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, 0),
179 CH_DEVICE(0x4002, 0),
180 CH_DEVICE(0x4003, 0),
181 CH_DEVICE(0x4004, 0),
182 CH_DEVICE(0x4005, 0),
183 CH_DEVICE(0x4006, 0),
184 CH_DEVICE(0x4007, 0),
185 CH_DEVICE(0x4008, 0),
186 CH_DEVICE(0x4009, 0),
187 CH_DEVICE(0x400a, 0),
191 #define FW_FNAME "cxgb4/t4fw.bin"
193 MODULE_DESCRIPTION(DRV_DESC);
194 MODULE_AUTHOR("Chelsio Communications");
195 MODULE_LICENSE("Dual BSD/GPL");
196 MODULE_VERSION(DRV_VERSION);
197 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
198 MODULE_FIRMWARE(FW_FNAME);
200 static int dflt_msg_enable = DFLT_MSG_ENABLE;
202 module_param(dflt_msg_enable, int, 0644);
203 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
206 * The driver uses the best interrupt scheme available on a platform in the
207 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
208 * of these schemes the driver may consider as follows:
210 * msi = 2: choose from among all three options
211 * msi = 1: only consider MSI and INTx interrupts
212 * msi = 0: force INTx interrupts
216 module_param(msi, int, 0644);
217 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
220 * Queue interrupt hold-off timer values. Queues default to the first of these
223 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
225 module_param_array(intr_holdoff, uint, NULL, 0644);
226 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
227 "0..4 in microseconds");
229 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
231 module_param_array(intr_cnt, uint, NULL, 0644);
232 MODULE_PARM_DESC(intr_cnt,
233 "thresholds 1..3 for queue interrupt packet counters");
237 #ifdef CONFIG_PCI_IOV
238 module_param(vf_acls, bool, 0644);
239 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
241 static unsigned int num_vf[4];
243 module_param_array(num_vf, uint, NULL, 0644);
244 MODULE_PARM_DESC(num_vf, "number of VFs for each of PFs 0-3");
247 static struct dentry *cxgb4_debugfs_root;
249 static LIST_HEAD(adapter_list);
250 static DEFINE_MUTEX(uld_mutex);
251 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
252 static const char *uld_str[] = { "RDMA", "iSCSI" };
254 static void link_report(struct net_device *dev)
256 if (!netif_carrier_ok(dev))
257 netdev_info(dev, "link down\n");
259 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
261 const char *s = "10Mbps";
262 const struct port_info *p = netdev_priv(dev);
264 switch (p->link_cfg.speed) {
276 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
281 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
283 struct net_device *dev = adapter->port[port_id];
285 /* Skip changes from disabled ports. */
286 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
288 netif_carrier_on(dev);
290 netif_carrier_off(dev);
296 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
298 static const char *mod_str[] = {
299 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
302 const struct net_device *dev = adap->port[port_id];
303 const struct port_info *pi = netdev_priv(dev);
305 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
306 netdev_info(dev, "port module unplugged\n");
307 else if (pi->mod_type < ARRAY_SIZE(mod_str))
308 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
312 * Configure the exact and hash address filters to handle a port's multicast
313 * and secondary unicast MAC addresses.
315 static int set_addr_filters(const struct net_device *dev, bool sleep)
323 const struct netdev_hw_addr *ha;
324 int uc_cnt = netdev_uc_count(dev);
325 int mc_cnt = netdev_mc_count(dev);
326 const struct port_info *pi = netdev_priv(dev);
327 unsigned int mb = pi->adapter->fn;
329 /* first do the secondary unicast addresses */
330 netdev_for_each_uc_addr(ha, dev) {
331 addr[naddr++] = ha->addr;
332 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
333 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
334 naddr, addr, filt_idx, &uhash, sleep);
343 /* next set up the multicast addresses */
344 netdev_for_each_mc_addr(ha, dev) {
345 addr[naddr++] = ha->addr;
346 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
347 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
348 naddr, addr, filt_idx, &mhash, sleep);
357 return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
358 uhash | mhash, sleep);
362 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
363 * If @mtu is -1 it is left unchanged.
365 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
368 struct port_info *pi = netdev_priv(dev);
370 ret = set_addr_filters(dev, sleep_ok);
372 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
373 (dev->flags & IFF_PROMISC) ? 1 : 0,
374 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
380 * link_start - enable a port
381 * @dev: the port to enable
383 * Performs the MAC and PHY actions needed to enable a port.
385 static int link_start(struct net_device *dev)
388 struct port_info *pi = netdev_priv(dev);
389 unsigned int mb = pi->adapter->fn;
392 * We do not set address filters and promiscuity here, the stack does
393 * that step explicitly.
395 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
396 pi->vlan_grp != NULL, true);
398 ret = t4_change_mac(pi->adapter, mb, pi->viid,
399 pi->xact_addr_filt, dev->dev_addr, true,
402 pi->xact_addr_filt = ret;
407 ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
410 ret = t4_enable_vi(pi->adapter, mb, pi->viid, true, true);
415 * Response queue handler for the FW event queue.
417 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
418 const struct pkt_gl *gl)
420 u8 opcode = ((const struct rss_header *)rsp)->opcode;
422 rsp++; /* skip RSS header */
423 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
424 const struct cpl_sge_egr_update *p = (void *)rsp;
425 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
426 struct sge_txq *txq = q->adap->sge.egr_map[qid];
429 if ((u8 *)txq < (u8 *)q->adap->sge.ethrxq) {
430 struct sge_eth_txq *eq;
432 eq = container_of(txq, struct sge_eth_txq, q);
433 netif_tx_wake_queue(eq->txq);
435 struct sge_ofld_txq *oq;
437 oq = container_of(txq, struct sge_ofld_txq, q);
438 tasklet_schedule(&oq->qresume_tsk);
440 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
441 const struct cpl_fw6_msg *p = (void *)rsp;
444 t4_handle_fw_rpl(q->adap, p->data);
445 } else if (opcode == CPL_L2T_WRITE_RPL) {
446 const struct cpl_l2t_write_rpl *p = (void *)rsp;
448 do_l2t_write_rpl(q->adap, p);
450 dev_err(q->adap->pdev_dev,
451 "unexpected CPL %#x on FW event queue\n", opcode);
456 * uldrx_handler - response queue handler for ULD queues
457 * @q: the response queue that received the packet
458 * @rsp: the response queue descriptor holding the offload message
459 * @gl: the gather list of packet fragments
461 * Deliver an ingress offload packet to a ULD. All processing is done by
462 * the ULD, we just maintain statistics.
464 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
465 const struct pkt_gl *gl)
467 struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
469 if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
475 else if (gl == CXGB4_MSG_AN)
482 static void disable_msi(struct adapter *adapter)
484 if (adapter->flags & USING_MSIX) {
485 pci_disable_msix(adapter->pdev);
486 adapter->flags &= ~USING_MSIX;
487 } else if (adapter->flags & USING_MSI) {
488 pci_disable_msi(adapter->pdev);
489 adapter->flags &= ~USING_MSI;
494 * Interrupt handler for non-data events used with MSI-X.
496 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
498 struct adapter *adap = cookie;
500 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
503 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
505 t4_slow_intr_handler(adap);
510 * Name the MSI-X interrupts.
512 static void name_msix_vecs(struct adapter *adap)
514 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc) - 1;
516 /* non-data interrupts */
517 snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
518 adap->msix_info[0].desc[n] = 0;
521 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq", adap->name);
522 adap->msix_info[1].desc[n] = 0;
524 /* Ethernet queues */
525 for_each_port(adap, j) {
526 struct net_device *d = adap->port[j];
527 const struct port_info *pi = netdev_priv(d);
529 for (i = 0; i < pi->nqsets; i++, msi_idx++) {
530 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
532 adap->msix_info[msi_idx].desc[n] = 0;
537 for_each_ofldrxq(&adap->sge, i) {
538 snprintf(adap->msix_info[msi_idx].desc, n, "%s-ofld%d",
540 adap->msix_info[msi_idx++].desc[n] = 0;
542 for_each_rdmarxq(&adap->sge, i) {
543 snprintf(adap->msix_info[msi_idx].desc, n, "%s-rdma%d",
545 adap->msix_info[msi_idx++].desc[n] = 0;
549 static int request_msix_queue_irqs(struct adapter *adap)
551 struct sge *s = &adap->sge;
552 int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi = 2;
554 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
555 adap->msix_info[1].desc, &s->fw_evtq);
559 for_each_ethrxq(s, ethqidx) {
560 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
561 adap->msix_info[msi].desc,
562 &s->ethrxq[ethqidx].rspq);
567 for_each_ofldrxq(s, ofldqidx) {
568 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
569 adap->msix_info[msi].desc,
570 &s->ofldrxq[ofldqidx].rspq);
575 for_each_rdmarxq(s, rdmaqidx) {
576 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
577 adap->msix_info[msi].desc,
578 &s->rdmarxq[rdmaqidx].rspq);
586 while (--rdmaqidx >= 0)
587 free_irq(adap->msix_info[--msi].vec,
588 &s->rdmarxq[rdmaqidx].rspq);
589 while (--ofldqidx >= 0)
590 free_irq(adap->msix_info[--msi].vec,
591 &s->ofldrxq[ofldqidx].rspq);
592 while (--ethqidx >= 0)
593 free_irq(adap->msix_info[--msi].vec, &s->ethrxq[ethqidx].rspq);
594 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
598 static void free_msix_queue_irqs(struct adapter *adap)
601 struct sge *s = &adap->sge;
603 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
604 for_each_ethrxq(s, i)
605 free_irq(adap->msix_info[msi++].vec, &s->ethrxq[i].rspq);
606 for_each_ofldrxq(s, i)
607 free_irq(adap->msix_info[msi++].vec, &s->ofldrxq[i].rspq);
608 for_each_rdmarxq(s, i)
609 free_irq(adap->msix_info[msi++].vec, &s->rdmarxq[i].rspq);
613 * write_rss - write the RSS table for a given port
615 * @queues: array of queue indices for RSS
617 * Sets up the portion of the HW RSS table for the port's VI to distribute
618 * packets to the Rx queues in @queues.
620 static int write_rss(const struct port_info *pi, const u16 *queues)
624 const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
626 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
630 /* map the queue indices to queue ids */
631 for (i = 0; i < pi->rss_size; i++, queues++)
632 rss[i] = q[*queues].rspq.abs_id;
634 err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
635 pi->rss_size, rss, pi->rss_size);
641 * setup_rss - configure RSS
644 * Sets up RSS for each port.
646 static int setup_rss(struct adapter *adap)
650 for_each_port(adap, i) {
651 const struct port_info *pi = adap2pinfo(adap, i);
653 err = write_rss(pi, pi->rss);
661 * Wait until all NAPI handlers are descheduled.
663 static void quiesce_rx(struct adapter *adap)
667 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
668 struct sge_rspq *q = adap->sge.ingr_map[i];
671 napi_disable(&q->napi);
676 * Enable NAPI scheduling and interrupt generation for all Rx queues.
678 static void enable_rx(struct adapter *adap)
682 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
683 struct sge_rspq *q = adap->sge.ingr_map[i];
688 napi_enable(&q->napi);
689 /* 0-increment GTS to start the timer and enable interrupts */
690 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
691 SEINTARM(q->intr_params) |
692 INGRESSQID(q->cntxt_id));
697 * setup_sge_queues - configure SGE Tx/Rx/response queues
700 * Determines how many sets of SGE queues to use and initializes them.
701 * We support multiple queue sets per port if we have MSI-X, otherwise
702 * just one queue set per port.
704 static int setup_sge_queues(struct adapter *adap)
706 int err, msi_idx, i, j;
707 struct sge *s = &adap->sge;
709 bitmap_zero(s->starving_fl, MAX_EGRQ);
710 bitmap_zero(s->txq_maperr, MAX_EGRQ);
712 if (adap->flags & USING_MSIX)
713 msi_idx = 1; /* vector 0 is for non-queue interrupts */
715 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
719 msi_idx = -((int)s->intrq.abs_id + 1);
722 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
723 msi_idx, NULL, fwevtq_handler);
725 freeout: t4_free_sge_resources(adap);
729 for_each_port(adap, i) {
730 struct net_device *dev = adap->port[i];
731 struct port_info *pi = netdev_priv(dev);
732 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
733 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
735 for (j = 0; j < pi->nqsets; j++, q++) {
738 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
744 memset(&q->stats, 0, sizeof(q->stats));
746 for (j = 0; j < pi->nqsets; j++, t++) {
747 err = t4_sge_alloc_eth_txq(adap, t, dev,
748 netdev_get_tx_queue(dev, j),
749 s->fw_evtq.cntxt_id);
755 j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
756 for_each_ofldrxq(s, i) {
757 struct sge_ofld_rxq *q = &s->ofldrxq[i];
758 struct net_device *dev = adap->port[i / j];
762 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
763 &q->fl, uldrx_handler);
766 memset(&q->stats, 0, sizeof(q->stats));
767 s->ofld_rxq[i] = q->rspq.abs_id;
768 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
769 s->fw_evtq.cntxt_id);
774 for_each_rdmarxq(s, i) {
775 struct sge_ofld_rxq *q = &s->rdmarxq[i];
779 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
780 msi_idx, &q->fl, uldrx_handler);
783 memset(&q->stats, 0, sizeof(q->stats));
784 s->rdma_rxq[i] = q->rspq.abs_id;
787 for_each_port(adap, i) {
789 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
790 * have RDMA queues, and that's the right value.
792 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
794 s->rdmarxq[i].rspq.cntxt_id);
799 t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
800 RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
801 QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
806 * Returns 0 if new FW was successfully loaded, a positive errno if a load was
807 * started but failed, and a negative errno if flash load couldn't start.
809 static int upgrade_fw(struct adapter *adap)
813 const struct fw_hdr *hdr;
814 const struct firmware *fw;
815 struct device *dev = adap->pdev_dev;
817 ret = request_firmware(&fw, FW_FNAME, dev);
819 dev_err(dev, "unable to load firmware image " FW_FNAME
820 ", error %d\n", ret);
824 hdr = (const struct fw_hdr *)fw->data;
825 vers = ntohl(hdr->fw_ver);
826 if (FW_HDR_FW_VER_MAJOR_GET(vers) != FW_VERSION_MAJOR) {
827 ret = -EINVAL; /* wrong major version, won't do */
832 * If the flash FW is unusable or we found something newer, load it.
834 if (FW_HDR_FW_VER_MAJOR_GET(adap->params.fw_vers) != FW_VERSION_MAJOR ||
835 vers > adap->params.fw_vers) {
836 ret = -t4_load_fw(adap, fw->data, fw->size);
838 dev_info(dev, "firmware upgraded to version %pI4 from "
839 FW_FNAME "\n", &hdr->fw_ver);
841 out: release_firmware(fw);
846 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
847 * The allocated memory is cleared.
849 void *t4_alloc_mem(size_t size)
851 void *p = kmalloc(size, GFP_KERNEL);
861 * Free memory allocated through alloc_mem().
863 void t4_free_mem(void *addr)
865 if (is_vmalloc_addr(addr))
871 static inline int is_offload(const struct adapter *adap)
873 return adap->params.offload;
877 * Implementation of ethtool operations.
880 static u32 get_msglevel(struct net_device *dev)
882 return netdev2adap(dev)->msg_enable;
885 static void set_msglevel(struct net_device *dev, u32 val)
887 netdev2adap(dev)->msg_enable = val;
890 static char stats_strings[][ETH_GSTRING_LEN] = {
893 "TxBroadcastFrames ",
894 "TxMulticastFrames ",
902 "TxFrames512To1023 ",
903 "TxFrames1024To1518 ",
904 "TxFrames1519ToMax ",
919 "RxBroadcastFrames ",
920 "RxMulticastFrames ",
934 "RxFrames512To1023 ",
935 "RxFrames1024To1518 ",
936 "RxFrames1519ToMax ",
948 "RxBG0FramesDropped ",
949 "RxBG1FramesDropped ",
950 "RxBG2FramesDropped ",
951 "RxBG3FramesDropped ",
966 static int get_sset_count(struct net_device *dev, int sset)
970 return ARRAY_SIZE(stats_strings);
976 #define T4_REGMAP_SIZE (160 * 1024)
978 static int get_regs_len(struct net_device *dev)
980 return T4_REGMAP_SIZE;
983 static int get_eeprom_len(struct net_device *dev)
988 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
990 struct adapter *adapter = netdev2adap(dev);
992 strcpy(info->driver, KBUILD_MODNAME);
993 strcpy(info->version, DRV_VERSION);
994 strcpy(info->bus_info, pci_name(adapter->pdev));
996 if (!adapter->params.fw_vers)
997 strcpy(info->fw_version, "N/A");
999 snprintf(info->fw_version, sizeof(info->fw_version),
1000 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1001 FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1002 FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1003 FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1004 FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1005 FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1006 FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1007 FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1008 FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1011 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1013 if (stringset == ETH_SS_STATS)
1014 memcpy(data, stats_strings, sizeof(stats_strings));
1018 * port stats maintained per queue of the port. They should be in the same
1019 * order as in stats_strings above.
1021 struct queue_port_stats {
1031 static void collect_sge_port_stats(const struct adapter *adap,
1032 const struct port_info *p, struct queue_port_stats *s)
1035 const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1036 const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1038 memset(s, 0, sizeof(*s));
1039 for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1041 s->tx_csum += tx->tx_cso;
1042 s->rx_csum += rx->stats.rx_cso;
1043 s->vlan_ex += rx->stats.vlan_ex;
1044 s->vlan_ins += tx->vlan_ins;
1045 s->gro_pkts += rx->stats.lro_pkts;
1046 s->gro_merged += rx->stats.lro_merged;
1050 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1053 struct port_info *pi = netdev_priv(dev);
1054 struct adapter *adapter = pi->adapter;
1056 t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1058 data += sizeof(struct port_stats) / sizeof(u64);
1059 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1063 * Return a version number to identify the type of adapter. The scheme is:
1064 * - bits 0..9: chip version
1065 * - bits 10..15: chip revision
1066 * - bits 16..23: register dump version
1068 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1070 return 4 | (ap->params.rev << 10) | (1 << 16);
1073 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1076 u32 *p = buf + start;
1078 for ( ; start <= end; start += sizeof(u32))
1079 *p++ = t4_read_reg(ap, start);
1082 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1085 static const unsigned int reg_ranges[] = {
1306 struct adapter *ap = netdev2adap(dev);
1308 regs->version = mk_adap_vers(ap);
1310 memset(buf, 0, T4_REGMAP_SIZE);
1311 for (i = 0; i < ARRAY_SIZE(reg_ranges); i += 2)
1312 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
1315 static int restart_autoneg(struct net_device *dev)
1317 struct port_info *p = netdev_priv(dev);
1319 if (!netif_running(dev))
1321 if (p->link_cfg.autoneg != AUTONEG_ENABLE)
1323 t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
1327 static int identify_port(struct net_device *dev, u32 data)
1329 struct adapter *adap = netdev2adap(dev);
1332 data = 2; /* default to 2 seconds */
1334 return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid,
1338 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
1342 if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
1343 type == FW_PORT_TYPE_BT_XAUI) {
1345 if (caps & FW_PORT_CAP_SPEED_100M)
1346 v |= SUPPORTED_100baseT_Full;
1347 if (caps & FW_PORT_CAP_SPEED_1G)
1348 v |= SUPPORTED_1000baseT_Full;
1349 if (caps & FW_PORT_CAP_SPEED_10G)
1350 v |= SUPPORTED_10000baseT_Full;
1351 } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
1352 v |= SUPPORTED_Backplane;
1353 if (caps & FW_PORT_CAP_SPEED_1G)
1354 v |= SUPPORTED_1000baseKX_Full;
1355 if (caps & FW_PORT_CAP_SPEED_10G)
1356 v |= SUPPORTED_10000baseKX4_Full;
1357 } else if (type == FW_PORT_TYPE_KR)
1358 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
1359 else if (type == FW_PORT_TYPE_BP_AP)
1360 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC;
1361 else if (type == FW_PORT_TYPE_FIBER_XFI ||
1362 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
1363 v |= SUPPORTED_FIBRE;
1365 if (caps & FW_PORT_CAP_ANEG)
1366 v |= SUPPORTED_Autoneg;
1370 static unsigned int to_fw_linkcaps(unsigned int caps)
1374 if (caps & ADVERTISED_100baseT_Full)
1375 v |= FW_PORT_CAP_SPEED_100M;
1376 if (caps & ADVERTISED_1000baseT_Full)
1377 v |= FW_PORT_CAP_SPEED_1G;
1378 if (caps & ADVERTISED_10000baseT_Full)
1379 v |= FW_PORT_CAP_SPEED_10G;
1383 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1385 const struct port_info *p = netdev_priv(dev);
1387 if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
1388 p->port_type == FW_PORT_TYPE_BT_XFI ||
1389 p->port_type == FW_PORT_TYPE_BT_XAUI)
1390 cmd->port = PORT_TP;
1391 else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
1392 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
1393 cmd->port = PORT_FIBRE;
1394 else if (p->port_type == FW_PORT_TYPE_SFP) {
1395 if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
1396 p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
1397 cmd->port = PORT_DA;
1399 cmd->port = PORT_FIBRE;
1401 cmd->port = PORT_OTHER;
1403 if (p->mdio_addr >= 0) {
1404 cmd->phy_address = p->mdio_addr;
1405 cmd->transceiver = XCVR_EXTERNAL;
1406 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
1407 MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
1409 cmd->phy_address = 0; /* not really, but no better option */
1410 cmd->transceiver = XCVR_INTERNAL;
1411 cmd->mdio_support = 0;
1414 cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
1415 cmd->advertising = from_fw_linkcaps(p->port_type,
1416 p->link_cfg.advertising);
1417 cmd->speed = netif_carrier_ok(dev) ? p->link_cfg.speed : 0;
1418 cmd->duplex = DUPLEX_FULL;
1419 cmd->autoneg = p->link_cfg.autoneg;
1425 static unsigned int speed_to_caps(int speed)
1427 if (speed == SPEED_100)
1428 return FW_PORT_CAP_SPEED_100M;
1429 if (speed == SPEED_1000)
1430 return FW_PORT_CAP_SPEED_1G;
1431 if (speed == SPEED_10000)
1432 return FW_PORT_CAP_SPEED_10G;
1436 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1439 struct port_info *p = netdev_priv(dev);
1440 struct link_config *lc = &p->link_cfg;
1442 if (cmd->duplex != DUPLEX_FULL) /* only full-duplex supported */
1445 if (!(lc->supported & FW_PORT_CAP_ANEG)) {
1447 * PHY offers a single speed. See if that's what's
1450 if (cmd->autoneg == AUTONEG_DISABLE &&
1451 (lc->supported & speed_to_caps(cmd->speed)))
1456 if (cmd->autoneg == AUTONEG_DISABLE) {
1457 cap = speed_to_caps(cmd->speed);
1459 if (!(lc->supported & cap) || cmd->speed == SPEED_1000 ||
1460 cmd->speed == SPEED_10000)
1462 lc->requested_speed = cap;
1463 lc->advertising = 0;
1465 cap = to_fw_linkcaps(cmd->advertising);
1466 if (!(lc->supported & cap))
1468 lc->requested_speed = 0;
1469 lc->advertising = cap | FW_PORT_CAP_ANEG;
1471 lc->autoneg = cmd->autoneg;
1473 if (netif_running(dev))
1474 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
1479 static void get_pauseparam(struct net_device *dev,
1480 struct ethtool_pauseparam *epause)
1482 struct port_info *p = netdev_priv(dev);
1484 epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1485 epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
1486 epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
1489 static int set_pauseparam(struct net_device *dev,
1490 struct ethtool_pauseparam *epause)
1492 struct port_info *p = netdev_priv(dev);
1493 struct link_config *lc = &p->link_cfg;
1495 if (epause->autoneg == AUTONEG_DISABLE)
1496 lc->requested_fc = 0;
1497 else if (lc->supported & FW_PORT_CAP_ANEG)
1498 lc->requested_fc = PAUSE_AUTONEG;
1502 if (epause->rx_pause)
1503 lc->requested_fc |= PAUSE_RX;
1504 if (epause->tx_pause)
1505 lc->requested_fc |= PAUSE_TX;
1506 if (netif_running(dev))
1507 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
1512 static u32 get_rx_csum(struct net_device *dev)
1514 struct port_info *p = netdev_priv(dev);
1516 return p->rx_offload & RX_CSO;
1519 static int set_rx_csum(struct net_device *dev, u32 data)
1521 struct port_info *p = netdev_priv(dev);
1524 p->rx_offload |= RX_CSO;
1526 p->rx_offload &= ~RX_CSO;
1530 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1532 const struct port_info *pi = netdev_priv(dev);
1533 const struct sge *s = &pi->adapter->sge;
1535 e->rx_max_pending = MAX_RX_BUFFERS;
1536 e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1537 e->rx_jumbo_max_pending = 0;
1538 e->tx_max_pending = MAX_TXQ_ENTRIES;
1540 e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
1541 e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1542 e->rx_jumbo_pending = 0;
1543 e->tx_pending = s->ethtxq[pi->first_qset].q.size;
1546 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1549 const struct port_info *pi = netdev_priv(dev);
1550 struct adapter *adapter = pi->adapter;
1551 struct sge *s = &adapter->sge;
1553 if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
1554 e->tx_pending > MAX_TXQ_ENTRIES ||
1555 e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1556 e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1557 e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
1560 if (adapter->flags & FULL_INIT_DONE)
1563 for (i = 0; i < pi->nqsets; ++i) {
1564 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
1565 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
1566 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
1571 static int closest_timer(const struct sge *s, int time)
1573 int i, delta, match = 0, min_delta = INT_MAX;
1575 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1576 delta = time - s->timer_val[i];
1579 if (delta < min_delta) {
1587 static int closest_thres(const struct sge *s, int thres)
1589 int i, delta, match = 0, min_delta = INT_MAX;
1591 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1592 delta = thres - s->counter_val[i];
1595 if (delta < min_delta) {
1604 * Return a queue's interrupt hold-off time in us. 0 means no timer.
1606 static unsigned int qtimer_val(const struct adapter *adap,
1607 const struct sge_rspq *q)
1609 unsigned int idx = q->intr_params >> 1;
1611 return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
1615 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1616 * @adap: the adapter
1618 * @us: the hold-off time in us, or 0 to disable timer
1619 * @cnt: the hold-off packet count, or 0 to disable counter
1621 * Sets an Rx queue's interrupt hold-off time and packet count. At least
1622 * one of the two needs to be enabled for the queue to generate interrupts.
1624 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
1625 unsigned int us, unsigned int cnt)
1627 if ((us | cnt) == 0)
1634 new_idx = closest_thres(&adap->sge, cnt);
1635 if (q->desc && q->pktcnt_idx != new_idx) {
1636 /* the queue has already been created, update it */
1637 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1638 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1639 FW_PARAMS_PARAM_YZ(q->cntxt_id);
1640 err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
1645 q->pktcnt_idx = new_idx;
1648 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1649 q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
1653 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1655 const struct port_info *pi = netdev_priv(dev);
1656 struct adapter *adap = pi->adapter;
1658 return set_rxq_intr_params(adap, &adap->sge.ethrxq[pi->first_qset].rspq,
1659 c->rx_coalesce_usecs, c->rx_max_coalesced_frames);
1662 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1664 const struct port_info *pi = netdev_priv(dev);
1665 const struct adapter *adap = pi->adapter;
1666 const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
1668 c->rx_coalesce_usecs = qtimer_val(adap, rq);
1669 c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
1670 adap->sge.counter_val[rq->pktcnt_idx] : 0;
1675 * Translate a physical EEPROM address to virtual. The first 1K is accessed
1676 * through virtual addresses starting at 31K, the rest is accessed through
1677 * virtual addresses starting at 0. This mapping is correct only for PF0.
1679 static int eeprom_ptov(unsigned int phys_addr)
1681 if (phys_addr < 1024)
1682 return phys_addr + (31 << 10);
1683 if (phys_addr < EEPROMSIZE)
1684 return phys_addr - 1024;
1689 * The next two routines implement eeprom read/write from physical addresses.
1690 * The physical->virtual translation is correct only for PF0.
1692 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
1694 int vaddr = eeprom_ptov(phys_addr);
1697 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
1698 return vaddr < 0 ? vaddr : 0;
1701 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
1703 int vaddr = eeprom_ptov(phys_addr);
1706 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
1707 return vaddr < 0 ? vaddr : 0;
1710 #define EEPROM_MAGIC 0x38E2F10C
1712 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
1716 struct adapter *adapter = netdev2adap(dev);
1718 u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
1722 e->magic = EEPROM_MAGIC;
1723 for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
1724 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
1727 memcpy(data, buf + e->offset, e->len);
1732 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
1737 u32 aligned_offset, aligned_len, *p;
1738 struct adapter *adapter = netdev2adap(dev);
1740 if (eeprom->magic != EEPROM_MAGIC)
1743 aligned_offset = eeprom->offset & ~3;
1744 aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
1746 if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
1748 * RMW possibly needed for first or last words.
1750 buf = kmalloc(aligned_len, GFP_KERNEL);
1753 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
1754 if (!err && aligned_len > 4)
1755 err = eeprom_rd_phys(adapter,
1756 aligned_offset + aligned_len - 4,
1757 (u32 *)&buf[aligned_len - 4]);
1760 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
1764 err = t4_seeprom_wp(adapter, false);
1768 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
1769 err = eeprom_wr_phys(adapter, aligned_offset, *p);
1770 aligned_offset += 4;
1774 err = t4_seeprom_wp(adapter, true);
1781 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
1784 const struct firmware *fw;
1785 struct adapter *adap = netdev2adap(netdev);
1787 ef->data[sizeof(ef->data) - 1] = '\0';
1788 ret = request_firmware(&fw, ef->data, adap->pdev_dev);
1792 ret = t4_load_fw(adap, fw->data, fw->size);
1793 release_firmware(fw);
1795 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
1799 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
1800 #define BCAST_CRC 0xa0ccc1a6
1802 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1804 wol->supported = WAKE_BCAST | WAKE_MAGIC;
1805 wol->wolopts = netdev2adap(dev)->wol;
1806 memset(&wol->sopass, 0, sizeof(wol->sopass));
1809 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1812 struct port_info *pi = netdev_priv(dev);
1814 if (wol->wolopts & ~WOL_SUPPORTED)
1816 t4_wol_magic_enable(pi->adapter, pi->tx_chan,
1817 (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
1818 if (wol->wolopts & WAKE_BCAST) {
1819 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
1822 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
1823 ~6ULL, ~0ULL, BCAST_CRC, true);
1825 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
1829 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
1831 static int set_tso(struct net_device *dev, u32 value)
1834 dev->features |= TSO_FLAGS;
1836 dev->features &= ~TSO_FLAGS;
1840 static int set_flags(struct net_device *dev, u32 flags)
1842 return ethtool_op_set_flags(dev, flags, ETH_FLAG_RXHASH);
1845 static int get_rss_table(struct net_device *dev, struct ethtool_rxfh_indir *p)
1847 const struct port_info *pi = netdev_priv(dev);
1848 unsigned int n = min_t(unsigned int, p->size, pi->rss_size);
1850 p->size = pi->rss_size;
1852 p->ring_index[n] = pi->rss[n];
1856 static int set_rss_table(struct net_device *dev,
1857 const struct ethtool_rxfh_indir *p)
1860 struct port_info *pi = netdev_priv(dev);
1862 if (p->size != pi->rss_size)
1864 for (i = 0; i < p->size; i++)
1865 if (p->ring_index[i] >= pi->nqsets)
1867 for (i = 0; i < p->size; i++)
1868 pi->rss[i] = p->ring_index[i];
1869 if (pi->adapter->flags & FULL_INIT_DONE)
1870 return write_rss(pi, pi->rss);
1874 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1877 const struct port_info *pi = netdev_priv(dev);
1879 switch (info->cmd) {
1880 case ETHTOOL_GRXFH: {
1881 unsigned int v = pi->rss_mode;
1884 switch (info->flow_type) {
1886 if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
1887 info->data = RXH_IP_SRC | RXH_IP_DST |
1888 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1889 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1890 info->data = RXH_IP_SRC | RXH_IP_DST;
1893 if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
1894 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
1895 info->data = RXH_IP_SRC | RXH_IP_DST |
1896 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1897 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1898 info->data = RXH_IP_SRC | RXH_IP_DST;
1901 case AH_ESP_V4_FLOW:
1903 if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1904 info->data = RXH_IP_SRC | RXH_IP_DST;
1907 if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
1908 info->data = RXH_IP_SRC | RXH_IP_DST |
1909 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1910 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1911 info->data = RXH_IP_SRC | RXH_IP_DST;
1914 if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
1915 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
1916 info->data = RXH_IP_SRC | RXH_IP_DST |
1917 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1918 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1919 info->data = RXH_IP_SRC | RXH_IP_DST;
1922 case AH_ESP_V6_FLOW:
1924 if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1925 info->data = RXH_IP_SRC | RXH_IP_DST;
1930 case ETHTOOL_GRXRINGS:
1931 info->data = pi->nqsets;
1937 static struct ethtool_ops cxgb_ethtool_ops = {
1938 .get_settings = get_settings,
1939 .set_settings = set_settings,
1940 .get_drvinfo = get_drvinfo,
1941 .get_msglevel = get_msglevel,
1942 .set_msglevel = set_msglevel,
1943 .get_ringparam = get_sge_param,
1944 .set_ringparam = set_sge_param,
1945 .get_coalesce = get_coalesce,
1946 .set_coalesce = set_coalesce,
1947 .get_eeprom_len = get_eeprom_len,
1948 .get_eeprom = get_eeprom,
1949 .set_eeprom = set_eeprom,
1950 .get_pauseparam = get_pauseparam,
1951 .set_pauseparam = set_pauseparam,
1952 .get_rx_csum = get_rx_csum,
1953 .set_rx_csum = set_rx_csum,
1954 .set_tx_csum = ethtool_op_set_tx_ipv6_csum,
1955 .set_sg = ethtool_op_set_sg,
1956 .get_link = ethtool_op_get_link,
1957 .get_strings = get_strings,
1958 .phys_id = identify_port,
1959 .nway_reset = restart_autoneg,
1960 .get_sset_count = get_sset_count,
1961 .get_ethtool_stats = get_stats,
1962 .get_regs_len = get_regs_len,
1963 .get_regs = get_regs,
1967 .set_flags = set_flags,
1968 .get_rxnfc = get_rxnfc,
1969 .get_rxfh_indir = get_rss_table,
1970 .set_rxfh_indir = set_rss_table,
1971 .flash_device = set_flash,
1978 static int mem_open(struct inode *inode, struct file *file)
1980 file->private_data = inode->i_private;
1984 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
1988 loff_t avail = file->f_path.dentry->d_inode->i_size;
1989 unsigned int mem = (uintptr_t)file->private_data & 3;
1990 struct adapter *adap = file->private_data - mem;
1996 if (count > avail - pos)
1997 count = avail - pos;
2005 ret = t4_mc_read(adap, pos, data, NULL);
2007 ret = t4_edc_read(adap, mem, pos, data, NULL);
2011 ofst = pos % sizeof(data);
2012 len = min(count, sizeof(data) - ofst);
2013 if (copy_to_user(buf, (u8 *)data + ofst, len))
2020 count = pos - *ppos;
2025 static const struct file_operations mem_debugfs_fops = {
2026 .owner = THIS_MODULE,
2029 .llseek = default_llseek,
2032 static void __devinit add_debugfs_mem(struct adapter *adap, const char *name,
2033 unsigned int idx, unsigned int size_mb)
2037 de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2038 (void *)adap + idx, &mem_debugfs_fops);
2039 if (de && de->d_inode)
2040 de->d_inode->i_size = size_mb << 20;
2043 static int __devinit setup_debugfs(struct adapter *adap)
2047 if (IS_ERR_OR_NULL(adap->debugfs_root))
2050 i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2051 if (i & EDRAM0_ENABLE)
2052 add_debugfs_mem(adap, "edc0", MEM_EDC0, 5);
2053 if (i & EDRAM1_ENABLE)
2054 add_debugfs_mem(adap, "edc1", MEM_EDC1, 5);
2055 if (i & EXT_MEM_ENABLE)
2056 add_debugfs_mem(adap, "mc", MEM_MC,
2057 EXT_MEM_SIZE_GET(t4_read_reg(adap, MA_EXT_MEMORY_BAR)));
2059 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2065 * upper-layer driver support
2069 * Allocate an active-open TID and set it to the supplied value.
2071 int cxgb4_alloc_atid(struct tid_info *t, void *data)
2075 spin_lock_bh(&t->atid_lock);
2077 union aopen_entry *p = t->afree;
2079 atid = p - t->atid_tab;
2084 spin_unlock_bh(&t->atid_lock);
2087 EXPORT_SYMBOL(cxgb4_alloc_atid);
2090 * Release an active-open TID.
2092 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
2094 union aopen_entry *p = &t->atid_tab[atid];
2096 spin_lock_bh(&t->atid_lock);
2100 spin_unlock_bh(&t->atid_lock);
2102 EXPORT_SYMBOL(cxgb4_free_atid);
2105 * Allocate a server TID and set it to the supplied value.
2107 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
2111 spin_lock_bh(&t->stid_lock);
2112 if (family == PF_INET) {
2113 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
2114 if (stid < t->nstids)
2115 __set_bit(stid, t->stid_bmap);
2119 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
2124 t->stid_tab[stid].data = data;
2125 stid += t->stid_base;
2128 spin_unlock_bh(&t->stid_lock);
2131 EXPORT_SYMBOL(cxgb4_alloc_stid);
2134 * Release a server TID.
2136 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
2138 stid -= t->stid_base;
2139 spin_lock_bh(&t->stid_lock);
2140 if (family == PF_INET)
2141 __clear_bit(stid, t->stid_bmap);
2143 bitmap_release_region(t->stid_bmap, stid, 2);
2144 t->stid_tab[stid].data = NULL;
2146 spin_unlock_bh(&t->stid_lock);
2148 EXPORT_SYMBOL(cxgb4_free_stid);
2151 * Populate a TID_RELEASE WR. Caller must properly size the skb.
2153 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
2156 struct cpl_tid_release *req;
2158 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
2159 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
2160 INIT_TP_WR(req, tid);
2161 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
2165 * Queue a TID release request and if necessary schedule a work queue to
2168 void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
2171 void **p = &t->tid_tab[tid];
2172 struct adapter *adap = container_of(t, struct adapter, tids);
2174 spin_lock_bh(&adap->tid_release_lock);
2175 *p = adap->tid_release_head;
2176 /* Low 2 bits encode the Tx channel number */
2177 adap->tid_release_head = (void **)((uintptr_t)p | chan);
2178 if (!adap->tid_release_task_busy) {
2179 adap->tid_release_task_busy = true;
2180 schedule_work(&adap->tid_release_task);
2182 spin_unlock_bh(&adap->tid_release_lock);
2184 EXPORT_SYMBOL(cxgb4_queue_tid_release);
2187 * Process the list of pending TID release requests.
2189 static void process_tid_release_list(struct work_struct *work)
2191 struct sk_buff *skb;
2192 struct adapter *adap;
2194 adap = container_of(work, struct adapter, tid_release_task);
2196 spin_lock_bh(&adap->tid_release_lock);
2197 while (adap->tid_release_head) {
2198 void **p = adap->tid_release_head;
2199 unsigned int chan = (uintptr_t)p & 3;
2200 p = (void *)p - chan;
2202 adap->tid_release_head = *p;
2204 spin_unlock_bh(&adap->tid_release_lock);
2206 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
2208 schedule_timeout_uninterruptible(1);
2210 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
2211 t4_ofld_send(adap, skb);
2212 spin_lock_bh(&adap->tid_release_lock);
2214 adap->tid_release_task_busy = false;
2215 spin_unlock_bh(&adap->tid_release_lock);
2219 * Release a TID and inform HW. If we are unable to allocate the release
2220 * message we defer to a work queue.
2222 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
2225 struct sk_buff *skb;
2226 struct adapter *adap = container_of(t, struct adapter, tids);
2228 old = t->tid_tab[tid];
2229 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
2231 t->tid_tab[tid] = NULL;
2232 mk_tid_release(skb, chan, tid);
2233 t4_ofld_send(adap, skb);
2235 cxgb4_queue_tid_release(t, chan, tid);
2237 atomic_dec(&t->tids_in_use);
2239 EXPORT_SYMBOL(cxgb4_remove_tid);
2242 * Allocate and initialize the TID tables. Returns 0 on success.
2244 static int tid_init(struct tid_info *t)
2247 unsigned int natids = t->natids;
2249 size = t->ntids * sizeof(*t->tid_tab) + natids * sizeof(*t->atid_tab) +
2250 t->nstids * sizeof(*t->stid_tab) +
2251 BITS_TO_LONGS(t->nstids) * sizeof(long);
2252 t->tid_tab = t4_alloc_mem(size);
2256 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
2257 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
2258 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids];
2259 spin_lock_init(&t->stid_lock);
2260 spin_lock_init(&t->atid_lock);
2262 t->stids_in_use = 0;
2264 t->atids_in_use = 0;
2265 atomic_set(&t->tids_in_use, 0);
2267 /* Setup the free list for atid_tab and clear the stid bitmap. */
2270 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
2271 t->afree = t->atid_tab;
2273 bitmap_zero(t->stid_bmap, t->nstids);
2278 * cxgb4_create_server - create an IP server
2280 * @stid: the server TID
2281 * @sip: local IP address to bind server to
2282 * @sport: the server's TCP port
2283 * @queue: queue to direct messages from this server to
2285 * Create an IP server for the given port and address.
2286 * Returns <0 on error and one of the %NET_XMIT_* values on success.
2288 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
2289 __be32 sip, __be16 sport, unsigned int queue)
2292 struct sk_buff *skb;
2293 struct adapter *adap;
2294 struct cpl_pass_open_req *req;
2296 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
2300 adap = netdev2adap(dev);
2301 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
2303 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
2304 req->local_port = sport;
2305 req->peer_port = htons(0);
2306 req->local_ip = sip;
2307 req->peer_ip = htonl(0);
2308 chan = netdev2pinfo(adap->sge.ingr_map[queue]->netdev)->tx_chan;
2309 req->opt0 = cpu_to_be64(TX_CHAN(chan));
2310 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
2311 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
2312 return t4_mgmt_tx(adap, skb);
2314 EXPORT_SYMBOL(cxgb4_create_server);
2317 * cxgb4_create_server6 - create an IPv6 server
2319 * @stid: the server TID
2320 * @sip: local IPv6 address to bind server to
2321 * @sport: the server's TCP port
2322 * @queue: queue to direct messages from this server to
2324 * Create an IPv6 server for the given port and address.
2325 * Returns <0 on error and one of the %NET_XMIT_* values on success.
2327 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
2328 const struct in6_addr *sip, __be16 sport,
2332 struct sk_buff *skb;
2333 struct adapter *adap;
2334 struct cpl_pass_open_req6 *req;
2336 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
2340 adap = netdev2adap(dev);
2341 req = (struct cpl_pass_open_req6 *)__skb_put(skb, sizeof(*req));
2343 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
2344 req->local_port = sport;
2345 req->peer_port = htons(0);
2346 req->local_ip_hi = *(__be64 *)(sip->s6_addr);
2347 req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
2348 req->peer_ip_hi = cpu_to_be64(0);
2349 req->peer_ip_lo = cpu_to_be64(0);
2350 chan = netdev2pinfo(adap->sge.ingr_map[queue]->netdev)->tx_chan;
2351 req->opt0 = cpu_to_be64(TX_CHAN(chan));
2352 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
2353 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
2354 return t4_mgmt_tx(adap, skb);
2356 EXPORT_SYMBOL(cxgb4_create_server6);
2359 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
2360 * @mtus: the HW MTU table
2361 * @mtu: the target MTU
2362 * @idx: index of selected entry in the MTU table
2364 * Returns the index and the value in the HW MTU table that is closest to
2365 * but does not exceed @mtu, unless @mtu is smaller than any value in the
2366 * table, in which case that smallest available value is selected.
2368 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
2373 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
2379 EXPORT_SYMBOL(cxgb4_best_mtu);
2382 * cxgb4_port_chan - get the HW channel of a port
2383 * @dev: the net device for the port
2385 * Return the HW Tx channel of the given port.
2387 unsigned int cxgb4_port_chan(const struct net_device *dev)
2389 return netdev2pinfo(dev)->tx_chan;
2391 EXPORT_SYMBOL(cxgb4_port_chan);
2394 * cxgb4_port_viid - get the VI id of a port
2395 * @dev: the net device for the port
2397 * Return the VI id of the given port.
2399 unsigned int cxgb4_port_viid(const struct net_device *dev)
2401 return netdev2pinfo(dev)->viid;
2403 EXPORT_SYMBOL(cxgb4_port_viid);
2406 * cxgb4_port_idx - get the index of a port
2407 * @dev: the net device for the port
2409 * Return the index of the given port.
2411 unsigned int cxgb4_port_idx(const struct net_device *dev)
2413 return netdev2pinfo(dev)->port_id;
2415 EXPORT_SYMBOL(cxgb4_port_idx);
2418 * cxgb4_netdev_by_hwid - return the net device of a HW port
2419 * @pdev: identifies the adapter
2420 * @id: the HW port id
2422 * Return the net device associated with the interface with the given HW
2425 struct net_device *cxgb4_netdev_by_hwid(struct pci_dev *pdev, unsigned int id)
2427 const struct adapter *adap = pci_get_drvdata(pdev);
2429 if (!adap || id >= NCHAN)
2431 id = adap->chan_map[id];
2432 return id < MAX_NPORTS ? adap->port[id] : NULL;
2434 EXPORT_SYMBOL(cxgb4_netdev_by_hwid);
2436 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
2437 struct tp_tcp_stats *v6)
2439 struct adapter *adap = pci_get_drvdata(pdev);
2441 spin_lock(&adap->stats_lock);
2442 t4_tp_get_tcp_stats(adap, v4, v6);
2443 spin_unlock(&adap->stats_lock);
2445 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
2447 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
2448 const unsigned int *pgsz_order)
2450 struct adapter *adap = netdev2adap(dev);
2452 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
2453 t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
2454 HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
2455 HPZ3(pgsz_order[3]));
2457 EXPORT_SYMBOL(cxgb4_iscsi_init);
2459 static struct pci_driver cxgb4_driver;
2461 static void check_neigh_update(struct neighbour *neigh)
2463 const struct device *parent;
2464 const struct net_device *netdev = neigh->dev;
2466 if (netdev->priv_flags & IFF_802_1Q_VLAN)
2467 netdev = vlan_dev_real_dev(netdev);
2468 parent = netdev->dev.parent;
2469 if (parent && parent->driver == &cxgb4_driver.driver)
2470 t4_l2t_update(dev_get_drvdata(parent), neigh);
2473 static int netevent_cb(struct notifier_block *nb, unsigned long event,
2477 case NETEVENT_NEIGH_UPDATE:
2478 check_neigh_update(data);
2480 case NETEVENT_PMTU_UPDATE:
2481 case NETEVENT_REDIRECT:
2488 static bool netevent_registered;
2489 static struct notifier_block cxgb4_netevent_nb = {
2490 .notifier_call = netevent_cb
2493 static void uld_attach(struct adapter *adap, unsigned int uld)
2496 struct cxgb4_lld_info lli;
2498 lli.pdev = adap->pdev;
2499 lli.l2t = adap->l2t;
2500 lli.tids = &adap->tids;
2501 lli.ports = adap->port;
2502 lli.vr = &adap->vres;
2503 lli.mtus = adap->params.mtus;
2504 if (uld == CXGB4_ULD_RDMA) {
2505 lli.rxq_ids = adap->sge.rdma_rxq;
2506 lli.nrxq = adap->sge.rdmaqs;
2507 } else if (uld == CXGB4_ULD_ISCSI) {
2508 lli.rxq_ids = adap->sge.ofld_rxq;
2509 lli.nrxq = adap->sge.ofldqsets;
2511 lli.ntxq = adap->sge.ofldqsets;
2512 lli.nchan = adap->params.nports;
2513 lli.nports = adap->params.nports;
2514 lli.wr_cred = adap->params.ofldq_wr_cred;
2515 lli.adapter_type = adap->params.rev;
2516 lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
2517 lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
2518 t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
2520 lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
2521 t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
2523 lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
2524 lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
2525 lli.fw_vers = adap->params.fw_vers;
2527 handle = ulds[uld].add(&lli);
2528 if (IS_ERR(handle)) {
2529 dev_warn(adap->pdev_dev,
2530 "could not attach to the %s driver, error %ld\n",
2531 uld_str[uld], PTR_ERR(handle));
2535 adap->uld_handle[uld] = handle;
2537 if (!netevent_registered) {
2538 register_netevent_notifier(&cxgb4_netevent_nb);
2539 netevent_registered = true;
2542 if (adap->flags & FULL_INIT_DONE)
2543 ulds[uld].state_change(handle, CXGB4_STATE_UP);
2546 static void attach_ulds(struct adapter *adap)
2550 mutex_lock(&uld_mutex);
2551 list_add_tail(&adap->list_node, &adapter_list);
2552 for (i = 0; i < CXGB4_ULD_MAX; i++)
2554 uld_attach(adap, i);
2555 mutex_unlock(&uld_mutex);
2558 static void detach_ulds(struct adapter *adap)
2562 mutex_lock(&uld_mutex);
2563 list_del(&adap->list_node);
2564 for (i = 0; i < CXGB4_ULD_MAX; i++)
2565 if (adap->uld_handle[i]) {
2566 ulds[i].state_change(adap->uld_handle[i],
2567 CXGB4_STATE_DETACH);
2568 adap->uld_handle[i] = NULL;
2570 if (netevent_registered && list_empty(&adapter_list)) {
2571 unregister_netevent_notifier(&cxgb4_netevent_nb);
2572 netevent_registered = false;
2574 mutex_unlock(&uld_mutex);
2577 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2581 mutex_lock(&uld_mutex);
2582 for (i = 0; i < CXGB4_ULD_MAX; i++)
2583 if (adap->uld_handle[i])
2584 ulds[i].state_change(adap->uld_handle[i], new_state);
2585 mutex_unlock(&uld_mutex);
2589 * cxgb4_register_uld - register an upper-layer driver
2590 * @type: the ULD type
2591 * @p: the ULD methods
2593 * Registers an upper-layer driver with this driver and notifies the ULD
2594 * about any presently available devices that support its type. Returns
2595 * %-EBUSY if a ULD of the same type is already registered.
2597 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
2600 struct adapter *adap;
2602 if (type >= CXGB4_ULD_MAX)
2604 mutex_lock(&uld_mutex);
2605 if (ulds[type].add) {
2610 list_for_each_entry(adap, &adapter_list, list_node)
2611 uld_attach(adap, type);
2612 out: mutex_unlock(&uld_mutex);
2615 EXPORT_SYMBOL(cxgb4_register_uld);
2618 * cxgb4_unregister_uld - unregister an upper-layer driver
2619 * @type: the ULD type
2621 * Unregisters an existing upper-layer driver.
2623 int cxgb4_unregister_uld(enum cxgb4_uld type)
2625 struct adapter *adap;
2627 if (type >= CXGB4_ULD_MAX)
2629 mutex_lock(&uld_mutex);
2630 list_for_each_entry(adap, &adapter_list, list_node)
2631 adap->uld_handle[type] = NULL;
2632 ulds[type].add = NULL;
2633 mutex_unlock(&uld_mutex);
2636 EXPORT_SYMBOL(cxgb4_unregister_uld);
2639 * cxgb_up - enable the adapter
2640 * @adap: adapter being enabled
2642 * Called when the first port is enabled, this function performs the
2643 * actions necessary to make an adapter operational, such as completing
2644 * the initialization of HW modules, and enabling interrupts.
2646 * Must be called with the rtnl lock held.
2648 static int cxgb_up(struct adapter *adap)
2652 err = setup_sge_queues(adap);
2655 err = setup_rss(adap);
2659 if (adap->flags & USING_MSIX) {
2660 name_msix_vecs(adap);
2661 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2662 adap->msix_info[0].desc, adap);
2666 err = request_msix_queue_irqs(adap);
2668 free_irq(adap->msix_info[0].vec, adap);
2672 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2673 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2680 t4_intr_enable(adap);
2681 adap->flags |= FULL_INIT_DONE;
2682 notify_ulds(adap, CXGB4_STATE_UP);
2686 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2688 t4_free_sge_resources(adap);
2692 static void cxgb_down(struct adapter *adapter)
2694 t4_intr_disable(adapter);
2695 cancel_work_sync(&adapter->tid_release_task);
2696 adapter->tid_release_task_busy = false;
2697 adapter->tid_release_head = NULL;
2699 if (adapter->flags & USING_MSIX) {
2700 free_msix_queue_irqs(adapter);
2701 free_irq(adapter->msix_info[0].vec, adapter);
2703 free_irq(adapter->pdev->irq, adapter);
2704 quiesce_rx(adapter);
2705 t4_sge_stop(adapter);
2706 t4_free_sge_resources(adapter);
2707 adapter->flags &= ~FULL_INIT_DONE;
2711 * net_device operations
2713 static int cxgb_open(struct net_device *dev)
2716 struct port_info *pi = netdev_priv(dev);
2717 struct adapter *adapter = pi->adapter;
2719 if (!(adapter->flags & FULL_INIT_DONE)) {
2720 err = cxgb_up(adapter);
2725 dev->real_num_tx_queues = pi->nqsets;
2726 err = link_start(dev);
2728 netif_tx_start_all_queues(dev);
2732 static int cxgb_close(struct net_device *dev)
2734 struct port_info *pi = netdev_priv(dev);
2735 struct adapter *adapter = pi->adapter;
2737 netif_tx_stop_all_queues(dev);
2738 netif_carrier_off(dev);
2739 return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
2742 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
2743 struct rtnl_link_stats64 *ns)
2745 struct port_stats stats;
2746 struct port_info *p = netdev_priv(dev);
2747 struct adapter *adapter = p->adapter;
2749 spin_lock(&adapter->stats_lock);
2750 t4_get_port_stats(adapter, p->tx_chan, &stats);
2751 spin_unlock(&adapter->stats_lock);
2753 ns->tx_bytes = stats.tx_octets;
2754 ns->tx_packets = stats.tx_frames;
2755 ns->rx_bytes = stats.rx_octets;
2756 ns->rx_packets = stats.rx_frames;
2757 ns->multicast = stats.rx_mcast_frames;
2759 /* detailed rx_errors */
2760 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2762 ns->rx_over_errors = 0;
2763 ns->rx_crc_errors = stats.rx_fcs_err;
2764 ns->rx_frame_errors = stats.rx_symbol_err;
2765 ns->rx_fifo_errors = stats.rx_ovflow0 + stats.rx_ovflow1 +
2766 stats.rx_ovflow2 + stats.rx_ovflow3 +
2767 stats.rx_trunc0 + stats.rx_trunc1 +
2768 stats.rx_trunc2 + stats.rx_trunc3;
2769 ns->rx_missed_errors = 0;
2771 /* detailed tx_errors */
2772 ns->tx_aborted_errors = 0;
2773 ns->tx_carrier_errors = 0;
2774 ns->tx_fifo_errors = 0;
2775 ns->tx_heartbeat_errors = 0;
2776 ns->tx_window_errors = 0;
2778 ns->tx_errors = stats.tx_error_frames;
2779 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2780 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2784 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2787 int ret = 0, prtad, devad;
2788 struct port_info *pi = netdev_priv(dev);
2789 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2793 if (pi->mdio_addr < 0)
2795 data->phy_id = pi->mdio_addr;
2799 if (mdio_phy_id_is_c45(data->phy_id)) {
2800 prtad = mdio_phy_id_prtad(data->phy_id);
2801 devad = mdio_phy_id_devad(data->phy_id);
2802 } else if (data->phy_id < 32) {
2803 prtad = data->phy_id;
2805 data->reg_num &= 0x1f;
2809 mbox = pi->adapter->fn;
2810 if (cmd == SIOCGMIIREG)
2811 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2812 data->reg_num, &data->val_out);
2814 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2815 data->reg_num, data->val_in);
2823 static void cxgb_set_rxmode(struct net_device *dev)
2825 /* unfortunately we can't return errors to the stack */
2826 set_rxmode(dev, -1, false);
2829 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2832 struct port_info *pi = netdev_priv(dev);
2834 if (new_mtu < 81 || new_mtu > MAX_MTU) /* accommodate SACK */
2836 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
2843 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2846 struct sockaddr *addr = p;
2847 struct port_info *pi = netdev_priv(dev);
2849 if (!is_valid_ether_addr(addr->sa_data))
2852 ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
2853 pi->xact_addr_filt, addr->sa_data, true, true);
2857 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2858 pi->xact_addr_filt = ret;
2862 static void vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
2864 struct port_info *pi = netdev_priv(dev);
2867 t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1, -1, -1, -1,
2871 #ifdef CONFIG_NET_POLL_CONTROLLER
2872 static void cxgb_netpoll(struct net_device *dev)
2874 struct port_info *pi = netdev_priv(dev);
2875 struct adapter *adap = pi->adapter;
2877 if (adap->flags & USING_MSIX) {
2879 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2881 for (i = pi->nqsets; i; i--, rx++)
2882 t4_sge_intr_msix(0, &rx->rspq);
2884 t4_intr_handler(adap)(0, adap);
2888 static const struct net_device_ops cxgb4_netdev_ops = {
2889 .ndo_open = cxgb_open,
2890 .ndo_stop = cxgb_close,
2891 .ndo_start_xmit = t4_eth_xmit,
2892 .ndo_get_stats64 = cxgb_get_stats,
2893 .ndo_set_rx_mode = cxgb_set_rxmode,
2894 .ndo_set_mac_address = cxgb_set_mac_addr,
2895 .ndo_validate_addr = eth_validate_addr,
2896 .ndo_do_ioctl = cxgb_ioctl,
2897 .ndo_change_mtu = cxgb_change_mtu,
2898 .ndo_vlan_rx_register = vlan_rx_register,
2899 #ifdef CONFIG_NET_POLL_CONTROLLER
2900 .ndo_poll_controller = cxgb_netpoll,
2904 void t4_fatal_err(struct adapter *adap)
2906 t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
2907 t4_intr_disable(adap);
2908 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
2911 static void setup_memwin(struct adapter *adap)
2915 bar0 = pci_resource_start(adap->pdev, 0); /* truncation intentional */
2916 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
2917 (bar0 + MEMWIN0_BASE) | BIR(0) |
2918 WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
2919 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
2920 (bar0 + MEMWIN1_BASE) | BIR(0) |
2921 WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
2922 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
2923 (bar0 + MEMWIN2_BASE) | BIR(0) |
2924 WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
2925 if (adap->vres.ocq.size) {
2926 unsigned int start, sz_kb;
2928 start = pci_resource_start(adap->pdev, 2) +
2929 OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
2930 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
2932 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
2933 start | BIR(1) | WINDOW(ilog2(sz_kb)));
2935 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
2936 adap->vres.ocq.start);
2938 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
2942 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
2947 /* get device capabilities */
2948 memset(c, 0, sizeof(*c));
2949 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2950 FW_CMD_REQUEST | FW_CMD_READ);
2951 c->retval_len16 = htonl(FW_LEN16(*c));
2952 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
2956 /* select capabilities we'll be using */
2957 if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
2959 c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
2961 c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
2962 } else if (vf_acls) {
2963 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
2966 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2967 FW_CMD_REQUEST | FW_CMD_WRITE);
2968 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
2972 ret = t4_config_glbl_rss(adap, adap->fn,
2973 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
2974 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
2975 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
2979 ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
2980 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
2986 /* tweak some settings */
2987 t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
2988 t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
2989 t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
2990 v = t4_read_reg(adap, TP_PIO_DATA);
2991 t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
2993 /* get basic stuff going */
2994 return t4_early_init(adap, adap->fn);
2998 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
3000 #define MAX_ATIDS 8192U
3003 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3005 static int adap_init0(struct adapter *adap)
3009 enum dev_state state;
3010 u32 params[7], val[7];
3011 struct fw_caps_config_cmd c;
3013 ret = t4_check_fw_version(adap);
3014 if (ret == -EINVAL || ret > 0) {
3015 if (upgrade_fw(adap) >= 0) /* recache FW version */
3016 ret = t4_check_fw_version(adap);
3021 /* contact FW, request master */
3022 ret = t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, &state);
3024 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
3030 ret = t4_fw_reset(adap, adap->fn, PIORSTMODE | PIORST);
3034 for (v = 0; v < SGE_NTIMERS - 1; v++)
3035 adap->sge.timer_val[v] = min(intr_holdoff[v], MAX_SGE_TIMERVAL);
3036 adap->sge.timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
3037 adap->sge.counter_val[0] = 1;
3038 for (v = 1; v < SGE_NCOUNTERS; v++)
3039 adap->sge.counter_val[v] = min(intr_cnt[v - 1],
3041 #define FW_PARAM_DEV(param) \
3042 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3043 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3045 params[0] = FW_PARAM_DEV(CCLK);
3046 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 1, params, val);
3049 adap->params.vpd.cclk = val[0];
3051 ret = adap_init1(adap, &c);
3055 #define FW_PARAM_PFVF(param) \
3056 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3057 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param) | \
3058 FW_PARAMS_PARAM_Y(adap->fn))
3060 params[0] = FW_PARAM_DEV(PORTVEC);
3061 params[1] = FW_PARAM_PFVF(L2T_START);
3062 params[2] = FW_PARAM_PFVF(L2T_END);
3063 params[3] = FW_PARAM_PFVF(FILTER_START);
3064 params[4] = FW_PARAM_PFVF(FILTER_END);
3065 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 5, params, val);
3069 adap->tids.ftid_base = val[3];
3070 adap->tids.nftids = val[4] - val[3] + 1;
3073 /* query offload-related parameters */
3074 params[0] = FW_PARAM_DEV(NTID);
3075 params[1] = FW_PARAM_PFVF(SERVER_START);
3076 params[2] = FW_PARAM_PFVF(SERVER_END);
3077 params[3] = FW_PARAM_PFVF(TDDP_START);
3078 params[4] = FW_PARAM_PFVF(TDDP_END);
3079 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3080 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3084 adap->tids.ntids = val[0];
3085 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
3086 adap->tids.stid_base = val[1];
3087 adap->tids.nstids = val[2] - val[1] + 1;
3088 adap->vres.ddp.start = val[3];
3089 adap->vres.ddp.size = val[4] - val[3] + 1;
3090 adap->params.ofldq_wr_cred = val[5];
3091 adap->params.offload = 1;
3094 params[0] = FW_PARAM_PFVF(STAG_START);
3095 params[1] = FW_PARAM_PFVF(STAG_END);
3096 params[2] = FW_PARAM_PFVF(RQ_START);
3097 params[3] = FW_PARAM_PFVF(RQ_END);
3098 params[4] = FW_PARAM_PFVF(PBL_START);
3099 params[5] = FW_PARAM_PFVF(PBL_END);
3100 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3104 adap->vres.stag.start = val[0];
3105 adap->vres.stag.size = val[1] - val[0] + 1;
3106 adap->vres.rq.start = val[2];
3107 adap->vres.rq.size = val[3] - val[2] + 1;
3108 adap->vres.pbl.start = val[4];
3109 adap->vres.pbl.size = val[5] - val[4] + 1;
3111 params[0] = FW_PARAM_PFVF(SQRQ_START);
3112 params[1] = FW_PARAM_PFVF(SQRQ_END);
3113 params[2] = FW_PARAM_PFVF(CQ_START);
3114 params[3] = FW_PARAM_PFVF(CQ_END);
3115 params[4] = FW_PARAM_PFVF(OCQ_START);
3116 params[5] = FW_PARAM_PFVF(OCQ_END);
3117 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3121 adap->vres.qp.start = val[0];
3122 adap->vres.qp.size = val[1] - val[0] + 1;
3123 adap->vres.cq.start = val[2];
3124 adap->vres.cq.size = val[3] - val[2] + 1;
3125 adap->vres.ocq.start = val[4];
3126 adap->vres.ocq.size = val[5] - val[4] + 1;
3129 params[0] = FW_PARAM_PFVF(ISCSI_START);
3130 params[1] = FW_PARAM_PFVF(ISCSI_END);
3131 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 2, params,
3135 adap->vres.iscsi.start = val[0];
3136 adap->vres.iscsi.size = val[1] - val[0] + 1;
3138 #undef FW_PARAM_PFVF
3141 adap->params.nports = hweight32(port_vec);
3142 adap->params.portvec = port_vec;
3143 adap->flags |= FW_OK;
3145 /* These are finalized by FW initialization, load their values now */
3146 v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
3147 adap->params.tp.tre = TIMERRESOLUTION_GET(v);
3148 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
3149 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3150 adap->params.b_wnd);
3152 #ifdef CONFIG_PCI_IOV
3154 * Provision resource limits for Virtual Functions. We currently
3155 * grant them all the same static resource limits except for the Port
3156 * Access Rights Mask which we're assigning based on the PF. All of
3157 * the static provisioning stuff for both the PF and VF really needs
3158 * to be managed in a persistent manner for each device which the
3159 * firmware controls.
3164 for (pf = 0; pf < ARRAY_SIZE(num_vf); pf++) {
3165 if (num_vf[pf] <= 0)
3168 /* VF numbering starts at 1! */
3169 for (vf = 1; vf <= num_vf[pf]; vf++) {
3170 ret = t4_cfg_pfvf(adap, adap->fn, pf, vf,
3171 VFRES_NEQ, VFRES_NETHCTRL,
3172 VFRES_NIQFLINT, VFRES_NIQ,
3173 VFRES_TC, VFRES_NVI,
3174 FW_PFVF_CMD_CMASK_MASK,
3175 pfvfres_pmask(adap, pf, vf),
3177 VFRES_R_CAPS, VFRES_WX_CAPS);
3179 dev_warn(adap->pdev_dev, "failed to "
3180 "provision pf/vf=%d/%d; "
3181 "err=%d\n", pf, vf, ret);
3191 * If a command timed out or failed with EIO FW does not operate within
3192 * its spec or something catastrophic happened to HW/FW, stop issuing
3195 bye: if (ret != -ETIMEDOUT && ret != -EIO)
3196 t4_fw_bye(adap, adap->fn);
3202 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
3203 pci_channel_state_t state)
3206 struct adapter *adap = pci_get_drvdata(pdev);
3212 adap->flags &= ~FW_OK;
3213 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
3214 for_each_port(adap, i) {
3215 struct net_device *dev = adap->port[i];
3217 netif_device_detach(dev);
3218 netif_carrier_off(dev);
3220 if (adap->flags & FULL_INIT_DONE)
3223 pci_disable_device(pdev);
3224 out: return state == pci_channel_io_perm_failure ?
3225 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
3228 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
3231 struct fw_caps_config_cmd c;
3232 struct adapter *adap = pci_get_drvdata(pdev);
3235 pci_restore_state(pdev);
3236 pci_save_state(pdev);
3237 return PCI_ERS_RESULT_RECOVERED;
3240 if (pci_enable_device(pdev)) {
3241 dev_err(&pdev->dev, "cannot reenable PCI device after reset\n");
3242 return PCI_ERS_RESULT_DISCONNECT;
3245 pci_set_master(pdev);
3246 pci_restore_state(pdev);
3247 pci_save_state(pdev);
3248 pci_cleanup_aer_uncorrect_error_status(pdev);
3250 if (t4_wait_dev_ready(adap) < 0)
3251 return PCI_ERS_RESULT_DISCONNECT;
3252 if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL))
3253 return PCI_ERS_RESULT_DISCONNECT;
3254 adap->flags |= FW_OK;
3255 if (adap_init1(adap, &c))
3256 return PCI_ERS_RESULT_DISCONNECT;
3258 for_each_port(adap, i) {
3259 struct port_info *p = adap2pinfo(adap, i);
3261 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
3264 return PCI_ERS_RESULT_DISCONNECT;
3266 p->xact_addr_filt = -1;
3269 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3270 adap->params.b_wnd);
3273 return PCI_ERS_RESULT_DISCONNECT;
3274 return PCI_ERS_RESULT_RECOVERED;
3277 static void eeh_resume(struct pci_dev *pdev)
3280 struct adapter *adap = pci_get_drvdata(pdev);
3286 for_each_port(adap, i) {
3287 struct net_device *dev = adap->port[i];
3289 if (netif_running(dev)) {
3291 cxgb_set_rxmode(dev);
3293 netif_device_attach(dev);
3298 static struct pci_error_handlers cxgb4_eeh = {
3299 .error_detected = eeh_err_detected,
3300 .slot_reset = eeh_slot_reset,
3301 .resume = eeh_resume,
3304 static inline bool is_10g_port(const struct link_config *lc)
3306 return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
3309 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
3310 unsigned int size, unsigned int iqe_size)
3312 q->intr_params = QINTR_TIMER_IDX(timer_idx) |
3313 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
3314 q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
3315 q->iqe_len = iqe_size;
3320 * Perform default configuration of DMA queues depending on the number and type
3321 * of ports we found and the number of available CPUs. Most settings can be
3322 * modified by the admin prior to actual use.
3324 static void __devinit cfg_queues(struct adapter *adap)
3326 struct sge *s = &adap->sge;
3327 int i, q10g = 0, n10g = 0, qidx = 0;
3329 for_each_port(adap, i)
3330 n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
3333 * We default to 1 queue per non-10G port and up to # of cores queues
3337 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
3338 if (q10g > num_online_cpus())
3339 q10g = num_online_cpus();
3341 for_each_port(adap, i) {
3342 struct port_info *pi = adap2pinfo(adap, i);
3344 pi->first_qset = qidx;
3345 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
3350 s->max_ethqsets = qidx; /* MSI-X may lower it later */
3352 if (is_offload(adap)) {
3354 * For offload we use 1 queue/channel if all ports are up to 1G,
3355 * otherwise we divide all available queues amongst the channels
3356 * capped by the number of available cores.
3359 i = min_t(int, ARRAY_SIZE(s->ofldrxq),
3361 s->ofldqsets = roundup(i, adap->params.nports);
3363 s->ofldqsets = adap->params.nports;
3364 /* For RDMA one Rx queue per channel suffices */
3365 s->rdmaqs = adap->params.nports;
3368 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
3369 struct sge_eth_rxq *r = &s->ethrxq[i];
3371 init_rspq(&r->rspq, 0, 0, 1024, 64);
3375 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
3376 s->ethtxq[i].q.size = 1024;
3378 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
3379 s->ctrlq[i].q.size = 512;
3381 for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
3382 s->ofldtxq[i].q.size = 1024;
3384 for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
3385 struct sge_ofld_rxq *r = &s->ofldrxq[i];
3387 init_rspq(&r->rspq, 0, 0, 1024, 64);
3388 r->rspq.uld = CXGB4_ULD_ISCSI;
3392 for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
3393 struct sge_ofld_rxq *r = &s->rdmarxq[i];
3395 init_rspq(&r->rspq, 0, 0, 511, 64);
3396 r->rspq.uld = CXGB4_ULD_RDMA;
3400 init_rspq(&s->fw_evtq, 6, 0, 512, 64);
3401 init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
3405 * Reduce the number of Ethernet queues across all ports to at most n.
3406 * n provides at least one queue per port.
3408 static void __devinit reduce_ethqs(struct adapter *adap, int n)
3411 struct port_info *pi;
3413 while (n < adap->sge.ethqsets)
3414 for_each_port(adap, i) {
3415 pi = adap2pinfo(adap, i);
3416 if (pi->nqsets > 1) {
3418 adap->sge.ethqsets--;
3419 if (adap->sge.ethqsets <= n)
3425 for_each_port(adap, i) {
3426 pi = adap2pinfo(adap, i);
3432 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
3433 #define EXTRA_VECS 2
3435 static int __devinit enable_msix(struct adapter *adap)
3438 int i, err, want, need;
3439 struct sge *s = &adap->sge;
3440 unsigned int nchan = adap->params.nports;
3441 struct msix_entry entries[MAX_INGQ + 1];
3443 for (i = 0; i < ARRAY_SIZE(entries); ++i)
3444 entries[i].entry = i;
3446 want = s->max_ethqsets + EXTRA_VECS;
3447 if (is_offload(adap)) {
3448 want += s->rdmaqs + s->ofldqsets;
3449 /* need nchan for each possible ULD */
3450 ofld_need = 2 * nchan;
3452 need = adap->params.nports + EXTRA_VECS + ofld_need;
3454 while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
3459 * Distribute available vectors to the various queue groups.
3460 * Every group gets its minimum requirement and NIC gets top
3461 * priority for leftovers.
3463 i = want - EXTRA_VECS - ofld_need;
3464 if (i < s->max_ethqsets) {
3465 s->max_ethqsets = i;
3466 if (i < s->ethqsets)
3467 reduce_ethqs(adap, i);
3469 if (is_offload(adap)) {
3470 i = want - EXTRA_VECS - s->max_ethqsets;
3471 i -= ofld_need - nchan;
3472 s->ofldqsets = (i / nchan) * nchan; /* round down */
3474 for (i = 0; i < want; ++i)
3475 adap->msix_info[i].vec = entries[i].vector;
3477 dev_info(adap->pdev_dev,
3478 "only %d MSI-X vectors left, not using MSI-X\n", err);
3484 static int __devinit init_rss(struct adapter *adap)
3488 for_each_port(adap, i) {
3489 struct port_info *pi = adap2pinfo(adap, i);
3491 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
3494 for (j = 0; j < pi->rss_size; j++)
3495 pi->rss[j] = j % pi->nqsets;
3500 static void __devinit print_port_info(struct adapter *adap)
3502 static const char *base[] = {
3503 "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
3504 "KX", "KR", "KR SFP+", "KR FEC"
3509 const char *spd = "";
3511 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
3513 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
3516 for_each_port(adap, i) {
3517 struct net_device *dev = adap->port[i];
3518 const struct port_info *pi = netdev_priv(dev);
3521 if (!test_bit(i, &adap->registered_device_map))
3524 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
3525 bufp += sprintf(bufp, "100/");
3526 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
3527 bufp += sprintf(bufp, "1000/");
3528 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
3529 bufp += sprintf(bufp, "10G/");
3532 sprintf(bufp, "BASE-%s", base[pi->port_type]);
3534 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
3535 adap->params.vpd.id, adap->params.rev,
3536 buf, is_offload(adap) ? "R" : "",
3537 adap->params.pci.width, spd,
3538 (adap->flags & USING_MSIX) ? " MSI-X" :
3539 (adap->flags & USING_MSI) ? " MSI" : "");
3540 if (adap->name == dev->name)
3541 netdev_info(dev, "S/N: %s, E/C: %s\n",
3542 adap->params.vpd.sn, adap->params.vpd.ec);
3547 * Free the following resources:
3548 * - memory used for tables
3551 * - resources FW is holding for us
3553 static void free_some_resources(struct adapter *adapter)
3557 t4_free_mem(adapter->l2t);
3558 t4_free_mem(adapter->tids.tid_tab);
3559 disable_msi(adapter);
3561 for_each_port(adapter, i)
3562 if (adapter->port[i]) {
3563 kfree(adap2pinfo(adapter, i)->rss);
3564 free_netdev(adapter->port[i]);
3566 if (adapter->flags & FW_OK)
3567 t4_fw_bye(adapter, adapter->fn);
3570 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
3571 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
3573 static int __devinit init_one(struct pci_dev *pdev,
3574 const struct pci_device_id *ent)
3577 struct port_info *pi;
3578 unsigned int highdma = 0;
3579 struct adapter *adapter = NULL;
3581 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
3583 err = pci_request_regions(pdev, KBUILD_MODNAME);
3585 /* Just info, some other driver may have claimed the device. */
3586 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
3590 /* We control everything through one PF */
3591 func = PCI_FUNC(pdev->devfn);
3592 if (func != ent->driver_data) {
3593 pci_save_state(pdev); /* to restore SR-IOV later */
3597 err = pci_enable_device(pdev);
3599 dev_err(&pdev->dev, "cannot enable PCI device\n");
3600 goto out_release_regions;
3603 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3604 highdma = NETIF_F_HIGHDMA;
3605 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3607 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
3608 "coherent allocations\n");
3609 goto out_disable_device;
3612 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3614 dev_err(&pdev->dev, "no usable DMA configuration\n");
3615 goto out_disable_device;
3619 pci_enable_pcie_error_reporting(pdev);
3620 pci_set_master(pdev);
3621 pci_save_state(pdev);
3623 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3626 goto out_disable_device;
3629 adapter->regs = pci_ioremap_bar(pdev, 0);
3630 if (!adapter->regs) {
3631 dev_err(&pdev->dev, "cannot map device registers\n");
3633 goto out_free_adapter;
3636 adapter->pdev = pdev;
3637 adapter->pdev_dev = &pdev->dev;
3639 adapter->name = pci_name(pdev);
3640 adapter->msg_enable = dflt_msg_enable;
3641 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
3643 spin_lock_init(&adapter->stats_lock);
3644 spin_lock_init(&adapter->tid_release_lock);
3646 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
3648 err = t4_prep_adapter(adapter);
3651 err = adap_init0(adapter);
3655 for_each_port(adapter, i) {
3656 struct net_device *netdev;
3658 netdev = alloc_etherdev_mq(sizeof(struct port_info),
3665 SET_NETDEV_DEV(netdev, &pdev->dev);
3667 adapter->port[i] = netdev;
3668 pi = netdev_priv(netdev);
3669 pi->adapter = adapter;
3670 pi->xact_addr_filt = -1;
3671 pi->rx_offload = RX_CSO;
3673 netif_carrier_off(netdev);
3674 netif_tx_stop_all_queues(netdev);
3675 netdev->irq = pdev->irq;
3677 netdev->features |= NETIF_F_SG | TSO_FLAGS;
3678 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3679 netdev->features |= NETIF_F_GRO | NETIF_F_RXHASH | highdma;
3680 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
3681 netdev->vlan_features = netdev->features & VLAN_FEAT;
3683 netdev->netdev_ops = &cxgb4_netdev_ops;
3684 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
3687 pci_set_drvdata(pdev, adapter);
3689 if (adapter->flags & FW_OK) {
3690 err = t4_port_init(adapter, func, func, 0);
3696 * Configure queues and allocate tables now, they can be needed as
3697 * soon as the first register_netdev completes.
3699 cfg_queues(adapter);
3701 adapter->l2t = t4_init_l2t();
3702 if (!adapter->l2t) {
3703 /* We tolerate a lack of L2T, giving up some functionality */
3704 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
3705 adapter->params.offload = 0;
3708 if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
3709 dev_warn(&pdev->dev, "could not allocate TID table, "
3711 adapter->params.offload = 0;
3714 /* See what interrupts we'll be using */
3715 if (msi > 1 && enable_msix(adapter) == 0)
3716 adapter->flags |= USING_MSIX;
3717 else if (msi > 0 && pci_enable_msi(pdev) == 0)
3718 adapter->flags |= USING_MSI;
3720 err = init_rss(adapter);
3725 * The card is now ready to go. If any errors occur during device
3726 * registration we do not fail the whole card but rather proceed only
3727 * with the ports we manage to register successfully. However we must
3728 * register at least one net device.
3730 for_each_port(adapter, i) {
3731 err = register_netdev(adapter->port[i]);
3733 dev_warn(&pdev->dev,
3734 "cannot register net device %s, skipping\n",
3735 adapter->port[i]->name);
3738 * Change the name we use for messages to the name of
3739 * the first successfully registered interface.
3741 if (!adapter->registered_device_map)
3742 adapter->name = adapter->port[i]->name;
3744 __set_bit(i, &adapter->registered_device_map);
3745 adapter->chan_map[adap2pinfo(adapter, i)->tx_chan] = i;
3748 if (!adapter->registered_device_map) {
3749 dev_err(&pdev->dev, "could not register any net devices\n");
3753 if (cxgb4_debugfs_root) {
3754 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
3755 cxgb4_debugfs_root);
3756 setup_debugfs(adapter);
3759 if (is_offload(adapter))
3760 attach_ulds(adapter);
3762 print_port_info(adapter);
3765 #ifdef CONFIG_PCI_IOV
3766 if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0)
3767 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
3768 dev_info(&pdev->dev,
3769 "instantiated %u virtual functions\n",
3775 free_some_resources(adapter);
3777 iounmap(adapter->regs);
3781 pci_disable_pcie_error_reporting(pdev);
3782 pci_disable_device(pdev);
3783 out_release_regions:
3784 pci_release_regions(pdev);
3785 pci_set_drvdata(pdev, NULL);
3789 static void __devexit remove_one(struct pci_dev *pdev)
3791 struct adapter *adapter = pci_get_drvdata(pdev);
3793 pci_disable_sriov(pdev);
3798 if (is_offload(adapter))
3799 detach_ulds(adapter);
3801 for_each_port(adapter, i)
3802 if (test_bit(i, &adapter->registered_device_map))
3803 unregister_netdev(adapter->port[i]);
3805 if (adapter->debugfs_root)
3806 debugfs_remove_recursive(adapter->debugfs_root);
3808 if (adapter->flags & FULL_INIT_DONE)
3811 free_some_resources(adapter);
3812 iounmap(adapter->regs);
3814 pci_disable_pcie_error_reporting(pdev);
3815 pci_disable_device(pdev);
3816 pci_release_regions(pdev);
3817 pci_set_drvdata(pdev, NULL);
3818 } else if (PCI_FUNC(pdev->devfn) > 0)
3819 pci_release_regions(pdev);
3822 static struct pci_driver cxgb4_driver = {
3823 .name = KBUILD_MODNAME,
3824 .id_table = cxgb4_pci_tbl,
3826 .remove = __devexit_p(remove_one),
3827 .err_handler = &cxgb4_eeh,
3830 static int __init cxgb4_init_module(void)
3834 /* Debugfs support is optional, just warn if this fails */
3835 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3836 if (!cxgb4_debugfs_root)
3837 pr_warning("could not create debugfs entry, continuing\n");
3839 ret = pci_register_driver(&cxgb4_driver);
3841 debugfs_remove(cxgb4_debugfs_root);
3845 static void __exit cxgb4_cleanup_module(void)
3847 pci_unregister_driver(&cxgb4_driver);
3848 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
3851 module_init(cxgb4_init_module);
3852 module_exit(cxgb4_cleanup_module);