxMapErr:", mapping_err);
1779 R("FLAllocErr:", fl.alloc_failed);
1780 R("FLLrgAlcErr:", fl.large_alloc_failed);
1781 R("FLStarving:", fl.starving);
1787 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1789 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
1790 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1791 evtq->unhandled_irqs);
1792 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
1793 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
1794 } else if (r == 1) {
1795 const struct sge_rspq *intrq = &adapter->sge.intrq;
1797 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
1798 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1799 intrq->unhandled_irqs);
1800 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
1801 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
1815 * Return the number of "entries" in our "file". We group the multi-Queue
1816 * sections with QPL Queue Sets per "entry". The sections of the output are:
1818 * Ethernet RX/TX Queue Sets
1819 * Firmware Event Queue
1820 * Forwarded Interrupt Queue (if in MSI mode)
1822 static int sge_qstats_entries(const struct adapter *adapter)
1824 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1825 ((adapter->flags & USING_MSI) != 0);
1828 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
1830 int entries = sge_qstats_entries(seq->private);
1832 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1835 static void sge_qstats_stop(struct seq_file *seq, void *v)
1839 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
1841 int entries = sge_qstats_entries(seq->private);
1844 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1847 static const struct seq_operations sge_qstats_seq_ops = {
1848 .start = sge_qstats_start,
1849 .next = sge_qstats_next,
1850 .stop = sge_qstats_stop,
1851 .show = sge_qstats_show
1854 static int sge_qstats_open(struct inode *inode, struct file *file)
1856 int res = seq_open(file, &sge_qstats_seq_ops);
1859 struct seq_file *seq = file->private_data;
1860 seq->private = inode->i_private;
1865 static const struct file_operations sge_qstats_proc_fops = {
1866 .owner = THIS_MODULE,
1867 .open = sge_qstats_open,
1869 .llseek = seq_lseek,
1870 .release = seq_release,
1874 * Show PCI-E SR-IOV Virtual Function Resource Limits.
1876 static int resources_show(struct seq_file *seq, void *v)
1878 struct adapter *adapter = seq->private;
1879 struct vf_resources *vfres = &adapter->params.vfres;
1881 #define S(desc, fmt, var) \
1882 seq_printf(seq, "%-60s " fmt "\n", \
1883 desc " (" #var "):", vfres->var)
1885 S("Virtual Interfaces", "%d", nvi);
1886 S("Egress Queues", "%d", neq);
1887 S("Ethernet Control", "%d", nethctrl);
1888 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
1889 S("Ingress Queues", "%d", niq);
1890 S("Traffic Class", "%d", tc);
1891 S("Port Access Rights Mask", "%#x", pmask);
1892 S("MAC Address Filters", "%d", nexactf);
1893 S("Firmware Command Read Capabilities", "%#x", r_caps);
1894 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
1901 static int resources_open(struct inode *inode, struct file *file)
1903 return single_open(file, resources_show, inode->i_private);
1906 static const struct file_operations resources_proc_fops = {
1907 .owner = THIS_MODULE,
1908 .open = resources_open,
1910 .llseek = seq_lseek,
1911 .release = single_release,
1915 * Show Virtual Interfaces.
1917 static int interfaces_show(struct seq_file *seq, void *v)
1919 if (v == SEQ_START_TOKEN) {
1920 seq_puts(seq, "Interface Port VIID\n");
1922 struct adapter *adapter = seq->private;
1923 int pidx = (uintptr_t)v - 2;
1924 struct net_device *dev = adapter->port[pidx];
1925 struct port_info *pi = netdev_priv(dev);
1927 seq_printf(seq, "%9s %4d %#5x\n",
1928 dev->name, pi->port_id, pi->viid);
1933 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
1935 return pos <= adapter->params.nports
1936 ? (void *)(uintptr_t)(pos + 1)
1940 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
1943 ? interfaces_get_idx(seq->private, *pos)
1947 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
1950 return interfaces_get_idx(seq->private, *pos);
1953 static void interfaces_stop(struct seq_file *seq, void *v)
1957 static const struct seq_operations interfaces_seq_ops = {
1958 .start = interfaces_start,
1959 .next = interfaces_next,
1960 .stop = interfaces_stop,
1961 .show = interfaces_show
1964 static int interfaces_open(struct inode *inode, struct file *file)
1966 int res = seq_open(file, &interfaces_seq_ops);
1969 struct seq_file *seq = file->private_data;
1970 seq->private = inode->i_private;
1975 static const struct file_operations interfaces_proc_fops = {
1976 .owner = THIS_MODULE,
1977 .open = interfaces_open,
1979 .llseek = seq_lseek,
1980 .release = seq_release,
1984 * /sys/kernel/debugfs/cxgb4vf/ files list.
1986 struct cxgb4vf_debugfs_entry {
1987 const char *name; /* name of debugfs node */
1988 mode_t mode; /* file system mode */
1989 const struct file_operations *fops;
1992 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
1993 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
1994 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
1995 { "resources", S_IRUGO, &resources_proc_fops },
1996 { "interfaces", S_IRUGO, &interfaces_proc_fops },
2000 * Module and device initialization and cleanup code.
2001 * ==================================================
2005 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2006 * directory (debugfs_root) has already been set up.
2008 static int __devinit setup_debugfs(struct adapter *adapter)
2012 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2015 * Debugfs support is best effort.
2017 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2018 (void)debugfs_create_file(debugfs_files[i].name,
2019 debugfs_files[i].mode,
2020 adapter->debugfs_root,
2022 debugfs_files[i].fops);
2028 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2029 * it to our caller to tear down the directory (debugfs_root).
2031 static void cleanup_debugfs(struct adapter *adapter)
2033 BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root));
2036 * Unlike our sister routine cleanup_proc(), we don't need to remove
2037 * individual entries because a call will be made to
2038 * debugfs_remove_recursive(). We just need to clean up any ancillary
2045 * Perform early "adapter" initialization. This is where we discover what
2046 * adapter parameters we're going to be using and initialize basic adapter
2049 static int __devinit adap_init0(struct adapter *adapter)
2051 struct vf_resources *vfres = &adapter->params.vfres;
2052 struct sge_params *sge_params = &adapter->params.sge;
2053 struct sge *s = &adapter->sge;
2054 unsigned int ethqsets;
2058 * Wait for the device to become ready before proceeding ...
2060 err = t4vf_wait_dev_ready(adapter);
2062 dev_err(adapter->pdev_dev, "device didn't become ready:"
2068 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2069 * 2.6.31 and later we can't call pci_reset_function() in order to
2070 * issue an FLR because of a self- deadlock on the device semaphore.
2071 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2072 * cases where they're needed -- for instance, some versions of KVM
2073 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2074 * use the firmware based reset in order to reset any per function
2077 err = t4vf_fw_reset(adapter);
2079 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2084 * Grab basic operational parameters. These will predominantly have
2085 * been set up by the Physical Function Driver or will be hard coded
2086 * into the adapter. We just have to live with them ... Note that
2087 * we _must_ get our VPD parameters before our SGE parameters because
2088 * we need to know the adapter's core clock from the VPD in order to
2089 * properly decode the SGE Timer Values.
2091 err = t4vf_get_dev_params(adapter);
2093 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2094 " device parameters: err=%d\n", err);
2097 err = t4vf_get_vpd_params(adapter);
2099 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2100 " VPD parameters: err=%d\n", err);
2103 err = t4vf_get_sge_params(adapter);
2105 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2106 " SGE parameters: err=%d\n", err);
2109 err = t4vf_get_rss_glb_config(adapter);
2111 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2112 " RSS parameters: err=%d\n", err);
2115 if (adapter->params.rss.mode !=
2116 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2117 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2118 " mode %d\n", adapter->params.rss.mode);
2121 err = t4vf_sge_init(adapter);
2123 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2129 * Retrieve our RX interrupt holdoff timer values and counter
2130 * threshold values from the SGE parameters.
2132 s->timer_val[0] = core_ticks_to_us(adapter,
2133 TIMERVALUE0_GET(sge_params->sge_timer_value_0_and_1));
2134 s->timer_val[1] = core_ticks_to_us(adapter,
2135 TIMERVALUE1_GET(sge_params->sge_timer_value_0_and_1));
2136 s->timer_val[2] = core_ticks_to_us(adapter,
2137 TIMERVALUE0_GET(sge_params->sge_timer_value_2_and_3));
2138 s->timer_val[3] = core_ticks_to_us(adapter,
2139 TIMERVALUE1_GET(sge_params->sge_timer_value_2_and_3));
2140 s->timer_val[4] = core_ticks_to_us(adapter,
2141 TIMERVALUE0_GET(sge_params->sge_timer_value_4_and_5));
2142 s->timer_val[5] = core_ticks_to_us(adapter,
2143 TIMERVALUE1_GET(sge_params->sge_timer_value_4_and_5));
2146 THRESHOLD_0_GET(sge_params->sge_ingress_rx_threshold);
2148 THRESHOLD_1_GET(sge_params->sge_ingress_rx_threshold);
2150 THRESHOLD_2_GET(sge_params->sge_ingress_rx_threshold);
2152 THRESHOLD_3_GET(sge_params->sge_ingress_rx_threshold);
2155 * Grab our Virtual Interface resource allocation, extract the
2156 * features that we're interested in and do a bit of sanity testing on
2159 err = t4vf_get_vfres(adapter);
2161 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2162 " resources: err=%d\n", err);
2167 * The number of "ports" which we support is equal to the number of
2168 * Virtual Interfaces with which we've been provisioned.
2170 adapter->params.nports = vfres->nvi;
2171 if (adapter->params.nports > MAX_NPORTS) {
2172 dev_warn(adapter->pdev_dev, "only using %d of %d allowed"
2173 " virtual interfaces\n", MAX_NPORTS,
2174 adapter->params.nports);
2175 adapter->params.nports = MAX_NPORTS;
2179 * We need to reserve a number of the ingress queues with Free List
2180 * and Interrupt capabilities for special interrupt purposes (like
2181 * asynchronous firmware messages, or forwarded interrupts if we're
2182 * using MSI). The rest of the FL/Intr-capable ingress queues will be
2183 * matched up one-for-one with Ethernet/Control egress queues in order
2184 * to form "Queue Sets" which will be aportioned between the "ports".
2185 * For each Queue Set, we'll need the ability to allocate two Egress
2186 * Contexts -- one for the Ingress Queue Free List and one for the TX
2189 ethqsets = vfres->niqflint - INGQ_EXTRAS;
2190 if (vfres->nethctrl != ethqsets) {
2191 dev_warn(adapter->pdev_dev, "unequal number of [available]"
2192 " ingress/egress queues (%d/%d); using minimum for"
2193 " number of Queue Sets\n", ethqsets, vfres->nethctrl);
2194 ethqsets = min(vfres->nethctrl, ethqsets);
2196 if (vfres->neq < ethqsets*2) {
2197 dev_warn(adapter->pdev_dev, "Not enough Egress Contexts (%d)"
2198 " to support Queue Sets (%d); reducing allowed Queue"
2199 " Sets\n", vfres->neq, ethqsets);
2200 ethqsets = vfres->neq/2;
2202 if (ethqsets > MAX_ETH_QSETS) {
2203 dev_warn(adapter->pdev_dev, "only using %d of %d allowed Queue"
2204 " Sets\n", MAX_ETH_QSETS, adapter->sge.max_ethqsets);
2205 ethqsets = MAX_ETH_QSETS;
2207 if (vfres->niq != 0 || vfres->neq > ethqsets*2) {
2208 dev_warn(adapter->pdev_dev, "unused resources niq/neq (%d/%d)"
2209 " ignored\n", vfres->niq, vfres->neq - ethqsets*2);
2211 adapter->sge.max_ethqsets = ethqsets;
2214 * Check for various parameter sanity issues. Most checks simply
2215 * result in us using fewer resources than our provissioning but we
2216 * do need at least one "port" with which to work ...
2218 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2219 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2220 " virtual interfaces (too few Queue Sets)\n",
2221 adapter->sge.max_ethqsets, adapter->params.nports);
2222 adapter->params.nports = adapter->sge.max_ethqsets;
2224 if (adapter->params.nports == 0) {
2225 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2232 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2233 u8 pkt_cnt_idx, unsigned int size,
2234 unsigned int iqe_size)
2236 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
2237 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0));
2238 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2241 rspq->iqe_len = iqe_size;
2246 * Perform default configuration of DMA queues depending on the number and
2247 * type of ports we found and the number of available CPUs. Most settings can
2248 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2249 * being brought up for the first time.
2251 static void __devinit cfg_queues(struct adapter *adapter)
2253 struct sge *s = &adapter->sge;
2254 int q10g, n10g, qidx, pidx, qs;
2258 * We should not be called till we know how many Queue Sets we can
2259 * support. In particular, this means that we need to know what kind
2260 * of interrupts we'll be using ...
2262 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2265 * Count the number of 10GbE Virtual Interfaces that we have.
2268 for_each_port(adapter, pidx)
2269 n10g += is_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2272 * We default to 1 queue per non-10G port and up to # of cores queues
2278 int n1g = (adapter->params.nports - n10g);
2279 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2280 if (q10g > num_online_cpus())
2281 q10g = num_online_cpus();
2285 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2286 * The layout will be established in setup_sge_queues() when the
2287 * adapter is brough up for the first time.
2290 for_each_port(adapter, pidx) {
2291 struct port_info *pi = adap2pinfo(adapter, pidx);
2293 pi->first_qset = qidx;
2294 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
2300 * The Ingress Queue Entry Size for our various Response Queues needs
2301 * to be big enough to accommodate the largest message we can receive
2302 * from the chip/firmware; which is 64 bytes ...
2307 * Set up default Queue Set parameters ... Start off with the
2308 * shortest interrupt holdoff timer.
2310 for (qs = 0; qs < s->max_ethqsets; qs++) {
2311 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2312 struct sge_eth_txq *txq = &s->ethtxq[qs];
2314 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2320 * The firmware event queue is used for link state changes and
2321 * notifications of TX DMA completions.
2323 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2326 * The forwarded interrupt queue is used when we're in MSI interrupt
2327 * mode. In this mode all interrupts associated with RX queues will
2328 * be forwarded to a single queue which we'll associate with our MSI
2329 * interrupt vector. The messages dropped in the forwarded interrupt
2330 * queue will indicate which ingress queue needs servicing ... This
2331 * queue needs to be large enough to accommodate all of the ingress
2332 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2333 * from equalling the CIDX if every ingress queue has an outstanding
2334 * interrupt). The queue doesn't need to be any larger because no
2335 * ingress queue will ever have more than one outstanding interrupt at
2338 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2343 * Reduce the number of Ethernet queues across all ports to at most n.
2344 * n provides at least one queue per port.
2346 static void __devinit reduce_ethqs(struct adapter *adapter, int n)
2349 struct port_info *pi;
2352 * While we have too many active Ether Queue Sets, interate across the
2353 * "ports" and reduce their individual Queue Set allocations.
2355 BUG_ON(n < adapter->params.nports);
2356 while (n < adapter->sge.ethqsets)
2357 for_each_port(adapter, i) {
2358 pi = adap2pinfo(adapter, i);
2359 if (pi->nqsets > 1) {
2361 adapter->sge.ethqsets--;
2362 if (adapter->sge.ethqsets <= n)
2368 * Reassign the starting Queue Sets for each of the "ports" ...
2371 for_each_port(adapter, i) {
2372 pi = adap2pinfo(adapter, i);
2379 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2380 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2381 * need. Minimally we need one for every Virtual Interface plus those needed
2382 * for our "extras". Note that this process may lower the maximum number of
2383 * allowed Queue Sets ...
2385 static int __devinit enable_msix(struct adapter *adapter)
2387 int i, err, want, need;
2388 struct msix_entry entries[MSIX_ENTRIES];
2389 struct sge *s = &adapter->sge;
2391 for (i = 0; i < MSIX_ENTRIES; ++i)
2392 entries[i].entry = i;
2395 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2396 * plus those needed for our "extras" (for example, the firmware
2397 * message queue). We _need_ at least one "Queue Set" per Virtual
2398 * Interface plus those needed for our "extras". So now we get to see
2399 * if the song is right ...
2401 want = s->max_ethqsets + MSIX_EXTRAS;
2402 need = adapter->params.nports + MSIX_EXTRAS;
2403 while ((err = pci_enable_msix(adapter->pdev, entries, want)) >= need)
2407 int nqsets = want - MSIX_EXTRAS;
2408 if (nqsets < s->max_ethqsets) {
2409 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2410 " for %d Queue Sets\n", nqsets);
2411 s->max_ethqsets = nqsets;
2412 if (nqsets < s->ethqsets)
2413 reduce_ethqs(adapter, nqsets);
2415 for (i = 0; i < want; ++i)
2416 adapter->msix_info[i].vec = entries[i].vector;
2417 } else if (err > 0) {
2418 pci_disable_msix(adapter->pdev);
2419 dev_info(adapter->pdev_dev, "only %d MSI-X vectors left,"
2420 " not using MSI-X\n", err);
2425 #ifdef HAVE_NET_DEVICE_OPS
2426 static const struct net_device_ops cxgb4vf_netdev_ops = {
2427 .ndo_open = cxgb4vf_open,
2428 .ndo_stop = cxgb4vf_stop,
2429 .ndo_start_xmit = t4vf_eth_xmit,
2430 .ndo_get_stats = cxgb4vf_get_stats,
2431 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2432 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2433 .ndo_validate_addr = eth_validate_addr,
2434 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2435 .ndo_change_mtu = cxgb4vf_change_mtu,
2436 .ndo_vlan_rx_register = cxgb4vf_vlan_rx_register,
2437 #ifdef CONFIG_NET_POLL_CONTROLLER
2438 .ndo_poll_controller = cxgb4vf_poll_controller,
2444 * "Probe" a device: initialize a device and construct all kernel and driver
2445 * state needed to manage the device. This routine is called "init_one" in
2448 static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
2449 const struct pci_device_id *ent)
2451 static int version_printed;
2456 struct adapter *adapter;
2457 struct port_info *pi;
2458 struct net_device *netdev;
2461 * Print our driver banner the first time we're called to initialize a
2464 if (version_printed == 0) {
2465 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
2466 version_printed = 1;
2470 * Initialize generic PCI device state.
2472 err = pci_enable_device(pdev);
2474 dev_err(&pdev->dev, "cannot enable PCI device\n");
2479 * Reserve PCI resources for the device. If we can't get them some
2480 * other driver may have already claimed the device ...
2482 err = pci_request_regions(pdev, KBUILD_MODNAME);
2484 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2485 goto err_disable_device;
2489 * Set up our DMA mask: try for 64-bit address masking first and
2490 * fall back to 32-bit if we can't get 64 bits ...
2492 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2494 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2496 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2497 " coherent allocations\n");
2498 goto err_release_regions;
2502 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2504 dev_err(&pdev->dev, "no usable DMA configuration\n");
2505 goto err_release_regions;
2511 * Enable bus mastering for the device ...
2513 pci_set_master(pdev);
2516 * Allocate our adapter data structure and attach it to the device.
2518 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2521 goto err_release_regions;
2523 pci_set_drvdata(pdev, adapter);
2524 adapter->pdev = pdev;
2525 adapter->pdev_dev = &pdev->dev;
2528 * Initialize SMP data synchronization resources.
2530 spin_lock_init(&adapter->stats_lock);
2533 * Map our I/O registers in BAR0.
2535 adapter->regs = pci_ioremap_bar(pdev, 0);
2536 if (!adapter->regs) {
2537 dev_err(&pdev->dev, "cannot map device registers\n");
2539 goto err_free_adapter;
2543 * Initialize adapter level features.
2545 adapter->name = pci_name(pdev);
2546 adapter->msg_enable = dflt_msg_enable;
2547 err = adap_init0(adapter);
2552 * Allocate our "adapter ports" and stitch everything together.
2554 pmask = adapter->params.vfres.pmask;
2555 for_each_port(adapter, pidx) {
2559 * We simplistically allocate our virtual interfaces
2560 * sequentially across the port numbers to which we have
2561 * access rights. This should be configurable in some manner
2566 port_id = ffs(pmask) - 1;
2567 pmask &= ~(1 << port_id);
2568 viid = t4vf_alloc_vi(adapter, port_id);
2570 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2571 " err=%d\n", port_id, viid);
2577 * Allocate our network device and stitch things together.
2579 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2581 if (netdev == NULL) {
2582 dev_err(&pdev->dev, "cannot allocate netdev for"
2583 " port %d\n", port_id);
2584 t4vf_free_vi(adapter, viid);
2588 adapter->port[pidx] = netdev;
2589 SET_NETDEV_DEV(netdev, &pdev->dev);
2590 pi = netdev_priv(netdev);
2591 pi->adapter = adapter;
2593 pi->port_id = port_id;
2597 * Initialize the starting state of our "port" and register
2600 pi->xact_addr_filt = -1;
2601 netif_carrier_off(netdev);
2602 netdev->irq = pdev->irq;
2604 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
2605 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2606 NETIF_F_HW_VLAN_TX | NETIF_F_RXCSUM;
2607 netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
2608 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2610 netdev->features = netdev->hw_features |
2613 netdev->features |= NETIF_F_HIGHDMA;
2615 #ifdef HAVE_NET_DEVICE_OPS
2616 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2618 netdev->vlan_rx_register = cxgb4vf_vlan_rx_register;
2619 netdev->open = cxgb4vf_open;
2620 netdev->stop = cxgb4vf_stop;
2621 netdev->hard_start_xmit = t4vf_eth_xmit;
2622 netdev->get_stats = cxgb4vf_get_stats;
2623 netdev->set_rx_mode = cxgb4vf_set_rxmode;
2624 netdev->do_ioctl = cxgb4vf_do_ioctl;
2625 netdev->change_mtu = cxgb4vf_change_mtu;
2626 netdev->set_mac_address = cxgb4vf_set_mac_addr;
2627 #ifdef CONFIG_NET_POLL_CONTROLLER
2628 netdev->poll_controller = cxgb4vf_poll_controller;
2631 SET_ETHTOOL_OPS(netdev, &cxgb4vf_ethtool_ops);
2634 * Initialize the hardware/software state for the port.
2636 err = t4vf_port_init(adapter, pidx);
2638 dev_err(&pdev->dev, "cannot initialize port %d\n",
2645 * The "card" is now ready to go. If any errors occur during device
2646 * registration we do not fail the whole "card" but rather proceed
2647 * only with the ports we manage to register successfully. However we
2648 * must register at least one net device.
2650 for_each_port(adapter, pidx) {
2651 netdev = adapter->port[pidx];
2655 err = register_netdev(netdev);
2657 dev_warn(&pdev->dev, "cannot register net device %s,"
2658 " skipping\n", netdev->name);
2662 set_bit(pidx, &adapter->registered_device_map);
2664 if (adapter->registered_device_map == 0) {
2665 dev_err(&pdev->dev, "could not register any net devices\n");
2670 * Set up our debugfs entries.
2672 if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) {
2673 adapter->debugfs_root =
2674 debugfs_create_dir(pci_name(pdev),
2675 cxgb4vf_debugfs_root);
2676 if (IS_ERR_OR_NULL(adapter->debugfs_root))
2677 dev_warn(&pdev->dev, "could not create debugfs"
2680 setup_debugfs(adapter);
2684 * See what interrupts we'll be using. If we've been configured to
2685 * use MSI-X interrupts, try to enable them but fall back to using
2686 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
2687 * get MSI interrupts we bail with the error.
2689 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
2690 adapter->flags |= USING_MSIX;
2692 err = pci_enable_msi(pdev);
2694 dev_err(&pdev->dev, "Unable to allocate %s interrupts;"
2696 msi == MSI_MSIX ? "MSI-X or MSI" : "MSI", err);
2697 goto err_free_debugfs;
2699 adapter->flags |= USING_MSI;
2703 * Now that we know how many "ports" we have and what their types are,
2704 * and how many Queue Sets we can support, we can configure our queue
2707 cfg_queues(adapter);
2710 * Print a short notice on the existence and configuration of the new
2711 * VF network device ...
2713 for_each_port(adapter, pidx) {
2714 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
2715 adapter->port[pidx]->name,
2716 (adapter->flags & USING_MSIX) ? "MSI-X" :
2717 (adapter->flags & USING_MSI) ? "MSI" : "");
2726 * Error recovery and exit code. Unwind state that's been created
2727 * so far and return the error.
2731 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
2732 cleanup_debugfs(adapter);
2733 debugfs_remove_recursive(adapter->debugfs_root);
2737 for_each_port(adapter, pidx) {
2738 netdev = adapter->port[pidx];
2741 pi = netdev_priv(netdev);
2742 t4vf_free_vi(adapter, pi->viid);
2743 if (test_bit(pidx, &adapter->registered_device_map))
2744 unregister_netdev(netdev);
2745 free_netdev(netdev);
2749 iounmap(adapter->regs);
2753 pci_set_drvdata(pdev, NULL);
2755 err_release_regions:
2756 pci_release_regions(pdev);
2757 pci_set_drvdata(pdev, NULL);
2758 pci_clear_master(pdev);
2761 pci_disable_device(pdev);
2767 * "Remove" a device: tear down all kernel and driver state created in the
2768 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
2769 * that this is called "remove_one" in the PF Driver.)
2771 static void __devexit cxgb4vf_pci_remove(struct pci_dev *pdev)
2773 struct adapter *adapter = pci_get_drvdata(pdev);
2776 * Tear down driver state associated with device.
2782 * Stop all of our activity. Unregister network port,
2783 * disable interrupts, etc.
2785 for_each_port(adapter, pidx)
2786 if (test_bit(pidx, &adapter->registered_device_map))
2787 unregister_netdev(adapter->port[pidx]);
2788 t4vf_sge_stop(adapter);
2789 if (adapter->flags & USING_MSIX) {
2790 pci_disable_msix(adapter->pdev);
2791 adapter->flags &= ~USING_MSIX;
2792 } else if (adapter->flags & USING_MSI) {
2793 pci_disable_msi(adapter->pdev);
2794 adapter->flags &= ~USING_MSI;
2798 * Tear down our debugfs entries.
2800 if (!IS_ERR_OR_NULL(adapter->debugfs_root)) {
2801 cleanup_debugfs(adapter);
2802 debugfs_remove_recursive(adapter->debugfs_root);
2806 * Free all of the various resources which we've acquired ...
2808 t4vf_free_sge_resources(adapter);
2809 for_each_port(adapter, pidx) {
2810 struct net_device *netdev = adapter->port[pidx];
2811 struct port_info *pi;
2816 pi = netdev_priv(netdev);
2817 t4vf_free_vi(adapter, pi->viid);
2818 free_netdev(netdev);
2820 iounmap(adapter->regs);
2822 pci_set_drvdata(pdev, NULL);
2826 * Disable the device and release its PCI resources.
2828 pci_disable_device(pdev);
2829 pci_clear_master(pdev);
2830 pci_release_regions(pdev);
2834 * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt
2837 static void __devexit cxgb4vf_pci_shutdown(struct pci_dev *pdev)
2839 struct adapter *adapter;
2842 adapter = pci_get_drvdata(pdev);
2847 * Disable all Virtual Interfaces. This will shut down the
2848 * delivery of all ingress packets into the chip for these
2849 * Virtual Interfaces.
2851 for_each_port(adapter, pidx) {
2852 struct net_device *netdev;
2853 struct port_info *pi;
2855 if (!test_bit(pidx, &adapter->registered_device_map))
2858 netdev = adapter->port[pidx];
2862 pi = netdev_priv(netdev);
2863 t4vf_enable_vi(adapter, pi->viid, false, false);
2867 * Free up all Queues which will prevent further DMA and
2868 * Interrupts allowing various internal pathways to drain.
2870 t4vf_free_sge_resources(adapter);
2874 * PCI Device registration data structures.
2876 #define CH_DEVICE(devid, idx) \
2877 { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
2879 static struct pci_device_id cxgb4vf_pci_tbl[] = {
2880 CH_DEVICE(0xb000, 0), /* PE10K FPGA */
2881 CH_DEVICE(0x4800, 0), /* T440-dbg */
2882 CH_DEVICE(0x4801, 0), /* T420-cr */
2883 CH_DEVICE(0x4802, 0), /* T422-cr */
2884 CH_DEVICE(0x4803, 0), /* T440-cr */
2885 CH_DEVICE(0x4804, 0), /* T420-bch */
2886 CH_DEVICE(0x4805, 0), /* T440-bch */
2887 CH_DEVICE(0x4806, 0), /* T460-ch */
2888 CH_DEVICE(0x4807, 0), /* T420-so */
2889 CH_DEVICE(0x4808, 0), /* T420-cx */
2890 CH_DEVICE(0x4809, 0), /* T420-bt */
2891 CH_DEVICE(0x480a, 0), /* T404-bt */
2895 MODULE_DESCRIPTION(DRV_DESC);
2896 MODULE_AUTHOR("Chelsio Communications");
2897 MODULE_LICENSE("Dual BSD/GPL");
2898 MODULE_VERSION(DRV_VERSION);
2899 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
2901 static struct pci_driver cxgb4vf_driver = {
2902 .name = KBUILD_MODNAME,
2903 .id_table = cxgb4vf_pci_tbl,
2904 .probe = cxgb4vf_pci_probe,
2905 .remove = __devexit_p(cxgb4vf_pci_remove),
2906 .shutdown = __devexit_p(cxgb4vf_pci_shutdown),
2910 * Initialize global driver state.
2912 static int __init cxgb4vf_module_init(void)
2917 * Vet our module parameters.
2919 if (msi != MSI_MSIX && msi != MSI_MSI) {
2920 printk(KERN_WARNING KBUILD_MODNAME
2921 ": bad module parameter msi=%d; must be %d"
2922 " (MSI-X or MSI) or %d (MSI)\n",
2923 msi, MSI_MSIX, MSI_MSI);
2927 /* Debugfs support is optional, just warn if this fails */
2928 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
2929 if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
2930 printk(KERN_WARNING KBUILD_MODNAME ": could not create"
2931 " debugfs entry, continuing\n");
2933 ret = pci_register_driver(&cxgb4vf_driver);
2934 if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root))
2935 debugfs_remove(cxgb4vf_debugfs_root);
2940 * Tear down global driver state.
2942 static void __exit cxgb4vf_module_exit(void)
2944 pci_unregister_driver(&cxgb4vf_driver);
2945 debugfs_remove(cxgb4vf_debugfs_root);
2948 module_init(cxgb4vf_module_init);
2949 module_exit(cxgb4vf_module_exit);