2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/bitmap.h>
38 #include <linux/crc32.h>
39 #include <linux/ctype.h>
40 #include <linux/debugfs.h>
41 #include <linux/err.h>
42 #include <linux/etherdevice.h>
43 #include <linux/firmware.h>
44 #include <linux/if_vlan.h>
45 #include <linux/init.h>
46 #include <linux/log2.h>
47 #include <linux/mdio.h>
48 #include <linux/module.h>
49 #include <linux/moduleparam.h>
50 #include <linux/mutex.h>
51 #include <linux/netdevice.h>
52 #include <linux/pci.h>
53 #include <linux/aer.h>
54 #include <linux/rtnetlink.h>
55 #include <linux/sched.h>
56 #include <linux/seq_file.h>
57 #include <linux/sockios.h>
58 #include <linux/vmalloc.h>
59 #include <linux/workqueue.h>
60 #include <net/neighbour.h>
61 #include <net/netevent.h>
62 #include <asm/uaccess.h>
70 #define DRV_VERSION "1.3.0-ko"
71 #define DRV_DESC "Chelsio T4 Network Driver"
74 * Max interrupt hold-off timer value in us. Queues fall back to this value
75 * under extreme memory pressure so it's largish to give the system time to
78 #define MAX_SGE_TIMERVAL 200U
82 * Virtual Function provisioning constants. We need two extra Ingress Queues
83 * with Interrupt capability to serve as the VF's Firmware Event Queue and
84 * Forwarded Interrupt Queue (when using MSI mode) -- neither will have Free
85 * Lists associated with them). For each Ethernet/Control Egress Queue and
86 * for each Free List, we need an Egress Context.
89 VFRES_NPORTS = 1, /* # of "ports" per VF */
90 VFRES_NQSETS = 2, /* # of "Queue Sets" per VF */
92 VFRES_NVI = VFRES_NPORTS, /* # of Virtual Interfaces */
93 VFRES_NETHCTRL = VFRES_NQSETS, /* # of EQs used for ETH or CTRL Qs */
94 VFRES_NIQFLINT = VFRES_NQSETS+2,/* # of ingress Qs/w Free List(s)/intr */
95 VFRES_NIQ = 0, /* # of non-fl/int ingress queues */
96 VFRES_NEQ = VFRES_NQSETS*2, /* # of egress queues */
97 VFRES_TC = 0, /* PCI-E traffic class */
98 VFRES_NEXACTF = 16, /* # of exact MPS filters */
100 VFRES_R_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF|FW_CMD_CAP_PORT,
101 VFRES_WX_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF,
105 * Provide a Port Access Rights Mask for the specified PF/VF. This is very
106 * static and likely not to be useful in the long run. We really need to
107 * implement some form of persistent configuration which the firmware
110 static unsigned int pfvfres_pmask(struct adapter *adapter,
111 unsigned int pf, unsigned int vf)
113 unsigned int portn, portvec;
116 * Give PF's access to all of the ports.
119 return FW_PFVF_CMD_PMASK_MASK;
122 * For VFs, we'll assign them access to the ports based purely on the
123 * PF. We assign active ports in order, wrapping around if there are
124 * fewer active ports than PFs: e.g. active port[pf % nports].
125 * Unfortunately the adapter's port_info structs haven't been
126 * initialized yet so we have to compute this.
128 if (adapter->params.nports == 0)
131 portn = pf % adapter->params.nports;
132 portvec = adapter->params.portvec;
135 * Isolate the lowest set bit in the port vector. If we're at
136 * the port number that we want, return that as the pmask.
137 * otherwise mask that bit out of the port vector and
138 * decrement our port number ...
140 unsigned int pmask = portvec ^ (portvec & (portvec-1));
151 MEMWIN0_APERTURE = 65536,
152 MEMWIN0_BASE = 0x30000,
153 MEMWIN1_APERTURE = 32768,
154 MEMWIN1_BASE = 0x28000,
155 MEMWIN2_APERTURE = 2048,
156 MEMWIN2_BASE = 0x1b800,
160 MAX_TXQ_ENTRIES = 16384,
161 MAX_CTRL_TXQ_ENTRIES = 1024,
162 MAX_RSPQ_ENTRIES = 16384,
163 MAX_RX_BUFFERS = 16384,
164 MIN_TXQ_ENTRIES = 32,
165 MIN_CTRL_TXQ_ENTRIES = 32,
166 MIN_RSPQ_ENTRIES = 128,
170 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
171 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
172 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
174 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
176 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
177 CH_DEVICE(0xa000, 0), /* PE10K */
178 CH_DEVICE(0x4001, -1),
179 CH_DEVICE(0x4002, -1),
180 CH_DEVICE(0x4003, -1),
181 CH_DEVICE(0x4004, -1),
182 CH_DEVICE(0x4005, -1),
183 CH_DEVICE(0x4006, -1),
184 CH_DEVICE(0x4007, -1),
185 CH_DEVICE(0x4008, -1),
186 CH_DEVICE(0x4009, -1),
187 CH_DEVICE(0x400a, -1),
188 CH_DEVICE(0x4401, 4),
189 CH_DEVICE(0x4402, 4),
190 CH_DEVICE(0x4403, 4),
191 CH_DEVICE(0x4404, 4),
192 CH_DEVICE(0x4405, 4),
193 CH_DEVICE(0x4406, 4),
194 CH_DEVICE(0x4407, 4),
195 CH_DEVICE(0x4408, 4),
196 CH_DEVICE(0x4409, 4),
197 CH_DEVICE(0x440a, 4),
201 #define FW_FNAME "cxgb4/t4fw.bin"
203 MODULE_DESCRIPTION(DRV_DESC);
204 MODULE_AUTHOR("Chelsio Communications");
205 MODULE_LICENSE("Dual BSD/GPL");
206 MODULE_VERSION(DRV_VERSION);
207 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
208 MODULE_FIRMWARE(FW_FNAME);
210 static int dflt_msg_enable = DFLT_MSG_ENABLE;
212 module_param(dflt_msg_enable, int, 0644);
213 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
216 * The driver uses the best interrupt scheme available on a platform in the
217 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
218 * of these schemes the driver may consider as follows:
220 * msi = 2: choose from among all three options
221 * msi = 1: only consider MSI and INTx interrupts
222 * msi = 0: force INTx interrupts
226 module_param(msi, int, 0644);
227 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
230 * Queue interrupt hold-off timer values. Queues default to the first of these
233 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
235 module_param_array(intr_holdoff, uint, NULL, 0644);
236 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
237 "0..4 in microseconds");
239 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
241 module_param_array(intr_cnt, uint, NULL, 0644);
242 MODULE_PARM_DESC(intr_cnt,
243 "thresholds 1..3 for queue interrupt packet counters");
247 #ifdef CONFIG_PCI_IOV
248 module_param(vf_acls, bool, 0644);
249 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
251 static unsigned int num_vf[4];
253 module_param_array(num_vf, uint, NULL, 0644);
254 MODULE_PARM_DESC(num_vf, "number of VFs for each of PFs 0-3");
257 static struct dentry *cxgb4_debugfs_root;
259 static LIST_HEAD(adapter_list);
260 static DEFINE_MUTEX(uld_mutex);
261 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
262 static const char *uld_str[] = { "RDMA", "iSCSI" };
264 static void link_report(struct net_device *dev)
266 if (!netif_carrier_ok(dev))
267 netdev_info(dev, "link down\n");
269 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
271 const char *s = "10Mbps";
272 const struct port_info *p = netdev_priv(dev);
274 switch (p->link_cfg.speed) {
286 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
291 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
293 struct net_device *dev = adapter->port[port_id];
295 /* Skip changes from disabled ports. */
296 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
298 netif_carrier_on(dev);
300 netif_carrier_off(dev);
306 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
308 static const char *mod_str[] = {
309 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
312 const struct net_device *dev = adap->port[port_id];
313 const struct port_info *pi = netdev_priv(dev);
315 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
316 netdev_info(dev, "port module unplugged\n");
317 else if (pi->mod_type < ARRAY_SIZE(mod_str))
318 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
322 * Configure the exact and hash address filters to handle a port's multicast
323 * and secondary unicast MAC addresses.
325 static int set_addr_filters(const struct net_device *dev, bool sleep)
333 const struct netdev_hw_addr *ha;
334 int uc_cnt = netdev_uc_count(dev);
335 int mc_cnt = netdev_mc_count(dev);
336 const struct port_info *pi = netdev_priv(dev);
337 unsigned int mb = pi->adapter->fn;
339 /* first do the secondary unicast addresses */
340 netdev_for_each_uc_addr(ha, dev) {
341 addr[naddr++] = ha->addr;
342 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
343 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
344 naddr, addr, filt_idx, &uhash, sleep);
353 /* next set up the multicast addresses */
354 netdev_for_each_mc_addr(ha, dev) {
355 addr[naddr++] = ha->addr;
356 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
357 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
358 naddr, addr, filt_idx, &mhash, sleep);
367 return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
368 uhash | mhash, sleep);
372 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
373 * If @mtu is -1 it is left unchanged.
375 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
378 struct port_info *pi = netdev_priv(dev);
380 ret = set_addr_filters(dev, sleep_ok);
382 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
383 (dev->flags & IFF_PROMISC) ? 1 : 0,
384 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
390 * link_start - enable a port
391 * @dev: the port to enable
393 * Performs the MAC and PHY actions needed to enable a port.
395 static int link_start(struct net_device *dev)
398 struct port_info *pi = netdev_priv(dev);
399 unsigned int mb = pi->adapter->fn;
402 * We do not set address filters and promiscuity here, the stack does
403 * that step explicitly.
405 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
406 pi->vlan_grp != NULL, true);
408 ret = t4_change_mac(pi->adapter, mb, pi->viid,
409 pi->xact_addr_filt, dev->dev_addr, true,
412 pi->xact_addr_filt = ret;
417 ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
420 ret = t4_enable_vi(pi->adapter, mb, pi->viid, true, true);
425 * Response queue handler for the FW event queue.
427 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
428 const struct pkt_gl *gl)
430 u8 opcode = ((const struct rss_header *)rsp)->opcode;
432 rsp++; /* skip RSS header */
433 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
434 const struct cpl_sge_egr_update *p = (void *)rsp;
435 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
438 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
440 if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
441 struct sge_eth_txq *eq;
443 eq = container_of(txq, struct sge_eth_txq, q);
444 netif_tx_wake_queue(eq->txq);
446 struct sge_ofld_txq *oq;
448 oq = container_of(txq, struct sge_ofld_txq, q);
449 tasklet_schedule(&oq->qresume_tsk);
451 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
452 const struct cpl_fw6_msg *p = (void *)rsp;
455 t4_handle_fw_rpl(q->adap, p->data);
456 } else if (opcode == CPL_L2T_WRITE_RPL) {
457 const struct cpl_l2t_write_rpl *p = (void *)rsp;
459 do_l2t_write_rpl(q->adap, p);
461 dev_err(q->adap->pdev_dev,
462 "unexpected CPL %#x on FW event queue\n", opcode);
467 * uldrx_handler - response queue handler for ULD queues
468 * @q: the response queue that received the packet
469 * @rsp: the response queue descriptor holding the offload message
470 * @gl: the gather list of packet fragments
472 * Deliver an ingress offload packet to a ULD. All processing is done by
473 * the ULD, we just maintain statistics.
475 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
476 const struct pkt_gl *gl)
478 struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
480 if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
486 else if (gl == CXGB4_MSG_AN)
493 static void disable_msi(struct adapter *adapter)
495 if (adapter->flags & USING_MSIX) {
496 pci_disable_msix(adapter->pdev);
497 adapter->flags &= ~USING_MSIX;
498 } else if (adapter->flags & USING_MSI) {
499 pci_disable_msi(adapter->pdev);
500 adapter->flags &= ~USING_MSI;
505 * Interrupt handler for non-data events used with MSI-X.
507 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
509 struct adapter *adap = cookie;
511 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
514 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
516 t4_slow_intr_handler(adap);
521 * Name the MSI-X interrupts.
523 static void name_msix_vecs(struct adapter *adap)
525 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc) - 1;
527 /* non-data interrupts */
528 snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
529 adap->msix_info[0].desc[n] = 0;
532 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq", adap->name);
533 adap->msix_info[1].desc[n] = 0;
535 /* Ethernet queues */
536 for_each_port(adap, j) {
537 struct net_device *d = adap->port[j];
538 const struct port_info *pi = netdev_priv(d);
540 for (i = 0; i < pi->nqsets; i++, msi_idx++) {
541 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
543 adap->msix_info[msi_idx].desc[n] = 0;
548 for_each_ofldrxq(&adap->sge, i) {
549 snprintf(adap->msix_info[msi_idx].desc, n, "%s-ofld%d",
551 adap->msix_info[msi_idx++].desc[n] = 0;
553 for_each_rdmarxq(&adap->sge, i) {
554 snprintf(adap->msix_info[msi_idx].desc, n, "%s-rdma%d",
556 adap->msix_info[msi_idx++].desc[n] = 0;
560 static int request_msix_queue_irqs(struct adapter *adap)
562 struct sge *s = &adap->sge;
563 int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi = 2;
565 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
566 adap->msix_info[1].desc, &s->fw_evtq);
570 for_each_ethrxq(s, ethqidx) {
571 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
572 adap->msix_info[msi].desc,
573 &s->ethrxq[ethqidx].rspq);
578 for_each_ofldrxq(s, ofldqidx) {
579 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
580 adap->msix_info[msi].desc,
581 &s->ofldrxq[ofldqidx].rspq);
586 for_each_rdmarxq(s, rdmaqidx) {
587 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
588 adap->msix_info[msi].desc,
589 &s->rdmarxq[rdmaqidx].rspq);
597 while (--rdmaqidx >= 0)
598 free_irq(adap->msix_info[--msi].vec,
599 &s->rdmarxq[rdmaqidx].rspq);
600 while (--ofldqidx >= 0)
601 free_irq(adap->msix_info[--msi].vec,
602 &s->ofldrxq[ofldqidx].rspq);
603 while (--ethqidx >= 0)
604 free_irq(adap->msix_info[--msi].vec, &s->ethrxq[ethqidx].rspq);
605 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
609 static void free_msix_queue_irqs(struct adapter *adap)
612 struct sge *s = &adap->sge;
614 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
615 for_each_ethrxq(s, i)
616 free_irq(adap->msix_info[msi++].vec, &s->ethrxq[i].rspq);
617 for_each_ofldrxq(s, i)
618 free_irq(adap->msix_info[msi++].vec, &s->ofldrxq[i].rspq);
619 for_each_rdmarxq(s, i)
620 free_irq(adap->msix_info[msi++].vec, &s->rdmarxq[i].rspq);
624 * write_rss - write the RSS table for a given port
626 * @queues: array of queue indices for RSS
628 * Sets up the portion of the HW RSS table for the port's VI to distribute
629 * packets to the Rx queues in @queues.
631 static int write_rss(const struct port_info *pi, const u16 *queues)
635 const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
637 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
641 /* map the queue indices to queue ids */
642 for (i = 0; i < pi->rss_size; i++, queues++)
643 rss[i] = q[*queues].rspq.abs_id;
645 err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
646 pi->rss_size, rss, pi->rss_size);
652 * setup_rss - configure RSS
655 * Sets up RSS for each port.
657 static int setup_rss(struct adapter *adap)
661 for_each_port(adap, i) {
662 const struct port_info *pi = adap2pinfo(adap, i);
664 err = write_rss(pi, pi->rss);
672 * Return the channel of the ingress queue with the given qid.
674 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
676 qid -= p->ingr_start;
677 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
681 * Wait until all NAPI handlers are descheduled.
683 static void quiesce_rx(struct adapter *adap)
687 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
688 struct sge_rspq *q = adap->sge.ingr_map[i];
691 napi_disable(&q->napi);
696 * Enable NAPI scheduling and interrupt generation for all Rx queues.
698 static void enable_rx(struct adapter *adap)
702 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
703 struct sge_rspq *q = adap->sge.ingr_map[i];
708 napi_enable(&q->napi);
709 /* 0-increment GTS to start the timer and enable interrupts */
710 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
711 SEINTARM(q->intr_params) |
712 INGRESSQID(q->cntxt_id));
717 * setup_sge_queues - configure SGE Tx/Rx/response queues
720 * Determines how many sets of SGE queues to use and initializes them.
721 * We support multiple queue sets per port if we have MSI-X, otherwise
722 * just one queue set per port.
724 static int setup_sge_queues(struct adapter *adap)
726 int err, msi_idx, i, j;
727 struct sge *s = &adap->sge;
729 bitmap_zero(s->starving_fl, MAX_EGRQ);
730 bitmap_zero(s->txq_maperr, MAX_EGRQ);
732 if (adap->flags & USING_MSIX)
733 msi_idx = 1; /* vector 0 is for non-queue interrupts */
735 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
739 msi_idx = -((int)s->intrq.abs_id + 1);
742 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
743 msi_idx, NULL, fwevtq_handler);
745 freeout: t4_free_sge_resources(adap);
749 for_each_port(adap, i) {
750 struct net_device *dev = adap->port[i];
751 struct port_info *pi = netdev_priv(dev);
752 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
753 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
755 for (j = 0; j < pi->nqsets; j++, q++) {
758 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
764 memset(&q->stats, 0, sizeof(q->stats));
766 for (j = 0; j < pi->nqsets; j++, t++) {
767 err = t4_sge_alloc_eth_txq(adap, t, dev,
768 netdev_get_tx_queue(dev, j),
769 s->fw_evtq.cntxt_id);
775 j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
776 for_each_ofldrxq(s, i) {
777 struct sge_ofld_rxq *q = &s->ofldrxq[i];
778 struct net_device *dev = adap->port[i / j];
782 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
783 &q->fl, uldrx_handler);
786 memset(&q->stats, 0, sizeof(q->stats));
787 s->ofld_rxq[i] = q->rspq.abs_id;
788 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
789 s->fw_evtq.cntxt_id);
794 for_each_rdmarxq(s, i) {
795 struct sge_ofld_rxq *q = &s->rdmarxq[i];
799 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
800 msi_idx, &q->fl, uldrx_handler);
803 memset(&q->stats, 0, sizeof(q->stats));
804 s->rdma_rxq[i] = q->rspq.abs_id;
807 for_each_port(adap, i) {
809 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
810 * have RDMA queues, and that's the right value.
812 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
814 s->rdmarxq[i].rspq.cntxt_id);
819 t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
820 RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
821 QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
826 * Returns 0 if new FW was successfully loaded, a positive errno if a load was
827 * started but failed, and a negative errno if flash load couldn't start.
829 static int upgrade_fw(struct adapter *adap)
833 const struct fw_hdr *hdr;
834 const struct firmware *fw;
835 struct device *dev = adap->pdev_dev;
837 ret = request_firmware(&fw, FW_FNAME, dev);
839 dev_err(dev, "unable to load firmware image " FW_FNAME
840 ", error %d\n", ret);
844 hdr = (const struct fw_hdr *)fw->data;
845 vers = ntohl(hdr->fw_ver);
846 if (FW_HDR_FW_VER_MAJOR_GET(vers) != FW_VERSION_MAJOR) {
847 ret = -EINVAL; /* wrong major version, won't do */
852 * If the flash FW is unusable or we found something newer, load it.
854 if (FW_HDR_FW_VER_MAJOR_GET(adap->params.fw_vers) != FW_VERSION_MAJOR ||
855 vers > adap->params.fw_vers) {
856 ret = -t4_load_fw(adap, fw->data, fw->size);
858 dev_info(dev, "firmware upgraded to version %pI4 from "
859 FW_FNAME "\n", &hdr->fw_ver);
861 out: release_firmware(fw);
866 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
867 * The allocated memory is cleared.
869 void *t4_alloc_mem(size_t size)
871 void *p = kmalloc(size, GFP_KERNEL);
881 * Free memory allocated through alloc_mem().
883 static void t4_free_mem(void *addr)
885 if (is_vmalloc_addr(addr))
891 static inline int is_offload(const struct adapter *adap)
893 return adap->params.offload;
897 * Implementation of ethtool operations.
900 static u32 get_msglevel(struct net_device *dev)
902 return netdev2adap(dev)->msg_enable;
905 static void set_msglevel(struct net_device *dev, u32 val)
907 netdev2adap(dev)->msg_enable = val;
910 static char stats_strings[][ETH_GSTRING_LEN] = {
913 "TxBroadcastFrames ",
914 "TxMulticastFrames ",
922 "TxFrames512To1023 ",
923 "TxFrames1024To1518 ",
924 "TxFrames1519ToMax ",
939 "RxBroadcastFrames ",
940 "RxMulticastFrames ",
954 "RxFrames512To1023 ",
955 "RxFrames1024To1518 ",
956 "RxFrames1519ToMax ",
968 "RxBG0FramesDropped ",
969 "RxBG1FramesDropped ",
970 "RxBG2FramesDropped ",
971 "RxBG3FramesDropped ",
986 static int get_sset_count(struct net_device *dev, int sset)
990 return ARRAY_SIZE(stats_strings);
996 #define T4_REGMAP_SIZE (160 * 1024)
998 static int get_regs_len(struct net_device *dev)
1000 return T4_REGMAP_SIZE;
1003 static int get_eeprom_len(struct net_device *dev)
1008 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1010 struct adapter *adapter = netdev2adap(dev);
1012 strcpy(info->driver, KBUILD_MODNAME);
1013 strcpy(info->version, DRV_VERSION);
1014 strcpy(info->bus_info, pci_name(adapter->pdev));
1016 if (!adapter->params.fw_vers)
1017 strcpy(info->fw_version, "N/A");
1019 snprintf(info->fw_version, sizeof(info->fw_version),
1020 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1021 FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1022 FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1023 FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1024 FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1025 FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1026 FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1027 FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1028 FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1031 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1033 if (stringset == ETH_SS_STATS)
1034 memcpy(data, stats_strings, sizeof(stats_strings));
1038 * port stats maintained per queue of the port. They should be in the same
1039 * order as in stats_strings above.
1041 struct queue_port_stats {
1051 static void collect_sge_port_stats(const struct adapter *adap,
1052 const struct port_info *p, struct queue_port_stats *s)
1055 const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1056 const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1058 memset(s, 0, sizeof(*s));
1059 for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1061 s->tx_csum += tx->tx_cso;
1062 s->rx_csum += rx->stats.rx_cso;
1063 s->vlan_ex += rx->stats.vlan_ex;
1064 s->vlan_ins += tx->vlan_ins;
1065 s->gro_pkts += rx->stats.lro_pkts;
1066 s->gro_merged += rx->stats.lro_merged;
1070 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1073 struct port_info *pi = netdev_priv(dev);
1074 struct adapter *adapter = pi->adapter;
1076 t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1078 data += sizeof(struct port_stats) / sizeof(u64);
1079 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1083 * Return a version number to identify the type of adapter. The scheme is:
1084 * - bits 0..9: chip version
1085 * - bits 10..15: chip revision
1086 * - bits 16..23: register dump version
1088 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1090 return 4 | (ap->params.rev << 10) | (1 << 16);
1093 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1096 u32 *p = buf + start;
1098 for ( ; start <= end; start += sizeof(u32))
1099 *p++ = t4_read_reg(ap, start);
1102 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1105 static const unsigned int reg_ranges[] = {
1326 struct adapter *ap = netdev2adap(dev);
1328 regs->version = mk_adap_vers(ap);
1330 memset(buf, 0, T4_REGMAP_SIZE);
1331 for (i = 0; i < ARRAY_SIZE(reg_ranges); i += 2)
1332 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
1335 static int restart_autoneg(struct net_device *dev)
1337 struct port_info *p = netdev_priv(dev);
1339 if (!netif_running(dev))
1341 if (p->link_cfg.autoneg != AUTONEG_ENABLE)
1343 t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
1347 static int identify_port(struct net_device *dev, u32 data)
1349 struct adapter *adap = netdev2adap(dev);
1352 data = 2; /* default to 2 seconds */
1354 return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid,
1358 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
1362 if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
1363 type == FW_PORT_TYPE_BT_XAUI) {
1365 if (caps & FW_PORT_CAP_SPEED_100M)
1366 v |= SUPPORTED_100baseT_Full;
1367 if (caps & FW_PORT_CAP_SPEED_1G)
1368 v |= SUPPORTED_1000baseT_Full;
1369 if (caps & FW_PORT_CAP_SPEED_10G)
1370 v |= SUPPORTED_10000baseT_Full;
1371 } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
1372 v |= SUPPORTED_Backplane;
1373 if (caps & FW_PORT_CAP_SPEED_1G)
1374 v |= SUPPORTED_1000baseKX_Full;
1375 if (caps & FW_PORT_CAP_SPEED_10G)
1376 v |= SUPPORTED_10000baseKX4_Full;
1377 } else if (type == FW_PORT_TYPE_KR)
1378 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
1379 else if (type == FW_PORT_TYPE_BP_AP)
1380 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC;
1381 else if (type == FW_PORT_TYPE_FIBER_XFI ||
1382 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
1383 v |= SUPPORTED_FIBRE;
1385 if (caps & FW_PORT_CAP_ANEG)
1386 v |= SUPPORTED_Autoneg;
1390 static unsigned int to_fw_linkcaps(unsigned int caps)
1394 if (caps & ADVERTISED_100baseT_Full)
1395 v |= FW_PORT_CAP_SPEED_100M;
1396 if (caps & ADVERTISED_1000baseT_Full)
1397 v |= FW_PORT_CAP_SPEED_1G;
1398 if (caps & ADVERTISED_10000baseT_Full)
1399 v |= FW_PORT_CAP_SPEED_10G;
1403 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1405 const struct port_info *p = netdev_priv(dev);
1407 if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
1408 p->port_type == FW_PORT_TYPE_BT_XFI ||
1409 p->port_type == FW_PORT_TYPE_BT_XAUI)
1410 cmd->port = PORT_TP;
1411 else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
1412 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
1413 cmd->port = PORT_FIBRE;
1414 else if (p->port_type == FW_PORT_TYPE_SFP) {
1415 if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
1416 p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
1417 cmd->port = PORT_DA;
1419 cmd->port = PORT_FIBRE;
1421 cmd->port = PORT_OTHER;
1423 if (p->mdio_addr >= 0) {
1424 cmd->phy_address = p->mdio_addr;
1425 cmd->transceiver = XCVR_EXTERNAL;
1426 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
1427 MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
1429 cmd->phy_address = 0; /* not really, but no better option */
1430 cmd->transceiver = XCVR_INTERNAL;
1431 cmd->mdio_support = 0;
1434 cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
1435 cmd->advertising = from_fw_linkcaps(p->port_type,
1436 p->link_cfg.advertising);
1437 cmd->speed = netif_carrier_ok(dev) ? p->link_cfg.speed : 0;
1438 cmd->duplex = DUPLEX_FULL;
1439 cmd->autoneg = p->link_cfg.autoneg;
1445 static unsigned int speed_to_caps(int speed)
1447 if (speed == SPEED_100)
1448 return FW_PORT_CAP_SPEED_100M;
1449 if (speed == SPEED_1000)
1450 return FW_PORT_CAP_SPEED_1G;
1451 if (speed == SPEED_10000)
1452 return FW_PORT_CAP_SPEED_10G;
1456 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1459 struct port_info *p = netdev_priv(dev);
1460 struct link_config *lc = &p->link_cfg;
1462 if (cmd->duplex != DUPLEX_FULL) /* only full-duplex supported */
1465 if (!(lc->supported & FW_PORT_CAP_ANEG)) {
1467 * PHY offers a single speed. See if that's what's
1470 if (cmd->autoneg == AUTONEG_DISABLE &&
1471 (lc->supported & speed_to_caps(cmd->speed)))
1476 if (cmd->autoneg == AUTONEG_DISABLE) {
1477 cap = speed_to_caps(cmd->speed);
1479 if (!(lc->supported & cap) || cmd->speed == SPEED_1000 ||
1480 cmd->speed == SPEED_10000)
1482 lc->requested_speed = cap;
1483 lc->advertising = 0;
1485 cap = to_fw_linkcaps(cmd->advertising);
1486 if (!(lc->supported & cap))
1488 lc->requested_speed = 0;
1489 lc->advertising = cap | FW_PORT_CAP_ANEG;
1491 lc->autoneg = cmd->autoneg;
1493 if (netif_running(dev))
1494 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
1499 static void get_pauseparam(struct net_device *dev,
1500 struct ethtool_pauseparam *epause)
1502 struct port_info *p = netdev_priv(dev);
1504 epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1505 epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
1506 epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
1509 static int set_pauseparam(struct net_device *dev,
1510 struct ethtool_pauseparam *epause)
1512 struct port_info *p = netdev_priv(dev);
1513 struct link_config *lc = &p->link_cfg;
1515 if (epause->autoneg == AUTONEG_DISABLE)
1516 lc->requested_fc = 0;
1517 else if (lc->supported & FW_PORT_CAP_ANEG)
1518 lc->requested_fc = PAUSE_AUTONEG;
1522 if (epause->rx_pause)
1523 lc->requested_fc |= PAUSE_RX;
1524 if (epause->tx_pause)
1525 lc->requested_fc |= PAUSE_TX;
1526 if (netif_running(dev))
1527 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
1532 static u32 get_rx_csum(struct net_device *dev)
1534 struct port_info *p = netdev_priv(dev);
1536 return p->rx_offload & RX_CSO;
1539 static int set_rx_csum(struct net_device *dev, u32 data)
1541 struct port_info *p = netdev_priv(dev);
1544 p->rx_offload |= RX_CSO;
1546 p->rx_offload &= ~RX_CSO;
1550 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1552 const struct port_info *pi = netdev_priv(dev);
1553 const struct sge *s = &pi->adapter->sge;
1555 e->rx_max_pending = MAX_RX_BUFFERS;
1556 e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1557 e->rx_jumbo_max_pending = 0;
1558 e->tx_max_pending = MAX_TXQ_ENTRIES;
1560 e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
1561 e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1562 e->rx_jumbo_pending = 0;
1563 e->tx_pending = s->ethtxq[pi->first_qset].q.size;
1566 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1569 const struct port_info *pi = netdev_priv(dev);
1570 struct adapter *adapter = pi->adapter;
1571 struct sge *s = &adapter->sge;
1573 if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
1574 e->tx_pending > MAX_TXQ_ENTRIES ||
1575 e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1576 e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1577 e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
1580 if (adapter->flags & FULL_INIT_DONE)
1583 for (i = 0; i < pi->nqsets; ++i) {
1584 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
1585 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
1586 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
1591 static int closest_timer(const struct sge *s, int time)
1593 int i, delta, match = 0, min_delta = INT_MAX;
1595 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1596 delta = time - s->timer_val[i];
1599 if (delta < min_delta) {
1607 static int closest_thres(const struct sge *s, int thres)
1609 int i, delta, match = 0, min_delta = INT_MAX;
1611 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1612 delta = thres - s->counter_val[i];
1615 if (delta < min_delta) {
1624 * Return a queue's interrupt hold-off time in us. 0 means no timer.
1626 static unsigned int qtimer_val(const struct adapter *adap,
1627 const struct sge_rspq *q)
1629 unsigned int idx = q->intr_params >> 1;
1631 return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
1635 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1636 * @adap: the adapter
1638 * @us: the hold-off time in us, or 0 to disable timer
1639 * @cnt: the hold-off packet count, or 0 to disable counter
1641 * Sets an Rx queue's interrupt hold-off time and packet count. At least
1642 * one of the two needs to be enabled for the queue to generate interrupts.
1644 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
1645 unsigned int us, unsigned int cnt)
1647 if ((us | cnt) == 0)
1654 new_idx = closest_thres(&adap->sge, cnt);
1655 if (q->desc && q->pktcnt_idx != new_idx) {
1656 /* the queue has already been created, update it */
1657 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1658 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1659 FW_PARAMS_PARAM_YZ(q->cntxt_id);
1660 err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
1665 q->pktcnt_idx = new_idx;
1668 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1669 q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
1673 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1675 const struct port_info *pi = netdev_priv(dev);
1676 struct adapter *adap = pi->adapter;
1678 return set_rxq_intr_params(adap, &adap->sge.ethrxq[pi->first_qset].rspq,
1679 c->rx_coalesce_usecs, c->rx_max_coalesced_frames);
1682 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1684 const struct port_info *pi = netdev_priv(dev);
1685 const struct adapter *adap = pi->adapter;
1686 const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
1688 c->rx_coalesce_usecs = qtimer_val(adap, rq);
1689 c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
1690 adap->sge.counter_val[rq->pktcnt_idx] : 0;
1695 * eeprom_ptov - translate a physical EEPROM address to virtual
1696 * @phys_addr: the physical EEPROM address
1697 * @fn: the PCI function number
1698 * @sz: size of function-specific area
1700 * Translate a physical EEPROM address to virtual. The first 1K is
1701 * accessed through virtual addresses starting at 31K, the rest is
1702 * accessed through virtual addresses starting at 0.
1704 * The mapping is as follows:
1705 * [0..1K) -> [31K..32K)
1706 * [1K..1K+A) -> [31K-A..31K)
1707 * [1K+A..ES) -> [0..ES-A-1K)
1709 * where A = @fn * @sz, and ES = EEPROM size.
1711 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
1714 if (phys_addr < 1024)
1715 return phys_addr + (31 << 10);
1716 if (phys_addr < 1024 + fn)
1717 return 31744 - fn + phys_addr - 1024;
1718 if (phys_addr < EEPROMSIZE)
1719 return phys_addr - 1024 - fn;
1724 * The next two routines implement eeprom read/write from physical addresses.
1726 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
1728 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
1731 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
1732 return vaddr < 0 ? vaddr : 0;
1735 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
1737 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
1740 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
1741 return vaddr < 0 ? vaddr : 0;
1744 #define EEPROM_MAGIC 0x38E2F10C
1746 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
1750 struct adapter *adapter = netdev2adap(dev);
1752 u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
1756 e->magic = EEPROM_MAGIC;
1757 for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
1758 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
1761 memcpy(data, buf + e->offset, e->len);
1766 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
1771 u32 aligned_offset, aligned_len, *p;
1772 struct adapter *adapter = netdev2adap(dev);
1774 if (eeprom->magic != EEPROM_MAGIC)
1777 aligned_offset = eeprom->offset & ~3;
1778 aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
1780 if (adapter->fn > 0) {
1781 u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
1783 if (aligned_offset < start ||
1784 aligned_offset + aligned_len > start + EEPROMPFSIZE)
1788 if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
1790 * RMW possibly needed for first or last words.
1792 buf = kmalloc(aligned_len, GFP_KERNEL);
1795 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
1796 if (!err && aligned_len > 4)
1797 err = eeprom_rd_phys(adapter,
1798 aligned_offset + aligned_len - 4,
1799 (u32 *)&buf[aligned_len - 4]);
1802 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
1806 err = t4_seeprom_wp(adapter, false);
1810 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
1811 err = eeprom_wr_phys(adapter, aligned_offset, *p);
1812 aligned_offset += 4;
1816 err = t4_seeprom_wp(adapter, true);
1823 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
1826 const struct firmware *fw;
1827 struct adapter *adap = netdev2adap(netdev);
1829 ef->data[sizeof(ef->data) - 1] = '\0';
1830 ret = request_firmware(&fw, ef->data, adap->pdev_dev);
1834 ret = t4_load_fw(adap, fw->data, fw->size);
1835 release_firmware(fw);
1837 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
1841 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
1842 #define BCAST_CRC 0xa0ccc1a6
1844 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1846 wol->supported = WAKE_BCAST | WAKE_MAGIC;
1847 wol->wolopts = netdev2adap(dev)->wol;
1848 memset(&wol->sopass, 0, sizeof(wol->sopass));
1851 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1854 struct port_info *pi = netdev_priv(dev);
1856 if (wol->wolopts & ~WOL_SUPPORTED)
1858 t4_wol_magic_enable(pi->adapter, pi->tx_chan,
1859 (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
1860 if (wol->wolopts & WAKE_BCAST) {
1861 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
1864 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
1865 ~6ULL, ~0ULL, BCAST_CRC, true);
1867 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
1871 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1873 static int set_tso(struct net_device *dev, u32 value)
1876 dev->features |= TSO_FLAGS;
1878 dev->features &= ~TSO_FLAGS;
1882 static int set_flags(struct net_device *dev, u32 flags)
1884 return ethtool_op_set_flags(dev, flags, ETH_FLAG_RXHASH);
1887 static int get_rss_table(struct net_device *dev, struct ethtool_rxfh_indir *p)
1889 const struct port_info *pi = netdev_priv(dev);
1890 unsigned int n = min_t(unsigned int, p->size, pi->rss_size);
1892 p->size = pi->rss_size;
1894 p->ring_index[n] = pi->rss[n];
1898 static int set_rss_table(struct net_device *dev,
1899 const struct ethtool_rxfh_indir *p)
1902 struct port_info *pi = netdev_priv(dev);
1904 if (p->size != pi->rss_size)
1906 for (i = 0; i < p->size; i++)
1907 if (p->ring_index[i] >= pi->nqsets)
1909 for (i = 0; i < p->size; i++)
1910 pi->rss[i] = p->ring_index[i];
1911 if (pi->adapter->flags & FULL_INIT_DONE)
1912 return write_rss(pi, pi->rss);
1916 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1919 const struct port_info *pi = netdev_priv(dev);
1921 switch (info->cmd) {
1922 case ETHTOOL_GRXFH: {
1923 unsigned int v = pi->rss_mode;
1926 switch (info->flow_type) {
1928 if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
1929 info->data = RXH_IP_SRC | RXH_IP_DST |
1930 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1931 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1932 info->data = RXH_IP_SRC | RXH_IP_DST;
1935 if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
1936 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
1937 info->data = RXH_IP_SRC | RXH_IP_DST |
1938 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1939 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1940 info->data = RXH_IP_SRC | RXH_IP_DST;
1943 case AH_ESP_V4_FLOW:
1945 if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1946 info->data = RXH_IP_SRC | RXH_IP_DST;
1949 if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
1950 info->data = RXH_IP_SRC | RXH_IP_DST |
1951 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1952 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1953 info->data = RXH_IP_SRC | RXH_IP_DST;
1956 if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
1957 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
1958 info->data = RXH_IP_SRC | RXH_IP_DST |
1959 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1960 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1961 info->data = RXH_IP_SRC | RXH_IP_DST;
1964 case AH_ESP_V6_FLOW:
1966 if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1967 info->data = RXH_IP_SRC | RXH_IP_DST;
1972 case ETHTOOL_GRXRINGS:
1973 info->data = pi->nqsets;
1979 static struct ethtool_ops cxgb_ethtool_ops = {
1980 .get_settings = get_settings,
1981 .set_settings = set_settings,
1982 .get_drvinfo = get_drvinfo,
1983 .get_msglevel = get_msglevel,
1984 .set_msglevel = set_msglevel,
1985 .get_ringparam = get_sge_param,
1986 .set_ringparam = set_sge_param,
1987 .get_coalesce = get_coalesce,
1988 .set_coalesce = set_coalesce,
1989 .get_eeprom_len = get_eeprom_len,
1990 .get_eeprom = get_eeprom,
1991 .set_eeprom = set_eeprom,
1992 .get_pauseparam = get_pauseparam,
1993 .set_pauseparam = set_pauseparam,
1994 .get_rx_csum = get_rx_csum,
1995 .set_rx_csum = set_rx_csum,
1996 .set_tx_csum = ethtool_op_set_tx_ipv6_csum,
1997 .set_sg = ethtool_op_set_sg,
1998 .get_link = ethtool_op_get_link,
1999 .get_strings = get_strings,
2000 .phys_id = identify_port,
2001 .nway_reset = restart_autoneg,
2002 .get_sset_count = get_sset_count,
2003 .get_ethtool_stats = get_stats,
2004 .get_regs_len = get_regs_len,
2005 .get_regs = get_regs,
2009 .set_flags = set_flags,
2010 .get_rxnfc = get_rxnfc,
2011 .get_rxfh_indir = get_rss_table,
2012 .set_rxfh_indir = set_rss_table,
2013 .flash_device = set_flash,
2020 static int mem_open(struct inode *inode, struct file *file)
2022 file->private_data = inode->i_private;
2026 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
2030 loff_t avail = file->f_path.dentry->d_inode->i_size;
2031 unsigned int mem = (uintptr_t)file->private_data & 3;
2032 struct adapter *adap = file->private_data - mem;
2038 if (count > avail - pos)
2039 count = avail - pos;
2047 ret = t4_mc_read(adap, pos, data, NULL);
2049 ret = t4_edc_read(adap, mem, pos, data, NULL);
2053 ofst = pos % sizeof(data);
2054 len = min(count, sizeof(data) - ofst);
2055 if (copy_to_user(buf, (u8 *)data + ofst, len))
2062 count = pos - *ppos;
2067 static const struct file_operations mem_debugfs_fops = {
2068 .owner = THIS_MODULE,
2071 .llseek = default_llseek,
2074 static void __devinit add_debugfs_mem(struct adapter *adap, const char *name,
2075 unsigned int idx, unsigned int size_mb)
2079 de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2080 (void *)adap + idx, &mem_debugfs_fops);
2081 if (de && de->d_inode)
2082 de->d_inode->i_size = size_mb << 20;
2085 static int __devinit setup_debugfs(struct adapter *adap)
2089 if (IS_ERR_OR_NULL(adap->debugfs_root))
2092 i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2093 if (i & EDRAM0_ENABLE)
2094 add_debugfs_mem(adap, "edc0", MEM_EDC0, 5);
2095 if (i & EDRAM1_ENABLE)
2096 add_debugfs_mem(adap, "edc1", MEM_EDC1, 5);
2097 if (i & EXT_MEM_ENABLE)
2098 add_debugfs_mem(adap, "mc", MEM_MC,
2099 EXT_MEM_SIZE_GET(t4_read_reg(adap, MA_EXT_MEMORY_BAR)));
2101 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2107 * upper-layer driver support
2111 * Allocate an active-open TID and set it to the supplied value.
2113 int cxgb4_alloc_atid(struct tid_info *t, void *data)
2117 spin_lock_bh(&t->atid_lock);
2119 union aopen_entry *p = t->afree;
2121 atid = p - t->atid_tab;
2126 spin_unlock_bh(&t->atid_lock);
2129 EXPORT_SYMBOL(cxgb4_alloc_atid);
2132 * Release an active-open TID.
2134 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
2136 union aopen_entry *p = &t->atid_tab[atid];
2138 spin_lock_bh(&t->atid_lock);
2142 spin_unlock_bh(&t->atid_lock);
2144 EXPORT_SYMBOL(cxgb4_free_atid);
2147 * Allocate a server TID and set it to the supplied value.
2149 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
2153 spin_lock_bh(&t->stid_lock);
2154 if (family == PF_INET) {
2155 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
2156 if (stid < t->nstids)
2157 __set_bit(stid, t->stid_bmap);
2161 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
2166 t->stid_tab[stid].data = data;
2167 stid += t->stid_base;
2170 spin_unlock_bh(&t->stid_lock);
2173 EXPORT_SYMBOL(cxgb4_alloc_stid);
2176 * Release a server TID.
2178 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
2180 stid -= t->stid_base;
2181 spin_lock_bh(&t->stid_lock);
2182 if (family == PF_INET)
2183 __clear_bit(stid, t->stid_bmap);
2185 bitmap_release_region(t->stid_bmap, stid, 2);
2186 t->stid_tab[stid].data = NULL;
2188 spin_unlock_bh(&t->stid_lock);
2190 EXPORT_SYMBOL(cxgb4_free_stid);
2193 * Populate a TID_RELEASE WR. Caller must properly size the skb.
2195 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
2198 struct cpl_tid_release *req;
2200 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
2201 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
2202 INIT_TP_WR(req, tid);
2203 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
2207 * Queue a TID release request and if necessary schedule a work queue to
2210 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
2213 void **p = &t->tid_tab[tid];
2214 struct adapter *adap = container_of(t, struct adapter, tids);
2216 spin_lock_bh(&adap->tid_release_lock);
2217 *p = adap->tid_release_head;
2218 /* Low 2 bits encode the Tx channel number */
2219 adap->tid_release_head = (void **)((uintptr_t)p | chan);
2220 if (!adap->tid_release_task_busy) {
2221 adap->tid_release_task_busy = true;
2222 schedule_work(&adap->tid_release_task);
2224 spin_unlock_bh(&adap->tid_release_lock);
2228 * Process the list of pending TID release requests.
2230 static void process_tid_release_list(struct work_struct *work)
2232 struct sk_buff *skb;
2233 struct adapter *adap;
2235 adap = container_of(work, struct adapter, tid_release_task);
2237 spin_lock_bh(&adap->tid_release_lock);
2238 while (adap->tid_release_head) {
2239 void **p = adap->tid_release_head;
2240 unsigned int chan = (uintptr_t)p & 3;
2241 p = (void *)p - chan;
2243 adap->tid_release_head = *p;
2245 spin_unlock_bh(&adap->tid_release_lock);
2247 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
2249 schedule_timeout_uninterruptible(1);
2251 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
2252 t4_ofld_send(adap, skb);
2253 spin_lock_bh(&adap->tid_release_lock);
2255 adap->tid_release_task_busy = false;
2256 spin_unlock_bh(&adap->tid_release_lock);
2260 * Release a TID and inform HW. If we are unable to allocate the release
2261 * message we defer to a work queue.
2263 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
2266 struct sk_buff *skb;
2267 struct adapter *adap = container_of(t, struct adapter, tids);
2269 old = t->tid_tab[tid];
2270 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
2272 t->tid_tab[tid] = NULL;
2273 mk_tid_release(skb, chan, tid);
2274 t4_ofld_send(adap, skb);
2276 cxgb4_queue_tid_release(t, chan, tid);
2278 atomic_dec(&t->tids_in_use);
2280 EXPORT_SYMBOL(cxgb4_remove_tid);
2283 * Allocate and initialize the TID tables. Returns 0 on success.
2285 static int tid_init(struct tid_info *t)
2288 unsigned int natids = t->natids;
2290 size = t->ntids * sizeof(*t->tid_tab) + natids * sizeof(*t->atid_tab) +
2291 t->nstids * sizeof(*t->stid_tab) +
2292 BITS_TO_LONGS(t->nstids) * sizeof(long);
2293 t->tid_tab = t4_alloc_mem(size);
2297 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
2298 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
2299 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids];
2300 spin_lock_init(&t->stid_lock);
2301 spin_lock_init(&t->atid_lock);
2303 t->stids_in_use = 0;
2305 t->atids_in_use = 0;
2306 atomic_set(&t->tids_in_use, 0);
2308 /* Setup the free list for atid_tab and clear the stid bitmap. */
2311 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
2312 t->afree = t->atid_tab;
2314 bitmap_zero(t->stid_bmap, t->nstids);
2319 * cxgb4_create_server - create an IP server
2321 * @stid: the server TID
2322 * @sip: local IP address to bind server to
2323 * @sport: the server's TCP port
2324 * @queue: queue to direct messages from this server to
2326 * Create an IP server for the given port and address.
2327 * Returns <0 on error and one of the %NET_XMIT_* values on success.
2329 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
2330 __be32 sip, __be16 sport, unsigned int queue)
2333 struct sk_buff *skb;
2334 struct adapter *adap;
2335 struct cpl_pass_open_req *req;
2337 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
2341 adap = netdev2adap(dev);
2342 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
2344 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
2345 req->local_port = sport;
2346 req->peer_port = htons(0);
2347 req->local_ip = sip;
2348 req->peer_ip = htonl(0);
2349 chan = rxq_to_chan(&adap->sge, queue);
2350 req->opt0 = cpu_to_be64(TX_CHAN(chan));
2351 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
2352 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
2353 return t4_mgmt_tx(adap, skb);
2355 EXPORT_SYMBOL(cxgb4_create_server);
2358 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
2359 * @mtus: the HW MTU table
2360 * @mtu: the target MTU
2361 * @idx: index of selected entry in the MTU table
2363 * Returns the index and the value in the HW MTU table that is closest to
2364 * but does not exceed @mtu, unless @mtu is smaller than any value in the
2365 * table, in which case that smallest available value is selected.
2367 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
2372 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
2378 EXPORT_SYMBOL(cxgb4_best_mtu);
2381 * cxgb4_port_chan - get the HW channel of a port
2382 * @dev: the net device for the port
2384 * Return the HW Tx channel of the given port.
2386 unsigned int cxgb4_port_chan(const struct net_device *dev)
2388 return netdev2pinfo(dev)->tx_chan;
2390 EXPORT_SYMBOL(cxgb4_port_chan);
2393 * cxgb4_port_viid - get the VI id of a port
2394 * @dev: the net device for the port
2396 * Return the VI id of the given port.
2398 unsigned int cxgb4_port_viid(const struct net_device *dev)
2400 return netdev2pinfo(dev)->viid;
2402 EXPORT_SYMBOL(cxgb4_port_viid);
2405 * cxgb4_port_idx - get the index of a port
2406 * @dev: the net device for the port
2408 * Return the index of the given port.
2410 unsigned int cxgb4_port_idx(const struct net_device *dev)
2412 return netdev2pinfo(dev)->port_id;
2414 EXPORT_SYMBOL(cxgb4_port_idx);
2416 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
2417 struct tp_tcp_stats *v6)
2419 struct adapter *adap = pci_get_drvdata(pdev);
2421 spin_lock(&adap->stats_lock);
2422 t4_tp_get_tcp_stats(adap, v4, v6);
2423 spin_unlock(&adap->stats_lock);
2425 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
2427 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
2428 const unsigned int *pgsz_order)
2430 struct adapter *adap = netdev2adap(dev);
2432 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
2433 t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
2434 HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
2435 HPZ3(pgsz_order[3]));
2437 EXPORT_SYMBOL(cxgb4_iscsi_init);
2439 static struct pci_driver cxgb4_driver;
2441 static void check_neigh_update(struct neighbour *neigh)
2443 const struct device *parent;
2444 const struct net_device *netdev = neigh->dev;
2446 if (netdev->priv_flags & IFF_802_1Q_VLAN)
2447 netdev = vlan_dev_real_dev(netdev);
2448 parent = netdev->dev.parent;
2449 if (parent && parent->driver == &cxgb4_driver.driver)
2450 t4_l2t_update(dev_get_drvdata(parent), neigh);
2453 static int netevent_cb(struct notifier_block *nb, unsigned long event,
2457 case NETEVENT_NEIGH_UPDATE:
2458 check_neigh_update(data);
2460 case NETEVENT_PMTU_UPDATE:
2461 case NETEVENT_REDIRECT:
2468 static bool netevent_registered;
2469 static struct notifier_block cxgb4_netevent_nb = {
2470 .notifier_call = netevent_cb
2473 static void uld_attach(struct adapter *adap, unsigned int uld)
2476 struct cxgb4_lld_info lli;
2478 lli.pdev = adap->pdev;
2479 lli.l2t = adap->l2t;
2480 lli.tids = &adap->tids;
2481 lli.ports = adap->port;
2482 lli.vr = &adap->vres;
2483 lli.mtus = adap->params.mtus;
2484 if (uld == CXGB4_ULD_RDMA) {
2485 lli.rxq_ids = adap->sge.rdma_rxq;
2486 lli.nrxq = adap->sge.rdmaqs;
2487 } else if (uld == CXGB4_ULD_ISCSI) {
2488 lli.rxq_ids = adap->sge.ofld_rxq;
2489 lli.nrxq = adap->sge.ofldqsets;
2491 lli.ntxq = adap->sge.ofldqsets;
2492 lli.nchan = adap->params.nports;
2493 lli.nports = adap->params.nports;
2494 lli.wr_cred = adap->params.ofldq_wr_cred;
2495 lli.adapter_type = adap->params.rev;
2496 lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
2497 lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
2498 t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
2500 lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
2501 t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
2503 lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
2504 lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
2505 lli.fw_vers = adap->params.fw_vers;
2507 handle = ulds[uld].add(&lli);
2508 if (IS_ERR(handle)) {
2509 dev_warn(adap->pdev_dev,
2510 "could not attach to the %s driver, error %ld\n",
2511 uld_str[uld], PTR_ERR(handle));
2515 adap->uld_handle[uld] = handle;
2517 if (!netevent_registered) {
2518 register_netevent_notifier(&cxgb4_netevent_nb);
2519 netevent_registered = true;
2522 if (adap->flags & FULL_INIT_DONE)
2523 ulds[uld].state_change(handle, CXGB4_STATE_UP);
2526 static void attach_ulds(struct adapter *adap)
2530 mutex_lock(&uld_mutex);
2531 list_add_tail(&adap->list_node, &adapter_list);
2532 for (i = 0; i < CXGB4_ULD_MAX; i++)
2534 uld_attach(adap, i);
2535 mutex_unlock(&uld_mutex);
2538 static void detach_ulds(struct adapter *adap)
2542 mutex_lock(&uld_mutex);
2543 list_del(&adap->list_node);
2544 for (i = 0; i < CXGB4_ULD_MAX; i++)
2545 if (adap->uld_handle[i]) {
2546 ulds[i].state_change(adap->uld_handle[i],
2547 CXGB4_STATE_DETACH);
2548 adap->uld_handle[i] = NULL;
2550 if (netevent_registered && list_empty(&adapter_list)) {
2551 unregister_netevent_notifier(&cxgb4_netevent_nb);
2552 netevent_registered = false;
2554 mutex_unlock(&uld_mutex);
2557 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2561 mutex_lock(&uld_mutex);
2562 for (i = 0; i < CXGB4_ULD_MAX; i++)
2563 if (adap->uld_handle[i])
2564 ulds[i].state_change(adap->uld_handle[i], new_state);
2565 mutex_unlock(&uld_mutex);
2569 * cxgb4_register_uld - register an upper-layer driver
2570 * @type: the ULD type
2571 * @p: the ULD methods
2573 * Registers an upper-layer driver with this driver and notifies the ULD
2574 * about any presently available devices that support its type. Returns
2575 * %-EBUSY if a ULD of the same type is already registered.
2577 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
2580 struct adapter *adap;
2582 if (type >= CXGB4_ULD_MAX)
2584 mutex_lock(&uld_mutex);
2585 if (ulds[type].add) {
2590 list_for_each_entry(adap, &adapter_list, list_node)
2591 uld_attach(adap, type);
2592 out: mutex_unlock(&uld_mutex);
2595 EXPORT_SYMBOL(cxgb4_register_uld);
2598 * cxgb4_unregister_uld - unregister an upper-layer driver
2599 * @type: the ULD type
2601 * Unregisters an existing upper-layer driver.
2603 int cxgb4_unregister_uld(enum cxgb4_uld type)
2605 struct adapter *adap;
2607 if (type >= CXGB4_ULD_MAX)
2609 mutex_lock(&uld_mutex);
2610 list_for_each_entry(adap, &adapter_list, list_node)
2611 adap->uld_handle[type] = NULL;
2612 ulds[type].add = NULL;
2613 mutex_unlock(&uld_mutex);
2616 EXPORT_SYMBOL(cxgb4_unregister_uld);
2619 * cxgb_up - enable the adapter
2620 * @adap: adapter being enabled
2622 * Called when the first port is enabled, this function performs the
2623 * actions necessary to make an adapter operational, such as completing
2624 * the initialization of HW modules, and enabling interrupts.
2626 * Must be called with the rtnl lock held.
2628 static int cxgb_up(struct adapter *adap)
2632 err = setup_sge_queues(adap);
2635 err = setup_rss(adap);
2639 if (adap->flags & USING_MSIX) {
2640 name_msix_vecs(adap);
2641 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2642 adap->msix_info[0].desc, adap);
2646 err = request_msix_queue_irqs(adap);
2648 free_irq(adap->msix_info[0].vec, adap);
2652 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2653 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2660 t4_intr_enable(adap);
2661 adap->flags |= FULL_INIT_DONE;
2662 notify_ulds(adap, CXGB4_STATE_UP);
2666 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2668 t4_free_sge_resources(adap);
2672 static void cxgb_down(struct adapter *adapter)
2674 t4_intr_disable(adapter);
2675 cancel_work_sync(&adapter->tid_release_task);
2676 adapter->tid_release_task_busy = false;
2677 adapter->tid_release_head = NULL;
2679 if (adapter->flags & USING_MSIX) {
2680 free_msix_queue_irqs(adapter);
2681 free_irq(adapter->msix_info[0].vec, adapter);
2683 free_irq(adapter->pdev->irq, adapter);
2684 quiesce_rx(adapter);
2685 t4_sge_stop(adapter);
2686 t4_free_sge_resources(adapter);
2687 adapter->flags &= ~FULL_INIT_DONE;
2691 * net_device operations
2693 static int cxgb_open(struct net_device *dev)
2696 struct port_info *pi = netdev_priv(dev);
2697 struct adapter *adapter = pi->adapter;
2699 if (!(adapter->flags & FULL_INIT_DONE)) {
2700 err = cxgb_up(adapter);
2705 netif_set_real_num_tx_queues(dev, pi->nqsets);
2706 err = netif_set_real_num_rx_queues(dev, pi->nqsets);
2709 err = link_start(dev);
2711 netif_tx_start_all_queues(dev);
2715 static int cxgb_close(struct net_device *dev)
2717 struct port_info *pi = netdev_priv(dev);
2718 struct adapter *adapter = pi->adapter;
2720 netif_tx_stop_all_queues(dev);
2721 netif_carrier_off(dev);
2722 return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
2725 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
2726 struct rtnl_link_stats64 *ns)
2728 struct port_stats stats;
2729 struct port_info *p = netdev_priv(dev);
2730 struct adapter *adapter = p->adapter;
2732 spin_lock(&adapter->stats_lock);
2733 t4_get_port_stats(adapter, p->tx_chan, &stats);
2734 spin_unlock(&adapter->stats_lock);
2736 ns->tx_bytes = stats.tx_octets;
2737 ns->tx_packets = stats.tx_frames;
2738 ns->rx_bytes = stats.rx_octets;
2739 ns->rx_packets = stats.rx_frames;
2740 ns->multicast = stats.rx_mcast_frames;
2742 /* detailed rx_errors */
2743 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2745 ns->rx_over_errors = 0;
2746 ns->rx_crc_errors = stats.rx_fcs_err;
2747 ns->rx_frame_errors = stats.rx_symbol_err;
2748 ns->rx_fifo_errors = stats.rx_ovflow0 + stats.rx_ovflow1 +
2749 stats.rx_ovflow2 + stats.rx_ovflow3 +
2750 stats.rx_trunc0 + stats.rx_trunc1 +
2751 stats.rx_trunc2 + stats.rx_trunc3;
2752 ns->rx_missed_errors = 0;
2754 /* detailed tx_errors */
2755 ns->tx_aborted_errors = 0;
2756 ns->tx_carrier_errors = 0;
2757 ns->tx_fifo_errors = 0;
2758 ns->tx_heartbeat_errors = 0;
2759 ns->tx_window_errors = 0;
2761 ns->tx_errors = stats.tx_error_frames;
2762 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2763 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2767 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2770 int ret = 0, prtad, devad;
2771 struct port_info *pi = netdev_priv(dev);
2772 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2776 if (pi->mdio_addr < 0)
2778 data->phy_id = pi->mdio_addr;
2782 if (mdio_phy_id_is_c45(data->phy_id)) {
2783 prtad = mdio_phy_id_prtad(data->phy_id);
2784 devad = mdio_phy_id_devad(data->phy_id);
2785 } else if (data->phy_id < 32) {
2786 prtad = data->phy_id;
2788 data->reg_num &= 0x1f;
2792 mbox = pi->adapter->fn;
2793 if (cmd == SIOCGMIIREG)
2794 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2795 data->reg_num, &data->val_out);
2797 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2798 data->reg_num, data->val_in);
2806 static void cxgb_set_rxmode(struct net_device *dev)
2808 /* unfortunately we can't return errors to the stack */
2809 set_rxmode(dev, -1, false);
2812 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2815 struct port_info *pi = netdev_priv(dev);
2817 if (new_mtu < 81 || new_mtu > MAX_MTU) /* accommodate SACK */
2819 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
2826 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2829 struct sockaddr *addr = p;
2830 struct port_info *pi = netdev_priv(dev);
2832 if (!is_valid_ether_addr(addr->sa_data))
2835 ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
2836 pi->xact_addr_filt, addr->sa_data, true, true);
2840 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2841 pi->xact_addr_filt = ret;
2845 static void vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
2847 struct port_info *pi = netdev_priv(dev);
2850 t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1, -1, -1, -1,
2854 #ifdef CONFIG_NET_POLL_CONTROLLER
2855 static void cxgb_netpoll(struct net_device *dev)
2857 struct port_info *pi = netdev_priv(dev);
2858 struct adapter *adap = pi->adapter;
2860 if (adap->flags & USING_MSIX) {
2862 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2864 for (i = pi->nqsets; i; i--, rx++)
2865 t4_sge_intr_msix(0, &rx->rspq);
2867 t4_intr_handler(adap)(0, adap);
2871 static const struct net_device_ops cxgb4_netdev_ops = {
2872 .ndo_open = cxgb_open,
2873 .ndo_stop = cxgb_close,
2874 .ndo_start_xmit = t4_eth_xmit,
2875 .ndo_get_stats64 = cxgb_get_stats,
2876 .ndo_set_rx_mode = cxgb_set_rxmode,
2877 .ndo_set_mac_address = cxgb_set_mac_addr,
2878 .ndo_validate_addr = eth_validate_addr,
2879 .ndo_do_ioctl = cxgb_ioctl,
2880 .ndo_change_mtu = cxgb_change_mtu,
2881 .ndo_vlan_rx_register = vlan_rx_register,
2882 #ifdef CONFIG_NET_POLL_CONTROLLER
2883 .ndo_poll_controller = cxgb_netpoll,
2887 void t4_fatal_err(struct adapter *adap)
2889 t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
2890 t4_intr_disable(adap);
2891 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
2894 static void setup_memwin(struct adapter *adap)
2898 bar0 = pci_resource_start(adap->pdev, 0); /* truncation intentional */
2899 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
2900 (bar0 + MEMWIN0_BASE) | BIR(0) |
2901 WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
2902 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
2903 (bar0 + MEMWIN1_BASE) | BIR(0) |
2904 WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
2905 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
2906 (bar0 + MEMWIN2_BASE) | BIR(0) |
2907 WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
2908 if (adap->vres.ocq.size) {
2909 unsigned int start, sz_kb;
2911 start = pci_resource_start(adap->pdev, 2) +
2912 OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
2913 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
2915 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
2916 start | BIR(1) | WINDOW(ilog2(sz_kb)));
2918 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
2919 adap->vres.ocq.start);
2921 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
2925 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
2930 /* get device capabilities */
2931 memset(c, 0, sizeof(*c));
2932 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2933 FW_CMD_REQUEST | FW_CMD_READ);
2934 c->retval_len16 = htonl(FW_LEN16(*c));
2935 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
2939 /* select capabilities we'll be using */
2940 if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
2942 c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
2944 c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
2945 } else if (vf_acls) {
2946 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
2949 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2950 FW_CMD_REQUEST | FW_CMD_WRITE);
2951 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
2955 ret = t4_config_glbl_rss(adap, adap->fn,
2956 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
2957 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
2958 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
2962 ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
2963 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
2969 /* tweak some settings */
2970 t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
2971 t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
2972 t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
2973 v = t4_read_reg(adap, TP_PIO_DATA);
2974 t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
2976 /* get basic stuff going */
2977 return t4_early_init(adap, adap->fn);
2981 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
2983 #define MAX_ATIDS 8192U
2986 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
2988 static int adap_init0(struct adapter *adap)
2992 enum dev_state state;
2993 u32 params[7], val[7];
2994 struct fw_caps_config_cmd c;
2996 ret = t4_check_fw_version(adap);
2997 if (ret == -EINVAL || ret > 0) {
2998 if (upgrade_fw(adap) >= 0) /* recache FW version */
2999 ret = t4_check_fw_version(adap);
3004 /* contact FW, request master */
3005 ret = t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, &state);
3007 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
3013 ret = t4_fw_reset(adap, adap->fn, PIORSTMODE | PIORST);
3017 for (v = 0; v < SGE_NTIMERS - 1; v++)
3018 adap->sge.timer_val[v] = min(intr_holdoff[v], MAX_SGE_TIMERVAL);
3019 adap->sge.timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
3020 adap->sge.counter_val[0] = 1;
3021 for (v = 1; v < SGE_NCOUNTERS; v++)
3022 adap->sge.counter_val[v] = min(intr_cnt[v - 1],
3024 #define FW_PARAM_DEV(param) \
3025 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3026 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3028 params[0] = FW_PARAM_DEV(CCLK);
3029 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 1, params, val);
3032 adap->params.vpd.cclk = val[0];
3034 ret = adap_init1(adap, &c);
3038 #define FW_PARAM_PFVF(param) \
3039 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3040 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param) | \
3041 FW_PARAMS_PARAM_Y(adap->fn))
3043 params[0] = FW_PARAM_DEV(PORTVEC);
3044 params[1] = FW_PARAM_PFVF(L2T_START);
3045 params[2] = FW_PARAM_PFVF(L2T_END);
3046 params[3] = FW_PARAM_PFVF(FILTER_START);
3047 params[4] = FW_PARAM_PFVF(FILTER_END);
3048 params[5] = FW_PARAM_PFVF(IQFLINT_START);
3049 params[6] = FW_PARAM_PFVF(EQ_START);
3050 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 7, params, val);
3054 adap->tids.ftid_base = val[3];
3055 adap->tids.nftids = val[4] - val[3] + 1;
3056 adap->sge.ingr_start = val[5];
3057 adap->sge.egr_start = val[6];
3060 /* query offload-related parameters */
3061 params[0] = FW_PARAM_DEV(NTID);
3062 params[1] = FW_PARAM_PFVF(SERVER_START);
3063 params[2] = FW_PARAM_PFVF(SERVER_END);
3064 params[3] = FW_PARAM_PFVF(TDDP_START);
3065 params[4] = FW_PARAM_PFVF(TDDP_END);
3066 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3067 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3071 adap->tids.ntids = val[0];
3072 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
3073 adap->tids.stid_base = val[1];
3074 adap->tids.nstids = val[2] - val[1] + 1;
3075 adap->vres.ddp.start = val[3];
3076 adap->vres.ddp.size = val[4] - val[3] + 1;
3077 adap->params.ofldq_wr_cred = val[5];
3078 adap->params.offload = 1;
3081 params[0] = FW_PARAM_PFVF(STAG_START);
3082 params[1] = FW_PARAM_PFVF(STAG_END);
3083 params[2] = FW_PARAM_PFVF(RQ_START);
3084 params[3] = FW_PARAM_PFVF(RQ_END);
3085 params[4] = FW_PARAM_PFVF(PBL_START);
3086 params[5] = FW_PARAM_PFVF(PBL_END);
3087 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3091 adap->vres.stag.start = val[0];
3092 adap->vres.stag.size = val[1] - val[0] + 1;
3093 adap->vres.rq.start = val[2];
3094 adap->vres.rq.size = val[3] - val[2] + 1;
3095 adap->vres.pbl.start = val[4];
3096 adap->vres.pbl.size = val[5] - val[4] + 1;
3098 params[0] = FW_PARAM_PFVF(SQRQ_START);
3099 params[1] = FW_PARAM_PFVF(SQRQ_END);
3100 params[2] = FW_PARAM_PFVF(CQ_START);
3101 params[3] = FW_PARAM_PFVF(CQ_END);
3102 params[4] = FW_PARAM_PFVF(OCQ_START);
3103 params[5] = FW_PARAM_PFVF(OCQ_END);
3104 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3108 adap->vres.qp.start = val[0];
3109 adap->vres.qp.size = val[1] - val[0] + 1;
3110 adap->vres.cq.start = val[2];
3111 adap->vres.cq.size = val[3] - val[2] + 1;
3112 adap->vres.ocq.start = val[4];
3113 adap->vres.ocq.size = val[5] - val[4] + 1;
3116 params[0] = FW_PARAM_PFVF(ISCSI_START);
3117 params[1] = FW_PARAM_PFVF(ISCSI_END);
3118 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 2, params,
3122 adap->vres.iscsi.start = val[0];
3123 adap->vres.iscsi.size = val[1] - val[0] + 1;
3125 #undef FW_PARAM_PFVF
3128 adap->params.nports = hweight32(port_vec);
3129 adap->params.portvec = port_vec;
3130 adap->flags |= FW_OK;
3132 /* These are finalized by FW initialization, load their values now */
3133 v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
3134 adap->params.tp.tre = TIMERRESOLUTION_GET(v);
3135 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
3136 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3137 adap->params.b_wnd);
3139 #ifdef CONFIG_PCI_IOV
3141 * Provision resource limits for Virtual Functions. We currently
3142 * grant them all the same static resource limits except for the Port
3143 * Access Rights Mask which we're assigning based on the PF. All of
3144 * the static provisioning stuff for both the PF and VF really needs
3145 * to be managed in a persistent manner for each device which the
3146 * firmware controls.
3151 for (pf = 0; pf < ARRAY_SIZE(num_vf); pf++) {
3152 if (num_vf[pf] <= 0)
3155 /* VF numbering starts at 1! */
3156 for (vf = 1; vf <= num_vf[pf]; vf++) {
3157 ret = t4_cfg_pfvf(adap, adap->fn, pf, vf,
3158 VFRES_NEQ, VFRES_NETHCTRL,
3159 VFRES_NIQFLINT, VFRES_NIQ,
3160 VFRES_TC, VFRES_NVI,
3161 FW_PFVF_CMD_CMASK_MASK,
3162 pfvfres_pmask(adap, pf, vf),
3164 VFRES_R_CAPS, VFRES_WX_CAPS);
3166 dev_warn(adap->pdev_dev, "failed to "
3167 "provision pf/vf=%d/%d; "
3168 "err=%d\n", pf, vf, ret);
3178 * If a command timed out or failed with EIO FW does not operate within
3179 * its spec or something catastrophic happened to HW/FW, stop issuing
3182 bye: if (ret != -ETIMEDOUT && ret != -EIO)
3183 t4_fw_bye(adap, adap->fn);
3189 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
3190 pci_channel_state_t state)
3193 struct adapter *adap = pci_get_drvdata(pdev);
3199 adap->flags &= ~FW_OK;
3200 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
3201 for_each_port(adap, i) {
3202 struct net_device *dev = adap->port[i];
3204 netif_device_detach(dev);
3205 netif_carrier_off(dev);
3207 if (adap->flags & FULL_INIT_DONE)
3210 pci_disable_device(pdev);
3211 out: return state == pci_channel_io_perm_failure ?
3212 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
3215 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
3218 struct fw_caps_config_cmd c;
3219 struct adapter *adap = pci_get_drvdata(pdev);
3222 pci_restore_state(pdev);
3223 pci_save_state(pdev);
3224 return PCI_ERS_RESULT_RECOVERED;
3227 if (pci_enable_device(pdev)) {
3228 dev_err(&pdev->dev, "cannot reenable PCI device after reset\n");
3229 return PCI_ERS_RESULT_DISCONNECT;
3232 pci_set_master(pdev);
3233 pci_restore_state(pdev);
3234 pci_save_state(pdev);
3235 pci_cleanup_aer_uncorrect_error_status(pdev);
3237 if (t4_wait_dev_ready(adap) < 0)
3238 return PCI_ERS_RESULT_DISCONNECT;
3239 if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL))
3240 return PCI_ERS_RESULT_DISCONNECT;
3241 adap->flags |= FW_OK;
3242 if (adap_init1(adap, &c))
3243 return PCI_ERS_RESULT_DISCONNECT;
3245 for_each_port(adap, i) {
3246 struct port_info *p = adap2pinfo(adap, i);
3248 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
3251 return PCI_ERS_RESULT_DISCONNECT;
3253 p->xact_addr_filt = -1;
3256 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3257 adap->params.b_wnd);
3260 return PCI_ERS_RESULT_DISCONNECT;
3261 return PCI_ERS_RESULT_RECOVERED;
3264 static void eeh_resume(struct pci_dev *pdev)
3267 struct adapter *adap = pci_get_drvdata(pdev);
3273 for_each_port(adap, i) {
3274 struct net_device *dev = adap->port[i];
3276 if (netif_running(dev)) {
3278 cxgb_set_rxmode(dev);
3280 netif_device_attach(dev);
3285 static struct pci_error_handlers cxgb4_eeh = {
3286 .error_detected = eeh_err_detected,
3287 .slot_reset = eeh_slot_reset,
3288 .resume = eeh_resume,
3291 static inline bool is_10g_port(const struct link_config *lc)
3293 return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
3296 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
3297 unsigned int size, unsigned int iqe_size)
3299 q->intr_params = QINTR_TIMER_IDX(timer_idx) |
3300 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
3301 q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
3302 q->iqe_len = iqe_size;
3307 * Perform default configuration of DMA queues depending on the number and type
3308 * of ports we found and the number of available CPUs. Most settings can be
3309 * modified by the admin prior to actual use.
3311 static void __devinit cfg_queues(struct adapter *adap)
3313 struct sge *s = &adap->sge;
3314 int i, q10g = 0, n10g = 0, qidx = 0;
3316 for_each_port(adap, i)
3317 n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
3320 * We default to 1 queue per non-10G port and up to # of cores queues
3324 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
3325 if (q10g > num_online_cpus())
3326 q10g = num_online_cpus();
3328 for_each_port(adap, i) {
3329 struct port_info *pi = adap2pinfo(adap, i);
3331 pi->first_qset = qidx;
3332 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
3337 s->max_ethqsets = qidx; /* MSI-X may lower it later */
3339 if (is_offload(adap)) {
3341 * For offload we use 1 queue/channel if all ports are up to 1G,
3342 * otherwise we divide all available queues amongst the channels
3343 * capped by the number of available cores.
3346 i = min_t(int, ARRAY_SIZE(s->ofldrxq),
3348 s->ofldqsets = roundup(i, adap->params.nports);
3350 s->ofldqsets = adap->params.nports;
3351 /* For RDMA one Rx queue per channel suffices */
3352 s->rdmaqs = adap->params.nports;
3355 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
3356 struct sge_eth_rxq *r = &s->ethrxq[i];
3358 init_rspq(&r->rspq, 0, 0, 1024, 64);
3362 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
3363 s->ethtxq[i].q.size = 1024;
3365 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
3366 s->ctrlq[i].q.size = 512;
3368 for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
3369 s->ofldtxq[i].q.size = 1024;
3371 for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
3372 struct sge_ofld_rxq *r = &s->ofldrxq[i];
3374 init_rspq(&r->rspq, 0, 0, 1024, 64);
3375 r->rspq.uld = CXGB4_ULD_ISCSI;
3379 for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
3380 struct sge_ofld_rxq *r = &s->rdmarxq[i];
3382 init_rspq(&r->rspq, 0, 0, 511, 64);
3383 r->rspq.uld = CXGB4_ULD_RDMA;
3387 init_rspq(&s->fw_evtq, 6, 0, 512, 64);
3388 init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
3392 * Reduce the number of Ethernet queues across all ports to at most n.
3393 * n provides at least one queue per port.
3395 static void __devinit reduce_ethqs(struct adapter *adap, int n)
3398 struct port_info *pi;
3400 while (n < adap->sge.ethqsets)
3401 for_each_port(adap, i) {
3402 pi = adap2pinfo(adap, i);
3403 if (pi->nqsets > 1) {
3405 adap->sge.ethqsets--;
3406 if (adap->sge.ethqsets <= n)
3412 for_each_port(adap, i) {
3413 pi = adap2pinfo(adap, i);
3419 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
3420 #define EXTRA_VECS 2
3422 static int __devinit enable_msix(struct adapter *adap)
3425 int i, err, want, need;
3426 struct sge *s = &adap->sge;
3427 unsigned int nchan = adap->params.nports;
3428 struct msix_entry entries[MAX_INGQ + 1];
3430 for (i = 0; i < ARRAY_SIZE(entries); ++i)
3431 entries[i].entry = i;
3433 want = s->max_ethqsets + EXTRA_VECS;
3434 if (is_offload(adap)) {
3435 want += s->rdmaqs + s->ofldqsets;
3436 /* need nchan for each possible ULD */
3437 ofld_need = 2 * nchan;
3439 need = adap->params.nports + EXTRA_VECS + ofld_need;
3441 while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
3446 * Distribute available vectors to the various queue groups.
3447 * Every group gets its minimum requirement and NIC gets top
3448 * priority for leftovers.
3450 i = want - EXTRA_VECS - ofld_need;
3451 if (i < s->max_ethqsets) {
3452 s->max_ethqsets = i;
3453 if (i < s->ethqsets)
3454 reduce_ethqs(adap, i);
3456 if (is_offload(adap)) {
3457 i = want - EXTRA_VECS - s->max_ethqsets;
3458 i -= ofld_need - nchan;
3459 s->ofldqsets = (i / nchan) * nchan; /* round down */
3461 for (i = 0; i < want; ++i)
3462 adap->msix_info[i].vec = entries[i].vector;
3464 dev_info(adap->pdev_dev,
3465 "only %d MSI-X vectors left, not using MSI-X\n", err);
3471 static int __devinit init_rss(struct adapter *adap)
3475 for_each_port(adap, i) {
3476 struct port_info *pi = adap2pinfo(adap, i);
3478 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
3481 for (j = 0; j < pi->rss_size; j++)
3482 pi->rss[j] = j % pi->nqsets;
3487 static void __devinit print_port_info(struct adapter *adap)
3489 static const char *base[] = {
3490 "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
3491 "KX", "KR", "KR SFP+", "KR FEC"
3496 const char *spd = "";
3498 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
3500 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
3503 for_each_port(adap, i) {
3504 struct net_device *dev = adap->port[i];
3505 const struct port_info *pi = netdev_priv(dev);
3508 if (!test_bit(i, &adap->registered_device_map))
3511 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
3512 bufp += sprintf(bufp, "100/");
3513 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
3514 bufp += sprintf(bufp, "1000/");
3515 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
3516 bufp += sprintf(bufp, "10G/");
3519 sprintf(bufp, "BASE-%s", base[pi->port_type]);
3521 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
3522 adap->params.vpd.id, adap->params.rev,
3523 buf, is_offload(adap) ? "R" : "",
3524 adap->params.pci.width, spd,
3525 (adap->flags & USING_MSIX) ? " MSI-X" :
3526 (adap->flags & USING_MSI) ? " MSI" : "");
3527 if (adap->name == dev->name)
3528 netdev_info(dev, "S/N: %s, E/C: %s\n",
3529 adap->params.vpd.sn, adap->params.vpd.ec);
3534 * Free the following resources:
3535 * - memory used for tables
3538 * - resources FW is holding for us
3540 static void free_some_resources(struct adapter *adapter)
3544 t4_free_mem(adapter->l2t);
3545 t4_free_mem(adapter->tids.tid_tab);
3546 disable_msi(adapter);
3548 for_each_port(adapter, i)
3549 if (adapter->port[i]) {
3550 kfree(adap2pinfo(adapter, i)->rss);
3551 free_netdev(adapter->port[i]);
3553 if (adapter->flags & FW_OK)
3554 t4_fw_bye(adapter, adapter->fn);
3557 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
3558 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
3560 static int __devinit init_one(struct pci_dev *pdev,
3561 const struct pci_device_id *ent)
3564 struct port_info *pi;
3565 unsigned int highdma = 0;
3566 struct adapter *adapter = NULL;
3568 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
3570 err = pci_request_regions(pdev, KBUILD_MODNAME);
3572 /* Just info, some other driver may have claimed the device. */
3573 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
3577 /* We control everything through one PF */
3578 func = PCI_FUNC(pdev->devfn);
3579 if (func != ent->driver_data) {
3580 pci_save_state(pdev); /* to restore SR-IOV later */
3584 err = pci_enable_device(pdev);
3586 dev_err(&pdev->dev, "cannot enable PCI device\n");
3587 goto out_release_regions;
3590 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3591 highdma = NETIF_F_HIGHDMA;
3592 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3594 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
3595 "coherent allocations\n");
3596 goto out_disable_device;
3599 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3601 dev_err(&pdev->dev, "no usable DMA configuration\n");
3602 goto out_disable_device;
3606 pci_enable_pcie_error_reporting(pdev);
3607 pci_set_master(pdev);
3608 pci_save_state(pdev);
3610 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3613 goto out_disable_device;
3616 adapter->regs = pci_ioremap_bar(pdev, 0);
3617 if (!adapter->regs) {
3618 dev_err(&pdev->dev, "cannot map device registers\n");
3620 goto out_free_adapter;
3623 adapter->pdev = pdev;
3624 adapter->pdev_dev = &pdev->dev;
3626 adapter->name = pci_name(pdev);
3627 adapter->msg_enable = dflt_msg_enable;
3628 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
3630 spin_lock_init(&adapter->stats_lock);
3631 spin_lock_init(&adapter->tid_release_lock);
3633 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
3635 err = t4_prep_adapter(adapter);
3638 err = adap_init0(adapter);
3642 for_each_port(adapter, i) {
3643 struct net_device *netdev;
3645 netdev = alloc_etherdev_mq(sizeof(struct port_info),
3652 SET_NETDEV_DEV(netdev, &pdev->dev);
3654 adapter->port[i] = netdev;
3655 pi = netdev_priv(netdev);
3656 pi->adapter = adapter;
3657 pi->xact_addr_filt = -1;
3658 pi->rx_offload = RX_CSO;
3660 netif_carrier_off(netdev);
3661 netif_tx_stop_all_queues(netdev);
3662 netdev->irq = pdev->irq;
3664 netdev->features |= NETIF_F_SG | TSO_FLAGS;
3665 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3666 netdev->features |= NETIF_F_GRO | NETIF_F_RXHASH | highdma;
3667 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
3668 netdev->vlan_features = netdev->features & VLAN_FEAT;
3670 netdev->netdev_ops = &cxgb4_netdev_ops;
3671 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
3674 pci_set_drvdata(pdev, adapter);
3676 if (adapter->flags & FW_OK) {
3677 err = t4_port_init(adapter, func, func, 0);
3683 * Configure queues and allocate tables now, they can be needed as
3684 * soon as the first register_netdev completes.
3686 cfg_queues(adapter);
3688 adapter->l2t = t4_init_l2t();
3689 if (!adapter->l2t) {
3690 /* We tolerate a lack of L2T, giving up some functionality */
3691 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
3692 adapter->params.offload = 0;
3695 if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
3696 dev_warn(&pdev->dev, "could not allocate TID table, "
3698 adapter->params.offload = 0;
3701 /* See what interrupts we'll be using */
3702 if (msi > 1 && enable_msix(adapter) == 0)
3703 adapter->flags |= USING_MSIX;
3704 else if (msi > 0 && pci_enable_msi(pdev) == 0)
3705 adapter->flags |= USING_MSI;
3707 err = init_rss(adapter);
3712 * The card is now ready to go. If any errors occur during device
3713 * registration we do not fail the whole card but rather proceed only
3714 * with the ports we manage to register successfully. However we must
3715 * register at least one net device.
3717 for_each_port(adapter, i) {
3718 err = register_netdev(adapter->port[i]);
3720 dev_warn(&pdev->dev,
3721 "cannot register net device %s, skipping\n",
3722 adapter->port[i]->name);
3725 * Change the name we use for messages to the name of
3726 * the first successfully registered interface.
3728 if (!adapter->registered_device_map)
3729 adapter->name = adapter->port[i]->name;
3731 __set_bit(i, &adapter->registered_device_map);
3732 adapter->chan_map[adap2pinfo(adapter, i)->tx_chan] = i;
3735 if (!adapter->registered_device_map) {
3736 dev_err(&pdev->dev, "could not register any net devices\n");
3740 if (cxgb4_debugfs_root) {
3741 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
3742 cxgb4_debugfs_root);
3743 setup_debugfs(adapter);
3746 if (is_offload(adapter))
3747 attach_ulds(adapter);
3749 print_port_info(adapter);
3752 #ifdef CONFIG_PCI_IOV
3753 if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0)
3754 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
3755 dev_info(&pdev->dev,
3756 "instantiated %u virtual functions\n",
3762 free_some_resources(adapter);
3764 iounmap(adapter->regs);
3768 pci_disable_pcie_error_reporting(pdev);
3769 pci_disable_device(pdev);
3770 out_release_regions:
3771 pci_release_regions(pdev);
3772 pci_set_drvdata(pdev, NULL);
3776 static void __devexit remove_one(struct pci_dev *pdev)
3778 struct adapter *adapter = pci_get_drvdata(pdev);
3780 pci_disable_sriov(pdev);
3785 if (is_offload(adapter))
3786 detach_ulds(adapter);
3788 for_each_port(adapter, i)
3789 if (test_bit(i, &adapter->registered_device_map))
3790 unregister_netdev(adapter->port[i]);
3792 if (adapter->debugfs_root)
3793 debugfs_remove_recursive(adapter->debugfs_root);
3795 if (adapter->flags & FULL_INIT_DONE)
3798 free_some_resources(adapter);
3799 iounmap(adapter->regs);
3801 pci_disable_pcie_error_reporting(pdev);
3802 pci_disable_device(pdev);
3803 pci_release_regions(pdev);
3804 pci_set_drvdata(pdev, NULL);
3806 pci_release_regions(pdev);
3809 static struct pci_driver cxgb4_driver = {
3810 .name = KBUILD_MODNAME,
3811 .id_table = cxgb4_pci_tbl,
3813 .remove = __devexit_p(remove_one),
3814 .err_handler = &cxgb4_eeh,
3817 static int __init cxgb4_init_module(void)
3821 /* Debugfs support is optional, just warn if this fails */
3822 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3823 if (!cxgb4_debugfs_root)
3824 pr_warning("could not create debugfs entry, continuing\n");
3826 ret = pci_register_driver(&cxgb4_driver);
3828 debugfs_remove(cxgb4_debugfs_root);
3832 static void __exit cxgb4_cleanup_module(void)
3834 pci_unregister_driver(&cxgb4_driver);
3835 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
3838 module_init(cxgb4_init_module);
3839 module_exit(cxgb4_cleanup_module);