spi-topcliff-pch: add recovery processing in case wait-event timeout
[zen-stable.git] / drivers / net / ethernet / chelsio / cxgb4 / cxgb4_main.c
blob9d76e59d9526164ca9cfc247769315d04b8c4fca
1 /*
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
14 * conditions are met:
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
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
32 * SOFTWARE.
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.h>
45 #include <linux/if_vlan.h>
46 #include <linux/init.h>
47 #include <linux/log2.h>
48 #include <linux/mdio.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/mutex.h>
52 #include <linux/netdevice.h>
53 #include <linux/pci.h>
54 #include <linux/aer.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/sched.h>
57 #include <linux/seq_file.h>
58 #include <linux/sockios.h>
59 #include <linux/vmalloc.h>
60 #include <linux/workqueue.h>
61 #include <net/neighbour.h>
62 #include <net/netevent.h>
63 #include <asm/uaccess.h>
65 #include "cxgb4.h"
66 #include "t4_regs.h"
67 #include "t4_msg.h"
68 #include "t4fw_api.h"
69 #include "l2t.h"
71 #define DRV_VERSION "1.3.0-ko"
72 #define DRV_DESC "Chelsio T4 Network Driver"
75 * Max interrupt hold-off timer value in us. Queues fall back to this value
76 * under extreme memory pressure so it's largish to give the system time to
77 * recover.
79 #define MAX_SGE_TIMERVAL 200U
81 #ifdef CONFIG_PCI_IOV
83 * Virtual Function provisioning constants. We need two extra Ingress Queues
84 * with Interrupt capability to serve as the VF's Firmware Event Queue and
85 * Forwarded Interrupt Queue (when using MSI mode) -- neither will have Free
86 * Lists associated with them). For each Ethernet/Control Egress Queue and
87 * for each Free List, we need an Egress Context.
89 enum {
90 VFRES_NPORTS = 1, /* # of "ports" per VF */
91 VFRES_NQSETS = 2, /* # of "Queue Sets" per VF */
93 VFRES_NVI = VFRES_NPORTS, /* # of Virtual Interfaces */
94 VFRES_NETHCTRL = VFRES_NQSETS, /* # of EQs used for ETH or CTRL Qs */
95 VFRES_NIQFLINT = VFRES_NQSETS+2,/* # of ingress Qs/w Free List(s)/intr */
96 VFRES_NIQ = 0, /* # of non-fl/int ingress queues */
97 VFRES_NEQ = VFRES_NQSETS*2, /* # of egress queues */
98 VFRES_TC = 0, /* PCI-E traffic class */
99 VFRES_NEXACTF = 16, /* # of exact MPS filters */
101 VFRES_R_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF|FW_CMD_CAP_PORT,
102 VFRES_WX_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF,
106 * Provide a Port Access Rights Mask for the specified PF/VF. This is very
107 * static and likely not to be useful in the long run. We really need to
108 * implement some form of persistent configuration which the firmware
109 * controls.
111 static unsigned int pfvfres_pmask(struct adapter *adapter,
112 unsigned int pf, unsigned int vf)
114 unsigned int portn, portvec;
117 * Give PF's access to all of the ports.
119 if (vf == 0)
120 return FW_PFVF_CMD_PMASK_MASK;
123 * For VFs, we'll assign them access to the ports based purely on the
124 * PF. We assign active ports in order, wrapping around if there are
125 * fewer active ports than PFs: e.g. active port[pf % nports].
126 * Unfortunately the adapter's port_info structs haven't been
127 * initialized yet so we have to compute this.
129 if (adapter->params.nports == 0)
130 return 0;
132 portn = pf % adapter->params.nports;
133 portvec = adapter->params.portvec;
134 for (;;) {
136 * Isolate the lowest set bit in the port vector. If we're at
137 * the port number that we want, return that as the pmask.
138 * otherwise mask that bit out of the port vector and
139 * decrement our port number ...
141 unsigned int pmask = portvec ^ (portvec & (portvec-1));
142 if (portn == 0)
143 return pmask;
144 portn--;
145 portvec &= ~pmask;
147 /*NOTREACHED*/
149 #endif
151 enum {
152 MEMWIN0_APERTURE = 65536,
153 MEMWIN0_BASE = 0x30000,
154 MEMWIN1_APERTURE = 32768,
155 MEMWIN1_BASE = 0x28000,
156 MEMWIN2_APERTURE = 2048,
157 MEMWIN2_BASE = 0x1b800,
160 enum {
161 MAX_TXQ_ENTRIES = 16384,
162 MAX_CTRL_TXQ_ENTRIES = 1024,
163 MAX_RSPQ_ENTRIES = 16384,
164 MAX_RX_BUFFERS = 16384,
165 MIN_TXQ_ENTRIES = 32,
166 MIN_CTRL_TXQ_ENTRIES = 32,
167 MIN_RSPQ_ENTRIES = 128,
168 MIN_FL_ENTRIES = 16
171 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
172 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
173 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
175 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
177 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
178 CH_DEVICE(0xa000, 0), /* PE10K */
179 CH_DEVICE(0x4001, -1),
180 CH_DEVICE(0x4002, -1),
181 CH_DEVICE(0x4003, -1),
182 CH_DEVICE(0x4004, -1),
183 CH_DEVICE(0x4005, -1),
184 CH_DEVICE(0x4006, -1),
185 CH_DEVICE(0x4007, -1),
186 CH_DEVICE(0x4008, -1),
187 CH_DEVICE(0x4009, -1),
188 CH_DEVICE(0x400a, -1),
189 CH_DEVICE(0x4401, 4),
190 CH_DEVICE(0x4402, 4),
191 CH_DEVICE(0x4403, 4),
192 CH_DEVICE(0x4404, 4),
193 CH_DEVICE(0x4405, 4),
194 CH_DEVICE(0x4406, 4),
195 CH_DEVICE(0x4407, 4),
196 CH_DEVICE(0x4408, 4),
197 CH_DEVICE(0x4409, 4),
198 CH_DEVICE(0x440a, 4),
199 CH_DEVICE(0x440d, 4),
200 CH_DEVICE(0x440e, 4),
201 { 0, }
204 #define FW_FNAME "cxgb4/t4fw.bin"
206 MODULE_DESCRIPTION(DRV_DESC);
207 MODULE_AUTHOR("Chelsio Communications");
208 MODULE_LICENSE("Dual BSD/GPL");
209 MODULE_VERSION(DRV_VERSION);
210 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
211 MODULE_FIRMWARE(FW_FNAME);
213 static int dflt_msg_enable = DFLT_MSG_ENABLE;
215 module_param(dflt_msg_enable, int, 0644);
216 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
219 * The driver uses the best interrupt scheme available on a platform in the
220 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
221 * of these schemes the driver may consider as follows:
223 * msi = 2: choose from among all three options
224 * msi = 1: only consider MSI and INTx interrupts
225 * msi = 0: force INTx interrupts
227 static int msi = 2;
229 module_param(msi, int, 0644);
230 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
233 * Queue interrupt hold-off timer values. Queues default to the first of these
234 * upon creation.
236 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
238 module_param_array(intr_holdoff, uint, NULL, 0644);
239 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
240 "0..4 in microseconds");
242 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
244 module_param_array(intr_cnt, uint, NULL, 0644);
245 MODULE_PARM_DESC(intr_cnt,
246 "thresholds 1..3 for queue interrupt packet counters");
248 static bool vf_acls;
250 #ifdef CONFIG_PCI_IOV
251 module_param(vf_acls, bool, 0644);
252 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
254 static unsigned int num_vf[4];
256 module_param_array(num_vf, uint, NULL, 0644);
257 MODULE_PARM_DESC(num_vf, "number of VFs for each of PFs 0-3");
258 #endif
260 static struct dentry *cxgb4_debugfs_root;
262 static LIST_HEAD(adapter_list);
263 static DEFINE_MUTEX(uld_mutex);
264 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
265 static const char *uld_str[] = { "RDMA", "iSCSI" };
267 static void link_report(struct net_device *dev)
269 if (!netif_carrier_ok(dev))
270 netdev_info(dev, "link down\n");
271 else {
272 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
274 const char *s = "10Mbps";
275 const struct port_info *p = netdev_priv(dev);
277 switch (p->link_cfg.speed) {
278 case SPEED_10000:
279 s = "10Gbps";
280 break;
281 case SPEED_1000:
282 s = "1000Mbps";
283 break;
284 case SPEED_100:
285 s = "100Mbps";
286 break;
289 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
290 fc[p->link_cfg.fc]);
294 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
296 struct net_device *dev = adapter->port[port_id];
298 /* Skip changes from disabled ports. */
299 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
300 if (link_stat)
301 netif_carrier_on(dev);
302 else
303 netif_carrier_off(dev);
305 link_report(dev);
309 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
311 static const char *mod_str[] = {
312 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
315 const struct net_device *dev = adap->port[port_id];
316 const struct port_info *pi = netdev_priv(dev);
318 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
319 netdev_info(dev, "port module unplugged\n");
320 else if (pi->mod_type < ARRAY_SIZE(mod_str))
321 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
325 * Configure the exact and hash address filters to handle a port's multicast
326 * and secondary unicast MAC addresses.
328 static int set_addr_filters(const struct net_device *dev, bool sleep)
330 u64 mhash = 0;
331 u64 uhash = 0;
332 bool free = true;
333 u16 filt_idx[7];
334 const u8 *addr[7];
335 int ret, naddr = 0;
336 const struct netdev_hw_addr *ha;
337 int uc_cnt = netdev_uc_count(dev);
338 int mc_cnt = netdev_mc_count(dev);
339 const struct port_info *pi = netdev_priv(dev);
340 unsigned int mb = pi->adapter->fn;
342 /* first do the secondary unicast addresses */
343 netdev_for_each_uc_addr(ha, dev) {
344 addr[naddr++] = ha->addr;
345 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
346 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
347 naddr, addr, filt_idx, &uhash, sleep);
348 if (ret < 0)
349 return ret;
351 free = false;
352 naddr = 0;
356 /* next set up the multicast addresses */
357 netdev_for_each_mc_addr(ha, dev) {
358 addr[naddr++] = ha->addr;
359 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
360 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
361 naddr, addr, filt_idx, &mhash, sleep);
362 if (ret < 0)
363 return ret;
365 free = false;
366 naddr = 0;
370 return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
371 uhash | mhash, sleep);
375 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
376 * If @mtu is -1 it is left unchanged.
378 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
380 int ret;
381 struct port_info *pi = netdev_priv(dev);
383 ret = set_addr_filters(dev, sleep_ok);
384 if (ret == 0)
385 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
386 (dev->flags & IFF_PROMISC) ? 1 : 0,
387 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
388 sleep_ok);
389 return ret;
393 * link_start - enable a port
394 * @dev: the port to enable
396 * Performs the MAC and PHY actions needed to enable a port.
398 static int link_start(struct net_device *dev)
400 int ret;
401 struct port_info *pi = netdev_priv(dev);
402 unsigned int mb = pi->adapter->fn;
405 * We do not set address filters and promiscuity here, the stack does
406 * that step explicitly.
408 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
409 !!(dev->features & NETIF_F_HW_VLAN_RX), true);
410 if (ret == 0) {
411 ret = t4_change_mac(pi->adapter, mb, pi->viid,
412 pi->xact_addr_filt, dev->dev_addr, true,
413 true);
414 if (ret >= 0) {
415 pi->xact_addr_filt = ret;
416 ret = 0;
419 if (ret == 0)
420 ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
421 &pi->link_cfg);
422 if (ret == 0)
423 ret = t4_enable_vi(pi->adapter, mb, pi->viid, true, true);
424 return ret;
428 * Response queue handler for the FW event queue.
430 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
431 const struct pkt_gl *gl)
433 u8 opcode = ((const struct rss_header *)rsp)->opcode;
435 rsp++; /* skip RSS header */
436 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
437 const struct cpl_sge_egr_update *p = (void *)rsp;
438 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
439 struct sge_txq *txq;
441 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
442 txq->restarts++;
443 if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
444 struct sge_eth_txq *eq;
446 eq = container_of(txq, struct sge_eth_txq, q);
447 netif_tx_wake_queue(eq->txq);
448 } else {
449 struct sge_ofld_txq *oq;
451 oq = container_of(txq, struct sge_ofld_txq, q);
452 tasklet_schedule(&oq->qresume_tsk);
454 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
455 const struct cpl_fw6_msg *p = (void *)rsp;
457 if (p->type == 0)
458 t4_handle_fw_rpl(q->adap, p->data);
459 } else if (opcode == CPL_L2T_WRITE_RPL) {
460 const struct cpl_l2t_write_rpl *p = (void *)rsp;
462 do_l2t_write_rpl(q->adap, p);
463 } else
464 dev_err(q->adap->pdev_dev,
465 "unexpected CPL %#x on FW event queue\n", opcode);
466 return 0;
470 * uldrx_handler - response queue handler for ULD queues
471 * @q: the response queue that received the packet
472 * @rsp: the response queue descriptor holding the offload message
473 * @gl: the gather list of packet fragments
475 * Deliver an ingress offload packet to a ULD. All processing is done by
476 * the ULD, we just maintain statistics.
478 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
479 const struct pkt_gl *gl)
481 struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
483 if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
484 rxq->stats.nomem++;
485 return -1;
487 if (gl == NULL)
488 rxq->stats.imm++;
489 else if (gl == CXGB4_MSG_AN)
490 rxq->stats.an++;
491 else
492 rxq->stats.pkts++;
493 return 0;
496 static void disable_msi(struct adapter *adapter)
498 if (adapter->flags & USING_MSIX) {
499 pci_disable_msix(adapter->pdev);
500 adapter->flags &= ~USING_MSIX;
501 } else if (adapter->flags & USING_MSI) {
502 pci_disable_msi(adapter->pdev);
503 adapter->flags &= ~USING_MSI;
508 * Interrupt handler for non-data events used with MSI-X.
510 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
512 struct adapter *adap = cookie;
514 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
515 if (v & PFSW) {
516 adap->swintr = 1;
517 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
519 t4_slow_intr_handler(adap);
520 return IRQ_HANDLED;
524 * Name the MSI-X interrupts.
526 static void name_msix_vecs(struct adapter *adap)
528 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
530 /* non-data interrupts */
531 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
533 /* FW events */
534 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
535 adap->port[0]->name);
537 /* Ethernet queues */
538 for_each_port(adap, j) {
539 struct net_device *d = adap->port[j];
540 const struct port_info *pi = netdev_priv(d);
542 for (i = 0; i < pi->nqsets; i++, msi_idx++)
543 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
544 d->name, i);
547 /* offload queues */
548 for_each_ofldrxq(&adap->sge, i)
549 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-ofld%d",
550 adap->port[0]->name, i);
552 for_each_rdmarxq(&adap->sge, i)
553 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
554 adap->port[0]->name, i);
557 static int request_msix_queue_irqs(struct adapter *adap)
559 struct sge *s = &adap->sge;
560 int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi = 2;
562 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
563 adap->msix_info[1].desc, &s->fw_evtq);
564 if (err)
565 return err;
567 for_each_ethrxq(s, ethqidx) {
568 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
569 adap->msix_info[msi].desc,
570 &s->ethrxq[ethqidx].rspq);
571 if (err)
572 goto unwind;
573 msi++;
575 for_each_ofldrxq(s, ofldqidx) {
576 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
577 adap->msix_info[msi].desc,
578 &s->ofldrxq[ofldqidx].rspq);
579 if (err)
580 goto unwind;
581 msi++;
583 for_each_rdmarxq(s, rdmaqidx) {
584 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
585 adap->msix_info[msi].desc,
586 &s->rdmarxq[rdmaqidx].rspq);
587 if (err)
588 goto unwind;
589 msi++;
591 return 0;
593 unwind:
594 while (--rdmaqidx >= 0)
595 free_irq(adap->msix_info[--msi].vec,
596 &s->rdmarxq[rdmaqidx].rspq);
597 while (--ofldqidx >= 0)
598 free_irq(adap->msix_info[--msi].vec,
599 &s->ofldrxq[ofldqidx].rspq);
600 while (--ethqidx >= 0)
601 free_irq(adap->msix_info[--msi].vec, &s->ethrxq[ethqidx].rspq);
602 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
603 return err;
606 static void free_msix_queue_irqs(struct adapter *adap)
608 int i, msi = 2;
609 struct sge *s = &adap->sge;
611 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
612 for_each_ethrxq(s, i)
613 free_irq(adap->msix_info[msi++].vec, &s->ethrxq[i].rspq);
614 for_each_ofldrxq(s, i)
615 free_irq(adap->msix_info[msi++].vec, &s->ofldrxq[i].rspq);
616 for_each_rdmarxq(s, i)
617 free_irq(adap->msix_info[msi++].vec, &s->rdmarxq[i].rspq);
621 * write_rss - write the RSS table for a given port
622 * @pi: the port
623 * @queues: array of queue indices for RSS
625 * Sets up the portion of the HW RSS table for the port's VI to distribute
626 * packets to the Rx queues in @queues.
628 static int write_rss(const struct port_info *pi, const u16 *queues)
630 u16 *rss;
631 int i, err;
632 const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
634 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
635 if (!rss)
636 return -ENOMEM;
638 /* map the queue indices to queue ids */
639 for (i = 0; i < pi->rss_size; i++, queues++)
640 rss[i] = q[*queues].rspq.abs_id;
642 err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
643 pi->rss_size, rss, pi->rss_size);
644 kfree(rss);
645 return err;
649 * setup_rss - configure RSS
650 * @adap: the adapter
652 * Sets up RSS for each port.
654 static int setup_rss(struct adapter *adap)
656 int i, err;
658 for_each_port(adap, i) {
659 const struct port_info *pi = adap2pinfo(adap, i);
661 err = write_rss(pi, pi->rss);
662 if (err)
663 return err;
665 return 0;
669 * Return the channel of the ingress queue with the given qid.
671 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
673 qid -= p->ingr_start;
674 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
678 * Wait until all NAPI handlers are descheduled.
680 static void quiesce_rx(struct adapter *adap)
682 int i;
684 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
685 struct sge_rspq *q = adap->sge.ingr_map[i];
687 if (q && q->handler)
688 napi_disable(&q->napi);
693 * Enable NAPI scheduling and interrupt generation for all Rx queues.
695 static void enable_rx(struct adapter *adap)
697 int i;
699 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
700 struct sge_rspq *q = adap->sge.ingr_map[i];
702 if (!q)
703 continue;
704 if (q->handler)
705 napi_enable(&q->napi);
706 /* 0-increment GTS to start the timer and enable interrupts */
707 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
708 SEINTARM(q->intr_params) |
709 INGRESSQID(q->cntxt_id));
714 * setup_sge_queues - configure SGE Tx/Rx/response queues
715 * @adap: the adapter
717 * Determines how many sets of SGE queues to use and initializes them.
718 * We support multiple queue sets per port if we have MSI-X, otherwise
719 * just one queue set per port.
721 static int setup_sge_queues(struct adapter *adap)
723 int err, msi_idx, i, j;
724 struct sge *s = &adap->sge;
726 bitmap_zero(s->starving_fl, MAX_EGRQ);
727 bitmap_zero(s->txq_maperr, MAX_EGRQ);
729 if (adap->flags & USING_MSIX)
730 msi_idx = 1; /* vector 0 is for non-queue interrupts */
731 else {
732 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
733 NULL, NULL);
734 if (err)
735 return err;
736 msi_idx = -((int)s->intrq.abs_id + 1);
739 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
740 msi_idx, NULL, fwevtq_handler);
741 if (err) {
742 freeout: t4_free_sge_resources(adap);
743 return err;
746 for_each_port(adap, i) {
747 struct net_device *dev = adap->port[i];
748 struct port_info *pi = netdev_priv(dev);
749 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
750 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
752 for (j = 0; j < pi->nqsets; j++, q++) {
753 if (msi_idx > 0)
754 msi_idx++;
755 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
756 msi_idx, &q->fl,
757 t4_ethrx_handler);
758 if (err)
759 goto freeout;
760 q->rspq.idx = j;
761 memset(&q->stats, 0, sizeof(q->stats));
763 for (j = 0; j < pi->nqsets; j++, t++) {
764 err = t4_sge_alloc_eth_txq(adap, t, dev,
765 netdev_get_tx_queue(dev, j),
766 s->fw_evtq.cntxt_id);
767 if (err)
768 goto freeout;
772 j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
773 for_each_ofldrxq(s, i) {
774 struct sge_ofld_rxq *q = &s->ofldrxq[i];
775 struct net_device *dev = adap->port[i / j];
777 if (msi_idx > 0)
778 msi_idx++;
779 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
780 &q->fl, uldrx_handler);
781 if (err)
782 goto freeout;
783 memset(&q->stats, 0, sizeof(q->stats));
784 s->ofld_rxq[i] = q->rspq.abs_id;
785 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
786 s->fw_evtq.cntxt_id);
787 if (err)
788 goto freeout;
791 for_each_rdmarxq(s, i) {
792 struct sge_ofld_rxq *q = &s->rdmarxq[i];
794 if (msi_idx > 0)
795 msi_idx++;
796 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
797 msi_idx, &q->fl, uldrx_handler);
798 if (err)
799 goto freeout;
800 memset(&q->stats, 0, sizeof(q->stats));
801 s->rdma_rxq[i] = q->rspq.abs_id;
804 for_each_port(adap, i) {
806 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
807 * have RDMA queues, and that's the right value.
809 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
810 s->fw_evtq.cntxt_id,
811 s->rdmarxq[i].rspq.cntxt_id);
812 if (err)
813 goto freeout;
816 t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
817 RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
818 QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
819 return 0;
823 * Returns 0 if new FW was successfully loaded, a positive errno if a load was
824 * started but failed, and a negative errno if flash load couldn't start.
826 static int upgrade_fw(struct adapter *adap)
828 int ret;
829 u32 vers;
830 const struct fw_hdr *hdr;
831 const struct firmware *fw;
832 struct device *dev = adap->pdev_dev;
834 ret = request_firmware(&fw, FW_FNAME, dev);
835 if (ret < 0) {
836 dev_err(dev, "unable to load firmware image " FW_FNAME
837 ", error %d\n", ret);
838 return ret;
841 hdr = (const struct fw_hdr *)fw->data;
842 vers = ntohl(hdr->fw_ver);
843 if (FW_HDR_FW_VER_MAJOR_GET(vers) != FW_VERSION_MAJOR) {
844 ret = -EINVAL; /* wrong major version, won't do */
845 goto out;
849 * If the flash FW is unusable or we found something newer, load it.
851 if (FW_HDR_FW_VER_MAJOR_GET(adap->params.fw_vers) != FW_VERSION_MAJOR ||
852 vers > adap->params.fw_vers) {
853 ret = -t4_load_fw(adap, fw->data, fw->size);
854 if (!ret)
855 dev_info(dev, "firmware upgraded to version %pI4 from "
856 FW_FNAME "\n", &hdr->fw_ver);
858 out: release_firmware(fw);
859 return ret;
863 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
864 * The allocated memory is cleared.
866 void *t4_alloc_mem(size_t size)
868 void *p = kzalloc(size, GFP_KERNEL);
870 if (!p)
871 p = vzalloc(size);
872 return p;
876 * Free memory allocated through alloc_mem().
878 static void t4_free_mem(void *addr)
880 if (is_vmalloc_addr(addr))
881 vfree(addr);
882 else
883 kfree(addr);
886 static inline int is_offload(const struct adapter *adap)
888 return adap->params.offload;
892 * Implementation of ethtool operations.
895 static u32 get_msglevel(struct net_device *dev)
897 return netdev2adap(dev)->msg_enable;
900 static void set_msglevel(struct net_device *dev, u32 val)
902 netdev2adap(dev)->msg_enable = val;
905 static char stats_strings[][ETH_GSTRING_LEN] = {
906 "TxOctetsOK ",
907 "TxFramesOK ",
908 "TxBroadcastFrames ",
909 "TxMulticastFrames ",
910 "TxUnicastFrames ",
911 "TxErrorFrames ",
913 "TxFrames64 ",
914 "TxFrames65To127 ",
915 "TxFrames128To255 ",
916 "TxFrames256To511 ",
917 "TxFrames512To1023 ",
918 "TxFrames1024To1518 ",
919 "TxFrames1519ToMax ",
921 "TxFramesDropped ",
922 "TxPauseFrames ",
923 "TxPPP0Frames ",
924 "TxPPP1Frames ",
925 "TxPPP2Frames ",
926 "TxPPP3Frames ",
927 "TxPPP4Frames ",
928 "TxPPP5Frames ",
929 "TxPPP6Frames ",
930 "TxPPP7Frames ",
932 "RxOctetsOK ",
933 "RxFramesOK ",
934 "RxBroadcastFrames ",
935 "RxMulticastFrames ",
936 "RxUnicastFrames ",
938 "RxFramesTooLong ",
939 "RxJabberErrors ",
940 "RxFCSErrors ",
941 "RxLengthErrors ",
942 "RxSymbolErrors ",
943 "RxRuntFrames ",
945 "RxFrames64 ",
946 "RxFrames65To127 ",
947 "RxFrames128To255 ",
948 "RxFrames256To511 ",
949 "RxFrames512To1023 ",
950 "RxFrames1024To1518 ",
951 "RxFrames1519ToMax ",
953 "RxPauseFrames ",
954 "RxPPP0Frames ",
955 "RxPPP1Frames ",
956 "RxPPP2Frames ",
957 "RxPPP3Frames ",
958 "RxPPP4Frames ",
959 "RxPPP5Frames ",
960 "RxPPP6Frames ",
961 "RxPPP7Frames ",
963 "RxBG0FramesDropped ",
964 "RxBG1FramesDropped ",
965 "RxBG2FramesDropped ",
966 "RxBG3FramesDropped ",
967 "RxBG0FramesTrunc ",
968 "RxBG1FramesTrunc ",
969 "RxBG2FramesTrunc ",
970 "RxBG3FramesTrunc ",
972 "TSO ",
973 "TxCsumOffload ",
974 "RxCsumGood ",
975 "VLANextractions ",
976 "VLANinsertions ",
977 "GROpackets ",
978 "GROmerged ",
981 static int get_sset_count(struct net_device *dev, int sset)
983 switch (sset) {
984 case ETH_SS_STATS:
985 return ARRAY_SIZE(stats_strings);
986 default:
987 return -EOPNOTSUPP;
991 #define T4_REGMAP_SIZE (160 * 1024)
993 static int get_regs_len(struct net_device *dev)
995 return T4_REGMAP_SIZE;
998 static int get_eeprom_len(struct net_device *dev)
1000 return EEPROMSIZE;
1003 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1005 struct adapter *adapter = netdev2adap(dev);
1007 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1008 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1009 strlcpy(info->bus_info, pci_name(adapter->pdev),
1010 sizeof(info->bus_info));
1012 if (adapter->params.fw_vers)
1013 snprintf(info->fw_version, sizeof(info->fw_version),
1014 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1015 FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1016 FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1017 FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1018 FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1019 FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1020 FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1021 FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1022 FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1025 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1027 if (stringset == ETH_SS_STATS)
1028 memcpy(data, stats_strings, sizeof(stats_strings));
1032 * port stats maintained per queue of the port. They should be in the same
1033 * order as in stats_strings above.
1035 struct queue_port_stats {
1036 u64 tso;
1037 u64 tx_csum;
1038 u64 rx_csum;
1039 u64 vlan_ex;
1040 u64 vlan_ins;
1041 u64 gro_pkts;
1042 u64 gro_merged;
1045 static void collect_sge_port_stats(const struct adapter *adap,
1046 const struct port_info *p, struct queue_port_stats *s)
1048 int i;
1049 const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1050 const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1052 memset(s, 0, sizeof(*s));
1053 for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1054 s->tso += tx->tso;
1055 s->tx_csum += tx->tx_cso;
1056 s->rx_csum += rx->stats.rx_cso;
1057 s->vlan_ex += rx->stats.vlan_ex;
1058 s->vlan_ins += tx->vlan_ins;
1059 s->gro_pkts += rx->stats.lro_pkts;
1060 s->gro_merged += rx->stats.lro_merged;
1064 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1065 u64 *data)
1067 struct port_info *pi = netdev_priv(dev);
1068 struct adapter *adapter = pi->adapter;
1070 t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1072 data += sizeof(struct port_stats) / sizeof(u64);
1073 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1077 * Return a version number to identify the type of adapter. The scheme is:
1078 * - bits 0..9: chip version
1079 * - bits 10..15: chip revision
1080 * - bits 16..23: register dump version
1082 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1084 return 4 | (ap->params.rev << 10) | (1 << 16);
1087 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1088 unsigned int end)
1090 u32 *p = buf + start;
1092 for ( ; start <= end; start += sizeof(u32))
1093 *p++ = t4_read_reg(ap, start);
1096 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1097 void *buf)
1099 static const unsigned int reg_ranges[] = {
1100 0x1008, 0x1108,
1101 0x1180, 0x11b4,
1102 0x11fc, 0x123c,
1103 0x1300, 0x173c,
1104 0x1800, 0x18fc,
1105 0x3000, 0x30d8,
1106 0x30e0, 0x5924,
1107 0x5960, 0x59d4,
1108 0x5a00, 0x5af8,
1109 0x6000, 0x6098,
1110 0x6100, 0x6150,
1111 0x6200, 0x6208,
1112 0x6240, 0x6248,
1113 0x6280, 0x6338,
1114 0x6370, 0x638c,
1115 0x6400, 0x643c,
1116 0x6500, 0x6524,
1117 0x6a00, 0x6a38,
1118 0x6a60, 0x6a78,
1119 0x6b00, 0x6b84,
1120 0x6bf0, 0x6c84,
1121 0x6cf0, 0x6d84,
1122 0x6df0, 0x6e84,
1123 0x6ef0, 0x6f84,
1124 0x6ff0, 0x7084,
1125 0x70f0, 0x7184,
1126 0x71f0, 0x7284,
1127 0x72f0, 0x7384,
1128 0x73f0, 0x7450,
1129 0x7500, 0x7530,
1130 0x7600, 0x761c,
1131 0x7680, 0x76cc,
1132 0x7700, 0x7798,
1133 0x77c0, 0x77fc,
1134 0x7900, 0x79fc,
1135 0x7b00, 0x7c38,
1136 0x7d00, 0x7efc,
1137 0x8dc0, 0x8e1c,
1138 0x8e30, 0x8e78,
1139 0x8ea0, 0x8f6c,
1140 0x8fc0, 0x9074,
1141 0x90fc, 0x90fc,
1142 0x9400, 0x9458,
1143 0x9600, 0x96bc,
1144 0x9800, 0x9808,
1145 0x9820, 0x983c,
1146 0x9850, 0x9864,
1147 0x9c00, 0x9c6c,
1148 0x9c80, 0x9cec,
1149 0x9d00, 0x9d6c,
1150 0x9d80, 0x9dec,
1151 0x9e00, 0x9e6c,
1152 0x9e80, 0x9eec,
1153 0x9f00, 0x9f6c,
1154 0x9f80, 0x9fec,
1155 0xd004, 0xd03c,
1156 0xdfc0, 0xdfe0,
1157 0xe000, 0xea7c,
1158 0xf000, 0x11190,
1159 0x19040, 0x1906c,
1160 0x19078, 0x19080,
1161 0x1908c, 0x19124,
1162 0x19150, 0x191b0,
1163 0x191d0, 0x191e8,
1164 0x19238, 0x1924c,
1165 0x193f8, 0x19474,
1166 0x19490, 0x194f8,
1167 0x19800, 0x19f30,
1168 0x1a000, 0x1a06c,
1169 0x1a0b0, 0x1a120,
1170 0x1a128, 0x1a138,
1171 0x1a190, 0x1a1c4,
1172 0x1a1fc, 0x1a1fc,
1173 0x1e040, 0x1e04c,
1174 0x1e284, 0x1e28c,
1175 0x1e2c0, 0x1e2c0,
1176 0x1e2e0, 0x1e2e0,
1177 0x1e300, 0x1e384,
1178 0x1e3c0, 0x1e3c8,
1179 0x1e440, 0x1e44c,
1180 0x1e684, 0x1e68c,
1181 0x1e6c0, 0x1e6c0,
1182 0x1e6e0, 0x1e6e0,
1183 0x1e700, 0x1e784,
1184 0x1e7c0, 0x1e7c8,
1185 0x1e840, 0x1e84c,
1186 0x1ea84, 0x1ea8c,
1187 0x1eac0, 0x1eac0,
1188 0x1eae0, 0x1eae0,
1189 0x1eb00, 0x1eb84,
1190 0x1ebc0, 0x1ebc8,
1191 0x1ec40, 0x1ec4c,
1192 0x1ee84, 0x1ee8c,
1193 0x1eec0, 0x1eec0,
1194 0x1eee0, 0x1eee0,
1195 0x1ef00, 0x1ef84,
1196 0x1efc0, 0x1efc8,
1197 0x1f040, 0x1f04c,
1198 0x1f284, 0x1f28c,
1199 0x1f2c0, 0x1f2c0,
1200 0x1f2e0, 0x1f2e0,
1201 0x1f300, 0x1f384,
1202 0x1f3c0, 0x1f3c8,
1203 0x1f440, 0x1f44c,
1204 0x1f684, 0x1f68c,
1205 0x1f6c0, 0x1f6c0,
1206 0x1f6e0, 0x1f6e0,
1207 0x1f700, 0x1f784,
1208 0x1f7c0, 0x1f7c8,
1209 0x1f840, 0x1f84c,
1210 0x1fa84, 0x1fa8c,
1211 0x1fac0, 0x1fac0,
1212 0x1fae0, 0x1fae0,
1213 0x1fb00, 0x1fb84,
1214 0x1fbc0, 0x1fbc8,
1215 0x1fc40, 0x1fc4c,
1216 0x1fe84, 0x1fe8c,
1217 0x1fec0, 0x1fec0,
1218 0x1fee0, 0x1fee0,
1219 0x1ff00, 0x1ff84,
1220 0x1ffc0, 0x1ffc8,
1221 0x20000, 0x2002c,
1222 0x20100, 0x2013c,
1223 0x20190, 0x201c8,
1224 0x20200, 0x20318,
1225 0x20400, 0x20528,
1226 0x20540, 0x20614,
1227 0x21000, 0x21040,
1228 0x2104c, 0x21060,
1229 0x210c0, 0x210ec,
1230 0x21200, 0x21268,
1231 0x21270, 0x21284,
1232 0x212fc, 0x21388,
1233 0x21400, 0x21404,
1234 0x21500, 0x21518,
1235 0x2152c, 0x2153c,
1236 0x21550, 0x21554,
1237 0x21600, 0x21600,
1238 0x21608, 0x21628,
1239 0x21630, 0x2163c,
1240 0x21700, 0x2171c,
1241 0x21780, 0x2178c,
1242 0x21800, 0x21c38,
1243 0x21c80, 0x21d7c,
1244 0x21e00, 0x21e04,
1245 0x22000, 0x2202c,
1246 0x22100, 0x2213c,
1247 0x22190, 0x221c8,
1248 0x22200, 0x22318,
1249 0x22400, 0x22528,
1250 0x22540, 0x22614,
1251 0x23000, 0x23040,
1252 0x2304c, 0x23060,
1253 0x230c0, 0x230ec,
1254 0x23200, 0x23268,
1255 0x23270, 0x23284,
1256 0x232fc, 0x23388,
1257 0x23400, 0x23404,
1258 0x23500, 0x23518,
1259 0x2352c, 0x2353c,
1260 0x23550, 0x23554,
1261 0x23600, 0x23600,
1262 0x23608, 0x23628,
1263 0x23630, 0x2363c,
1264 0x23700, 0x2371c,
1265 0x23780, 0x2378c,
1266 0x23800, 0x23c38,
1267 0x23c80, 0x23d7c,
1268 0x23e00, 0x23e04,
1269 0x24000, 0x2402c,
1270 0x24100, 0x2413c,
1271 0x24190, 0x241c8,
1272 0x24200, 0x24318,
1273 0x24400, 0x24528,
1274 0x24540, 0x24614,
1275 0x25000, 0x25040,
1276 0x2504c, 0x25060,
1277 0x250c0, 0x250ec,
1278 0x25200, 0x25268,
1279 0x25270, 0x25284,
1280 0x252fc, 0x25388,
1281 0x25400, 0x25404,
1282 0x25500, 0x25518,
1283 0x2552c, 0x2553c,
1284 0x25550, 0x25554,
1285 0x25600, 0x25600,
1286 0x25608, 0x25628,
1287 0x25630, 0x2563c,
1288 0x25700, 0x2571c,
1289 0x25780, 0x2578c,
1290 0x25800, 0x25c38,
1291 0x25c80, 0x25d7c,
1292 0x25e00, 0x25e04,
1293 0x26000, 0x2602c,
1294 0x26100, 0x2613c,
1295 0x26190, 0x261c8,
1296 0x26200, 0x26318,
1297 0x26400, 0x26528,
1298 0x26540, 0x26614,
1299 0x27000, 0x27040,
1300 0x2704c, 0x27060,
1301 0x270c0, 0x270ec,
1302 0x27200, 0x27268,
1303 0x27270, 0x27284,
1304 0x272fc, 0x27388,
1305 0x27400, 0x27404,
1306 0x27500, 0x27518,
1307 0x2752c, 0x2753c,
1308 0x27550, 0x27554,
1309 0x27600, 0x27600,
1310 0x27608, 0x27628,
1311 0x27630, 0x2763c,
1312 0x27700, 0x2771c,
1313 0x27780, 0x2778c,
1314 0x27800, 0x27c38,
1315 0x27c80, 0x27d7c,
1316 0x27e00, 0x27e04
1319 int i;
1320 struct adapter *ap = netdev2adap(dev);
1322 regs->version = mk_adap_vers(ap);
1324 memset(buf, 0, T4_REGMAP_SIZE);
1325 for (i = 0; i < ARRAY_SIZE(reg_ranges); i += 2)
1326 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
1329 static int restart_autoneg(struct net_device *dev)
1331 struct port_info *p = netdev_priv(dev);
1333 if (!netif_running(dev))
1334 return -EAGAIN;
1335 if (p->link_cfg.autoneg != AUTONEG_ENABLE)
1336 return -EINVAL;
1337 t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
1338 return 0;
1341 static int identify_port(struct net_device *dev,
1342 enum ethtool_phys_id_state state)
1344 unsigned int val;
1345 struct adapter *adap = netdev2adap(dev);
1347 if (state == ETHTOOL_ID_ACTIVE)
1348 val = 0xffff;
1349 else if (state == ETHTOOL_ID_INACTIVE)
1350 val = 0;
1351 else
1352 return -EINVAL;
1354 return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid, val);
1357 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
1359 unsigned int v = 0;
1361 if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
1362 type == FW_PORT_TYPE_BT_XAUI) {
1363 v |= SUPPORTED_TP;
1364 if (caps & FW_PORT_CAP_SPEED_100M)
1365 v |= SUPPORTED_100baseT_Full;
1366 if (caps & FW_PORT_CAP_SPEED_1G)
1367 v |= SUPPORTED_1000baseT_Full;
1368 if (caps & FW_PORT_CAP_SPEED_10G)
1369 v |= SUPPORTED_10000baseT_Full;
1370 } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
1371 v |= SUPPORTED_Backplane;
1372 if (caps & FW_PORT_CAP_SPEED_1G)
1373 v |= SUPPORTED_1000baseKX_Full;
1374 if (caps & FW_PORT_CAP_SPEED_10G)
1375 v |= SUPPORTED_10000baseKX4_Full;
1376 } else if (type == FW_PORT_TYPE_KR)
1377 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
1378 else if (type == FW_PORT_TYPE_BP_AP)
1379 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
1380 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full;
1381 else if (type == FW_PORT_TYPE_BP4_AP)
1382 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
1383 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full |
1384 SUPPORTED_10000baseKX4_Full;
1385 else if (type == FW_PORT_TYPE_FIBER_XFI ||
1386 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
1387 v |= SUPPORTED_FIBRE;
1389 if (caps & FW_PORT_CAP_ANEG)
1390 v |= SUPPORTED_Autoneg;
1391 return v;
1394 static unsigned int to_fw_linkcaps(unsigned int caps)
1396 unsigned int v = 0;
1398 if (caps & ADVERTISED_100baseT_Full)
1399 v |= FW_PORT_CAP_SPEED_100M;
1400 if (caps & ADVERTISED_1000baseT_Full)
1401 v |= FW_PORT_CAP_SPEED_1G;
1402 if (caps & ADVERTISED_10000baseT_Full)
1403 v |= FW_PORT_CAP_SPEED_10G;
1404 return v;
1407 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1409 const struct port_info *p = netdev_priv(dev);
1411 if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
1412 p->port_type == FW_PORT_TYPE_BT_XFI ||
1413 p->port_type == FW_PORT_TYPE_BT_XAUI)
1414 cmd->port = PORT_TP;
1415 else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
1416 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
1417 cmd->port = PORT_FIBRE;
1418 else if (p->port_type == FW_PORT_TYPE_SFP) {
1419 if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
1420 p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
1421 cmd->port = PORT_DA;
1422 else
1423 cmd->port = PORT_FIBRE;
1424 } else
1425 cmd->port = PORT_OTHER;
1427 if (p->mdio_addr >= 0) {
1428 cmd->phy_address = p->mdio_addr;
1429 cmd->transceiver = XCVR_EXTERNAL;
1430 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
1431 MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
1432 } else {
1433 cmd->phy_address = 0; /* not really, but no better option */
1434 cmd->transceiver = XCVR_INTERNAL;
1435 cmd->mdio_support = 0;
1438 cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
1439 cmd->advertising = from_fw_linkcaps(p->port_type,
1440 p->link_cfg.advertising);
1441 ethtool_cmd_speed_set(cmd,
1442 netif_carrier_ok(dev) ? p->link_cfg.speed : 0);
1443 cmd->duplex = DUPLEX_FULL;
1444 cmd->autoneg = p->link_cfg.autoneg;
1445 cmd->maxtxpkt = 0;
1446 cmd->maxrxpkt = 0;
1447 return 0;
1450 static unsigned int speed_to_caps(int speed)
1452 if (speed == SPEED_100)
1453 return FW_PORT_CAP_SPEED_100M;
1454 if (speed == SPEED_1000)
1455 return FW_PORT_CAP_SPEED_1G;
1456 if (speed == SPEED_10000)
1457 return FW_PORT_CAP_SPEED_10G;
1458 return 0;
1461 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1463 unsigned int cap;
1464 struct port_info *p = netdev_priv(dev);
1465 struct link_config *lc = &p->link_cfg;
1466 u32 speed = ethtool_cmd_speed(cmd);
1468 if (cmd->duplex != DUPLEX_FULL) /* only full-duplex supported */
1469 return -EINVAL;
1471 if (!(lc->supported & FW_PORT_CAP_ANEG)) {
1473 * PHY offers a single speed. See if that's what's
1474 * being requested.
1476 if (cmd->autoneg == AUTONEG_DISABLE &&
1477 (lc->supported & speed_to_caps(speed)))
1478 return 0;
1479 return -EINVAL;
1482 if (cmd->autoneg == AUTONEG_DISABLE) {
1483 cap = speed_to_caps(speed);
1485 if (!(lc->supported & cap) || (speed == SPEED_1000) ||
1486 (speed == SPEED_10000))
1487 return -EINVAL;
1488 lc->requested_speed = cap;
1489 lc->advertising = 0;
1490 } else {
1491 cap = to_fw_linkcaps(cmd->advertising);
1492 if (!(lc->supported & cap))
1493 return -EINVAL;
1494 lc->requested_speed = 0;
1495 lc->advertising = cap | FW_PORT_CAP_ANEG;
1497 lc->autoneg = cmd->autoneg;
1499 if (netif_running(dev))
1500 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
1501 lc);
1502 return 0;
1505 static void get_pauseparam(struct net_device *dev,
1506 struct ethtool_pauseparam *epause)
1508 struct port_info *p = netdev_priv(dev);
1510 epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1511 epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
1512 epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
1515 static int set_pauseparam(struct net_device *dev,
1516 struct ethtool_pauseparam *epause)
1518 struct port_info *p = netdev_priv(dev);
1519 struct link_config *lc = &p->link_cfg;
1521 if (epause->autoneg == AUTONEG_DISABLE)
1522 lc->requested_fc = 0;
1523 else if (lc->supported & FW_PORT_CAP_ANEG)
1524 lc->requested_fc = PAUSE_AUTONEG;
1525 else
1526 return -EINVAL;
1528 if (epause->rx_pause)
1529 lc->requested_fc |= PAUSE_RX;
1530 if (epause->tx_pause)
1531 lc->requested_fc |= PAUSE_TX;
1532 if (netif_running(dev))
1533 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
1534 lc);
1535 return 0;
1538 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1540 const struct port_info *pi = netdev_priv(dev);
1541 const struct sge *s = &pi->adapter->sge;
1543 e->rx_max_pending = MAX_RX_BUFFERS;
1544 e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1545 e->rx_jumbo_max_pending = 0;
1546 e->tx_max_pending = MAX_TXQ_ENTRIES;
1548 e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
1549 e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1550 e->rx_jumbo_pending = 0;
1551 e->tx_pending = s->ethtxq[pi->first_qset].q.size;
1554 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1556 int i;
1557 const struct port_info *pi = netdev_priv(dev);
1558 struct adapter *adapter = pi->adapter;
1559 struct sge *s = &adapter->sge;
1561 if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
1562 e->tx_pending > MAX_TXQ_ENTRIES ||
1563 e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1564 e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1565 e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
1566 return -EINVAL;
1568 if (adapter->flags & FULL_INIT_DONE)
1569 return -EBUSY;
1571 for (i = 0; i < pi->nqsets; ++i) {
1572 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
1573 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
1574 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
1576 return 0;
1579 static int closest_timer(const struct sge *s, int time)
1581 int i, delta, match = 0, min_delta = INT_MAX;
1583 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1584 delta = time - s->timer_val[i];
1585 if (delta < 0)
1586 delta = -delta;
1587 if (delta < min_delta) {
1588 min_delta = delta;
1589 match = i;
1592 return match;
1595 static int closest_thres(const struct sge *s, int thres)
1597 int i, delta, match = 0, min_delta = INT_MAX;
1599 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1600 delta = thres - s->counter_val[i];
1601 if (delta < 0)
1602 delta = -delta;
1603 if (delta < min_delta) {
1604 min_delta = delta;
1605 match = i;
1608 return match;
1612 * Return a queue's interrupt hold-off time in us. 0 means no timer.
1614 static unsigned int qtimer_val(const struct adapter *adap,
1615 const struct sge_rspq *q)
1617 unsigned int idx = q->intr_params >> 1;
1619 return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
1623 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
1624 * @adap: the adapter
1625 * @q: the Rx queue
1626 * @us: the hold-off time in us, or 0 to disable timer
1627 * @cnt: the hold-off packet count, or 0 to disable counter
1629 * Sets an Rx queue's interrupt hold-off time and packet count. At least
1630 * one of the two needs to be enabled for the queue to generate interrupts.
1632 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
1633 unsigned int us, unsigned int cnt)
1635 if ((us | cnt) == 0)
1636 cnt = 1;
1638 if (cnt) {
1639 int err;
1640 u32 v, new_idx;
1642 new_idx = closest_thres(&adap->sge, cnt);
1643 if (q->desc && q->pktcnt_idx != new_idx) {
1644 /* the queue has already been created, update it */
1645 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1646 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1647 FW_PARAMS_PARAM_YZ(q->cntxt_id);
1648 err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
1649 &new_idx);
1650 if (err)
1651 return err;
1653 q->pktcnt_idx = new_idx;
1656 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1657 q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
1658 return 0;
1661 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1663 const struct port_info *pi = netdev_priv(dev);
1664 struct adapter *adap = pi->adapter;
1666 return set_rxq_intr_params(adap, &adap->sge.ethrxq[pi->first_qset].rspq,
1667 c->rx_coalesce_usecs, c->rx_max_coalesced_frames);
1670 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1672 const struct port_info *pi = netdev_priv(dev);
1673 const struct adapter *adap = pi->adapter;
1674 const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
1676 c->rx_coalesce_usecs = qtimer_val(adap, rq);
1677 c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
1678 adap->sge.counter_val[rq->pktcnt_idx] : 0;
1679 return 0;
1683 * eeprom_ptov - translate a physical EEPROM address to virtual
1684 * @phys_addr: the physical EEPROM address
1685 * @fn: the PCI function number
1686 * @sz: size of function-specific area
1688 * Translate a physical EEPROM address to virtual. The first 1K is
1689 * accessed through virtual addresses starting at 31K, the rest is
1690 * accessed through virtual addresses starting at 0.
1692 * The mapping is as follows:
1693 * [0..1K) -> [31K..32K)
1694 * [1K..1K+A) -> [31K-A..31K)
1695 * [1K+A..ES) -> [0..ES-A-1K)
1697 * where A = @fn * @sz, and ES = EEPROM size.
1699 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
1701 fn *= sz;
1702 if (phys_addr < 1024)
1703 return phys_addr + (31 << 10);
1704 if (phys_addr < 1024 + fn)
1705 return 31744 - fn + phys_addr - 1024;
1706 if (phys_addr < EEPROMSIZE)
1707 return phys_addr - 1024 - fn;
1708 return -EINVAL;
1712 * The next two routines implement eeprom read/write from physical addresses.
1714 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
1716 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
1718 if (vaddr >= 0)
1719 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
1720 return vaddr < 0 ? vaddr : 0;
1723 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
1725 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
1727 if (vaddr >= 0)
1728 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
1729 return vaddr < 0 ? vaddr : 0;
1732 #define EEPROM_MAGIC 0x38E2F10C
1734 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
1735 u8 *data)
1737 int i, err = 0;
1738 struct adapter *adapter = netdev2adap(dev);
1740 u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
1741 if (!buf)
1742 return -ENOMEM;
1744 e->magic = EEPROM_MAGIC;
1745 for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
1746 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
1748 if (!err)
1749 memcpy(data, buf + e->offset, e->len);
1750 kfree(buf);
1751 return err;
1754 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
1755 u8 *data)
1757 u8 *buf;
1758 int err = 0;
1759 u32 aligned_offset, aligned_len, *p;
1760 struct adapter *adapter = netdev2adap(dev);
1762 if (eeprom->magic != EEPROM_MAGIC)
1763 return -EINVAL;
1765 aligned_offset = eeprom->offset & ~3;
1766 aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
1768 if (adapter->fn > 0) {
1769 u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
1771 if (aligned_offset < start ||
1772 aligned_offset + aligned_len > start + EEPROMPFSIZE)
1773 return -EPERM;
1776 if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
1778 * RMW possibly needed for first or last words.
1780 buf = kmalloc(aligned_len, GFP_KERNEL);
1781 if (!buf)
1782 return -ENOMEM;
1783 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
1784 if (!err && aligned_len > 4)
1785 err = eeprom_rd_phys(adapter,
1786 aligned_offset + aligned_len - 4,
1787 (u32 *)&buf[aligned_len - 4]);
1788 if (err)
1789 goto out;
1790 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
1791 } else
1792 buf = data;
1794 err = t4_seeprom_wp(adapter, false);
1795 if (err)
1796 goto out;
1798 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
1799 err = eeprom_wr_phys(adapter, aligned_offset, *p);
1800 aligned_offset += 4;
1803 if (!err)
1804 err = t4_seeprom_wp(adapter, true);
1805 out:
1806 if (buf != data)
1807 kfree(buf);
1808 return err;
1811 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
1813 int ret;
1814 const struct firmware *fw;
1815 struct adapter *adap = netdev2adap(netdev);
1817 ef->data[sizeof(ef->data) - 1] = '\0';
1818 ret = request_firmware(&fw, ef->data, adap->pdev_dev);
1819 if (ret < 0)
1820 return ret;
1822 ret = t4_load_fw(adap, fw->data, fw->size);
1823 release_firmware(fw);
1824 if (!ret)
1825 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
1826 return ret;
1829 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
1830 #define BCAST_CRC 0xa0ccc1a6
1832 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1834 wol->supported = WAKE_BCAST | WAKE_MAGIC;
1835 wol->wolopts = netdev2adap(dev)->wol;
1836 memset(&wol->sopass, 0, sizeof(wol->sopass));
1839 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1841 int err = 0;
1842 struct port_info *pi = netdev_priv(dev);
1844 if (wol->wolopts & ~WOL_SUPPORTED)
1845 return -EINVAL;
1846 t4_wol_magic_enable(pi->adapter, pi->tx_chan,
1847 (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
1848 if (wol->wolopts & WAKE_BCAST) {
1849 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
1850 ~0ULL, 0, false);
1851 if (!err)
1852 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
1853 ~6ULL, ~0ULL, BCAST_CRC, true);
1854 } else
1855 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
1856 return err;
1859 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1861 const struct port_info *pi = netdev_priv(dev);
1862 netdev_features_t changed = dev->features ^ features;
1863 int err;
1865 if (!(changed & NETIF_F_HW_VLAN_RX))
1866 return 0;
1868 err = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1,
1869 -1, -1, -1,
1870 !!(features & NETIF_F_HW_VLAN_RX), true);
1871 if (unlikely(err))
1872 dev->features = features ^ NETIF_F_HW_VLAN_RX;
1873 return err;
1876 static u32 get_rss_table_size(struct net_device *dev)
1878 const struct port_info *pi = netdev_priv(dev);
1880 return pi->rss_size;
1883 static int get_rss_table(struct net_device *dev, u32 *p)
1885 const struct port_info *pi = netdev_priv(dev);
1886 unsigned int n = pi->rss_size;
1888 while (n--)
1889 p[n] = pi->rss[n];
1890 return 0;
1893 static int set_rss_table(struct net_device *dev, const u32 *p)
1895 unsigned int i;
1896 struct port_info *pi = netdev_priv(dev);
1898 for (i = 0; i < pi->rss_size; i++)
1899 pi->rss[i] = p[i];
1900 if (pi->adapter->flags & FULL_INIT_DONE)
1901 return write_rss(pi, pi->rss);
1902 return 0;
1905 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1906 u32 *rules)
1908 const struct port_info *pi = netdev_priv(dev);
1910 switch (info->cmd) {
1911 case ETHTOOL_GRXFH: {
1912 unsigned int v = pi->rss_mode;
1914 info->data = 0;
1915 switch (info->flow_type) {
1916 case TCP_V4_FLOW:
1917 if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
1918 info->data = RXH_IP_SRC | RXH_IP_DST |
1919 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1920 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1921 info->data = RXH_IP_SRC | RXH_IP_DST;
1922 break;
1923 case UDP_V4_FLOW:
1924 if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
1925 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
1926 info->data = RXH_IP_SRC | RXH_IP_DST |
1927 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1928 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1929 info->data = RXH_IP_SRC | RXH_IP_DST;
1930 break;
1931 case SCTP_V4_FLOW:
1932 case AH_ESP_V4_FLOW:
1933 case IPV4_FLOW:
1934 if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
1935 info->data = RXH_IP_SRC | RXH_IP_DST;
1936 break;
1937 case TCP_V6_FLOW:
1938 if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
1939 info->data = RXH_IP_SRC | RXH_IP_DST |
1940 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1941 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1942 info->data = RXH_IP_SRC | RXH_IP_DST;
1943 break;
1944 case UDP_V6_FLOW:
1945 if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
1946 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
1947 info->data = RXH_IP_SRC | RXH_IP_DST |
1948 RXH_L4_B_0_1 | RXH_L4_B_2_3;
1949 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1950 info->data = RXH_IP_SRC | RXH_IP_DST;
1951 break;
1952 case SCTP_V6_FLOW:
1953 case AH_ESP_V6_FLOW:
1954 case IPV6_FLOW:
1955 if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
1956 info->data = RXH_IP_SRC | RXH_IP_DST;
1957 break;
1959 return 0;
1961 case ETHTOOL_GRXRINGS:
1962 info->data = pi->nqsets;
1963 return 0;
1965 return -EOPNOTSUPP;
1968 static const struct ethtool_ops cxgb_ethtool_ops = {
1969 .get_settings = get_settings,
1970 .set_settings = set_settings,
1971 .get_drvinfo = get_drvinfo,
1972 .get_msglevel = get_msglevel,
1973 .set_msglevel = set_msglevel,
1974 .get_ringparam = get_sge_param,
1975 .set_ringparam = set_sge_param,
1976 .get_coalesce = get_coalesce,
1977 .set_coalesce = set_coalesce,
1978 .get_eeprom_len = get_eeprom_len,
1979 .get_eeprom = get_eeprom,
1980 .set_eeprom = set_eeprom,
1981 .get_pauseparam = get_pauseparam,
1982 .set_pauseparam = set_pauseparam,
1983 .get_link = ethtool_op_get_link,
1984 .get_strings = get_strings,
1985 .set_phys_id = identify_port,
1986 .nway_reset = restart_autoneg,
1987 .get_sset_count = get_sset_count,
1988 .get_ethtool_stats = get_stats,
1989 .get_regs_len = get_regs_len,
1990 .get_regs = get_regs,
1991 .get_wol = get_wol,
1992 .set_wol = set_wol,
1993 .get_rxnfc = get_rxnfc,
1994 .get_rxfh_indir_size = get_rss_table_size,
1995 .get_rxfh_indir = get_rss_table,
1996 .set_rxfh_indir = set_rss_table,
1997 .flash_device = set_flash,
2001 * debugfs support
2004 static int mem_open(struct inode *inode, struct file *file)
2006 file->private_data = inode->i_private;
2007 return 0;
2010 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
2011 loff_t *ppos)
2013 loff_t pos = *ppos;
2014 loff_t avail = file->f_path.dentry->d_inode->i_size;
2015 unsigned int mem = (uintptr_t)file->private_data & 3;
2016 struct adapter *adap = file->private_data - mem;
2018 if (pos < 0)
2019 return -EINVAL;
2020 if (pos >= avail)
2021 return 0;
2022 if (count > avail - pos)
2023 count = avail - pos;
2025 while (count) {
2026 size_t len;
2027 int ret, ofst;
2028 __be32 data[16];
2030 if (mem == MEM_MC)
2031 ret = t4_mc_read(adap, pos, data, NULL);
2032 else
2033 ret = t4_edc_read(adap, mem, pos, data, NULL);
2034 if (ret)
2035 return ret;
2037 ofst = pos % sizeof(data);
2038 len = min(count, sizeof(data) - ofst);
2039 if (copy_to_user(buf, (u8 *)data + ofst, len))
2040 return -EFAULT;
2042 buf += len;
2043 pos += len;
2044 count -= len;
2046 count = pos - *ppos;
2047 *ppos = pos;
2048 return count;
2051 static const struct file_operations mem_debugfs_fops = {
2052 .owner = THIS_MODULE,
2053 .open = mem_open,
2054 .read = mem_read,
2055 .llseek = default_llseek,
2058 static void __devinit add_debugfs_mem(struct adapter *adap, const char *name,
2059 unsigned int idx, unsigned int size_mb)
2061 struct dentry *de;
2063 de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2064 (void *)adap + idx, &mem_debugfs_fops);
2065 if (de && de->d_inode)
2066 de->d_inode->i_size = size_mb << 20;
2069 static int __devinit setup_debugfs(struct adapter *adap)
2071 int i;
2073 if (IS_ERR_OR_NULL(adap->debugfs_root))
2074 return -1;
2076 i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2077 if (i & EDRAM0_ENABLE)
2078 add_debugfs_mem(adap, "edc0", MEM_EDC0, 5);
2079 if (i & EDRAM1_ENABLE)
2080 add_debugfs_mem(adap, "edc1", MEM_EDC1, 5);
2081 if (i & EXT_MEM_ENABLE)
2082 add_debugfs_mem(adap, "mc", MEM_MC,
2083 EXT_MEM_SIZE_GET(t4_read_reg(adap, MA_EXT_MEMORY_BAR)));
2084 if (adap->l2t)
2085 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2086 &t4_l2t_fops);
2087 return 0;
2091 * upper-layer driver support
2095 * Allocate an active-open TID and set it to the supplied value.
2097 int cxgb4_alloc_atid(struct tid_info *t, void *data)
2099 int atid = -1;
2101 spin_lock_bh(&t->atid_lock);
2102 if (t->afree) {
2103 union aopen_entry *p = t->afree;
2105 atid = p - t->atid_tab;
2106 t->afree = p->next;
2107 p->data = data;
2108 t->atids_in_use++;
2110 spin_unlock_bh(&t->atid_lock);
2111 return atid;
2113 EXPORT_SYMBOL(cxgb4_alloc_atid);
2116 * Release an active-open TID.
2118 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
2120 union aopen_entry *p = &t->atid_tab[atid];
2122 spin_lock_bh(&t->atid_lock);
2123 p->next = t->afree;
2124 t->afree = p;
2125 t->atids_in_use--;
2126 spin_unlock_bh(&t->atid_lock);
2128 EXPORT_SYMBOL(cxgb4_free_atid);
2131 * Allocate a server TID and set it to the supplied value.
2133 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
2135 int stid;
2137 spin_lock_bh(&t->stid_lock);
2138 if (family == PF_INET) {
2139 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
2140 if (stid < t->nstids)
2141 __set_bit(stid, t->stid_bmap);
2142 else
2143 stid = -1;
2144 } else {
2145 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
2146 if (stid < 0)
2147 stid = -1;
2149 if (stid >= 0) {
2150 t->stid_tab[stid].data = data;
2151 stid += t->stid_base;
2152 t->stids_in_use++;
2154 spin_unlock_bh(&t->stid_lock);
2155 return stid;
2157 EXPORT_SYMBOL(cxgb4_alloc_stid);
2160 * Release a server TID.
2162 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
2164 stid -= t->stid_base;
2165 spin_lock_bh(&t->stid_lock);
2166 if (family == PF_INET)
2167 __clear_bit(stid, t->stid_bmap);
2168 else
2169 bitmap_release_region(t->stid_bmap, stid, 2);
2170 t->stid_tab[stid].data = NULL;
2171 t->stids_in_use--;
2172 spin_unlock_bh(&t->stid_lock);
2174 EXPORT_SYMBOL(cxgb4_free_stid);
2177 * Populate a TID_RELEASE WR. Caller must properly size the skb.
2179 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
2180 unsigned int tid)
2182 struct cpl_tid_release *req;
2184 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
2185 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
2186 INIT_TP_WR(req, tid);
2187 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
2191 * Queue a TID release request and if necessary schedule a work queue to
2192 * process it.
2194 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
2195 unsigned int tid)
2197 void **p = &t->tid_tab[tid];
2198 struct adapter *adap = container_of(t, struct adapter, tids);
2200 spin_lock_bh(&adap->tid_release_lock);
2201 *p = adap->tid_release_head;
2202 /* Low 2 bits encode the Tx channel number */
2203 adap->tid_release_head = (void **)((uintptr_t)p | chan);
2204 if (!adap->tid_release_task_busy) {
2205 adap->tid_release_task_busy = true;
2206 schedule_work(&adap->tid_release_task);
2208 spin_unlock_bh(&adap->tid_release_lock);
2212 * Process the list of pending TID release requests.
2214 static void process_tid_release_list(struct work_struct *work)
2216 struct sk_buff *skb;
2217 struct adapter *adap;
2219 adap = container_of(work, struct adapter, tid_release_task);
2221 spin_lock_bh(&adap->tid_release_lock);
2222 while (adap->tid_release_head) {
2223 void **p = adap->tid_release_head;
2224 unsigned int chan = (uintptr_t)p & 3;
2225 p = (void *)p - chan;
2227 adap->tid_release_head = *p;
2228 *p = NULL;
2229 spin_unlock_bh(&adap->tid_release_lock);
2231 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
2232 GFP_KERNEL)))
2233 schedule_timeout_uninterruptible(1);
2235 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
2236 t4_ofld_send(adap, skb);
2237 spin_lock_bh(&adap->tid_release_lock);
2239 adap->tid_release_task_busy = false;
2240 spin_unlock_bh(&adap->tid_release_lock);
2244 * Release a TID and inform HW. If we are unable to allocate the release
2245 * message we defer to a work queue.
2247 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
2249 void *old;
2250 struct sk_buff *skb;
2251 struct adapter *adap = container_of(t, struct adapter, tids);
2253 old = t->tid_tab[tid];
2254 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
2255 if (likely(skb)) {
2256 t->tid_tab[tid] = NULL;
2257 mk_tid_release(skb, chan, tid);
2258 t4_ofld_send(adap, skb);
2259 } else
2260 cxgb4_queue_tid_release(t, chan, tid);
2261 if (old)
2262 atomic_dec(&t->tids_in_use);
2264 EXPORT_SYMBOL(cxgb4_remove_tid);
2267 * Allocate and initialize the TID tables. Returns 0 on success.
2269 static int tid_init(struct tid_info *t)
2271 size_t size;
2272 unsigned int natids = t->natids;
2274 size = t->ntids * sizeof(*t->tid_tab) + natids * sizeof(*t->atid_tab) +
2275 t->nstids * sizeof(*t->stid_tab) +
2276 BITS_TO_LONGS(t->nstids) * sizeof(long);
2277 t->tid_tab = t4_alloc_mem(size);
2278 if (!t->tid_tab)
2279 return -ENOMEM;
2281 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
2282 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
2283 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids];
2284 spin_lock_init(&t->stid_lock);
2285 spin_lock_init(&t->atid_lock);
2287 t->stids_in_use = 0;
2288 t->afree = NULL;
2289 t->atids_in_use = 0;
2290 atomic_set(&t->tids_in_use, 0);
2292 /* Setup the free list for atid_tab and clear the stid bitmap. */
2293 if (natids) {
2294 while (--natids)
2295 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
2296 t->afree = t->atid_tab;
2298 bitmap_zero(t->stid_bmap, t->nstids);
2299 return 0;
2303 * cxgb4_create_server - create an IP server
2304 * @dev: the device
2305 * @stid: the server TID
2306 * @sip: local IP address to bind server to
2307 * @sport: the server's TCP port
2308 * @queue: queue to direct messages from this server to
2310 * Create an IP server for the given port and address.
2311 * Returns <0 on error and one of the %NET_XMIT_* values on success.
2313 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
2314 __be32 sip, __be16 sport, unsigned int queue)
2316 unsigned int chan;
2317 struct sk_buff *skb;
2318 struct adapter *adap;
2319 struct cpl_pass_open_req *req;
2321 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
2322 if (!skb)
2323 return -ENOMEM;
2325 adap = netdev2adap(dev);
2326 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
2327 INIT_TP_WR(req, 0);
2328 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
2329 req->local_port = sport;
2330 req->peer_port = htons(0);
2331 req->local_ip = sip;
2332 req->peer_ip = htonl(0);
2333 chan = rxq_to_chan(&adap->sge, queue);
2334 req->opt0 = cpu_to_be64(TX_CHAN(chan));
2335 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
2336 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
2337 return t4_mgmt_tx(adap, skb);
2339 EXPORT_SYMBOL(cxgb4_create_server);
2342 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
2343 * @mtus: the HW MTU table
2344 * @mtu: the target MTU
2345 * @idx: index of selected entry in the MTU table
2347 * Returns the index and the value in the HW MTU table that is closest to
2348 * but does not exceed @mtu, unless @mtu is smaller than any value in the
2349 * table, in which case that smallest available value is selected.
2351 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
2352 unsigned int *idx)
2354 unsigned int i = 0;
2356 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
2357 ++i;
2358 if (idx)
2359 *idx = i;
2360 return mtus[i];
2362 EXPORT_SYMBOL(cxgb4_best_mtu);
2365 * cxgb4_port_chan - get the HW channel of a port
2366 * @dev: the net device for the port
2368 * Return the HW Tx channel of the given port.
2370 unsigned int cxgb4_port_chan(const struct net_device *dev)
2372 return netdev2pinfo(dev)->tx_chan;
2374 EXPORT_SYMBOL(cxgb4_port_chan);
2377 * cxgb4_port_viid - get the VI id of a port
2378 * @dev: the net device for the port
2380 * Return the VI id of the given port.
2382 unsigned int cxgb4_port_viid(const struct net_device *dev)
2384 return netdev2pinfo(dev)->viid;
2386 EXPORT_SYMBOL(cxgb4_port_viid);
2389 * cxgb4_port_idx - get the index of a port
2390 * @dev: the net device for the port
2392 * Return the index of the given port.
2394 unsigned int cxgb4_port_idx(const struct net_device *dev)
2396 return netdev2pinfo(dev)->port_id;
2398 EXPORT_SYMBOL(cxgb4_port_idx);
2400 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
2401 struct tp_tcp_stats *v6)
2403 struct adapter *adap = pci_get_drvdata(pdev);
2405 spin_lock(&adap->stats_lock);
2406 t4_tp_get_tcp_stats(adap, v4, v6);
2407 spin_unlock(&adap->stats_lock);
2409 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
2411 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
2412 const unsigned int *pgsz_order)
2414 struct adapter *adap = netdev2adap(dev);
2416 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
2417 t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
2418 HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
2419 HPZ3(pgsz_order[3]));
2421 EXPORT_SYMBOL(cxgb4_iscsi_init);
2423 static struct pci_driver cxgb4_driver;
2425 static void check_neigh_update(struct neighbour *neigh)
2427 const struct device *parent;
2428 const struct net_device *netdev = neigh->dev;
2430 if (netdev->priv_flags & IFF_802_1Q_VLAN)
2431 netdev = vlan_dev_real_dev(netdev);
2432 parent = netdev->dev.parent;
2433 if (parent && parent->driver == &cxgb4_driver.driver)
2434 t4_l2t_update(dev_get_drvdata(parent), neigh);
2437 static int netevent_cb(struct notifier_block *nb, unsigned long event,
2438 void *data)
2440 switch (event) {
2441 case NETEVENT_NEIGH_UPDATE:
2442 check_neigh_update(data);
2443 break;
2444 case NETEVENT_REDIRECT:
2445 default:
2446 break;
2448 return 0;
2451 static bool netevent_registered;
2452 static struct notifier_block cxgb4_netevent_nb = {
2453 .notifier_call = netevent_cb
2456 static void uld_attach(struct adapter *adap, unsigned int uld)
2458 void *handle;
2459 struct cxgb4_lld_info lli;
2461 lli.pdev = adap->pdev;
2462 lli.l2t = adap->l2t;
2463 lli.tids = &adap->tids;
2464 lli.ports = adap->port;
2465 lli.vr = &adap->vres;
2466 lli.mtus = adap->params.mtus;
2467 if (uld == CXGB4_ULD_RDMA) {
2468 lli.rxq_ids = adap->sge.rdma_rxq;
2469 lli.nrxq = adap->sge.rdmaqs;
2470 } else if (uld == CXGB4_ULD_ISCSI) {
2471 lli.rxq_ids = adap->sge.ofld_rxq;
2472 lli.nrxq = adap->sge.ofldqsets;
2474 lli.ntxq = adap->sge.ofldqsets;
2475 lli.nchan = adap->params.nports;
2476 lli.nports = adap->params.nports;
2477 lli.wr_cred = adap->params.ofldq_wr_cred;
2478 lli.adapter_type = adap->params.rev;
2479 lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
2480 lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
2481 t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
2482 (adap->fn * 4));
2483 lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
2484 t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
2485 (adap->fn * 4));
2486 lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
2487 lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
2488 lli.fw_vers = adap->params.fw_vers;
2490 handle = ulds[uld].add(&lli);
2491 if (IS_ERR(handle)) {
2492 dev_warn(adap->pdev_dev,
2493 "could not attach to the %s driver, error %ld\n",
2494 uld_str[uld], PTR_ERR(handle));
2495 return;
2498 adap->uld_handle[uld] = handle;
2500 if (!netevent_registered) {
2501 register_netevent_notifier(&cxgb4_netevent_nb);
2502 netevent_registered = true;
2505 if (adap->flags & FULL_INIT_DONE)
2506 ulds[uld].state_change(handle, CXGB4_STATE_UP);
2509 static void attach_ulds(struct adapter *adap)
2511 unsigned int i;
2513 mutex_lock(&uld_mutex);
2514 list_add_tail(&adap->list_node, &adapter_list);
2515 for (i = 0; i < CXGB4_ULD_MAX; i++)
2516 if (ulds[i].add)
2517 uld_attach(adap, i);
2518 mutex_unlock(&uld_mutex);
2521 static void detach_ulds(struct adapter *adap)
2523 unsigned int i;
2525 mutex_lock(&uld_mutex);
2526 list_del(&adap->list_node);
2527 for (i = 0; i < CXGB4_ULD_MAX; i++)
2528 if (adap->uld_handle[i]) {
2529 ulds[i].state_change(adap->uld_handle[i],
2530 CXGB4_STATE_DETACH);
2531 adap->uld_handle[i] = NULL;
2533 if (netevent_registered && list_empty(&adapter_list)) {
2534 unregister_netevent_notifier(&cxgb4_netevent_nb);
2535 netevent_registered = false;
2537 mutex_unlock(&uld_mutex);
2540 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2542 unsigned int i;
2544 mutex_lock(&uld_mutex);
2545 for (i = 0; i < CXGB4_ULD_MAX; i++)
2546 if (adap->uld_handle[i])
2547 ulds[i].state_change(adap->uld_handle[i], new_state);
2548 mutex_unlock(&uld_mutex);
2552 * cxgb4_register_uld - register an upper-layer driver
2553 * @type: the ULD type
2554 * @p: the ULD methods
2556 * Registers an upper-layer driver with this driver and notifies the ULD
2557 * about any presently available devices that support its type. Returns
2558 * %-EBUSY if a ULD of the same type is already registered.
2560 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
2562 int ret = 0;
2563 struct adapter *adap;
2565 if (type >= CXGB4_ULD_MAX)
2566 return -EINVAL;
2567 mutex_lock(&uld_mutex);
2568 if (ulds[type].add) {
2569 ret = -EBUSY;
2570 goto out;
2572 ulds[type] = *p;
2573 list_for_each_entry(adap, &adapter_list, list_node)
2574 uld_attach(adap, type);
2575 out: mutex_unlock(&uld_mutex);
2576 return ret;
2578 EXPORT_SYMBOL(cxgb4_register_uld);
2581 * cxgb4_unregister_uld - unregister an upper-layer driver
2582 * @type: the ULD type
2584 * Unregisters an existing upper-layer driver.
2586 int cxgb4_unregister_uld(enum cxgb4_uld type)
2588 struct adapter *adap;
2590 if (type >= CXGB4_ULD_MAX)
2591 return -EINVAL;
2592 mutex_lock(&uld_mutex);
2593 list_for_each_entry(adap, &adapter_list, list_node)
2594 adap->uld_handle[type] = NULL;
2595 ulds[type].add = NULL;
2596 mutex_unlock(&uld_mutex);
2597 return 0;
2599 EXPORT_SYMBOL(cxgb4_unregister_uld);
2602 * cxgb_up - enable the adapter
2603 * @adap: adapter being enabled
2605 * Called when the first port is enabled, this function performs the
2606 * actions necessary to make an adapter operational, such as completing
2607 * the initialization of HW modules, and enabling interrupts.
2609 * Must be called with the rtnl lock held.
2611 static int cxgb_up(struct adapter *adap)
2613 int err;
2615 err = setup_sge_queues(adap);
2616 if (err)
2617 goto out;
2618 err = setup_rss(adap);
2619 if (err)
2620 goto freeq;
2622 if (adap->flags & USING_MSIX) {
2623 name_msix_vecs(adap);
2624 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2625 adap->msix_info[0].desc, adap);
2626 if (err)
2627 goto irq_err;
2629 err = request_msix_queue_irqs(adap);
2630 if (err) {
2631 free_irq(adap->msix_info[0].vec, adap);
2632 goto irq_err;
2634 } else {
2635 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2636 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2637 adap->port[0]->name, adap);
2638 if (err)
2639 goto irq_err;
2641 enable_rx(adap);
2642 t4_sge_start(adap);
2643 t4_intr_enable(adap);
2644 adap->flags |= FULL_INIT_DONE;
2645 notify_ulds(adap, CXGB4_STATE_UP);
2646 out:
2647 return err;
2648 irq_err:
2649 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2650 freeq:
2651 t4_free_sge_resources(adap);
2652 goto out;
2655 static void cxgb_down(struct adapter *adapter)
2657 t4_intr_disable(adapter);
2658 cancel_work_sync(&adapter->tid_release_task);
2659 adapter->tid_release_task_busy = false;
2660 adapter->tid_release_head = NULL;
2662 if (adapter->flags & USING_MSIX) {
2663 free_msix_queue_irqs(adapter);
2664 free_irq(adapter->msix_info[0].vec, adapter);
2665 } else
2666 free_irq(adapter->pdev->irq, adapter);
2667 quiesce_rx(adapter);
2668 t4_sge_stop(adapter);
2669 t4_free_sge_resources(adapter);
2670 adapter->flags &= ~FULL_INIT_DONE;
2674 * net_device operations
2676 static int cxgb_open(struct net_device *dev)
2678 int err;
2679 struct port_info *pi = netdev_priv(dev);
2680 struct adapter *adapter = pi->adapter;
2682 netif_carrier_off(dev);
2684 if (!(adapter->flags & FULL_INIT_DONE)) {
2685 err = cxgb_up(adapter);
2686 if (err < 0)
2687 return err;
2690 err = link_start(dev);
2691 if (!err)
2692 netif_tx_start_all_queues(dev);
2693 return err;
2696 static int cxgb_close(struct net_device *dev)
2698 struct port_info *pi = netdev_priv(dev);
2699 struct adapter *adapter = pi->adapter;
2701 netif_tx_stop_all_queues(dev);
2702 netif_carrier_off(dev);
2703 return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
2706 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
2707 struct rtnl_link_stats64 *ns)
2709 struct port_stats stats;
2710 struct port_info *p = netdev_priv(dev);
2711 struct adapter *adapter = p->adapter;
2713 spin_lock(&adapter->stats_lock);
2714 t4_get_port_stats(adapter, p->tx_chan, &stats);
2715 spin_unlock(&adapter->stats_lock);
2717 ns->tx_bytes = stats.tx_octets;
2718 ns->tx_packets = stats.tx_frames;
2719 ns->rx_bytes = stats.rx_octets;
2720 ns->rx_packets = stats.rx_frames;
2721 ns->multicast = stats.rx_mcast_frames;
2723 /* detailed rx_errors */
2724 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2725 stats.rx_runt;
2726 ns->rx_over_errors = 0;
2727 ns->rx_crc_errors = stats.rx_fcs_err;
2728 ns->rx_frame_errors = stats.rx_symbol_err;
2729 ns->rx_fifo_errors = stats.rx_ovflow0 + stats.rx_ovflow1 +
2730 stats.rx_ovflow2 + stats.rx_ovflow3 +
2731 stats.rx_trunc0 + stats.rx_trunc1 +
2732 stats.rx_trunc2 + stats.rx_trunc3;
2733 ns->rx_missed_errors = 0;
2735 /* detailed tx_errors */
2736 ns->tx_aborted_errors = 0;
2737 ns->tx_carrier_errors = 0;
2738 ns->tx_fifo_errors = 0;
2739 ns->tx_heartbeat_errors = 0;
2740 ns->tx_window_errors = 0;
2742 ns->tx_errors = stats.tx_error_frames;
2743 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2744 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2745 return ns;
2748 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2750 unsigned int mbox;
2751 int ret = 0, prtad, devad;
2752 struct port_info *pi = netdev_priv(dev);
2753 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2755 switch (cmd) {
2756 case SIOCGMIIPHY:
2757 if (pi->mdio_addr < 0)
2758 return -EOPNOTSUPP;
2759 data->phy_id = pi->mdio_addr;
2760 break;
2761 case SIOCGMIIREG:
2762 case SIOCSMIIREG:
2763 if (mdio_phy_id_is_c45(data->phy_id)) {
2764 prtad = mdio_phy_id_prtad(data->phy_id);
2765 devad = mdio_phy_id_devad(data->phy_id);
2766 } else if (data->phy_id < 32) {
2767 prtad = data->phy_id;
2768 devad = 0;
2769 data->reg_num &= 0x1f;
2770 } else
2771 return -EINVAL;
2773 mbox = pi->adapter->fn;
2774 if (cmd == SIOCGMIIREG)
2775 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2776 data->reg_num, &data->val_out);
2777 else
2778 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2779 data->reg_num, data->val_in);
2780 break;
2781 default:
2782 return -EOPNOTSUPP;
2784 return ret;
2787 static void cxgb_set_rxmode(struct net_device *dev)
2789 /* unfortunately we can't return errors to the stack */
2790 set_rxmode(dev, -1, false);
2793 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2795 int ret;
2796 struct port_info *pi = netdev_priv(dev);
2798 if (new_mtu < 81 || new_mtu > MAX_MTU) /* accommodate SACK */
2799 return -EINVAL;
2800 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
2801 -1, -1, -1, true);
2802 if (!ret)
2803 dev->mtu = new_mtu;
2804 return ret;
2807 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2809 int ret;
2810 struct sockaddr *addr = p;
2811 struct port_info *pi = netdev_priv(dev);
2813 if (!is_valid_ether_addr(addr->sa_data))
2814 return -EINVAL;
2816 ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
2817 pi->xact_addr_filt, addr->sa_data, true, true);
2818 if (ret < 0)
2819 return ret;
2821 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2822 pi->xact_addr_filt = ret;
2823 return 0;
2826 #ifdef CONFIG_NET_POLL_CONTROLLER
2827 static void cxgb_netpoll(struct net_device *dev)
2829 struct port_info *pi = netdev_priv(dev);
2830 struct adapter *adap = pi->adapter;
2832 if (adap->flags & USING_MSIX) {
2833 int i;
2834 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2836 for (i = pi->nqsets; i; i--, rx++)
2837 t4_sge_intr_msix(0, &rx->rspq);
2838 } else
2839 t4_intr_handler(adap)(0, adap);
2841 #endif
2843 static const struct net_device_ops cxgb4_netdev_ops = {
2844 .ndo_open = cxgb_open,
2845 .ndo_stop = cxgb_close,
2846 .ndo_start_xmit = t4_eth_xmit,
2847 .ndo_get_stats64 = cxgb_get_stats,
2848 .ndo_set_rx_mode = cxgb_set_rxmode,
2849 .ndo_set_mac_address = cxgb_set_mac_addr,
2850 .ndo_set_features = cxgb_set_features,
2851 .ndo_validate_addr = eth_validate_addr,
2852 .ndo_do_ioctl = cxgb_ioctl,
2853 .ndo_change_mtu = cxgb_change_mtu,
2854 #ifdef CONFIG_NET_POLL_CONTROLLER
2855 .ndo_poll_controller = cxgb_netpoll,
2856 #endif
2859 void t4_fatal_err(struct adapter *adap)
2861 t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
2862 t4_intr_disable(adap);
2863 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
2866 static void setup_memwin(struct adapter *adap)
2868 u32 bar0;
2870 bar0 = pci_resource_start(adap->pdev, 0); /* truncation intentional */
2871 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
2872 (bar0 + MEMWIN0_BASE) | BIR(0) |
2873 WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
2874 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
2875 (bar0 + MEMWIN1_BASE) | BIR(0) |
2876 WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
2877 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
2878 (bar0 + MEMWIN2_BASE) | BIR(0) |
2879 WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
2880 if (adap->vres.ocq.size) {
2881 unsigned int start, sz_kb;
2883 start = pci_resource_start(adap->pdev, 2) +
2884 OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
2885 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
2886 t4_write_reg(adap,
2887 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
2888 start | BIR(1) | WINDOW(ilog2(sz_kb)));
2889 t4_write_reg(adap,
2890 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
2891 adap->vres.ocq.start);
2892 t4_read_reg(adap,
2893 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
2897 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
2899 u32 v;
2900 int ret;
2902 /* get device capabilities */
2903 memset(c, 0, sizeof(*c));
2904 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2905 FW_CMD_REQUEST | FW_CMD_READ);
2906 c->retval_len16 = htonl(FW_LEN16(*c));
2907 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
2908 if (ret < 0)
2909 return ret;
2911 /* select capabilities we'll be using */
2912 if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
2913 if (!vf_acls)
2914 c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
2915 else
2916 c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
2917 } else if (vf_acls) {
2918 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
2919 return ret;
2921 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2922 FW_CMD_REQUEST | FW_CMD_WRITE);
2923 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
2924 if (ret < 0)
2925 return ret;
2927 ret = t4_config_glbl_rss(adap, adap->fn,
2928 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
2929 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
2930 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
2931 if (ret < 0)
2932 return ret;
2934 ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
2935 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
2936 if (ret < 0)
2937 return ret;
2939 t4_sge_init(adap);
2941 /* tweak some settings */
2942 t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
2943 t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
2944 t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
2945 v = t4_read_reg(adap, TP_PIO_DATA);
2946 t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
2948 /* get basic stuff going */
2949 return t4_early_init(adap, adap->fn);
2953 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
2955 #define MAX_ATIDS 8192U
2958 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
2960 static int adap_init0(struct adapter *adap)
2962 int ret;
2963 u32 v, port_vec;
2964 enum dev_state state;
2965 u32 params[7], val[7];
2966 struct fw_caps_config_cmd c;
2968 ret = t4_check_fw_version(adap);
2969 if (ret == -EINVAL || ret > 0) {
2970 if (upgrade_fw(adap) >= 0) /* recache FW version */
2971 ret = t4_check_fw_version(adap);
2973 if (ret < 0)
2974 return ret;
2976 /* contact FW, request master */
2977 ret = t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, &state);
2978 if (ret < 0) {
2979 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
2980 ret);
2981 return ret;
2984 /* reset device */
2985 ret = t4_fw_reset(adap, adap->fn, PIORSTMODE | PIORST);
2986 if (ret < 0)
2987 goto bye;
2989 for (v = 0; v < SGE_NTIMERS - 1; v++)
2990 adap->sge.timer_val[v] = min(intr_holdoff[v], MAX_SGE_TIMERVAL);
2991 adap->sge.timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
2992 adap->sge.counter_val[0] = 1;
2993 for (v = 1; v < SGE_NCOUNTERS; v++)
2994 adap->sge.counter_val[v] = min(intr_cnt[v - 1],
2995 THRESHOLD_3_MASK);
2996 #define FW_PARAM_DEV(param) \
2997 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
2998 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3000 params[0] = FW_PARAM_DEV(CCLK);
3001 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 1, params, val);
3002 if (ret < 0)
3003 goto bye;
3004 adap->params.vpd.cclk = val[0];
3006 ret = adap_init1(adap, &c);
3007 if (ret < 0)
3008 goto bye;
3010 #define FW_PARAM_PFVF(param) \
3011 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3012 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param) | \
3013 FW_PARAMS_PARAM_Y(adap->fn))
3015 params[0] = FW_PARAM_DEV(PORTVEC);
3016 params[1] = FW_PARAM_PFVF(L2T_START);
3017 params[2] = FW_PARAM_PFVF(L2T_END);
3018 params[3] = FW_PARAM_PFVF(FILTER_START);
3019 params[4] = FW_PARAM_PFVF(FILTER_END);
3020 params[5] = FW_PARAM_PFVF(IQFLINT_START);
3021 params[6] = FW_PARAM_PFVF(EQ_START);
3022 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 7, params, val);
3023 if (ret < 0)
3024 goto bye;
3025 port_vec = val[0];
3026 adap->tids.ftid_base = val[3];
3027 adap->tids.nftids = val[4] - val[3] + 1;
3028 adap->sge.ingr_start = val[5];
3029 adap->sge.egr_start = val[6];
3031 if (c.ofldcaps) {
3032 /* query offload-related parameters */
3033 params[0] = FW_PARAM_DEV(NTID);
3034 params[1] = FW_PARAM_PFVF(SERVER_START);
3035 params[2] = FW_PARAM_PFVF(SERVER_END);
3036 params[3] = FW_PARAM_PFVF(TDDP_START);
3037 params[4] = FW_PARAM_PFVF(TDDP_END);
3038 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3039 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3040 val);
3041 if (ret < 0)
3042 goto bye;
3043 adap->tids.ntids = val[0];
3044 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
3045 adap->tids.stid_base = val[1];
3046 adap->tids.nstids = val[2] - val[1] + 1;
3047 adap->vres.ddp.start = val[3];
3048 adap->vres.ddp.size = val[4] - val[3] + 1;
3049 adap->params.ofldq_wr_cred = val[5];
3050 adap->params.offload = 1;
3052 if (c.rdmacaps) {
3053 params[0] = FW_PARAM_PFVF(STAG_START);
3054 params[1] = FW_PARAM_PFVF(STAG_END);
3055 params[2] = FW_PARAM_PFVF(RQ_START);
3056 params[3] = FW_PARAM_PFVF(RQ_END);
3057 params[4] = FW_PARAM_PFVF(PBL_START);
3058 params[5] = FW_PARAM_PFVF(PBL_END);
3059 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3060 val);
3061 if (ret < 0)
3062 goto bye;
3063 adap->vres.stag.start = val[0];
3064 adap->vres.stag.size = val[1] - val[0] + 1;
3065 adap->vres.rq.start = val[2];
3066 adap->vres.rq.size = val[3] - val[2] + 1;
3067 adap->vres.pbl.start = val[4];
3068 adap->vres.pbl.size = val[5] - val[4] + 1;
3070 params[0] = FW_PARAM_PFVF(SQRQ_START);
3071 params[1] = FW_PARAM_PFVF(SQRQ_END);
3072 params[2] = FW_PARAM_PFVF(CQ_START);
3073 params[3] = FW_PARAM_PFVF(CQ_END);
3074 params[4] = FW_PARAM_PFVF(OCQ_START);
3075 params[5] = FW_PARAM_PFVF(OCQ_END);
3076 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 6, params,
3077 val);
3078 if (ret < 0)
3079 goto bye;
3080 adap->vres.qp.start = val[0];
3081 adap->vres.qp.size = val[1] - val[0] + 1;
3082 adap->vres.cq.start = val[2];
3083 adap->vres.cq.size = val[3] - val[2] + 1;
3084 adap->vres.ocq.start = val[4];
3085 adap->vres.ocq.size = val[5] - val[4] + 1;
3087 if (c.iscsicaps) {
3088 params[0] = FW_PARAM_PFVF(ISCSI_START);
3089 params[1] = FW_PARAM_PFVF(ISCSI_END);
3090 ret = t4_query_params(adap, adap->fn, adap->fn, 0, 2, params,
3091 val);
3092 if (ret < 0)
3093 goto bye;
3094 adap->vres.iscsi.start = val[0];
3095 adap->vres.iscsi.size = val[1] - val[0] + 1;
3097 #undef FW_PARAM_PFVF
3098 #undef FW_PARAM_DEV
3100 adap->params.nports = hweight32(port_vec);
3101 adap->params.portvec = port_vec;
3102 adap->flags |= FW_OK;
3104 /* These are finalized by FW initialization, load their values now */
3105 v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
3106 adap->params.tp.tre = TIMERRESOLUTION_GET(v);
3107 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
3108 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3109 adap->params.b_wnd);
3111 #ifdef CONFIG_PCI_IOV
3113 * Provision resource limits for Virtual Functions. We currently
3114 * grant them all the same static resource limits except for the Port
3115 * Access Rights Mask which we're assigning based on the PF. All of
3116 * the static provisioning stuff for both the PF and VF really needs
3117 * to be managed in a persistent manner for each device which the
3118 * firmware controls.
3121 int pf, vf;
3123 for (pf = 0; pf < ARRAY_SIZE(num_vf); pf++) {
3124 if (num_vf[pf] <= 0)
3125 continue;
3127 /* VF numbering starts at 1! */
3128 for (vf = 1; vf <= num_vf[pf]; vf++) {
3129 ret = t4_cfg_pfvf(adap, adap->fn, pf, vf,
3130 VFRES_NEQ, VFRES_NETHCTRL,
3131 VFRES_NIQFLINT, VFRES_NIQ,
3132 VFRES_TC, VFRES_NVI,
3133 FW_PFVF_CMD_CMASK_MASK,
3134 pfvfres_pmask(adap, pf, vf),
3135 VFRES_NEXACTF,
3136 VFRES_R_CAPS, VFRES_WX_CAPS);
3137 if (ret < 0)
3138 dev_warn(adap->pdev_dev, "failed to "
3139 "provision pf/vf=%d/%d; "
3140 "err=%d\n", pf, vf, ret);
3144 #endif
3146 setup_memwin(adap);
3147 return 0;
3150 * If a command timed out or failed with EIO FW does not operate within
3151 * its spec or something catastrophic happened to HW/FW, stop issuing
3152 * commands.
3154 bye: if (ret != -ETIMEDOUT && ret != -EIO)
3155 t4_fw_bye(adap, adap->fn);
3156 return ret;
3159 /* EEH callbacks */
3161 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
3162 pci_channel_state_t state)
3164 int i;
3165 struct adapter *adap = pci_get_drvdata(pdev);
3167 if (!adap)
3168 goto out;
3170 rtnl_lock();
3171 adap->flags &= ~FW_OK;
3172 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
3173 for_each_port(adap, i) {
3174 struct net_device *dev = adap->port[i];
3176 netif_device_detach(dev);
3177 netif_carrier_off(dev);
3179 if (adap->flags & FULL_INIT_DONE)
3180 cxgb_down(adap);
3181 rtnl_unlock();
3182 pci_disable_device(pdev);
3183 out: return state == pci_channel_io_perm_failure ?
3184 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
3187 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
3189 int i, ret;
3190 struct fw_caps_config_cmd c;
3191 struct adapter *adap = pci_get_drvdata(pdev);
3193 if (!adap) {
3194 pci_restore_state(pdev);
3195 pci_save_state(pdev);
3196 return PCI_ERS_RESULT_RECOVERED;
3199 if (pci_enable_device(pdev)) {
3200 dev_err(&pdev->dev, "cannot reenable PCI device after reset\n");
3201 return PCI_ERS_RESULT_DISCONNECT;
3204 pci_set_master(pdev);
3205 pci_restore_state(pdev);
3206 pci_save_state(pdev);
3207 pci_cleanup_aer_uncorrect_error_status(pdev);
3209 if (t4_wait_dev_ready(adap) < 0)
3210 return PCI_ERS_RESULT_DISCONNECT;
3211 if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL))
3212 return PCI_ERS_RESULT_DISCONNECT;
3213 adap->flags |= FW_OK;
3214 if (adap_init1(adap, &c))
3215 return PCI_ERS_RESULT_DISCONNECT;
3217 for_each_port(adap, i) {
3218 struct port_info *p = adap2pinfo(adap, i);
3220 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
3221 NULL, NULL);
3222 if (ret < 0)
3223 return PCI_ERS_RESULT_DISCONNECT;
3224 p->viid = ret;
3225 p->xact_addr_filt = -1;
3228 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3229 adap->params.b_wnd);
3230 setup_memwin(adap);
3231 if (cxgb_up(adap))
3232 return PCI_ERS_RESULT_DISCONNECT;
3233 return PCI_ERS_RESULT_RECOVERED;
3236 static void eeh_resume(struct pci_dev *pdev)
3238 int i;
3239 struct adapter *adap = pci_get_drvdata(pdev);
3241 if (!adap)
3242 return;
3244 rtnl_lock();
3245 for_each_port(adap, i) {
3246 struct net_device *dev = adap->port[i];
3248 if (netif_running(dev)) {
3249 link_start(dev);
3250 cxgb_set_rxmode(dev);
3252 netif_device_attach(dev);
3254 rtnl_unlock();
3257 static struct pci_error_handlers cxgb4_eeh = {
3258 .error_detected = eeh_err_detected,
3259 .slot_reset = eeh_slot_reset,
3260 .resume = eeh_resume,
3263 static inline bool is_10g_port(const struct link_config *lc)
3265 return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
3268 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
3269 unsigned int size, unsigned int iqe_size)
3271 q->intr_params = QINTR_TIMER_IDX(timer_idx) |
3272 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
3273 q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
3274 q->iqe_len = iqe_size;
3275 q->size = size;
3279 * Perform default configuration of DMA queues depending on the number and type
3280 * of ports we found and the number of available CPUs. Most settings can be
3281 * modified by the admin prior to actual use.
3283 static void __devinit cfg_queues(struct adapter *adap)
3285 struct sge *s = &adap->sge;
3286 int i, q10g = 0, n10g = 0, qidx = 0;
3288 for_each_port(adap, i)
3289 n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
3292 * We default to 1 queue per non-10G port and up to # of cores queues
3293 * per 10G port.
3295 if (n10g)
3296 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
3297 if (q10g > num_online_cpus())
3298 q10g = num_online_cpus();
3300 for_each_port(adap, i) {
3301 struct port_info *pi = adap2pinfo(adap, i);
3303 pi->first_qset = qidx;
3304 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
3305 qidx += pi->nqsets;
3308 s->ethqsets = qidx;
3309 s->max_ethqsets = qidx; /* MSI-X may lower it later */
3311 if (is_offload(adap)) {
3313 * For offload we use 1 queue/channel if all ports are up to 1G,
3314 * otherwise we divide all available queues amongst the channels
3315 * capped by the number of available cores.
3317 if (n10g) {
3318 i = min_t(int, ARRAY_SIZE(s->ofldrxq),
3319 num_online_cpus());
3320 s->ofldqsets = roundup(i, adap->params.nports);
3321 } else
3322 s->ofldqsets = adap->params.nports;
3323 /* For RDMA one Rx queue per channel suffices */
3324 s->rdmaqs = adap->params.nports;
3327 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
3328 struct sge_eth_rxq *r = &s->ethrxq[i];
3330 init_rspq(&r->rspq, 0, 0, 1024, 64);
3331 r->fl.size = 72;
3334 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
3335 s->ethtxq[i].q.size = 1024;
3337 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
3338 s->ctrlq[i].q.size = 512;
3340 for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
3341 s->ofldtxq[i].q.size = 1024;
3343 for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
3344 struct sge_ofld_rxq *r = &s->ofldrxq[i];
3346 init_rspq(&r->rspq, 0, 0, 1024, 64);
3347 r->rspq.uld = CXGB4_ULD_ISCSI;
3348 r->fl.size = 72;
3351 for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
3352 struct sge_ofld_rxq *r = &s->rdmarxq[i];
3354 init_rspq(&r->rspq, 0, 0, 511, 64);
3355 r->rspq.uld = CXGB4_ULD_RDMA;
3356 r->fl.size = 72;
3359 init_rspq(&s->fw_evtq, 6, 0, 512, 64);
3360 init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
3364 * Reduce the number of Ethernet queues across all ports to at most n.
3365 * n provides at least one queue per port.
3367 static void __devinit reduce_ethqs(struct adapter *adap, int n)
3369 int i;
3370 struct port_info *pi;
3372 while (n < adap->sge.ethqsets)
3373 for_each_port(adap, i) {
3374 pi = adap2pinfo(adap, i);
3375 if (pi->nqsets > 1) {
3376 pi->nqsets--;
3377 adap->sge.ethqsets--;
3378 if (adap->sge.ethqsets <= n)
3379 break;
3383 n = 0;
3384 for_each_port(adap, i) {
3385 pi = adap2pinfo(adap, i);
3386 pi->first_qset = n;
3387 n += pi->nqsets;
3391 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
3392 #define EXTRA_VECS 2
3394 static int __devinit enable_msix(struct adapter *adap)
3396 int ofld_need = 0;
3397 int i, err, want, need;
3398 struct sge *s = &adap->sge;
3399 unsigned int nchan = adap->params.nports;
3400 struct msix_entry entries[MAX_INGQ + 1];
3402 for (i = 0; i < ARRAY_SIZE(entries); ++i)
3403 entries[i].entry = i;
3405 want = s->max_ethqsets + EXTRA_VECS;
3406 if (is_offload(adap)) {
3407 want += s->rdmaqs + s->ofldqsets;
3408 /* need nchan for each possible ULD */
3409 ofld_need = 2 * nchan;
3411 need = adap->params.nports + EXTRA_VECS + ofld_need;
3413 while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
3414 want = err;
3416 if (!err) {
3418 * Distribute available vectors to the various queue groups.
3419 * Every group gets its minimum requirement and NIC gets top
3420 * priority for leftovers.
3422 i = want - EXTRA_VECS - ofld_need;
3423 if (i < s->max_ethqsets) {
3424 s->max_ethqsets = i;
3425 if (i < s->ethqsets)
3426 reduce_ethqs(adap, i);
3428 if (is_offload(adap)) {
3429 i = want - EXTRA_VECS - s->max_ethqsets;
3430 i -= ofld_need - nchan;
3431 s->ofldqsets = (i / nchan) * nchan; /* round down */
3433 for (i = 0; i < want; ++i)
3434 adap->msix_info[i].vec = entries[i].vector;
3435 } else if (err > 0)
3436 dev_info(adap->pdev_dev,
3437 "only %d MSI-X vectors left, not using MSI-X\n", err);
3438 return err;
3441 #undef EXTRA_VECS
3443 static int __devinit init_rss(struct adapter *adap)
3445 unsigned int i, j;
3447 for_each_port(adap, i) {
3448 struct port_info *pi = adap2pinfo(adap, i);
3450 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
3451 if (!pi->rss)
3452 return -ENOMEM;
3453 for (j = 0; j < pi->rss_size; j++)
3454 pi->rss[j] = ethtool_rxfh_indir_default(j, pi->nqsets);
3456 return 0;
3459 static void __devinit print_port_info(const struct net_device *dev)
3461 static const char *base[] = {
3462 "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
3463 "KX", "KR", "R SFP+", "KR/KX", "KR/KX/KX4"
3466 char buf[80];
3467 char *bufp = buf;
3468 const char *spd = "";
3469 const struct port_info *pi = netdev_priv(dev);
3470 const struct adapter *adap = pi->adapter;
3472 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
3473 spd = " 2.5 GT/s";
3474 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
3475 spd = " 5 GT/s";
3477 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
3478 bufp += sprintf(bufp, "100/");
3479 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
3480 bufp += sprintf(bufp, "1000/");
3481 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
3482 bufp += sprintf(bufp, "10G/");
3483 if (bufp != buf)
3484 --bufp;
3485 sprintf(bufp, "BASE-%s", base[pi->port_type]);
3487 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
3488 adap->params.vpd.id, adap->params.rev, buf,
3489 is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
3490 (adap->flags & USING_MSIX) ? " MSI-X" :
3491 (adap->flags & USING_MSI) ? " MSI" : "");
3492 netdev_info(dev, "S/N: %s, E/C: %s\n",
3493 adap->params.vpd.sn, adap->params.vpd.ec);
3496 static void __devinit enable_pcie_relaxed_ordering(struct pci_dev *dev)
3498 u16 v;
3499 int pos;
3501 pos = pci_pcie_cap(dev);
3502 if (pos > 0) {
3503 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL, &v);
3504 v |= PCI_EXP_DEVCTL_RELAX_EN;
3505 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL, v);
3510 * Free the following resources:
3511 * - memory used for tables
3512 * - MSI/MSI-X
3513 * - net devices
3514 * - resources FW is holding for us
3516 static void free_some_resources(struct adapter *adapter)
3518 unsigned int i;
3520 t4_free_mem(adapter->l2t);
3521 t4_free_mem(adapter->tids.tid_tab);
3522 disable_msi(adapter);
3524 for_each_port(adapter, i)
3525 if (adapter->port[i]) {
3526 kfree(adap2pinfo(adapter, i)->rss);
3527 free_netdev(adapter->port[i]);
3529 if (adapter->flags & FW_OK)
3530 t4_fw_bye(adapter, adapter->fn);
3533 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
3534 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
3535 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
3537 static int __devinit init_one(struct pci_dev *pdev,
3538 const struct pci_device_id *ent)
3540 int func, i, err;
3541 struct port_info *pi;
3542 bool highdma = false;
3543 struct adapter *adapter = NULL;
3545 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
3547 err = pci_request_regions(pdev, KBUILD_MODNAME);
3548 if (err) {
3549 /* Just info, some other driver may have claimed the device. */
3550 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
3551 return err;
3554 /* We control everything through one PF */
3555 func = PCI_FUNC(pdev->devfn);
3556 if (func != ent->driver_data) {
3557 pci_save_state(pdev); /* to restore SR-IOV later */
3558 goto sriov;
3561 err = pci_enable_device(pdev);
3562 if (err) {
3563 dev_err(&pdev->dev, "cannot enable PCI device\n");
3564 goto out_release_regions;
3567 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3568 highdma = true;
3569 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3570 if (err) {
3571 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
3572 "coherent allocations\n");
3573 goto out_disable_device;
3575 } else {
3576 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3577 if (err) {
3578 dev_err(&pdev->dev, "no usable DMA configuration\n");
3579 goto out_disable_device;
3583 pci_enable_pcie_error_reporting(pdev);
3584 enable_pcie_relaxed_ordering(pdev);
3585 pci_set_master(pdev);
3586 pci_save_state(pdev);
3588 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3589 if (!adapter) {
3590 err = -ENOMEM;
3591 goto out_disable_device;
3594 adapter->regs = pci_ioremap_bar(pdev, 0);
3595 if (!adapter->regs) {
3596 dev_err(&pdev->dev, "cannot map device registers\n");
3597 err = -ENOMEM;
3598 goto out_free_adapter;
3601 adapter->pdev = pdev;
3602 adapter->pdev_dev = &pdev->dev;
3603 adapter->fn = func;
3604 adapter->msg_enable = dflt_msg_enable;
3605 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
3607 spin_lock_init(&adapter->stats_lock);
3608 spin_lock_init(&adapter->tid_release_lock);
3610 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
3612 err = t4_prep_adapter(adapter);
3613 if (err)
3614 goto out_unmap_bar;
3615 err = adap_init0(adapter);
3616 if (err)
3617 goto out_unmap_bar;
3619 for_each_port(adapter, i) {
3620 struct net_device *netdev;
3622 netdev = alloc_etherdev_mq(sizeof(struct port_info),
3623 MAX_ETH_QSETS);
3624 if (!netdev) {
3625 err = -ENOMEM;
3626 goto out_free_dev;
3629 SET_NETDEV_DEV(netdev, &pdev->dev);
3631 adapter->port[i] = netdev;
3632 pi = netdev_priv(netdev);
3633 pi->adapter = adapter;
3634 pi->xact_addr_filt = -1;
3635 pi->port_id = i;
3636 netdev->irq = pdev->irq;
3638 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
3639 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3640 NETIF_F_RXCSUM | NETIF_F_RXHASH |
3641 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
3642 if (highdma)
3643 netdev->hw_features |= NETIF_F_HIGHDMA;
3644 netdev->features |= netdev->hw_features;
3645 netdev->vlan_features = netdev->features & VLAN_FEAT;
3647 netdev->priv_flags |= IFF_UNICAST_FLT;
3649 netdev->netdev_ops = &cxgb4_netdev_ops;
3650 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
3653 pci_set_drvdata(pdev, adapter);
3655 if (adapter->flags & FW_OK) {
3656 err = t4_port_init(adapter, func, func, 0);
3657 if (err)
3658 goto out_free_dev;
3662 * Configure queues and allocate tables now, they can be needed as
3663 * soon as the first register_netdev completes.
3665 cfg_queues(adapter);
3667 adapter->l2t = t4_init_l2t();
3668 if (!adapter->l2t) {
3669 /* We tolerate a lack of L2T, giving up some functionality */
3670 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
3671 adapter->params.offload = 0;
3674 if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
3675 dev_warn(&pdev->dev, "could not allocate TID table, "
3676 "continuing\n");
3677 adapter->params.offload = 0;
3680 /* See what interrupts we'll be using */
3681 if (msi > 1 && enable_msix(adapter) == 0)
3682 adapter->flags |= USING_MSIX;
3683 else if (msi > 0 && pci_enable_msi(pdev) == 0)
3684 adapter->flags |= USING_MSI;
3686 err = init_rss(adapter);
3687 if (err)
3688 goto out_free_dev;
3691 * The card is now ready to go. If any errors occur during device
3692 * registration we do not fail the whole card but rather proceed only
3693 * with the ports we manage to register successfully. However we must
3694 * register at least one net device.
3696 for_each_port(adapter, i) {
3697 pi = adap2pinfo(adapter, i);
3698 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
3699 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
3701 err = register_netdev(adapter->port[i]);
3702 if (err)
3703 break;
3704 adapter->chan_map[pi->tx_chan] = i;
3705 print_port_info(adapter->port[i]);
3707 if (i == 0) {
3708 dev_err(&pdev->dev, "could not register any net devices\n");
3709 goto out_free_dev;
3711 if (err) {
3712 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
3713 err = 0;
3716 if (cxgb4_debugfs_root) {
3717 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
3718 cxgb4_debugfs_root);
3719 setup_debugfs(adapter);
3722 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
3723 pdev->needs_freset = 1;
3725 if (is_offload(adapter))
3726 attach_ulds(adapter);
3728 sriov:
3729 #ifdef CONFIG_PCI_IOV
3730 if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0)
3731 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
3732 dev_info(&pdev->dev,
3733 "instantiated %u virtual functions\n",
3734 num_vf[func]);
3735 #endif
3736 return 0;
3738 out_free_dev:
3739 free_some_resources(adapter);
3740 out_unmap_bar:
3741 iounmap(adapter->regs);
3742 out_free_adapter:
3743 kfree(adapter);
3744 out_disable_device:
3745 pci_disable_pcie_error_reporting(pdev);
3746 pci_disable_device(pdev);
3747 out_release_regions:
3748 pci_release_regions(pdev);
3749 pci_set_drvdata(pdev, NULL);
3750 return err;
3753 static void __devexit remove_one(struct pci_dev *pdev)
3755 struct adapter *adapter = pci_get_drvdata(pdev);
3757 pci_disable_sriov(pdev);
3759 if (adapter) {
3760 int i;
3762 if (is_offload(adapter))
3763 detach_ulds(adapter);
3765 for_each_port(adapter, i)
3766 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
3767 unregister_netdev(adapter->port[i]);
3769 if (adapter->debugfs_root)
3770 debugfs_remove_recursive(adapter->debugfs_root);
3772 if (adapter->flags & FULL_INIT_DONE)
3773 cxgb_down(adapter);
3775 free_some_resources(adapter);
3776 iounmap(adapter->regs);
3777 kfree(adapter);
3778 pci_disable_pcie_error_reporting(pdev);
3779 pci_disable_device(pdev);
3780 pci_release_regions(pdev);
3781 pci_set_drvdata(pdev, NULL);
3782 } else
3783 pci_release_regions(pdev);
3786 static struct pci_driver cxgb4_driver = {
3787 .name = KBUILD_MODNAME,
3788 .id_table = cxgb4_pci_tbl,
3789 .probe = init_one,
3790 .remove = __devexit_p(remove_one),
3791 .err_handler = &cxgb4_eeh,
3794 static int __init cxgb4_init_module(void)
3796 int ret;
3798 /* Debugfs support is optional, just warn if this fails */
3799 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3800 if (!cxgb4_debugfs_root)
3801 pr_warning("could not create debugfs entry, continuing\n");
3803 ret = pci_register_driver(&cxgb4_driver);
3804 if (ret < 0)
3805 debugfs_remove(cxgb4_debugfs_root);
3806 return ret;
3809 static void __exit cxgb4_cleanup_module(void)
3811 pci_unregister_driver(&cxgb4_driver);
3812 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
3815 module_init(cxgb4_init_module);
3816 module_exit(cxgb4_cleanup_module);