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, starting
818 * at a specified offset within the list, into an array of addrss pointers and
819 * return the number collected.
821 static inline unsigned int collect_netdev_uc_list_addrs(const struct net_device *dev,
824 unsigned int maxaddrs)
826 unsigned int index = 0;
827 unsigned int naddr = 0;
828 const struct netdev_hw_addr *ha;
830 for_each_dev_addr(dev, ha)
831 if (index++ >= offset) {
832 addr[naddr++] = ha->addr;
833 if (naddr >= maxaddrs)
840 * Collect up to maxaddrs worth of a netdevice's multicast addresses, starting
841 * at a specified offset within the list, into an array of addrss pointers and
842 * return the number collected.
844 static inline unsigned int collect_netdev_mc_list_addrs(const struct net_device *dev,
847 unsigned int maxaddrs)
849 unsigned int index = 0;
850 unsigned int naddr = 0;
851 const struct netdev_hw_addr *ha;
853 netdev_for_each_mc_addr(ha, dev)
854 if (index++ >= offset) {
855 addr[naddr++] = ha->addr;
856 if (naddr >= maxaddrs)
863 * Configure the exact and hash address filters to handle a port's multicast
864 * and secondary unicast MAC addresses.
866 static int set_addr_filters(const struct net_device *dev, bool sleep)
871 unsigned int offset, naddr;
874 const struct port_info *pi = netdev_priv(dev);
876 /* first do the secondary unicast addresses */
877 for (offset = 0; ; offset += naddr) {
878 naddr = collect_netdev_uc_list_addrs(dev, addr, offset,
883 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
884 naddr, addr, NULL, &uhash, sleep);
891 /* next set up the multicast addresses */
892 for (offset = 0; ; offset += naddr) {
893 naddr = collect_netdev_mc_list_addrs(dev, addr, offset,
898 ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free,
899 naddr, addr, NULL, &mhash, sleep);
905 return t4vf_set_addr_hash(pi->adapter, pi->viid, uhash != 0,
906 uhash | mhash, sleep);
910 * Set RX properties of a port, such as promiscruity, address filters, and MTU.
911 * If @mtu is -1 it is left unchanged.
913 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
916 struct port_info *pi = netdev_priv(dev);
918 ret = set_addr_filters(dev, sleep_ok);
920 ret = t4vf_set_rxmode(pi->adapter, pi->viid, -1,
921 (dev->flags & IFF_PROMISC) != 0,
922 (dev->flags & IFF_ALLMULTI) != 0,
928 * Set the current receive modes on the device.
930 static void cxgb4vf_set_rxmode(struct net_device *dev)
932 /* unfortunately we can't return errors to the stack */
933 set_rxmode(dev, -1, false);
937 * Find the entry in the interrupt holdoff timer value array which comes
938 * closest to the specified interrupt holdoff value.
940 static int closest_timer(const struct sge *s, int us)
942 int i, timer_idx = 0, min_delta = INT_MAX;
944 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
945 int delta = us - s->timer_val[i];
948 if (delta < min_delta) {
956 static int closest_thres(const struct sge *s, int thres)
958 int i, delta, pktcnt_idx = 0, min_delta = INT_MAX;
960 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
961 delta = thres - s->counter_val[i];
964 if (delta < min_delta) {
973 * Return a queue's interrupt hold-off time in us. 0 means no timer.
975 static unsigned int qtimer_val(const struct adapter *adapter,
976 const struct sge_rspq *rspq)
978 unsigned int timer_idx = QINTR_TIMER_IDX_GET(rspq->intr_params);
980 return timer_idx < SGE_NTIMERS
981 ? adapter->sge.timer_val[timer_idx]
986 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
987 * @adapter: the adapter
988 * @rspq: the RX response queue
989 * @us: the hold-off time in us, or 0 to disable timer
990 * @cnt: the hold-off packet count, or 0 to disable counter
992 * Sets an RX response queue's interrupt hold-off time and packet count.
993 * At least one of the two needs to be enabled for the queue to generate
996 static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq,
997 unsigned int us, unsigned int cnt)
999 unsigned int timer_idx;
1002 * If both the interrupt holdoff timer and count are specified as
1003 * zero, default to a holdoff count of 1 ...
1005 if ((us | cnt) == 0)
1009 * If an interrupt holdoff count has been specified, then find the
1010 * closest configured holdoff count and use that. If the response
1011 * queue has already been created, then update its queue context
1018 pktcnt_idx = closest_thres(&adapter->sge, cnt);
1019 if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) {
1020 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1022 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1023 FW_PARAMS_PARAM_YZ(rspq->cntxt_id);
1024 err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx);
1028 rspq->pktcnt_idx = pktcnt_idx;
1032 * Compute the closest holdoff timer index from the supplied holdoff
1035 timer_idx = (us == 0
1036 ? SGE_TIMER_RSTRT_CNTR
1037 : closest_timer(&adapter->sge, us));
1040 * Update the response queue's interrupt coalescing parameters and
1043 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
1044 (cnt > 0 ? QINTR_CNT_EN : 0));
1049 * Return a version number to identify the type of adapter. The scheme is:
1050 * - bits 0..9: chip version
1051 * - bits 10..15: chip revision
1053 static inline unsigned int mk_adap_vers(const struct adapter *adapter)
1056 * Chip version 4, revision 0x3f (cxgb4vf).
1058 return 4 | (0x3f << 10);
1062 * Execute the specified ioctl command.
1064 static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1070 * The VF Driver doesn't have access to any of the other
1071 * common Ethernet device ioctl()'s (like reading/writing
1072 * PHY registers, etc.
1083 * Change the device's MTU.
1085 static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu)
1088 struct port_info *pi = netdev_priv(dev);
1090 /* accommodate SACK */
1094 ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu,
1095 -1, -1, -1, -1, true);
1102 * Change the devices MAC address.
1104 static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr)
1107 struct sockaddr *addr = _addr;
1108 struct port_info *pi = netdev_priv(dev);
1110 if (!is_valid_ether_addr(addr->sa_data))
1113 ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt,
1114 addr->sa_data, true);
1118 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1119 pi->xact_addr_filt = ret;
1123 #ifdef CONFIG_NET_POLL_CONTROLLER
1125 * Poll all of our receive queues. This is called outside of normal interrupt
1128 static void cxgb4vf_poll_controller(struct net_device *dev)
1130 struct port_info *pi = netdev_priv(dev);
1131 struct adapter *adapter = pi->adapter;
1133 if (adapter->flags & USING_MSIX) {
1134 struct sge_eth_rxq *rxq;
1137 rxq = &adapter->sge.ethrxq[pi->first_qset];
1138 for (nqsets = pi->nqsets; nqsets; nqsets--) {
1139 t4vf_sge_intr_msix(0, &rxq->rspq);
1143 t4vf_intr_handler(adapter)(0, adapter);
1148 * Ethtool operations.
1149 * ===================
1151 * Note that we don't support any ethtool operations which change the physical
1152 * state of the port to which we're linked.
1156 * Return current port link settings.
1158 static int cxgb4vf_get_settings(struct net_device *dev,
1159 struct ethtool_cmd *cmd)
1161 const struct port_info *pi = netdev_priv(dev);
1163 cmd->supported = pi->link_cfg.supported;
1164 cmd->advertising = pi->link_cfg.advertising;
1165 cmd->speed = netif_carrier_ok(dev) ? pi->link_cfg.speed : -1;
1166 cmd->duplex = DUPLEX_FULL;
1168 cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1169 cmd->phy_address = pi->port_id;
1170 cmd->transceiver = XCVR_EXTERNAL;
1171 cmd->autoneg = pi->link_cfg.autoneg;
1178 * Return our driver information.
1180 static void cxgb4vf_get_drvinfo(struct net_device *dev,
1181 struct ethtool_drvinfo *drvinfo)
1183 struct adapter *adapter = netdev2adap(dev);
1185 strcpy(drvinfo->driver, KBUILD_MODNAME);
1186 strcpy(drvinfo->version, DRV_VERSION);
1187 strcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)));
1188 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
1189 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1190 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.fwrev),
1191 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.fwrev),
1192 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.fwrev),
1193 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.fwrev),
1194 FW_HDR_FW_VER_MAJOR_GET(adapter->params.dev.tprev),
1195 FW_HDR_FW_VER_MINOR_GET(adapter->params.dev.tprev),
1196 FW_HDR_FW_VER_MICRO_GET(adapter->params.dev.tprev),
1197 FW_HDR_FW_VER_BUILD_GET(adapter->params.dev.tprev));
1201 * Return current adapter message level.
1203 static u32 cxgb4vf_get_msglevel(struct net_device *dev)
1205 return netdev2adap(dev)->msg_enable;
1209 * Set current adapter message level.
1211 static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel)
1213 netdev2adap(dev)->msg_enable = msglevel;
1217 * Return the device's current Queue Set ring size parameters along with the
1218 * allowed maximum values. Since ethtool doesn't understand the concept of
1219 * multi-queue devices, we just return the current values associated with the
1222 static void cxgb4vf_get_ringparam(struct net_device *dev,
1223 struct ethtool_ringparam *rp)
1225 const struct port_info *pi = netdev_priv(dev);
1226 const struct sge *s = &pi->adapter->sge;
1228 rp->rx_max_pending = MAX_RX_BUFFERS;
1229 rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1230 rp->rx_jumbo_max_pending = 0;
1231 rp->tx_max_pending = MAX_TXQ_ENTRIES;
1233 rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID;
1234 rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1235 rp->rx_jumbo_pending = 0;
1236 rp->tx_pending = s->ethtxq[pi->first_qset].q.size;
1240 * Set the Queue Set ring size parameters for the device. Again, since
1241 * ethtool doesn't allow for the concept of multiple queues per device, we'll
1242 * apply these new values across all of the Queue Sets associated with the
1243 * device -- after vetting them of course!
1245 static int cxgb4vf_set_ringparam(struct net_device *dev,
1246 struct ethtool_ringparam *rp)
1248 const struct port_info *pi = netdev_priv(dev);
1249 struct adapter *adapter = pi->adapter;
1250 struct sge *s = &adapter->sge;
1253 if (rp->rx_pending > MAX_RX_BUFFERS ||
1254 rp->rx_jumbo_pending ||
1255 rp->tx_pending > MAX_TXQ_ENTRIES ||
1256 rp->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1257 rp->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1258 rp->rx_pending < MIN_FL_ENTRIES ||
1259 rp->tx_pending < MIN_TXQ_ENTRIES)
1262 if (adapter->flags & FULL_INIT_DONE)
1265 for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) {
1266 s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID;
1267 s->ethrxq[qs].rspq.size = rp->rx_mini_pending;
1268 s->ethtxq[qs].q.size = rp->tx_pending;
1274 * Return the interrupt holdoff timer and count for the first Queue Set on the
1275 * device. Our extension ioctl() (the cxgbtool interface) allows the
1276 * interrupt holdoff timer to be read on all of the device's Queue Sets.
1278 static int cxgb4vf_get_coalesce(struct net_device *dev,
1279 struct ethtool_coalesce *coalesce)
1281 const struct port_info *pi = netdev_priv(dev);
1282 const struct adapter *adapter = pi->adapter;
1283 const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq;
1285 coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq);
1286 coalesce->rx_max_coalesced_frames =
1287 ((rspq->intr_params & QINTR_CNT_EN)
1288 ? adapter->sge.counter_val[rspq->pktcnt_idx]
1294 * Set the RX interrupt holdoff timer and count for the first Queue Set on the
1295 * interface. Our extension ioctl() (the cxgbtool interface) allows us to set
1296 * the interrupt holdoff timer on any of the device's Queue Sets.
1298 static int cxgb4vf_set_coalesce(struct net_device *dev,
1299 struct ethtool_coalesce *coalesce)
1301 const struct port_info *pi = netdev_priv(dev);
1302 struct adapter *adapter = pi->adapter;
1304 return set_rxq_intr_params(adapter,
1305 &adapter->sge.ethrxq[pi->first_qset].rspq,
1306 coalesce->rx_coalesce_usecs,
1307 coalesce->rx_max_coalesced_frames);
1311 * Report current port link pause parameter settings.
1313 static void cxgb4vf_get_pauseparam(struct net_device *dev,
1314 struct ethtool_pauseparam *pauseparam)
1316 struct port_info *pi = netdev_priv(dev);
1318 pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1319 pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0;
1320 pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0;
1324 * Return whether RX Checksum Offloading is currently enabled for the device.
1326 static u32 cxgb4vf_get_rx_csum(struct net_device *dev)
1328 struct port_info *pi = netdev_priv(dev);
1330 return (pi->rx_offload & RX_CSO) != 0;
1334 * Turn RX Checksum Offloading on or off for the device.
1336 static int cxgb4vf_set_rx_csum(struct net_device *dev, u32 csum)
1338 struct port_info *pi = netdev_priv(dev);
1341 pi->rx_offload |= RX_CSO;
1343 pi->rx_offload &= ~RX_CSO;
1348 * Identify the port by blinking the port's LED.
1350 static int cxgb4vf_phys_id(struct net_device *dev, u32 id)
1352 struct port_info *pi = netdev_priv(dev);
1354 return t4vf_identify_port(pi->adapter, pi->viid, 5);
1358 * Port stats maintained per queue of the port.
1360 struct queue_port_stats {
1371 * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that
1372 * these need to match the order of statistics returned by
1373 * t4vf_get_port_stats().
1375 static const char stats_strings[][ETH_GSTRING_LEN] = {
1377 * These must match the layout of the t4vf_port_stats structure.
1379 "TxBroadcastBytes ",
1380 "TxBroadcastFrames ",
1381 "TxMulticastBytes ",
1382 "TxMulticastFrames ",
1388 "RxBroadcastBytes ",
1389 "RxBroadcastFrames ",
1390 "RxMulticastBytes ",
1391 "RxMulticastFrames ",
1397 * These are accumulated per-queue statistics and must match the
1398 * order of the fields in the queue_port_stats structure.
1410 * Return the number of statistics in the specified statistics set.
1412 static int cxgb4vf_get_sset_count(struct net_device *dev, int sset)
1416 return ARRAY_SIZE(stats_strings);
1424 * Return the strings for the specified statistics set.
1426 static void cxgb4vf_get_strings(struct net_device *dev,
1432 memcpy(data, stats_strings, sizeof(stats_strings));
1438 * Small utility routine to accumulate queue statistics across the queues of
1441 static void collect_sge_port_stats(const struct adapter *adapter,
1442 const struct port_info *pi,
1443 struct queue_port_stats *stats)
1445 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset];
1446 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset];
1449 memset(stats, 0, sizeof(*stats));
1450 for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) {
1451 stats->tso += txq->tso;
1452 stats->tx_csum += txq->tx_cso;
1453 stats->rx_csum += rxq->stats.rx_cso;
1454 stats->vlan_ex += rxq->stats.vlan_ex;
1455 stats->vlan_ins += txq->vlan_ins;
1456 stats->lro_pkts += rxq->stats.lro_pkts;
1457 stats->lro_merged += rxq->stats.lro_merged;
1462 * Return the ETH_SS_STATS statistics set.
1464 static void cxgb4vf_get_ethtool_stats(struct net_device *dev,
1465 struct ethtool_stats *stats,
1468 struct port_info *pi = netdev2pinfo(dev);
1469 struct adapter *adapter = pi->adapter;
1470 int err = t4vf_get_port_stats(adapter, pi->pidx,
1471 (struct t4vf_port_stats *)data);
1473 memset(data, 0, sizeof(struct t4vf_port_stats));
1475 data += sizeof(struct t4vf_port_stats) / sizeof(u64);
1476 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1480 * Return the size of our register map.
1482 static int cxgb4vf_get_regs_len(struct net_device *dev)
1484 return T4VF_REGMAP_SIZE;
1488 * Dump a block of registers, start to end inclusive, into a buffer.
1490 static void reg_block_dump(struct adapter *adapter, void *regbuf,
1491 unsigned int start, unsigned int end)
1493 u32 *bp = regbuf + start - T4VF_REGMAP_START;
1495 for ( ; start <= end; start += sizeof(u32)) {
1497 * Avoid reading the Mailbox Control register since that
1498 * can trigger a Mailbox Ownership Arbitration cycle and
1499 * interfere with communication with the firmware.
1501 if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL)
1504 *bp++ = t4_read_reg(adapter, start);
1509 * Copy our entire register map into the provided buffer.
1511 static void cxgb4vf_get_regs(struct net_device *dev,
1512 struct ethtool_regs *regs,
1515 struct adapter *adapter = netdev2adap(dev);
1517 regs->version = mk_adap_vers(adapter);
1520 * Fill in register buffer with our register map.
1522 memset(regbuf, 0, T4VF_REGMAP_SIZE);
1524 reg_block_dump(adapter, regbuf,
1525 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST,
1526 T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST);
1527 reg_block_dump(adapter, regbuf,
1528 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST,
1529 T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST);
1530 reg_block_dump(adapter, regbuf,
1531 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST,
1532 T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_LAST);
1533 reg_block_dump(adapter, regbuf,
1534 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST,
1535 T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST);
1537 reg_block_dump(adapter, regbuf,
1538 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST,
1539 T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST);
1543 * Report current Wake On LAN settings.
1545 static void cxgb4vf_get_wol(struct net_device *dev,
1546 struct ethtool_wolinfo *wol)
1550 memset(&wol->sopass, 0, sizeof(wol->sopass));
1554 * TCP Segmentation Offload flags which we support.
1556 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1559 * Set TCP Segmentation Offloading feature capabilities.
1561 static int cxgb4vf_set_tso(struct net_device *dev, u32 tso)
1564 dev->features |= TSO_FLAGS;
1566 dev->features &= ~TSO_FLAGS;
1570 static struct ethtool_ops cxgb4vf_ethtool_ops = {
1571 .get_settings = cxgb4vf_get_settings,
1572 .get_drvinfo = cxgb4vf_get_drvinfo,
1573 .get_msglevel = cxgb4vf_get_msglevel,
1574 .set_msglevel = cxgb4vf_set_msglevel,
1575 .get_ringparam = cxgb4vf_get_ringparam,
1576 .set_ringparam = cxgb4vf_set_ringparam,
1577 .get_coalesce = cxgb4vf_get_coalesce,
1578 .set_coalesce = cxgb4vf_set_coalesce,
1579 .get_pauseparam = cxgb4vf_get_pauseparam,
1580 .get_rx_csum = cxgb4vf_get_rx_csum,
1581 .set_rx_csum = cxgb4vf_set_rx_csum,
1582 .set_tx_csum = ethtool_op_set_tx_ipv6_csum,
1583 .set_sg = ethtool_op_set_sg,
1584 .get_link = ethtool_op_get_link,
1585 .get_strings = cxgb4vf_get_strings,
1586 .phys_id = cxgb4vf_phys_id,
1587 .get_sset_count = cxgb4vf_get_sset_count,
1588 .get_ethtool_stats = cxgb4vf_get_ethtool_stats,
1589 .get_regs_len = cxgb4vf_get_regs_len,
1590 .get_regs = cxgb4vf_get_regs,
1591 .get_wol = cxgb4vf_get_wol,
1592 .set_tso = cxgb4vf_set_tso,
1596 * /sys/kernel/debug/cxgb4vf support code and data.
1597 * ================================================
1601 * Show SGE Queue Set information. We display QPL Queues Sets per line.
1605 static int sge_qinfo_show(struct seq_file *seq, void *v)
1607 struct adapter *adapter = seq->private;
1608 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1609 int qs, r = (uintptr_t)v - 1;
1612 seq_putc(seq, '\n');
1614 #define S3(fmt_spec, s, v) \
1616 seq_printf(seq, "%-12s", s); \
1617 for (qs = 0; qs < n; ++qs) \
1618 seq_printf(seq, " %16" fmt_spec, v); \
1619 seq_putc(seq, '\n'); \
1621 #define S(s, v) S3("s", s, v)
1622 #define T(s, v) S3("u", s, txq[qs].v)
1623 #define R(s, v) S3("u", s, rxq[qs].v)
1625 if (r < eth_entries) {
1626 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1627 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1628 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1630 S("QType:", "Ethernet");
1632 (rxq[qs].rspq.netdev
1633 ? rxq[qs].rspq.netdev->name
1636 (rxq[qs].rspq.netdev
1637 ? ((struct port_info *)
1638 netdev_priv(rxq[qs].rspq.netdev))->port_id
1640 T("TxQ ID:", q.abs_id);
1641 T("TxQ size:", q.size);
1642 T("TxQ inuse:", q.in_use);
1643 T("TxQ PIdx:", q.pidx);
1644 T("TxQ CIdx:", q.cidx);
1645 R("RspQ ID:", rspq.abs_id);
1646 R("RspQ size:", rspq.size);
1647 R("RspQE size:", rspq.iqe_len);
1648 S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq));
1649 S3("u", "Intr pktcnt:",
1650 adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]);
1651 R("RspQ CIdx:", rspq.cidx);
1652 R("RspQ Gen:", rspq.gen);
1653 R("FL ID:", fl.abs_id);
1654 R("FL size:", fl.size - MIN_FL_RESID);
1655 R("FL avail:", fl.avail);
1656 R("FL PIdx:", fl.pidx);
1657 R("FL CIdx:", fl.cidx);
1663 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1665 seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue");
1666 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id);
1667 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1668 qtimer_val(adapter, evtq));
1669 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1670 adapter->sge.counter_val[evtq->pktcnt_idx]);
1671 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx);
1672 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen);
1673 } else if (r == 1) {
1674 const struct sge_rspq *intrq = &adapter->sge.intrq;
1676 seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue");
1677 seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id);
1678 seq_printf(seq, "%-12s %16u\n", "Intr delay:",
1679 qtimer_val(adapter, intrq));
1680 seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:",
1681 adapter->sge.counter_val[intrq->pktcnt_idx]);
1682 seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx);
1683 seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen);
1695 * Return the number of "entries" in our "file". We group the multi-Queue
1696 * sections with QPL Queue Sets per "entry". The sections of the output are:
1698 * Ethernet RX/TX Queue Sets
1699 * Firmware Event Queue
1700 * Forwarded Interrupt Queue (if in MSI mode)
1702 static int sge_queue_entries(const struct adapter *adapter)
1704 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1705 ((adapter->flags & USING_MSI) != 0);
1708 static void *sge_queue_start(struct seq_file *seq, loff_t *pos)
1710 int entries = sge_queue_entries(seq->private);
1712 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1715 static void sge_queue_stop(struct seq_file *seq, void *v)
1719 static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos)
1721 int entries = sge_queue_entries(seq->private);
1724 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1727 static const struct seq_operations sge_qinfo_seq_ops = {
1728 .start = sge_queue_start,
1729 .next = sge_queue_next,
1730 .stop = sge_queue_stop,
1731 .show = sge_qinfo_show
1734 static int sge_qinfo_open(struct inode *inode, struct file *file)
1736 int res = seq_open(file, &sge_qinfo_seq_ops);
1739 struct seq_file *seq = file->private_data;
1740 seq->private = inode->i_private;
1745 static const struct file_operations sge_qinfo_debugfs_fops = {
1746 .owner = THIS_MODULE,
1747 .open = sge_qinfo_open,
1749 .llseek = seq_lseek,
1750 .release = seq_release,
1754 * Show SGE Queue Set statistics. We display QPL Queues Sets per line.
1758 static int sge_qstats_show(struct seq_file *seq, void *v)
1760 struct adapter *adapter = seq->private;
1761 int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL);
1762 int qs, r = (uintptr_t)v - 1;
1765 seq_putc(seq, '\n');
1767 #define S3(fmt, s, v) \
1769 seq_printf(seq, "%-16s", s); \
1770 for (qs = 0; qs < n; ++qs) \
1771 seq_printf(seq, " %8" fmt, v); \
1772 seq_putc(seq, '\n'); \
1774 #define S(s, v) S3("s", s, v)
1776 #define T3(fmt, s, v) S3(fmt, s, txq[qs].v)
1777 #define T(s, v) T3("lu", s, v)
1779 #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v)
1780 #define R(s, v) R3("lu", s, v)
1782 if (r < eth_entries) {
1783 const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL];
1784 const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL];
1785 int n = min(QPL, adapter->sge.ethqsets - QPL * r);
1787 S("QType:", "Ethernet");
1789 (rxq[qs].rspq.netdev
1790 ? rxq[qs].rspq.netdev->name
1792 R3("u", "RspQNullInts:", rspq.unhandled_irqs);
1793 R("RxPackets:", stats.pkts);
1794 R("RxCSO:", stats.rx_cso);
1795 R("VLANxtract:", stats.vlan_ex);
1796 R("LROmerged:", stats.lro_merged);
1797 R("LROpackets:", stats.lro_pkts);
1798 R("RxDrops:", stats.rx_drops);
1800 T("TxCSO:", tx_cso);
1801 T("VLANins:", vlan_ins);
1802 T("TxQFull:", q.stops);
1803 T("TxQRestarts:", q.restarts);
1804 T("TxMapErr:", mapping_err);
1805 R("FLAllocErr:", fl.alloc_failed);
1806 R("FLLrgAlcErr:", fl.large_alloc_failed);
1807 R("FLStarving:", fl.starving);
1813 const struct sge_rspq *evtq = &adapter->sge.fw_evtq;
1815 seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue");
1816 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1817 evtq->unhandled_irqs);
1818 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx);
1819 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen);
1820 } else if (r == 1) {
1821 const struct sge_rspq *intrq = &adapter->sge.intrq;
1823 seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue");
1824 seq_printf(seq, "%-16s %8u\n", "RspQNullInts:",
1825 intrq->unhandled_irqs);
1826 seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx);
1827 seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen);
1841 * Return the number of "entries" in our "file". We group the multi-Queue
1842 * sections with QPL Queue Sets per "entry". The sections of the output are:
1844 * Ethernet RX/TX Queue Sets
1845 * Firmware Event Queue
1846 * Forwarded Interrupt Queue (if in MSI mode)
1848 static int sge_qstats_entries(const struct adapter *adapter)
1850 return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 +
1851 ((adapter->flags & USING_MSI) != 0);
1854 static void *sge_qstats_start(struct seq_file *seq, loff_t *pos)
1856 int entries = sge_qstats_entries(seq->private);
1858 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1861 static void sge_qstats_stop(struct seq_file *seq, void *v)
1865 static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos)
1867 int entries = sge_qstats_entries(seq->private);
1870 return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL;
1873 static const struct seq_operations sge_qstats_seq_ops = {
1874 .start = sge_qstats_start,
1875 .next = sge_qstats_next,
1876 .stop = sge_qstats_stop,
1877 .show = sge_qstats_show
1880 static int sge_qstats_open(struct inode *inode, struct file *file)
1882 int res = seq_open(file, &sge_qstats_seq_ops);
1885 struct seq_file *seq = file->private_data;
1886 seq->private = inode->i_private;
1891 static const struct file_operations sge_qstats_proc_fops = {
1892 .owner = THIS_MODULE,
1893 .open = sge_qstats_open,
1895 .llseek = seq_lseek,
1896 .release = seq_release,
1900 * Show PCI-E SR-IOV Virtual Function Resource Limits.
1902 static int resources_show(struct seq_file *seq, void *v)
1904 struct adapter *adapter = seq->private;
1905 struct vf_resources *vfres = &adapter->params.vfres;
1907 #define S(desc, fmt, var) \
1908 seq_printf(seq, "%-60s " fmt "\n", \
1909 desc " (" #var "):", vfres->var)
1911 S("Virtual Interfaces", "%d", nvi);
1912 S("Egress Queues", "%d", neq);
1913 S("Ethernet Control", "%d", nethctrl);
1914 S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint);
1915 S("Ingress Queues", "%d", niq);
1916 S("Traffic Class", "%d", tc);
1917 S("Port Access Rights Mask", "%#x", pmask);
1918 S("MAC Address Filters", "%d", nexactf);
1919 S("Firmware Command Read Capabilities", "%#x", r_caps);
1920 S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps);
1927 static int resources_open(struct inode *inode, struct file *file)
1929 return single_open(file, resources_show, inode->i_private);
1932 static const struct file_operations resources_proc_fops = {
1933 .owner = THIS_MODULE,
1934 .open = resources_open,
1936 .llseek = seq_lseek,
1937 .release = single_release,
1941 * Show Virtual Interfaces.
1943 static int interfaces_show(struct seq_file *seq, void *v)
1945 if (v == SEQ_START_TOKEN) {
1946 seq_puts(seq, "Interface Port VIID\n");
1948 struct adapter *adapter = seq->private;
1949 int pidx = (uintptr_t)v - 2;
1950 struct net_device *dev = adapter->port[pidx];
1951 struct port_info *pi = netdev_priv(dev);
1953 seq_printf(seq, "%9s %4d %#5x\n",
1954 dev->name, pi->port_id, pi->viid);
1959 static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos)
1961 return pos <= adapter->params.nports
1962 ? (void *)(uintptr_t)(pos + 1)
1966 static void *interfaces_start(struct seq_file *seq, loff_t *pos)
1969 ? interfaces_get_idx(seq->private, *pos)
1973 static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos)
1976 return interfaces_get_idx(seq->private, *pos);
1979 static void interfaces_stop(struct seq_file *seq, void *v)
1983 static const struct seq_operations interfaces_seq_ops = {
1984 .start = interfaces_start,
1985 .next = interfaces_next,
1986 .stop = interfaces_stop,
1987 .show = interfaces_show
1990 static int interfaces_open(struct inode *inode, struct file *file)
1992 int res = seq_open(file, &interfaces_seq_ops);
1995 struct seq_file *seq = file->private_data;
1996 seq->private = inode->i_private;
2001 static const struct file_operations interfaces_proc_fops = {
2002 .owner = THIS_MODULE,
2003 .open = interfaces_open,
2005 .llseek = seq_lseek,
2006 .release = seq_release,
2010 * /sys/kernel/debugfs/cxgb4vf/ files list.
2012 struct cxgb4vf_debugfs_entry {
2013 const char *name; /* name of debugfs node */
2014 mode_t mode; /* file system mode */
2015 const struct file_operations *fops;
2018 static struct cxgb4vf_debugfs_entry debugfs_files[] = {
2019 { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops },
2020 { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops },
2021 { "resources", S_IRUGO, &resources_proc_fops },
2022 { "interfaces", S_IRUGO, &interfaces_proc_fops },
2026 * Module and device initialization and cleanup code.
2027 * ==================================================
2031 * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the
2032 * directory (debugfs_root) has already been set up.
2034 static int __devinit setup_debugfs(struct adapter *adapter)
2038 BUG_ON(adapter->debugfs_root == NULL);
2041 * Debugfs support is best effort.
2043 for (i = 0; i < ARRAY_SIZE(debugfs_files); i++)
2044 (void)debugfs_create_file(debugfs_files[i].name,
2045 debugfs_files[i].mode,
2046 adapter->debugfs_root,
2048 debugfs_files[i].fops);
2054 * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave
2055 * it to our caller to tear down the directory (debugfs_root).
2057 static void cleanup_debugfs(struct adapter *adapter)
2059 BUG_ON(adapter->debugfs_root == NULL);
2062 * Unlike our sister routine cleanup_proc(), we don't need to remove
2063 * individual entries because a call will be made to
2064 * debugfs_remove_recursive(). We just need to clean up any ancillary
2071 * Perform early "adapter" initialization. This is where we discover what
2072 * adapter parameters we're going to be using and initialize basic adapter
2075 static int __devinit adap_init0(struct adapter *adapter)
2077 struct vf_resources *vfres = &adapter->params.vfres;
2078 struct sge_params *sge_params = &adapter->params.sge;
2079 struct sge *s = &adapter->sge;
2080 unsigned int ethqsets;
2084 * Wait for the device to become ready before proceeding ...
2086 err = t4vf_wait_dev_ready(adapter);
2088 dev_err(adapter->pdev_dev, "device didn't become ready:"
2094 * Some environments do not properly handle PCIE FLRs -- e.g. in Linux
2095 * 2.6.31 and later we can't call pci_reset_function() in order to
2096 * issue an FLR because of a self- deadlock on the device semaphore.
2097 * Meanwhile, the OS infrastructure doesn't issue FLRs in all the
2098 * cases where they're needed -- for instance, some versions of KVM
2099 * fail to reset "Assigned Devices" when the VM reboots. Therefore we
2100 * use the firmware based reset in order to reset any per function
2103 err = t4vf_fw_reset(adapter);
2105 dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err);
2110 * Grab basic operational parameters. These will predominantly have
2111 * been set up by the Physical Function Driver or will be hard coded
2112 * into the adapter. We just have to live with them ... Note that
2113 * we _must_ get our VPD parameters before our SGE parameters because
2114 * we need to know the adapter's core clock from the VPD in order to
2115 * properly decode the SGE Timer Values.
2117 err = t4vf_get_dev_params(adapter);
2119 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2120 " device parameters: err=%d\n", err);
2123 err = t4vf_get_vpd_params(adapter);
2125 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2126 " VPD parameters: err=%d\n", err);
2129 err = t4vf_get_sge_params(adapter);
2131 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2132 " SGE parameters: err=%d\n", err);
2135 err = t4vf_get_rss_glb_config(adapter);
2137 dev_err(adapter->pdev_dev, "unable to retrieve adapter"
2138 " RSS parameters: err=%d\n", err);
2141 if (adapter->params.rss.mode !=
2142 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) {
2143 dev_err(adapter->pdev_dev, "unable to operate with global RSS"
2144 " mode %d\n", adapter->params.rss.mode);
2147 err = t4vf_sge_init(adapter);
2149 dev_err(adapter->pdev_dev, "unable to use adapter parameters:"
2155 * Retrieve our RX interrupt holdoff timer values and counter
2156 * threshold values from the SGE parameters.
2158 s->timer_val[0] = core_ticks_to_us(adapter,
2159 TIMERVALUE0_GET(sge_params->sge_timer_value_0_and_1));
2160 s->timer_val[1] = core_ticks_to_us(adapter,
2161 TIMERVALUE1_GET(sge_params->sge_timer_value_0_and_1));
2162 s->timer_val[2] = core_ticks_to_us(adapter,
2163 TIMERVALUE0_GET(sge_params->sge_timer_value_2_and_3));
2164 s->timer_val[3] = core_ticks_to_us(adapter,
2165 TIMERVALUE1_GET(sge_params->sge_timer_value_2_and_3));
2166 s->timer_val[4] = core_ticks_to_us(adapter,
2167 TIMERVALUE0_GET(sge_params->sge_timer_value_4_and_5));
2168 s->timer_val[5] = core_ticks_to_us(adapter,
2169 TIMERVALUE1_GET(sge_params->sge_timer_value_4_and_5));
2172 THRESHOLD_0_GET(sge_params->sge_ingress_rx_threshold);
2174 THRESHOLD_1_GET(sge_params->sge_ingress_rx_threshold);
2176 THRESHOLD_2_GET(sge_params->sge_ingress_rx_threshold);
2178 THRESHOLD_3_GET(sge_params->sge_ingress_rx_threshold);
2181 * Grab our Virtual Interface resource allocation, extract the
2182 * features that we're interested in and do a bit of sanity testing on
2185 err = t4vf_get_vfres(adapter);
2187 dev_err(adapter->pdev_dev, "unable to get virtual interface"
2188 " resources: err=%d\n", err);
2193 * The number of "ports" which we support is equal to the number of
2194 * Virtual Interfaces with which we've been provisioned.
2196 adapter->params.nports = vfres->nvi;
2197 if (adapter->params.nports > MAX_NPORTS) {
2198 dev_warn(adapter->pdev_dev, "only using %d of %d allowed"
2199 " virtual interfaces\n", MAX_NPORTS,
2200 adapter->params.nports);
2201 adapter->params.nports = MAX_NPORTS;
2205 * We need to reserve a number of the ingress queues with Free List
2206 * and Interrupt capabilities for special interrupt purposes (like
2207 * asynchronous firmware messages, or forwarded interrupts if we're
2208 * using MSI). The rest of the FL/Intr-capable ingress queues will be
2209 * matched up one-for-one with Ethernet/Control egress queues in order
2210 * to form "Queue Sets" which will be aportioned between the "ports".
2211 * For each Queue Set, we'll need the ability to allocate two Egress
2212 * Contexts -- one for the Ingress Queue Free List and one for the TX
2215 ethqsets = vfres->niqflint - INGQ_EXTRAS;
2216 if (vfres->nethctrl != ethqsets) {
2217 dev_warn(adapter->pdev_dev, "unequal number of [available]"
2218 " ingress/egress queues (%d/%d); using minimum for"
2219 " number of Queue Sets\n", ethqsets, vfres->nethctrl);
2220 ethqsets = min(vfres->nethctrl, ethqsets);
2222 if (vfres->neq < ethqsets*2) {
2223 dev_warn(adapter->pdev_dev, "Not enough Egress Contexts (%d)"
2224 " to support Queue Sets (%d); reducing allowed Queue"
2225 " Sets\n", vfres->neq, ethqsets);
2226 ethqsets = vfres->neq/2;
2228 if (ethqsets > MAX_ETH_QSETS) {
2229 dev_warn(adapter->pdev_dev, "only using %d of %d allowed Queue"
2230 " Sets\n", MAX_ETH_QSETS, adapter->sge.max_ethqsets);
2231 ethqsets = MAX_ETH_QSETS;
2233 if (vfres->niq != 0 || vfres->neq > ethqsets*2) {
2234 dev_warn(adapter->pdev_dev, "unused resources niq/neq (%d/%d)"
2235 " ignored\n", vfres->niq, vfres->neq - ethqsets*2);
2237 adapter->sge.max_ethqsets = ethqsets;
2240 * Check for various parameter sanity issues. Most checks simply
2241 * result in us using fewer resources than our provissioning but we
2242 * do need at least one "port" with which to work ...
2244 if (adapter->sge.max_ethqsets < adapter->params.nports) {
2245 dev_warn(adapter->pdev_dev, "only using %d of %d available"
2246 " virtual interfaces (too few Queue Sets)\n",
2247 adapter->sge.max_ethqsets, adapter->params.nports);
2248 adapter->params.nports = adapter->sge.max_ethqsets;
2250 if (adapter->params.nports == 0) {
2251 dev_err(adapter->pdev_dev, "no virtual interfaces configured/"
2258 static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx,
2259 u8 pkt_cnt_idx, unsigned int size,
2260 unsigned int iqe_size)
2262 rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) |
2263 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0));
2264 rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS
2267 rspq->iqe_len = iqe_size;
2272 * Perform default configuration of DMA queues depending on the number and
2273 * type of ports we found and the number of available CPUs. Most settings can
2274 * be modified by the admin via ethtool and cxgbtool prior to the adapter
2275 * being brought up for the first time.
2277 static void __devinit cfg_queues(struct adapter *adapter)
2279 struct sge *s = &adapter->sge;
2280 int q10g, n10g, qidx, pidx, qs;
2284 * We should not be called till we know how many Queue Sets we can
2285 * support. In particular, this means that we need to know what kind
2286 * of interrupts we'll be using ...
2288 BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
2291 * Count the number of 10GbE Virtual Interfaces that we have.
2294 for_each_port(adapter, pidx)
2295 n10g += is_10g_port(&adap2pinfo(adapter, pidx)->link_cfg);
2298 * We default to 1 queue per non-10G port and up to # of cores queues
2304 int n1g = (adapter->params.nports - n10g);
2305 q10g = (adapter->sge.max_ethqsets - n1g) / n10g;
2306 if (q10g > num_online_cpus())
2307 q10g = num_online_cpus();
2311 * Allocate the "Queue Sets" to the various Virtual Interfaces.
2312 * The layout will be established in setup_sge_queues() when the
2313 * adapter is brough up for the first time.
2316 for_each_port(adapter, pidx) {
2317 struct port_info *pi = adap2pinfo(adapter, pidx);
2319 pi->first_qset = qidx;
2320 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
2326 * The Ingress Queue Entry Size for our various Response Queues needs
2327 * to be big enough to accommodate the largest message we can receive
2328 * from the chip/firmware; which is 64 bytes ...
2333 * Set up default Queue Set parameters ... Start off with the
2334 * shortest interrupt holdoff timer.
2336 for (qs = 0; qs < s->max_ethqsets; qs++) {
2337 struct sge_eth_rxq *rxq = &s->ethrxq[qs];
2338 struct sge_eth_txq *txq = &s->ethtxq[qs];
2340 init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size);
2346 * The firmware event queue is used for link state changes and
2347 * notifications of TX DMA completions.
2349 init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size);
2352 * The forwarded interrupt queue is used when we're in MSI interrupt
2353 * mode. In this mode all interrupts associated with RX queues will
2354 * be forwarded to a single queue which we'll associate with our MSI
2355 * interrupt vector. The messages dropped in the forwarded interrupt
2356 * queue will indicate which ingress queue needs servicing ... This
2357 * queue needs to be large enough to accommodate all of the ingress
2358 * queues which are forwarding their interrupt (+1 to prevent the PIDX
2359 * from equalling the CIDX if every ingress queue has an outstanding
2360 * interrupt). The queue doesn't need to be any larger because no
2361 * ingress queue will ever have more than one outstanding interrupt at
2364 init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1,
2369 * Reduce the number of Ethernet queues across all ports to at most n.
2370 * n provides at least one queue per port.
2372 static void __devinit reduce_ethqs(struct adapter *adapter, int n)
2375 struct port_info *pi;
2378 * While we have too many active Ether Queue Sets, interate across the
2379 * "ports" and reduce their individual Queue Set allocations.
2381 BUG_ON(n < adapter->params.nports);
2382 while (n < adapter->sge.ethqsets)
2383 for_each_port(adapter, i) {
2384 pi = adap2pinfo(adapter, i);
2385 if (pi->nqsets > 1) {
2387 adapter->sge.ethqsets--;
2388 if (adapter->sge.ethqsets <= n)
2394 * Reassign the starting Queue Sets for each of the "ports" ...
2397 for_each_port(adapter, i) {
2398 pi = adap2pinfo(adapter, i);
2405 * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally
2406 * we get a separate MSI-X vector for every "Queue Set" plus any extras we
2407 * need. Minimally we need one for every Virtual Interface plus those needed
2408 * for our "extras". Note that this process may lower the maximum number of
2409 * allowed Queue Sets ...
2411 static int __devinit enable_msix(struct adapter *adapter)
2413 int i, err, want, need;
2414 struct msix_entry entries[MSIX_ENTRIES];
2415 struct sge *s = &adapter->sge;
2417 for (i = 0; i < MSIX_ENTRIES; ++i)
2418 entries[i].entry = i;
2421 * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets"
2422 * plus those needed for our "extras" (for example, the firmware
2423 * message queue). We _need_ at least one "Queue Set" per Virtual
2424 * Interface plus those needed for our "extras". So now we get to see
2425 * if the song is right ...
2427 want = s->max_ethqsets + MSIX_EXTRAS;
2428 need = adapter->params.nports + MSIX_EXTRAS;
2429 while ((err = pci_enable_msix(adapter->pdev, entries, want)) >= need)
2433 int nqsets = want - MSIX_EXTRAS;
2434 if (nqsets < s->max_ethqsets) {
2435 dev_warn(adapter->pdev_dev, "only enough MSI-X vectors"
2436 " for %d Queue Sets\n", nqsets);
2437 s->max_ethqsets = nqsets;
2438 if (nqsets < s->ethqsets)
2439 reduce_ethqs(adapter, nqsets);
2441 for (i = 0; i < want; ++i)
2442 adapter->msix_info[i].vec = entries[i].vector;
2443 } else if (err > 0) {
2444 pci_disable_msix(adapter->pdev);
2445 dev_info(adapter->pdev_dev, "only %d MSI-X vectors left,"
2446 " not using MSI-X\n", err);
2451 #ifdef HAVE_NET_DEVICE_OPS
2452 static const struct net_device_ops cxgb4vf_netdev_ops = {
2453 .ndo_open = cxgb4vf_open,
2454 .ndo_stop = cxgb4vf_stop,
2455 .ndo_start_xmit = t4vf_eth_xmit,
2456 .ndo_get_stats = cxgb4vf_get_stats,
2457 .ndo_set_rx_mode = cxgb4vf_set_rxmode,
2458 .ndo_set_mac_address = cxgb4vf_set_mac_addr,
2459 .ndo_validate_addr = eth_validate_addr,
2460 .ndo_do_ioctl = cxgb4vf_do_ioctl,
2461 .ndo_change_mtu = cxgb4vf_change_mtu,
2462 .ndo_vlan_rx_register = cxgb4vf_vlan_rx_register,
2463 #ifdef CONFIG_NET_POLL_CONTROLLER
2464 .ndo_poll_controller = cxgb4vf_poll_controller,
2470 * "Probe" a device: initialize a device and construct all kernel and driver
2471 * state needed to manage the device. This routine is called "init_one" in
2474 static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
2475 const struct pci_device_id *ent)
2477 static int version_printed;
2482 struct adapter *adapter;
2483 struct port_info *pi;
2484 struct net_device *netdev;
2487 * Vet our module parameters.
2489 if (msi != MSI_MSIX && msi != MSI_MSI) {
2490 dev_err(&pdev->dev, "bad module parameter msi=%d; must be %d"
2491 " (MSI-X or MSI) or %d (MSI)\n", msi, MSI_MSIX,
2498 * Print our driver banner the first time we're called to initialize a
2501 if (version_printed == 0) {
2502 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
2503 version_printed = 1;
2507 * Initialize generic PCI device state.
2509 err = pci_enable_device(pdev);
2511 dev_err(&pdev->dev, "cannot enable PCI device\n");
2516 * Reserve PCI resources for the device. If we can't get them some
2517 * other driver may have already claimed the device ...
2519 err = pci_request_regions(pdev, KBUILD_MODNAME);
2521 dev_err(&pdev->dev, "cannot obtain PCI resources\n");
2522 goto err_disable_device;
2526 * Set up our DMA mask: try for 64-bit address masking first and
2527 * fall back to 32-bit if we can't get 64 bits ...
2529 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2531 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2533 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for"
2534 " coherent allocations\n");
2535 goto err_release_regions;
2539 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2541 dev_err(&pdev->dev, "no usable DMA configuration\n");
2542 goto err_release_regions;
2548 * Enable bus mastering for the device ...
2550 pci_set_master(pdev);
2553 * Allocate our adapter data structure and attach it to the device.
2555 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2558 goto err_release_regions;
2560 pci_set_drvdata(pdev, adapter);
2561 adapter->pdev = pdev;
2562 adapter->pdev_dev = &pdev->dev;
2565 * Initialize SMP data synchronization resources.
2567 spin_lock_init(&adapter->stats_lock);
2570 * Map our I/O registers in BAR0.
2572 adapter->regs = pci_ioremap_bar(pdev, 0);
2573 if (!adapter->regs) {
2574 dev_err(&pdev->dev, "cannot map device registers\n");
2576 goto err_free_adapter;
2580 * Initialize adapter level features.
2582 adapter->name = pci_name(pdev);
2583 adapter->msg_enable = dflt_msg_enable;
2584 err = adap_init0(adapter);
2589 * Allocate our "adapter ports" and stitch everything together.
2591 pmask = adapter->params.vfres.pmask;
2592 for_each_port(adapter, pidx) {
2596 * We simplistically allocate our virtual interfaces
2597 * sequentially across the port numbers to which we have
2598 * access rights. This should be configurable in some manner
2603 port_id = ffs(pmask) - 1;
2604 pmask &= ~(1 << port_id);
2605 viid = t4vf_alloc_vi(adapter, port_id);
2607 dev_err(&pdev->dev, "cannot allocate VI for port %d:"
2608 " err=%d\n", port_id, viid);
2614 * Allocate our network device and stitch things together.
2616 netdev = alloc_etherdev_mq(sizeof(struct port_info),
2618 if (netdev == NULL) {
2619 dev_err(&pdev->dev, "cannot allocate netdev for"
2620 " port %d\n", port_id);
2621 t4vf_free_vi(adapter, viid);
2625 adapter->port[pidx] = netdev;
2626 SET_NETDEV_DEV(netdev, &pdev->dev);
2627 pi = netdev_priv(netdev);
2628 pi->adapter = adapter;
2630 pi->port_id = port_id;
2634 * Initialize the starting state of our "port" and register
2637 pi->xact_addr_filt = -1;
2638 pi->rx_offload = RX_CSO;
2639 netif_carrier_off(netdev);
2640 netdev->irq = pdev->irq;
2642 netdev->features = (NETIF_F_SG | TSO_FLAGS |
2643 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2644 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
2647 netdev->features |= NETIF_F_HIGHDMA;
2648 netdev->vlan_features =
2650 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX));
2652 #ifdef HAVE_NET_DEVICE_OPS
2653 netdev->netdev_ops = &cxgb4vf_netdev_ops;
2655 netdev->vlan_rx_register = cxgb4vf_vlan_rx_register;
2656 netdev->open = cxgb4vf_open;
2657 netdev->stop = cxgb4vf_stop;
2658 netdev->hard_start_xmit = t4vf_eth_xmit;
2659 netdev->get_stats = cxgb4vf_get_stats;
2660 netdev->set_rx_mode = cxgb4vf_set_rxmode;
2661 netdev->do_ioctl = cxgb4vf_do_ioctl;
2662 netdev->change_mtu = cxgb4vf_change_mtu;
2663 netdev->set_mac_address = cxgb4vf_set_mac_addr;
2664 #ifdef CONFIG_NET_POLL_CONTROLLER
2665 netdev->poll_controller = cxgb4vf_poll_controller;
2668 SET_ETHTOOL_OPS(netdev, &cxgb4vf_ethtool_ops);
2671 * Initialize the hardware/software state for the port.
2673 err = t4vf_port_init(adapter, pidx);
2675 dev_err(&pdev->dev, "cannot initialize port %d\n",
2682 * The "card" is now ready to go. If any errors occur during device
2683 * registration we do not fail the whole "card" but rather proceed
2684 * only with the ports we manage to register successfully. However we
2685 * must register at least one net device.
2687 for_each_port(adapter, pidx) {
2688 netdev = adapter->port[pidx];
2692 err = register_netdev(netdev);
2694 dev_warn(&pdev->dev, "cannot register net device %s,"
2695 " skipping\n", netdev->name);
2699 set_bit(pidx, &adapter->registered_device_map);
2701 if (adapter->registered_device_map == 0) {
2702 dev_err(&pdev->dev, "could not register any net devices\n");
2707 * Set up our debugfs entries.
2709 if (cxgb4vf_debugfs_root) {
2710 adapter->debugfs_root =
2711 debugfs_create_dir(pci_name(pdev),
2712 cxgb4vf_debugfs_root);
2713 if (adapter->debugfs_root == NULL)
2714 dev_warn(&pdev->dev, "could not create debugfs"
2717 setup_debugfs(adapter);
2721 * See what interrupts we'll be using. If we've been configured to
2722 * use MSI-X interrupts, try to enable them but fall back to using
2723 * MSI interrupts if we can't enable MSI-X interrupts. If we can't
2724 * get MSI interrupts we bail with the error.
2726 if (msi == MSI_MSIX && enable_msix(adapter) == 0)
2727 adapter->flags |= USING_MSIX;
2729 err = pci_enable_msi(pdev);
2731 dev_err(&pdev->dev, "Unable to allocate %s interrupts;"
2733 msi == MSI_MSIX ? "MSI-X or MSI" : "MSI", err);
2734 goto err_free_debugfs;
2736 adapter->flags |= USING_MSI;
2740 * Now that we know how many "ports" we have and what their types are,
2741 * and how many Queue Sets we can support, we can configure our queue
2744 cfg_queues(adapter);
2747 * Print a short notice on the existance and configuration of the new
2748 * VF network device ...
2750 for_each_port(adapter, pidx) {
2751 dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n",
2752 adapter->port[pidx]->name,
2753 (adapter->flags & USING_MSIX) ? "MSI-X" :
2754 (adapter->flags & USING_MSI) ? "MSI" : "");
2763 * Error recovery and exit code. Unwind state that's been created
2764 * so far and return the error.
2768 if (adapter->debugfs_root) {
2769 cleanup_debugfs(adapter);
2770 debugfs_remove_recursive(adapter->debugfs_root);
2774 for_each_port(adapter, pidx) {
2775 netdev = adapter->port[pidx];
2778 pi = netdev_priv(netdev);
2779 t4vf_free_vi(adapter, pi->viid);
2780 if (test_bit(pidx, &adapter->registered_device_map))
2781 unregister_netdev(netdev);
2782 free_netdev(netdev);
2786 iounmap(adapter->regs);
2790 pci_set_drvdata(pdev, NULL);
2792 err_release_regions:
2793 pci_release_regions(pdev);
2794 pci_set_drvdata(pdev, NULL);
2795 pci_clear_master(pdev);
2798 pci_disable_device(pdev);
2805 * "Remove" a device: tear down all kernel and driver state created in the
2806 * "probe" routine and quiesce the device (disable interrupts, etc.). (Note
2807 * that this is called "remove_one" in the PF Driver.)
2809 static void __devexit cxgb4vf_pci_remove(struct pci_dev *pdev)
2811 struct adapter *adapter = pci_get_drvdata(pdev);
2814 * Tear down driver state associated with device.
2820 * Stop all of our activity. Unregister network port,
2821 * disable interrupts, etc.
2823 for_each_port(adapter, pidx)
2824 if (test_bit(pidx, &adapter->registered_device_map))
2825 unregister_netdev(adapter->port[pidx]);
2826 t4vf_sge_stop(adapter);
2827 if (adapter->flags & USING_MSIX) {
2828 pci_disable_msix(adapter->pdev);
2829 adapter->flags &= ~USING_MSIX;
2830 } else if (adapter->flags & USING_MSI) {
2831 pci_disable_msi(adapter->pdev);
2832 adapter->flags &= ~USING_MSI;
2836 * Tear down our debugfs entries.
2838 if (adapter->debugfs_root) {
2839 cleanup_debugfs(adapter);
2840 debugfs_remove_recursive(adapter->debugfs_root);
2844 * Free all of the various resources which we've acquired ...
2846 t4vf_free_sge_resources(adapter);
2847 for_each_port(adapter, pidx) {
2848 struct net_device *netdev = adapter->port[pidx];
2849 struct port_info *pi;
2854 pi = netdev_priv(netdev);
2855 t4vf_free_vi(adapter, pi->viid);
2856 free_netdev(netdev);
2858 iounmap(adapter->regs);
2860 pci_set_drvdata(pdev, NULL);
2864 * Disable the device and release its PCI resources.
2866 pci_disable_device(pdev);
2867 pci_clear_master(pdev);
2868 pci_release_regions(pdev);
2872 * PCI Device registration data structures.
2874 #define CH_DEVICE(devid, idx) \
2875 { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
2877 static struct pci_device_id cxgb4vf_pci_tbl[] = {
2878 CH_DEVICE(0xb000, 0), /* PE10K FPGA */
2879 CH_DEVICE(0x4800, 0), /* T440-dbg */
2880 CH_DEVICE(0x4801, 0), /* T420-cr */
2881 CH_DEVICE(0x4802, 0), /* T422-cr */
2882 CH_DEVICE(0x4803, 0), /* T440-cr */
2883 CH_DEVICE(0x4804, 0), /* T420-bch */
2884 CH_DEVICE(0x4805, 0), /* T440-bch */
2885 CH_DEVICE(0x4806, 0), /* T460-ch */
2886 CH_DEVICE(0x4807, 0), /* T420-so */
2887 CH_DEVICE(0x4808, 0), /* T420-cx */
2888 CH_DEVICE(0x4809, 0), /* T420-bt */
2889 CH_DEVICE(0x480a, 0), /* T404-bt */
2893 MODULE_DESCRIPTION(DRV_DESC);
2894 MODULE_AUTHOR("Chelsio Communications");
2895 MODULE_LICENSE("Dual BSD/GPL");
2896 MODULE_VERSION(DRV_VERSION);
2897 MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl);
2899 static struct pci_driver cxgb4vf_driver = {
2900 .name = KBUILD_MODNAME,
2901 .id_table = cxgb4vf_pci_tbl,
2902 .probe = cxgb4vf_pci_probe,
2903 .remove = __devexit_p(cxgb4vf_pci_remove),
2907 * Initialize global driver state.
2909 static int __init cxgb4vf_module_init(void)
2913 /* Debugfs support is optional, just warn if this fails */
2914 cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
2915 if (!cxgb4vf_debugfs_root)
2916 printk(KERN_WARNING KBUILD_MODNAME ": could not create"
2917 " debugfs entry, continuing\n");
2919 ret = pci_register_driver(&cxgb4vf_driver);
2921 debugfs_remove(cxgb4vf_debugfs_root);
2926 * Tear down global driver state.
2928 static void __exit cxgb4vf_module_exit(void)
2930 pci_unregister_driver(&cxgb4vf_driver);
2931 debugfs_remove(cxgb4vf_debugfs_root);
2934 module_init(cxgb4vf_module_init);
2935 module_exit(cxgb4vf_module_exit);