]>
Commit | Line | Data |
---|---|---|
be839e39 CL |
1 | /* |
2 | * This file is part of the Chelsio T4 PCI-E SR-IOV Virtual Function Ethernet | |
3 | * driver for Linux. | |
4 | * | |
5 | * Copyright (c) 2009-2010 Chelsio Communications, Inc. All rights reserved. | |
6 | * | |
7 | * This software is available to you under a choice of one of two | |
8 | * licenses. You may choose to be licensed under the terms of the GNU | |
9 | * General Public License (GPL) Version 2, available from the file | |
10 | * COPYING in the main directory of this source tree, or the | |
11 | * OpenIB.org BSD license below: | |
12 | * | |
13 | * Redistribution and use in source and binary forms, with or | |
14 | * without modification, are permitted provided that the following | |
15 | * conditions are met: | |
16 | * | |
17 | * - Redistributions of source code must retain the above | |
18 | * copyright notice, this list of conditions and the following | |
19 | * disclaimer. | |
20 | * | |
21 | * - Redistributions in binary form must reproduce the above | |
22 | * copyright notice, this list of conditions and the following | |
23 | * disclaimer in the documentation and/or other materials | |
24 | * provided with the distribution. | |
25 | * | |
26 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
27 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
28 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
29 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
30 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
31 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
32 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
33 | * SOFTWARE. | |
34 | */ | |
35 | ||
428ac43f JP |
36 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
37 | ||
be839e39 CL |
38 | #include <linux/module.h> |
39 | #include <linux/moduleparam.h> | |
40 | #include <linux/init.h> | |
41 | #include <linux/pci.h> | |
42 | #include <linux/dma-mapping.h> | |
43 | #include <linux/netdevice.h> | |
44 | #include <linux/etherdevice.h> | |
45 | #include <linux/debugfs.h> | |
46 | #include <linux/ethtool.h> | |
5ad24def | 47 | #include <linux/mdio.h> |
be839e39 CL |
48 | |
49 | #include "t4vf_common.h" | |
50 | #include "t4vf_defs.h" | |
51 | ||
52 | #include "../cxgb4/t4_regs.h" | |
53 | #include "../cxgb4/t4_msg.h" | |
54 | ||
55 | /* | |
56 | * Generic information about the driver. | |
57 | */ | |
622c62b5 | 58 | #define DRV_VERSION "2.0.0-ko" |
52a5f846 | 59 | #define DRV_DESC "Chelsio T4/T5/T6 Virtual Function (VF) Network Driver" |
be839e39 CL |
60 | |
61 | /* | |
62 | * Module Parameters. | |
63 | * ================== | |
64 | */ | |
65 | ||
66 | /* | |
67 | * Default ethtool "message level" for adapters. | |
68 | */ | |
69 | #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \ | |
70 | NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\ | |
71 | NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR) | |
72 | ||
be839e39 CL |
73 | /* |
74 | * The driver uses the best interrupt scheme available on a platform in the | |
75 | * order MSI-X then MSI. This parameter determines which of these schemes the | |
76 | * driver may consider as follows: | |
77 | * | |
78 | * msi = 2: choose from among MSI-X and MSI | |
79 | * msi = 1: only consider MSI interrupts | |
80 | * | |
81 | * Note that unlike the Physical Function driver, this Virtual Function driver | |
82 | * does _not_ support legacy INTx interrupts (this limitation is mandated by | |
83 | * the PCI-E SR-IOV standard). | |
84 | */ | |
85 | #define MSI_MSIX 2 | |
86 | #define MSI_MSI 1 | |
87 | #define MSI_DEFAULT MSI_MSIX | |
88 | ||
89 | static int msi = MSI_DEFAULT; | |
90 | ||
91 | module_param(msi, int, 0644); | |
92 | MODULE_PARM_DESC(msi, "whether to use MSI-X or MSI"); | |
93 | ||
94 | /* | |
95 | * Fundamental constants. | |
96 | * ====================== | |
97 | */ | |
98 | ||
99 | enum { | |
100 | MAX_TXQ_ENTRIES = 16384, | |
101 | MAX_RSPQ_ENTRIES = 16384, | |
102 | MAX_RX_BUFFERS = 16384, | |
103 | ||
104 | MIN_TXQ_ENTRIES = 32, | |
105 | MIN_RSPQ_ENTRIES = 128, | |
106 | MIN_FL_ENTRIES = 16, | |
107 | ||
108 | /* | |
109 | * For purposes of manipulating the Free List size we need to | |
110 | * recognize that Free Lists are actually Egress Queues (the host | |
111 | * produces free buffers which the hardware consumes), Egress Queues | |
112 | * indices are all in units of Egress Context Units bytes, and free | |
113 | * list entries are 64-bit PCI DMA addresses. And since the state of | |
114 | * the Producer Index == the Consumer Index implies an EMPTY list, we | |
115 | * always have at least one Egress Unit's worth of Free List entries | |
116 | * unused. See sge.c for more details ... | |
117 | */ | |
118 | EQ_UNIT = SGE_EQ_IDXSIZE, | |
119 | FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64), | |
120 | MIN_FL_RESID = FL_PER_EQ_UNIT, | |
121 | }; | |
122 | ||
123 | /* | |
124 | * Global driver state. | |
125 | * ==================== | |
126 | */ | |
127 | ||
128 | static struct dentry *cxgb4vf_debugfs_root; | |
129 | ||
130 | /* | |
131 | * OS "Callback" functions. | |
132 | * ======================== | |
133 | */ | |
134 | ||
135 | /* | |
136 | * The link status has changed on the indicated "port" (Virtual Interface). | |
137 | */ | |
138 | void t4vf_os_link_changed(struct adapter *adapter, int pidx, int link_ok) | |
139 | { | |
140 | struct net_device *dev = adapter->port[pidx]; | |
141 | ||
142 | /* | |
143 | * If the port is disabled or the current recorded "link up" | |
144 | * status matches the new status, just return. | |
145 | */ | |
146 | if (!netif_running(dev) || link_ok == netif_carrier_ok(dev)) | |
147 | return; | |
148 | ||
149 | /* | |
150 | * Tell the OS that the link status has changed and print a short | |
151 | * informative message on the console about the event. | |
152 | */ | |
153 | if (link_ok) { | |
154 | const char *s; | |
155 | const char *fc; | |
156 | const struct port_info *pi = netdev_priv(dev); | |
157 | ||
158 | netif_carrier_on(dev); | |
159 | ||
160 | switch (pi->link_cfg.speed) { | |
5e78f7fd GG |
161 | case 100: |
162 | s = "100Mbps"; | |
163 | break; | |
164 | case 1000: | |
165 | s = "1Gbps"; | |
897d55df | 166 | break; |
897d55df | 167 | case 10000: |
be839e39 CL |
168 | s = "10Gbps"; |
169 | break; | |
5e78f7fd GG |
170 | case 25000: |
171 | s = "25Gbps"; | |
be839e39 | 172 | break; |
5e78f7fd GG |
173 | case 40000: |
174 | s = "40Gbps"; | |
175 | break; | |
176 | case 100000: | |
177 | s = "100Gbps"; | |
be839e39 CL |
178 | break; |
179 | ||
180 | default: | |
181 | s = "unknown"; | |
182 | break; | |
183 | } | |
184 | ||
185 | switch (pi->link_cfg.fc) { | |
186 | case PAUSE_RX: | |
187 | fc = "RX"; | |
188 | break; | |
189 | ||
190 | case PAUSE_TX: | |
191 | fc = "TX"; | |
192 | break; | |
193 | ||
194 | case PAUSE_RX|PAUSE_TX: | |
195 | fc = "RX/TX"; | |
196 | break; | |
197 | ||
198 | default: | |
199 | fc = "no"; | |
200 | break; | |
201 | } | |
202 | ||
428ac43f | 203 | netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s, fc); |
be839e39 CL |
204 | } else { |
205 | netif_carrier_off(dev); | |
428ac43f | 206 | netdev_info(dev, "link down\n"); |
be839e39 CL |
207 | } |
208 | } | |
209 | ||
5ad24def HS |
210 | /* |
211 | * THe port module type has changed on the indicated "port" (Virtual | |
212 | * Interface). | |
213 | */ | |
214 | void t4vf_os_portmod_changed(struct adapter *adapter, int pidx) | |
215 | { | |
216 | static const char * const mod_str[] = { | |
217 | NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM" | |
218 | }; | |
219 | const struct net_device *dev = adapter->port[pidx]; | |
220 | const struct port_info *pi = netdev_priv(dev); | |
221 | ||
222 | if (pi->mod_type == FW_PORT_MOD_TYPE_NONE) | |
223 | dev_info(adapter->pdev_dev, "%s: port module unplugged\n", | |
224 | dev->name); | |
225 | else if (pi->mod_type < ARRAY_SIZE(mod_str)) | |
226 | dev_info(adapter->pdev_dev, "%s: %s port module inserted\n", | |
227 | dev->name, mod_str[pi->mod_type]); | |
228 | else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED) | |
229 | dev_info(adapter->pdev_dev, "%s: unsupported optical port " | |
230 | "module inserted\n", dev->name); | |
231 | else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN) | |
232 | dev_info(adapter->pdev_dev, "%s: unknown port module inserted," | |
233 | "forcing TWINAX\n", dev->name); | |
234 | else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR) | |
235 | dev_info(adapter->pdev_dev, "%s: transceiver module error\n", | |
236 | dev->name); | |
237 | else | |
238 | dev_info(adapter->pdev_dev, "%s: unknown module type %d " | |
239 | "inserted\n", dev->name, pi->mod_type); | |
240 | } | |
241 | ||
be839e39 CL |
242 | /* |
243 | * Net device operations. | |
244 | * ====================== | |
245 | */ | |
246 | ||
be839e39 | 247 | |
87737663 | 248 | |
be839e39 CL |
249 | |
250 | /* | |
251 | * Perform the MAC and PHY actions needed to enable a "port" (Virtual | |
252 | * Interface). | |
253 | */ | |
254 | static int link_start(struct net_device *dev) | |
255 | { | |
256 | int ret; | |
257 | struct port_info *pi = netdev_priv(dev); | |
258 | ||
259 | /* | |
260 | * We do not set address filters and promiscuity here, the stack does | |
87737663 | 261 | * that step explicitly. Enable vlan accel. |
be839e39 | 262 | */ |
87737663 | 263 | ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, 1, |
be839e39 CL |
264 | true); |
265 | if (ret == 0) { | |
266 | ret = t4vf_change_mac(pi->adapter, pi->viid, | |
267 | pi->xact_addr_filt, dev->dev_addr, true); | |
268 | if (ret >= 0) { | |
269 | pi->xact_addr_filt = ret; | |
270 | ret = 0; | |
271 | } | |
272 | } | |
273 | ||
274 | /* | |
275 | * We don't need to actually "start the link" itself since the | |
276 | * firmware will do that for us when the first Virtual Interface | |
277 | * is enabled on a port. | |
278 | */ | |
279 | if (ret == 0) | |
280 | ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true); | |
281 | return ret; | |
282 | } | |
283 | ||
284 | /* | |
285 | * Name the MSI-X interrupts. | |
286 | */ | |
287 | static void name_msix_vecs(struct adapter *adapter) | |
288 | { | |
289 | int namelen = sizeof(adapter->msix_info[0].desc) - 1; | |
290 | int pidx; | |
291 | ||
292 | /* | |
293 | * Firmware events. | |
294 | */ | |
295 | snprintf(adapter->msix_info[MSIX_FW].desc, namelen, | |
296 | "%s-FWeventq", adapter->name); | |
297 | adapter->msix_info[MSIX_FW].desc[namelen] = 0; | |
298 | ||
299 | /* | |
300 | * Ethernet queues. | |
301 | */ | |
302 | for_each_port(adapter, pidx) { | |
303 | struct net_device *dev = adapter->port[pidx]; | |
304 | const struct port_info *pi = netdev_priv(dev); | |
305 | int qs, msi; | |
306 | ||
caedda35 | 307 | for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) { |
be839e39 CL |
308 | snprintf(adapter->msix_info[msi].desc, namelen, |
309 | "%s-%d", dev->name, qs); | |
310 | adapter->msix_info[msi].desc[namelen] = 0; | |
311 | } | |
312 | } | |
313 | } | |
314 | ||
315 | /* | |
316 | * Request all of our MSI-X resources. | |
317 | */ | |
318 | static int request_msix_queue_irqs(struct adapter *adapter) | |
319 | { | |
320 | struct sge *s = &adapter->sge; | |
321 | int rxq, msi, err; | |
322 | ||
323 | /* | |
324 | * Firmware events. | |
325 | */ | |
326 | err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix, | |
327 | 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq); | |
328 | if (err) | |
329 | return err; | |
330 | ||
331 | /* | |
332 | * Ethernet queues. | |
333 | */ | |
caedda35 | 334 | msi = MSIX_IQFLINT; |
be839e39 CL |
335 | for_each_ethrxq(s, rxq) { |
336 | err = request_irq(adapter->msix_info[msi].vec, | |
337 | t4vf_sge_intr_msix, 0, | |
338 | adapter->msix_info[msi].desc, | |
339 | &s->ethrxq[rxq].rspq); | |
340 | if (err) | |
341 | goto err_free_irqs; | |
342 | msi++; | |
343 | } | |
344 | return 0; | |
345 | ||
346 | err_free_irqs: | |
347 | while (--rxq >= 0) | |
348 | free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq); | |
349 | free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq); | |
350 | return err; | |
351 | } | |
352 | ||
353 | /* | |
354 | * Free our MSI-X resources. | |
355 | */ | |
356 | static void free_msix_queue_irqs(struct adapter *adapter) | |
357 | { | |
358 | struct sge *s = &adapter->sge; | |
359 | int rxq, msi; | |
360 | ||
361 | free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq); | |
caedda35 | 362 | msi = MSIX_IQFLINT; |
be839e39 CL |
363 | for_each_ethrxq(s, rxq) |
364 | free_irq(adapter->msix_info[msi++].vec, | |
365 | &s->ethrxq[rxq].rspq); | |
366 | } | |
367 | ||
368 | /* | |
369 | * Turn on NAPI and start up interrupts on a response queue. | |
370 | */ | |
371 | static void qenable(struct sge_rspq *rspq) | |
372 | { | |
373 | napi_enable(&rspq->napi); | |
374 | ||
375 | /* | |
376 | * 0-increment the Going To Sleep register to start the timer and | |
377 | * enable interrupts. | |
378 | */ | |
379 | t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS, | |
f612b815 HS |
380 | CIDXINC_V(0) | |
381 | SEINTARM_V(rspq->intr_params) | | |
382 | INGRESSQID_V(rspq->cntxt_id)); | |
be839e39 CL |
383 | } |
384 | ||
385 | /* | |
386 | * Enable NAPI scheduling and interrupt generation for all Receive Queues. | |
387 | */ | |
388 | static void enable_rx(struct adapter *adapter) | |
389 | { | |
390 | int rxq; | |
391 | struct sge *s = &adapter->sge; | |
392 | ||
393 | for_each_ethrxq(s, rxq) | |
394 | qenable(&s->ethrxq[rxq].rspq); | |
395 | qenable(&s->fw_evtq); | |
396 | ||
397 | /* | |
398 | * The interrupt queue doesn't use NAPI so we do the 0-increment of | |
399 | * its Going To Sleep register here to get it started. | |
400 | */ | |
401 | if (adapter->flags & USING_MSI) | |
402 | t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS, | |
f612b815 HS |
403 | CIDXINC_V(0) | |
404 | SEINTARM_V(s->intrq.intr_params) | | |
405 | INGRESSQID_V(s->intrq.cntxt_id)); | |
be839e39 CL |
406 | |
407 | } | |
408 | ||
409 | /* | |
410 | * Wait until all NAPI handlers are descheduled. | |
411 | */ | |
412 | static void quiesce_rx(struct adapter *adapter) | |
413 | { | |
414 | struct sge *s = &adapter->sge; | |
415 | int rxq; | |
416 | ||
417 | for_each_ethrxq(s, rxq) | |
418 | napi_disable(&s->ethrxq[rxq].rspq.napi); | |
419 | napi_disable(&s->fw_evtq.napi); | |
420 | } | |
421 | ||
422 | /* | |
423 | * Response queue handler for the firmware event queue. | |
424 | */ | |
425 | static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp, | |
426 | const struct pkt_gl *gl) | |
427 | { | |
428 | /* | |
429 | * Extract response opcode and get pointer to CPL message body. | |
430 | */ | |
431 | struct adapter *adapter = rspq->adapter; | |
432 | u8 opcode = ((const struct rss_header *)rsp)->opcode; | |
433 | void *cpl = (void *)(rsp + 1); | |
434 | ||
435 | switch (opcode) { | |
436 | case CPL_FW6_MSG: { | |
437 | /* | |
438 | * We've received an asynchronous message from the firmware. | |
439 | */ | |
440 | const struct cpl_fw6_msg *fw_msg = cpl; | |
441 | if (fw_msg->type == FW6_TYPE_CMD_RPL) | |
442 | t4vf_handle_fw_rpl(adapter, fw_msg->data); | |
443 | break; | |
444 | } | |
445 | ||
94dace10 VP |
446 | case CPL_FW4_MSG: { |
447 | /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG. | |
448 | */ | |
449 | const struct cpl_sge_egr_update *p = (void *)(rsp + 3); | |
6c53e938 | 450 | opcode = CPL_OPCODE_G(ntohl(p->opcode_qid)); |
94dace10 VP |
451 | if (opcode != CPL_SGE_EGR_UPDATE) { |
452 | dev_err(adapter->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n" | |
453 | , opcode); | |
454 | break; | |
455 | } | |
456 | cpl = (void *)p; | |
457 | /*FALLTHROUGH*/ | |
458 | } | |
459 | ||
be839e39 CL |
460 | case CPL_SGE_EGR_UPDATE: { |
461 | /* | |
7f9dd2fa CL |
462 | * We've received an Egress Queue Status Update message. We |
463 | * get these, if the SGE is configured to send these when the | |
464 | * firmware passes certain points in processing our TX | |
465 | * Ethernet Queue or if we make an explicit request for one. | |
466 | * We use these updates to determine when we may need to | |
467 | * restart a TX Ethernet Queue which was stopped for lack of | |
468 | * free TX Queue Descriptors ... | |
be839e39 | 469 | */ |
64699336 | 470 | const struct cpl_sge_egr_update *p = cpl; |
bdc590b9 | 471 | unsigned int qid = EGR_QID_G(be32_to_cpu(p->opcode_qid)); |
be839e39 CL |
472 | struct sge *s = &adapter->sge; |
473 | struct sge_txq *tq; | |
474 | struct sge_eth_txq *txq; | |
475 | unsigned int eq_idx; | |
be839e39 CL |
476 | |
477 | /* | |
478 | * Perform sanity checking on the Queue ID to make sure it | |
479 | * really refers to one of our TX Ethernet Egress Queues which | |
480 | * is active and matches the queue's ID. None of these error | |
481 | * conditions should ever happen so we may want to either make | |
482 | * them fatal and/or conditionalized under DEBUG. | |
483 | */ | |
484 | eq_idx = EQ_IDX(s, qid); | |
485 | if (unlikely(eq_idx >= MAX_EGRQ)) { | |
486 | dev_err(adapter->pdev_dev, | |
487 | "Egress Update QID %d out of range\n", qid); | |
488 | break; | |
489 | } | |
490 | tq = s->egr_map[eq_idx]; | |
491 | if (unlikely(tq == NULL)) { | |
492 | dev_err(adapter->pdev_dev, | |
493 | "Egress Update QID %d TXQ=NULL\n", qid); | |
494 | break; | |
495 | } | |
496 | txq = container_of(tq, struct sge_eth_txq, q); | |
497 | if (unlikely(tq->abs_id != qid)) { | |
498 | dev_err(adapter->pdev_dev, | |
499 | "Egress Update QID %d refers to TXQ %d\n", | |
500 | qid, tq->abs_id); | |
501 | break; | |
502 | } | |
503 | ||
be839e39 CL |
504 | /* |
505 | * Restart a stopped TX Queue which has less than half of its | |
506 | * TX ring in use ... | |
507 | */ | |
508 | txq->q.restarts++; | |
509 | netif_tx_wake_queue(txq->txq); | |
510 | break; | |
511 | } | |
512 | ||
513 | default: | |
514 | dev_err(adapter->pdev_dev, | |
515 | "unexpected CPL %#x on FW event queue\n", opcode); | |
516 | } | |
517 | ||
518 | return 0; | |
519 | } | |
520 | ||
521 | /* | |
522 | * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues | |
523 | * to use and initializes them. We support multiple "Queue Sets" per port if | |
524 | * we have MSI-X, otherwise just one queue set per port. | |
525 | */ | |
526 | static int setup_sge_queues(struct adapter *adapter) | |
527 | { | |
528 | struct sge *s = &adapter->sge; | |
529 | int err, pidx, msix; | |
530 | ||
531 | /* | |
532 | * Clear "Queue Set" Free List Starving and TX Queue Mapping Error | |
533 | * state. | |
534 | */ | |
535 | bitmap_zero(s->starving_fl, MAX_EGRQ); | |
536 | ||
537 | /* | |
538 | * If we're using MSI interrupt mode we need to set up a "forwarded | |
539 | * interrupt" queue which we'll set up with our MSI vector. The rest | |
540 | * of the ingress queues will be set up to forward their interrupts to | |
541 | * this queue ... This must be first since t4vf_sge_alloc_rxq() uses | |
542 | * the intrq's queue ID as the interrupt forwarding queue for the | |
543 | * subsequent calls ... | |
544 | */ | |
545 | if (adapter->flags & USING_MSI) { | |
546 | err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false, | |
547 | adapter->port[0], 0, NULL, NULL); | |
548 | if (err) | |
549 | goto err_free_queues; | |
550 | } | |
551 | ||
552 | /* | |
553 | * Allocate our ingress queue for asynchronous firmware messages. | |
554 | */ | |
555 | err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0], | |
556 | MSIX_FW, NULL, fwevtq_handler); | |
557 | if (err) | |
558 | goto err_free_queues; | |
559 | ||
560 | /* | |
561 | * Allocate each "port"'s initial Queue Sets. These can be changed | |
562 | * later on ... up to the point where any interface on the adapter is | |
563 | * brought up at which point lots of things get nailed down | |
564 | * permanently ... | |
565 | */ | |
caedda35 | 566 | msix = MSIX_IQFLINT; |
be839e39 CL |
567 | for_each_port(adapter, pidx) { |
568 | struct net_device *dev = adapter->port[pidx]; | |
569 | struct port_info *pi = netdev_priv(dev); | |
570 | struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset]; | |
571 | struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset]; | |
be839e39 CL |
572 | int qs; |
573 | ||
c8639a82 | 574 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { |
be839e39 CL |
575 | err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false, |
576 | dev, msix++, | |
577 | &rxq->fl, t4vf_ethrx_handler); | |
578 | if (err) | |
579 | goto err_free_queues; | |
580 | ||
581 | err = t4vf_sge_alloc_eth_txq(adapter, txq, dev, | |
582 | netdev_get_tx_queue(dev, qs), | |
583 | s->fw_evtq.cntxt_id); | |
584 | if (err) | |
585 | goto err_free_queues; | |
586 | ||
587 | rxq->rspq.idx = qs; | |
588 | memset(&rxq->stats, 0, sizeof(rxq->stats)); | |
589 | } | |
590 | } | |
591 | ||
592 | /* | |
593 | * Create the reverse mappings for the queues. | |
594 | */ | |
595 | s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id; | |
596 | s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id; | |
597 | IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq; | |
598 | for_each_port(adapter, pidx) { | |
599 | struct net_device *dev = adapter->port[pidx]; | |
600 | struct port_info *pi = netdev_priv(dev); | |
601 | struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset]; | |
602 | struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset]; | |
be839e39 CL |
603 | int qs; |
604 | ||
c8639a82 | 605 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { |
be839e39 CL |
606 | IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq; |
607 | EQ_MAP(s, txq->q.abs_id) = &txq->q; | |
608 | ||
609 | /* | |
610 | * The FW_IQ_CMD doesn't return the Absolute Queue IDs | |
611 | * for Free Lists but since all of the Egress Queues | |
612 | * (including Free Lists) have Relative Queue IDs | |
613 | * which are computed as Absolute - Base Queue ID, we | |
614 | * can synthesize the Absolute Queue IDs for the Free | |
615 | * Lists. This is useful for debugging purposes when | |
616 | * we want to dump Queue Contexts via the PF Driver. | |
617 | */ | |
618 | rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base; | |
619 | EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl; | |
620 | } | |
621 | } | |
622 | return 0; | |
623 | ||
624 | err_free_queues: | |
625 | t4vf_free_sge_resources(adapter); | |
626 | return err; | |
627 | } | |
628 | ||
629 | /* | |
630 | * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive | |
631 | * queues. We configure the RSS CPU lookup table to distribute to the number | |
632 | * of HW receive queues, and the response queue lookup table to narrow that | |
633 | * down to the response queues actually configured for each "port" (Virtual | |
634 | * Interface). We always configure the RSS mapping for all ports since the | |
635 | * mapping table has plenty of entries. | |
636 | */ | |
637 | static int setup_rss(struct adapter *adapter) | |
638 | { | |
639 | int pidx; | |
640 | ||
641 | for_each_port(adapter, pidx) { | |
642 | struct port_info *pi = adap2pinfo(adapter, pidx); | |
643 | struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset]; | |
644 | u16 rss[MAX_PORT_QSETS]; | |
645 | int qs, err; | |
646 | ||
647 | for (qs = 0; qs < pi->nqsets; qs++) | |
648 | rss[qs] = rxq[qs].rspq.abs_id; | |
649 | ||
650 | err = t4vf_config_rss_range(adapter, pi->viid, | |
651 | 0, pi->rss_size, rss, pi->nqsets); | |
652 | if (err) | |
653 | return err; | |
654 | ||
655 | /* | |
656 | * Perform Global RSS Mode-specific initialization. | |
657 | */ | |
658 | switch (adapter->params.rss.mode) { | |
659 | case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL: | |
660 | /* | |
661 | * If Tunnel All Lookup isn't specified in the global | |
662 | * RSS Configuration, then we need to specify a | |
663 | * default Ingress Queue for any ingress packets which | |
664 | * aren't hashed. We'll use our first ingress queue | |
665 | * ... | |
666 | */ | |
667 | if (!adapter->params.rss.u.basicvirtual.tnlalllookup) { | |
668 | union rss_vi_config config; | |
669 | err = t4vf_read_rss_vi_config(adapter, | |
670 | pi->viid, | |
671 | &config); | |
672 | if (err) | |
673 | return err; | |
674 | config.basicvirtual.defaultq = | |
675 | rxq[0].rspq.abs_id; | |
676 | err = t4vf_write_rss_vi_config(adapter, | |
677 | pi->viid, | |
678 | &config); | |
679 | if (err) | |
680 | return err; | |
681 | } | |
682 | break; | |
683 | } | |
684 | } | |
685 | ||
686 | return 0; | |
687 | } | |
688 | ||
689 | /* | |
690 | * Bring the adapter up. Called whenever we go from no "ports" open to having | |
691 | * one open. This function performs the actions necessary to make an adapter | |
692 | * operational, such as completing the initialization of HW modules, and | |
693 | * enabling interrupts. Must be called with the rtnl lock held. (Note that | |
694 | * this is called "cxgb_up" in the PF Driver.) | |
695 | */ | |
696 | static int adapter_up(struct adapter *adapter) | |
697 | { | |
698 | int err; | |
699 | ||
700 | /* | |
701 | * If this is the first time we've been called, perform basic | |
702 | * adapter setup. Once we've done this, many of our adapter | |
703 | * parameters can no longer be changed ... | |
704 | */ | |
705 | if ((adapter->flags & FULL_INIT_DONE) == 0) { | |
706 | err = setup_sge_queues(adapter); | |
707 | if (err) | |
708 | return err; | |
709 | err = setup_rss(adapter); | |
710 | if (err) { | |
711 | t4vf_free_sge_resources(adapter); | |
712 | return err; | |
713 | } | |
714 | ||
715 | if (adapter->flags & USING_MSIX) | |
716 | name_msix_vecs(adapter); | |
717 | adapter->flags |= FULL_INIT_DONE; | |
718 | } | |
719 | ||
720 | /* | |
721 | * Acquire our interrupt resources. We only support MSI-X and MSI. | |
722 | */ | |
723 | BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0); | |
724 | if (adapter->flags & USING_MSIX) | |
725 | err = request_msix_queue_irqs(adapter); | |
726 | else | |
727 | err = request_irq(adapter->pdev->irq, | |
728 | t4vf_intr_handler(adapter), 0, | |
729 | adapter->name, adapter); | |
730 | if (err) { | |
731 | dev_err(adapter->pdev_dev, "request_irq failed, err %d\n", | |
732 | err); | |
733 | return err; | |
734 | } | |
735 | ||
736 | /* | |
737 | * Enable NAPI ingress processing and return success. | |
738 | */ | |
739 | enable_rx(adapter); | |
740 | t4vf_sge_start(adapter); | |
fe5d2709 HS |
741 | |
742 | /* Initialize hash mac addr list*/ | |
743 | INIT_LIST_HEAD(&adapter->mac_hlist); | |
be839e39 CL |
744 | return 0; |
745 | } | |
746 | ||
747 | /* | |
748 | * Bring the adapter down. Called whenever the last "port" (Virtual | |
749 | * Interface) closed. (Note that this routine is called "cxgb_down" in the PF | |
750 | * Driver.) | |
751 | */ | |
752 | static void adapter_down(struct adapter *adapter) | |
753 | { | |
754 | /* | |
755 | * Free interrupt resources. | |
756 | */ | |
757 | if (adapter->flags & USING_MSIX) | |
758 | free_msix_queue_irqs(adapter); | |
759 | else | |
760 | free_irq(adapter->pdev->irq, adapter); | |
761 | ||
762 | /* | |
763 | * Wait for NAPI handlers to finish. | |
764 | */ | |
765 | quiesce_rx(adapter); | |
766 | } | |
767 | ||
768 | /* | |
769 | * Start up a net device. | |
770 | */ | |
771 | static int cxgb4vf_open(struct net_device *dev) | |
772 | { | |
773 | int err; | |
774 | struct port_info *pi = netdev_priv(dev); | |
775 | struct adapter *adapter = pi->adapter; | |
776 | ||
777 | /* | |
778 | * If this is the first interface that we're opening on the "adapter", | |
779 | * bring the "adapter" up now. | |
780 | */ | |
781 | if (adapter->open_device_map == 0) { | |
782 | err = adapter_up(adapter); | |
783 | if (err) | |
784 | return err; | |
785 | } | |
786 | ||
787 | /* | |
788 | * Note that this interface is up and start everything up ... | |
789 | */ | |
e7a3795f CL |
790 | err = link_start(dev); |
791 | if (err) | |
343a8d13 CL |
792 | goto err_unwind; |
793 | ||
be839e39 | 794 | netif_tx_start_all_queues(dev); |
343a8d13 | 795 | set_bit(pi->port_id, &adapter->open_device_map); |
be839e39 | 796 | return 0; |
343a8d13 CL |
797 | |
798 | err_unwind: | |
799 | if (adapter->open_device_map == 0) | |
800 | adapter_down(adapter); | |
801 | return err; | |
be839e39 CL |
802 | } |
803 | ||
804 | /* | |
805 | * Shut down a net device. This routine is called "cxgb_close" in the PF | |
806 | * Driver ... | |
807 | */ | |
808 | static int cxgb4vf_stop(struct net_device *dev) | |
809 | { | |
be839e39 CL |
810 | struct port_info *pi = netdev_priv(dev); |
811 | struct adapter *adapter = pi->adapter; | |
812 | ||
813 | netif_tx_stop_all_queues(dev); | |
814 | netif_carrier_off(dev); | |
343a8d13 | 815 | t4vf_enable_vi(adapter, pi->viid, false, false); |
be839e39 CL |
816 | pi->link_cfg.link_ok = 0; |
817 | ||
818 | clear_bit(pi->port_id, &adapter->open_device_map); | |
819 | if (adapter->open_device_map == 0) | |
820 | adapter_down(adapter); | |
821 | return 0; | |
822 | } | |
823 | ||
824 | /* | |
825 | * Translate our basic statistics into the standard "ifconfig" statistics. | |
826 | */ | |
827 | static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev) | |
828 | { | |
829 | struct t4vf_port_stats stats; | |
830 | struct port_info *pi = netdev2pinfo(dev); | |
831 | struct adapter *adapter = pi->adapter; | |
832 | struct net_device_stats *ns = &dev->stats; | |
833 | int err; | |
834 | ||
835 | spin_lock(&adapter->stats_lock); | |
836 | err = t4vf_get_port_stats(adapter, pi->pidx, &stats); | |
837 | spin_unlock(&adapter->stats_lock); | |
838 | ||
839 | memset(ns, 0, sizeof(*ns)); | |
840 | if (err) | |
841 | return ns; | |
842 | ||
843 | ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes + | |
844 | stats.tx_ucast_bytes + stats.tx_offload_bytes); | |
845 | ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames + | |
846 | stats.tx_ucast_frames + stats.tx_offload_frames); | |
847 | ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes + | |
848 | stats.rx_ucast_bytes); | |
849 | ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames + | |
850 | stats.rx_ucast_frames); | |
851 | ns->multicast = stats.rx_mcast_frames; | |
852 | ns->tx_errors = stats.tx_drop_frames; | |
853 | ns->rx_errors = stats.rx_err_frames; | |
854 | ||
855 | return ns; | |
856 | } | |
857 | ||
fe5d2709 HS |
858 | static inline int cxgb4vf_set_addr_hash(struct port_info *pi) |
859 | { | |
860 | struct adapter *adapter = pi->adapter; | |
861 | u64 vec = 0; | |
862 | bool ucast = false; | |
863 | struct hash_mac_addr *entry; | |
864 | ||
865 | /* Calculate the hash vector for the updated list and program it */ | |
866 | list_for_each_entry(entry, &adapter->mac_hlist, list) { | |
867 | ucast |= is_unicast_ether_addr(entry->addr); | |
868 | vec |= (1ULL << hash_mac_addr(entry->addr)); | |
869 | } | |
870 | return t4vf_set_addr_hash(adapter, pi->viid, ucast, vec, false); | |
871 | } | |
872 | ||
873 | static int cxgb4vf_mac_sync(struct net_device *netdev, const u8 *mac_addr) | |
be839e39 | 874 | { |
fe5d2709 HS |
875 | struct port_info *pi = netdev_priv(netdev); |
876 | struct adapter *adapter = pi->adapter; | |
877 | int ret; | |
be839e39 CL |
878 | u64 mhash = 0; |
879 | u64 uhash = 0; | |
fe5d2709 HS |
880 | bool free = false; |
881 | bool ucast = is_unicast_ether_addr(mac_addr); | |
882 | const u8 *maclist[1] = {mac_addr}; | |
883 | struct hash_mac_addr *new_entry; | |
884 | ||
885 | ret = t4vf_alloc_mac_filt(adapter, pi->viid, free, 1, maclist, | |
886 | NULL, ucast ? &uhash : &mhash, false); | |
887 | if (ret < 0) | |
888 | goto out; | |
889 | /* if hash != 0, then add the addr to hash addr list | |
890 | * so on the end we will calculate the hash for the | |
891 | * list and program it | |
892 | */ | |
893 | if (uhash || mhash) { | |
894 | new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC); | |
895 | if (!new_entry) | |
896 | return -ENOMEM; | |
897 | ether_addr_copy(new_entry->addr, mac_addr); | |
898 | list_add_tail(&new_entry->list, &adapter->mac_hlist); | |
899 | ret = cxgb4vf_set_addr_hash(pi); | |
be839e39 | 900 | } |
fe5d2709 HS |
901 | out: |
902 | return ret < 0 ? ret : 0; | |
903 | } | |
be839e39 | 904 | |
fe5d2709 HS |
905 | static int cxgb4vf_mac_unsync(struct net_device *netdev, const u8 *mac_addr) |
906 | { | |
907 | struct port_info *pi = netdev_priv(netdev); | |
908 | struct adapter *adapter = pi->adapter; | |
909 | int ret; | |
910 | const u8 *maclist[1] = {mac_addr}; | |
911 | struct hash_mac_addr *entry, *tmp; | |
42eb59d3 | 912 | |
fe5d2709 HS |
913 | /* If the MAC address to be removed is in the hash addr |
914 | * list, delete it from the list and update hash vector | |
915 | */ | |
916 | list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist, list) { | |
917 | if (ether_addr_equal(entry->addr, mac_addr)) { | |
918 | list_del(&entry->list); | |
919 | kfree(entry); | |
920 | return cxgb4vf_set_addr_hash(pi); | |
921 | } | |
be839e39 CL |
922 | } |
923 | ||
fe5d2709 HS |
924 | ret = t4vf_free_mac_filt(adapter, pi->viid, 1, maclist, false); |
925 | return ret < 0 ? -EINVAL : 0; | |
be839e39 CL |
926 | } |
927 | ||
928 | /* | |
929 | * Set RX properties of a port, such as promiscruity, address filters, and MTU. | |
930 | * If @mtu is -1 it is left unchanged. | |
931 | */ | |
932 | static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok) | |
933 | { | |
be839e39 CL |
934 | struct port_info *pi = netdev_priv(dev); |
935 | ||
d01f7abc HS |
936 | __dev_uc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync); |
937 | __dev_mc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync); | |
fe5d2709 HS |
938 | return t4vf_set_rxmode(pi->adapter, pi->viid, -1, |
939 | (dev->flags & IFF_PROMISC) != 0, | |
940 | (dev->flags & IFF_ALLMULTI) != 0, | |
941 | 1, -1, sleep_ok); | |
be839e39 CL |
942 | } |
943 | ||
944 | /* | |
945 | * Set the current receive modes on the device. | |
946 | */ | |
947 | static void cxgb4vf_set_rxmode(struct net_device *dev) | |
948 | { | |
949 | /* unfortunately we can't return errors to the stack */ | |
950 | set_rxmode(dev, -1, false); | |
951 | } | |
952 | ||
953 | /* | |
954 | * Find the entry in the interrupt holdoff timer value array which comes | |
955 | * closest to the specified interrupt holdoff value. | |
956 | */ | |
957 | static int closest_timer(const struct sge *s, int us) | |
958 | { | |
959 | int i, timer_idx = 0, min_delta = INT_MAX; | |
960 | ||
961 | for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) { | |
962 | int delta = us - s->timer_val[i]; | |
963 | if (delta < 0) | |
964 | delta = -delta; | |
965 | if (delta < min_delta) { | |
966 | min_delta = delta; | |
967 | timer_idx = i; | |
968 | } | |
969 | } | |
970 | return timer_idx; | |
971 | } | |
972 | ||
973 | static int closest_thres(const struct sge *s, int thres) | |
974 | { | |
975 | int i, delta, pktcnt_idx = 0, min_delta = INT_MAX; | |
976 | ||
977 | for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) { | |
978 | delta = thres - s->counter_val[i]; | |
979 | if (delta < 0) | |
980 | delta = -delta; | |
981 | if (delta < min_delta) { | |
982 | min_delta = delta; | |
983 | pktcnt_idx = i; | |
984 | } | |
985 | } | |
986 | return pktcnt_idx; | |
987 | } | |
988 | ||
989 | /* | |
990 | * Return a queue's interrupt hold-off time in us. 0 means no timer. | |
991 | */ | |
992 | static unsigned int qtimer_val(const struct adapter *adapter, | |
993 | const struct sge_rspq *rspq) | |
994 | { | |
1ecc7b7a | 995 | unsigned int timer_idx = QINTR_TIMER_IDX_G(rspq->intr_params); |
be839e39 CL |
996 | |
997 | return timer_idx < SGE_NTIMERS | |
998 | ? adapter->sge.timer_val[timer_idx] | |
999 | : 0; | |
1000 | } | |
1001 | ||
1002 | /** | |
1003 | * set_rxq_intr_params - set a queue's interrupt holdoff parameters | |
1004 | * @adapter: the adapter | |
1005 | * @rspq: the RX response queue | |
1006 | * @us: the hold-off time in us, or 0 to disable timer | |
1007 | * @cnt: the hold-off packet count, or 0 to disable counter | |
1008 | * | |
1009 | * Sets an RX response queue's interrupt hold-off time and packet count. | |
1010 | * At least one of the two needs to be enabled for the queue to generate | |
1011 | * interrupts. | |
1012 | */ | |
1013 | static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq, | |
1014 | unsigned int us, unsigned int cnt) | |
1015 | { | |
1016 | unsigned int timer_idx; | |
1017 | ||
1018 | /* | |
1019 | * If both the interrupt holdoff timer and count are specified as | |
1020 | * zero, default to a holdoff count of 1 ... | |
1021 | */ | |
1022 | if ((us | cnt) == 0) | |
1023 | cnt = 1; | |
1024 | ||
1025 | /* | |
1026 | * If an interrupt holdoff count has been specified, then find the | |
1027 | * closest configured holdoff count and use that. If the response | |
1028 | * queue has already been created, then update its queue context | |
1029 | * parameters ... | |
1030 | */ | |
1031 | if (cnt) { | |
1032 | int err; | |
1033 | u32 v, pktcnt_idx; | |
1034 | ||
1035 | pktcnt_idx = closest_thres(&adapter->sge, cnt); | |
1036 | if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) { | |
5167865a HS |
1037 | v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) | |
1038 | FW_PARAMS_PARAM_X_V( | |
be839e39 | 1039 | FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) | |
5167865a | 1040 | FW_PARAMS_PARAM_YZ_V(rspq->cntxt_id); |
be839e39 CL |
1041 | err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx); |
1042 | if (err) | |
1043 | return err; | |
1044 | } | |
1045 | rspq->pktcnt_idx = pktcnt_idx; | |
1046 | } | |
1047 | ||
1048 | /* | |
1049 | * Compute the closest holdoff timer index from the supplied holdoff | |
1050 | * timer value. | |
1051 | */ | |
1052 | timer_idx = (us == 0 | |
1053 | ? SGE_TIMER_RSTRT_CNTR | |
1054 | : closest_timer(&adapter->sge, us)); | |
1055 | ||
1056 | /* | |
1057 | * Update the response queue's interrupt coalescing parameters and | |
1058 | * return success. | |
1059 | */ | |
1ecc7b7a HS |
1060 | rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) | |
1061 | QINTR_CNT_EN_V(cnt > 0)); | |
be839e39 CL |
1062 | return 0; |
1063 | } | |
1064 | ||
1065 | /* | |
1066 | * Return a version number to identify the type of adapter. The scheme is: | |
1067 | * - bits 0..9: chip version | |
1068 | * - bits 10..15: chip revision | |
1069 | */ | |
1070 | static inline unsigned int mk_adap_vers(const struct adapter *adapter) | |
1071 | { | |
1072 | /* | |
1073 | * Chip version 4, revision 0x3f (cxgb4vf). | |
1074 | */ | |
70ee3666 | 1075 | return CHELSIO_CHIP_VERSION(adapter->params.chip) | (0x3f << 10); |
be839e39 CL |
1076 | } |
1077 | ||
1078 | /* | |
1079 | * Execute the specified ioctl command. | |
1080 | */ | |
1081 | static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd) | |
1082 | { | |
1083 | int ret = 0; | |
1084 | ||
1085 | switch (cmd) { | |
1086 | /* | |
1087 | * The VF Driver doesn't have access to any of the other | |
1088 | * common Ethernet device ioctl()'s (like reading/writing | |
1089 | * PHY registers, etc. | |
1090 | */ | |
1091 | ||
1092 | default: | |
1093 | ret = -EOPNOTSUPP; | |
1094 | break; | |
1095 | } | |
1096 | return ret; | |
1097 | } | |
1098 | ||
1099 | /* | |
1100 | * Change the device's MTU. | |
1101 | */ | |
1102 | static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu) | |
1103 | { | |
1104 | int ret; | |
1105 | struct port_info *pi = netdev_priv(dev); | |
1106 | ||
be839e39 CL |
1107 | ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu, |
1108 | -1, -1, -1, -1, true); | |
1109 | if (!ret) | |
1110 | dev->mtu = new_mtu; | |
1111 | return ret; | |
1112 | } | |
1113 | ||
c8f44aff MM |
1114 | static netdev_features_t cxgb4vf_fix_features(struct net_device *dev, |
1115 | netdev_features_t features) | |
87737663 JP |
1116 | { |
1117 | /* | |
1118 | * Since there is no support for separate rx/tx vlan accel | |
1119 | * enable/disable make sure tx flag is always in same state as rx. | |
1120 | */ | |
f646968f PM |
1121 | if (features & NETIF_F_HW_VLAN_CTAG_RX) |
1122 | features |= NETIF_F_HW_VLAN_CTAG_TX; | |
87737663 | 1123 | else |
f646968f | 1124 | features &= ~NETIF_F_HW_VLAN_CTAG_TX; |
87737663 JP |
1125 | |
1126 | return features; | |
1127 | } | |
1128 | ||
c8f44aff MM |
1129 | static int cxgb4vf_set_features(struct net_device *dev, |
1130 | netdev_features_t features) | |
87737663 JP |
1131 | { |
1132 | struct port_info *pi = netdev_priv(dev); | |
c8f44aff | 1133 | netdev_features_t changed = dev->features ^ features; |
87737663 | 1134 | |
f646968f | 1135 | if (changed & NETIF_F_HW_VLAN_CTAG_RX) |
87737663 | 1136 | t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1, |
f646968f | 1137 | features & NETIF_F_HW_VLAN_CTAG_TX, 0); |
87737663 JP |
1138 | |
1139 | return 0; | |
1140 | } | |
1141 | ||
be839e39 CL |
1142 | /* |
1143 | * Change the devices MAC address. | |
1144 | */ | |
1145 | static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr) | |
1146 | { | |
1147 | int ret; | |
1148 | struct sockaddr *addr = _addr; | |
1149 | struct port_info *pi = netdev_priv(dev); | |
1150 | ||
1151 | if (!is_valid_ether_addr(addr->sa_data)) | |
504f9b5a | 1152 | return -EADDRNOTAVAIL; |
be839e39 CL |
1153 | |
1154 | ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt, | |
1155 | addr->sa_data, true); | |
1156 | if (ret < 0) | |
1157 | return ret; | |
1158 | ||
1159 | memcpy(dev->dev_addr, addr->sa_data, dev->addr_len); | |
1160 | pi->xact_addr_filt = ret; | |
1161 | return 0; | |
1162 | } | |
1163 | ||
be839e39 CL |
1164 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1165 | /* | |
1166 | * Poll all of our receive queues. This is called outside of normal interrupt | |
1167 | * context. | |
1168 | */ | |
1169 | static void cxgb4vf_poll_controller(struct net_device *dev) | |
1170 | { | |
1171 | struct port_info *pi = netdev_priv(dev); | |
1172 | struct adapter *adapter = pi->adapter; | |
1173 | ||
1174 | if (adapter->flags & USING_MSIX) { | |
1175 | struct sge_eth_rxq *rxq; | |
1176 | int nqsets; | |
1177 | ||
1178 | rxq = &adapter->sge.ethrxq[pi->first_qset]; | |
1179 | for (nqsets = pi->nqsets; nqsets; nqsets--) { | |
1180 | t4vf_sge_intr_msix(0, &rxq->rspq); | |
1181 | rxq++; | |
1182 | } | |
1183 | } else | |
1184 | t4vf_intr_handler(adapter)(0, adapter); | |
1185 | } | |
1186 | #endif | |
1187 | ||
1188 | /* | |
1189 | * Ethtool operations. | |
1190 | * =================== | |
1191 | * | |
1192 | * Note that we don't support any ethtool operations which change the physical | |
1193 | * state of the port to which we're linked. | |
1194 | */ | |
1195 | ||
eb97ad99 GG |
1196 | /** |
1197 | * from_fw_port_mod_type - translate Firmware Port/Module type to Ethtool | |
1198 | * @port_type: Firmware Port Type | |
1199 | * @mod_type: Firmware Module Type | |
1200 | * | |
1201 | * Translate Firmware Port/Module type to Ethtool Port Type. | |
1202 | */ | |
1203 | static int from_fw_port_mod_type(enum fw_port_type port_type, | |
1204 | enum fw_port_module_type mod_type) | |
be839e39 | 1205 | { |
eb97ad99 GG |
1206 | if (port_type == FW_PORT_TYPE_BT_SGMII || |
1207 | port_type == FW_PORT_TYPE_BT_XFI || | |
1208 | port_type == FW_PORT_TYPE_BT_XAUI) { | |
1209 | return PORT_TP; | |
1210 | } else if (port_type == FW_PORT_TYPE_FIBER_XFI || | |
1211 | port_type == FW_PORT_TYPE_FIBER_XAUI) { | |
1212 | return PORT_FIBRE; | |
1213 | } else if (port_type == FW_PORT_TYPE_SFP || | |
1214 | port_type == FW_PORT_TYPE_QSFP_10G || | |
1215 | port_type == FW_PORT_TYPE_QSA || | |
1216 | port_type == FW_PORT_TYPE_QSFP) { | |
1217 | if (mod_type == FW_PORT_MOD_TYPE_LR || | |
1218 | mod_type == FW_PORT_MOD_TYPE_SR || | |
1219 | mod_type == FW_PORT_MOD_TYPE_ER || | |
1220 | mod_type == FW_PORT_MOD_TYPE_LRM) | |
1221 | return PORT_FIBRE; | |
1222 | else if (mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE || | |
1223 | mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE) | |
1224 | return PORT_DA; | |
1225 | else | |
1226 | return PORT_OTHER; | |
5ad24def HS |
1227 | } |
1228 | ||
eb97ad99 | 1229 | return PORT_OTHER; |
5ad24def | 1230 | } |
be839e39 | 1231 | |
eb97ad99 GG |
1232 | /** |
1233 | * fw_caps_to_lmm - translate Firmware to ethtool Link Mode Mask | |
1234 | * @port_type: Firmware Port Type | |
1235 | * @fw_caps: Firmware Port Capabilities | |
1236 | * @link_mode_mask: ethtool Link Mode Mask | |
1237 | * | |
1238 | * Translate a Firmware Port Capabilities specification to an ethtool | |
1239 | * Link Mode Mask. | |
1240 | */ | |
1241 | static void fw_caps_to_lmm(enum fw_port_type port_type, | |
1242 | unsigned int fw_caps, | |
1243 | unsigned long *link_mode_mask) | |
5ad24def | 1244 | { |
eb97ad99 GG |
1245 | #define SET_LMM(__lmm_name) __set_bit(ETHTOOL_LINK_MODE_ ## __lmm_name\ |
1246 | ## _BIT, link_mode_mask) | |
1247 | ||
1248 | #define FW_CAPS_TO_LMM(__fw_name, __lmm_name) \ | |
1249 | do { \ | |
1250 | if (fw_caps & FW_PORT_CAP_ ## __fw_name) \ | |
1251 | SET_LMM(__lmm_name); \ | |
1252 | } while (0) | |
1253 | ||
1254 | switch (port_type) { | |
1255 | case FW_PORT_TYPE_BT_SGMII: | |
1256 | case FW_PORT_TYPE_BT_XFI: | |
1257 | case FW_PORT_TYPE_BT_XAUI: | |
1258 | SET_LMM(TP); | |
1259 | FW_CAPS_TO_LMM(SPEED_100M, 100baseT_Full); | |
1260 | FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full); | |
1261 | FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full); | |
1262 | break; | |
1263 | ||
1264 | case FW_PORT_TYPE_KX4: | |
1265 | case FW_PORT_TYPE_KX: | |
1266 | SET_LMM(Backplane); | |
1267 | FW_CAPS_TO_LMM(SPEED_1G, 1000baseKX_Full); | |
1268 | FW_CAPS_TO_LMM(SPEED_10G, 10000baseKX4_Full); | |
1269 | break; | |
1270 | ||
1271 | case FW_PORT_TYPE_KR: | |
1272 | SET_LMM(Backplane); | |
1273 | SET_LMM(10000baseKR_Full); | |
1274 | break; | |
1275 | ||
1276 | case FW_PORT_TYPE_BP_AP: | |
1277 | SET_LMM(Backplane); | |
1278 | SET_LMM(10000baseR_FEC); | |
1279 | SET_LMM(10000baseKR_Full); | |
1280 | SET_LMM(1000baseKX_Full); | |
1281 | break; | |
1282 | ||
1283 | case FW_PORT_TYPE_BP4_AP: | |
1284 | SET_LMM(Backplane); | |
1285 | SET_LMM(10000baseR_FEC); | |
1286 | SET_LMM(10000baseKR_Full); | |
1287 | SET_LMM(1000baseKX_Full); | |
1288 | SET_LMM(10000baseKX4_Full); | |
1289 | break; | |
1290 | ||
1291 | case FW_PORT_TYPE_FIBER_XFI: | |
1292 | case FW_PORT_TYPE_FIBER_XAUI: | |
1293 | case FW_PORT_TYPE_SFP: | |
1294 | case FW_PORT_TYPE_QSFP_10G: | |
1295 | case FW_PORT_TYPE_QSA: | |
1296 | SET_LMM(FIBRE); | |
1297 | FW_CAPS_TO_LMM(SPEED_1G, 1000baseT_Full); | |
1298 | FW_CAPS_TO_LMM(SPEED_10G, 10000baseT_Full); | |
1299 | break; | |
1300 | ||
1301 | case FW_PORT_TYPE_BP40_BA: | |
1302 | case FW_PORT_TYPE_QSFP: | |
1303 | SET_LMM(FIBRE); | |
1304 | SET_LMM(40000baseSR4_Full); | |
1305 | break; | |
5ad24def | 1306 | |
eb97ad99 GG |
1307 | case FW_PORT_TYPE_CR_QSFP: |
1308 | case FW_PORT_TYPE_SFP28: | |
1309 | SET_LMM(FIBRE); | |
1310 | SET_LMM(25000baseCR_Full); | |
1311 | break; | |
1312 | ||
1313 | case FW_PORT_TYPE_KR4_100G: | |
1314 | case FW_PORT_TYPE_CR4_QSFP: | |
1315 | SET_LMM(FIBRE); | |
1316 | SET_LMM(100000baseCR4_Full); | |
1317 | break; | |
1318 | ||
1319 | default: | |
1320 | break; | |
1321 | } | |
1322 | ||
1323 | FW_CAPS_TO_LMM(ANEG, Autoneg); | |
1324 | FW_CAPS_TO_LMM(802_3_PAUSE, Pause); | |
1325 | FW_CAPS_TO_LMM(802_3_ASM_DIR, Asym_Pause); | |
1326 | ||
1327 | #undef FW_CAPS_TO_LMM | |
1328 | #undef SET_LMM | |
1329 | } | |
1330 | ||
1331 | static int cxgb4vf_get_link_ksettings(struct net_device *dev, | |
1332 | struct ethtool_link_ksettings | |
1333 | *link_ksettings) | |
1334 | { | |
1335 | const struct port_info *pi = netdev_priv(dev); | |
1336 | struct ethtool_link_settings *base = &link_ksettings->base; | |
1337 | ||
1338 | ethtool_link_ksettings_zero_link_mode(link_ksettings, supported); | |
1339 | ethtool_link_ksettings_zero_link_mode(link_ksettings, advertising); | |
1340 | ethtool_link_ksettings_zero_link_mode(link_ksettings, lp_advertising); | |
1341 | ||
1342 | base->port = from_fw_port_mod_type(pi->port_type, pi->mod_type); | |
1343 | ||
1344 | if (pi->mdio_addr >= 0) { | |
1345 | base->phy_address = pi->mdio_addr; | |
1346 | base->mdio_support = (pi->port_type == FW_PORT_TYPE_BT_SGMII | |
1347 | ? ETH_MDIO_SUPPORTS_C22 | |
1348 | : ETH_MDIO_SUPPORTS_C45); | |
5ad24def | 1349 | } else { |
eb97ad99 GG |
1350 | base->phy_address = 255; |
1351 | base->mdio_support = 0; | |
5ad24def HS |
1352 | } |
1353 | ||
eb97ad99 GG |
1354 | fw_caps_to_lmm(pi->port_type, pi->link_cfg.supported, |
1355 | link_ksettings->link_modes.supported); | |
1356 | fw_caps_to_lmm(pi->port_type, pi->link_cfg.advertising, | |
1357 | link_ksettings->link_modes.advertising); | |
1358 | fw_caps_to_lmm(pi->port_type, pi->link_cfg.lp_advertising, | |
1359 | link_ksettings->link_modes.lp_advertising); | |
1360 | ||
1361 | if (netif_carrier_ok(dev)) { | |
1362 | base->speed = pi->link_cfg.speed; | |
1363 | base->duplex = DUPLEX_FULL; | |
1364 | } else { | |
1365 | base->speed = SPEED_UNKNOWN; | |
1366 | base->duplex = DUPLEX_UNKNOWN; | |
1367 | } | |
1368 | ||
1369 | base->autoneg = pi->link_cfg.autoneg; | |
1370 | if (pi->link_cfg.supported & FW_PORT_CAP_ANEG) | |
1371 | ethtool_link_ksettings_add_link_mode(link_ksettings, | |
1372 | supported, Autoneg); | |
1373 | if (pi->link_cfg.autoneg) | |
1374 | ethtool_link_ksettings_add_link_mode(link_ksettings, | |
1375 | advertising, Autoneg); | |
1376 | ||
be839e39 CL |
1377 | return 0; |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * Return our driver information. | |
1382 | */ | |
1383 | static void cxgb4vf_get_drvinfo(struct net_device *dev, | |
1384 | struct ethtool_drvinfo *drvinfo) | |
1385 | { | |
1386 | struct adapter *adapter = netdev2adap(dev); | |
1387 | ||
23020ab3 RJ |
1388 | strlcpy(drvinfo->driver, KBUILD_MODNAME, sizeof(drvinfo->driver)); |
1389 | strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); | |
1390 | strlcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)), | |
1391 | sizeof(drvinfo->bus_info)); | |
be839e39 CL |
1392 | snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), |
1393 | "%u.%u.%u.%u, TP %u.%u.%u.%u", | |
b2e1a3f0 HS |
1394 | FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.fwrev), |
1395 | FW_HDR_FW_VER_MINOR_G(adapter->params.dev.fwrev), | |
1396 | FW_HDR_FW_VER_MICRO_G(adapter->params.dev.fwrev), | |
1397 | FW_HDR_FW_VER_BUILD_G(adapter->params.dev.fwrev), | |
1398 | FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.tprev), | |
1399 | FW_HDR_FW_VER_MINOR_G(adapter->params.dev.tprev), | |
1400 | FW_HDR_FW_VER_MICRO_G(adapter->params.dev.tprev), | |
1401 | FW_HDR_FW_VER_BUILD_G(adapter->params.dev.tprev)); | |
be839e39 CL |
1402 | } |
1403 | ||
1404 | /* | |
1405 | * Return current adapter message level. | |
1406 | */ | |
1407 | static u32 cxgb4vf_get_msglevel(struct net_device *dev) | |
1408 | { | |
1409 | return netdev2adap(dev)->msg_enable; | |
1410 | } | |
1411 | ||
1412 | /* | |
1413 | * Set current adapter message level. | |
1414 | */ | |
1415 | static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel) | |
1416 | { | |
1417 | netdev2adap(dev)->msg_enable = msglevel; | |
1418 | } | |
1419 | ||
1420 | /* | |
1421 | * Return the device's current Queue Set ring size parameters along with the | |
1422 | * allowed maximum values. Since ethtool doesn't understand the concept of | |
1423 | * multi-queue devices, we just return the current values associated with the | |
1424 | * first Queue Set. | |
1425 | */ | |
1426 | static void cxgb4vf_get_ringparam(struct net_device *dev, | |
1427 | struct ethtool_ringparam *rp) | |
1428 | { | |
1429 | const struct port_info *pi = netdev_priv(dev); | |
1430 | const struct sge *s = &pi->adapter->sge; | |
1431 | ||
1432 | rp->rx_max_pending = MAX_RX_BUFFERS; | |
1433 | rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES; | |
1434 | rp->rx_jumbo_max_pending = 0; | |
1435 | rp->tx_max_pending = MAX_TXQ_ENTRIES; | |
1436 | ||
1437 | rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID; | |
1438 | rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size; | |
1439 | rp->rx_jumbo_pending = 0; | |
1440 | rp->tx_pending = s->ethtxq[pi->first_qset].q.size; | |
1441 | } | |
1442 | ||
1443 | /* | |
1444 | * Set the Queue Set ring size parameters for the device. Again, since | |
1445 | * ethtool doesn't allow for the concept of multiple queues per device, we'll | |
1446 | * apply these new values across all of the Queue Sets associated with the | |
1447 | * device -- after vetting them of course! | |
1448 | */ | |
1449 | static int cxgb4vf_set_ringparam(struct net_device *dev, | |
1450 | struct ethtool_ringparam *rp) | |
1451 | { | |
1452 | const struct port_info *pi = netdev_priv(dev); | |
1453 | struct adapter *adapter = pi->adapter; | |
1454 | struct sge *s = &adapter->sge; | |
1455 | int qs; | |
1456 | ||
1457 | if (rp->rx_pending > MAX_RX_BUFFERS || | |
1458 | rp->rx_jumbo_pending || | |
1459 | rp->tx_pending > MAX_TXQ_ENTRIES || | |
1460 | rp->rx_mini_pending > MAX_RSPQ_ENTRIES || | |
1461 | rp->rx_mini_pending < MIN_RSPQ_ENTRIES || | |
1462 | rp->rx_pending < MIN_FL_ENTRIES || | |
1463 | rp->tx_pending < MIN_TXQ_ENTRIES) | |
1464 | return -EINVAL; | |
1465 | ||
1466 | if (adapter->flags & FULL_INIT_DONE) | |
1467 | return -EBUSY; | |
1468 | ||
1469 | for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) { | |
1470 | s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID; | |
1471 | s->ethrxq[qs].rspq.size = rp->rx_mini_pending; | |
1472 | s->ethtxq[qs].q.size = rp->tx_pending; | |
1473 | } | |
1474 | return 0; | |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * Return the interrupt holdoff timer and count for the first Queue Set on the | |
1479 | * device. Our extension ioctl() (the cxgbtool interface) allows the | |
1480 | * interrupt holdoff timer to be read on all of the device's Queue Sets. | |
1481 | */ | |
1482 | static int cxgb4vf_get_coalesce(struct net_device *dev, | |
1483 | struct ethtool_coalesce *coalesce) | |
1484 | { | |
1485 | const struct port_info *pi = netdev_priv(dev); | |
1486 | const struct adapter *adapter = pi->adapter; | |
1487 | const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq; | |
1488 | ||
1489 | coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq); | |
1490 | coalesce->rx_max_coalesced_frames = | |
1ecc7b7a | 1491 | ((rspq->intr_params & QINTR_CNT_EN_F) |
be839e39 CL |
1492 | ? adapter->sge.counter_val[rspq->pktcnt_idx] |
1493 | : 0); | |
1494 | return 0; | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * Set the RX interrupt holdoff timer and count for the first Queue Set on the | |
1499 | * interface. Our extension ioctl() (the cxgbtool interface) allows us to set | |
1500 | * the interrupt holdoff timer on any of the device's Queue Sets. | |
1501 | */ | |
1502 | static int cxgb4vf_set_coalesce(struct net_device *dev, | |
1503 | struct ethtool_coalesce *coalesce) | |
1504 | { | |
1505 | const struct port_info *pi = netdev_priv(dev); | |
1506 | struct adapter *adapter = pi->adapter; | |
1507 | ||
1508 | return set_rxq_intr_params(adapter, | |
1509 | &adapter->sge.ethrxq[pi->first_qset].rspq, | |
1510 | coalesce->rx_coalesce_usecs, | |
1511 | coalesce->rx_max_coalesced_frames); | |
1512 | } | |
1513 | ||
1514 | /* | |
1515 | * Report current port link pause parameter settings. | |
1516 | */ | |
1517 | static void cxgb4vf_get_pauseparam(struct net_device *dev, | |
1518 | struct ethtool_pauseparam *pauseparam) | |
1519 | { | |
1520 | struct port_info *pi = netdev_priv(dev); | |
1521 | ||
1522 | pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0; | |
1523 | pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0; | |
1524 | pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0; | |
1525 | } | |
1526 | ||
be839e39 CL |
1527 | /* |
1528 | * Identify the port by blinking the port's LED. | |
1529 | */ | |
857a3d0f DM |
1530 | static int cxgb4vf_phys_id(struct net_device *dev, |
1531 | enum ethtool_phys_id_state state) | |
be839e39 | 1532 | { |
857a3d0f | 1533 | unsigned int val; |
be839e39 CL |
1534 | struct port_info *pi = netdev_priv(dev); |
1535 | ||
857a3d0f DM |
1536 | if (state == ETHTOOL_ID_ACTIVE) |
1537 | val = 0xffff; | |
1538 | else if (state == ETHTOOL_ID_INACTIVE) | |
1539 | val = 0; | |
1540 | else | |
1541 | return -EINVAL; | |
1542 | ||
1543 | return t4vf_identify_port(pi->adapter, pi->viid, val); | |
be839e39 CL |
1544 | } |
1545 | ||
1546 | /* | |
1547 | * Port stats maintained per queue of the port. | |
1548 | */ | |
1549 | struct queue_port_stats { | |
1550 | u64 tso; | |
1551 | u64 tx_csum; | |
1552 | u64 rx_csum; | |
1553 | u64 vlan_ex; | |
1554 | u64 vlan_ins; | |
f12fe353 CL |
1555 | u64 lro_pkts; |
1556 | u64 lro_merged; | |
be839e39 CL |
1557 | }; |
1558 | ||
1559 | /* | |
1560 | * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that | |
1561 | * these need to match the order of statistics returned by | |
1562 | * t4vf_get_port_stats(). | |
1563 | */ | |
1564 | static const char stats_strings[][ETH_GSTRING_LEN] = { | |
1565 | /* | |
1566 | * These must match the layout of the t4vf_port_stats structure. | |
1567 | */ | |
1568 | "TxBroadcastBytes ", | |
1569 | "TxBroadcastFrames ", | |
1570 | "TxMulticastBytes ", | |
1571 | "TxMulticastFrames ", | |
1572 | "TxUnicastBytes ", | |
1573 | "TxUnicastFrames ", | |
1574 | "TxDroppedFrames ", | |
1575 | "TxOffloadBytes ", | |
1576 | "TxOffloadFrames ", | |
1577 | "RxBroadcastBytes ", | |
1578 | "RxBroadcastFrames ", | |
1579 | "RxMulticastBytes ", | |
1580 | "RxMulticastFrames ", | |
1581 | "RxUnicastBytes ", | |
1582 | "RxUnicastFrames ", | |
1583 | "RxErrorFrames ", | |
1584 | ||
1585 | /* | |
1586 | * These are accumulated per-queue statistics and must match the | |
1587 | * order of the fields in the queue_port_stats structure. | |
1588 | */ | |
1589 | "TSO ", | |
1590 | "TxCsumOffload ", | |
1591 | "RxCsumGood ", | |
1592 | "VLANextractions ", | |
1593 | "VLANinsertions ", | |
f12fe353 CL |
1594 | "GROPackets ", |
1595 | "GROMerged ", | |
be839e39 CL |
1596 | }; |
1597 | ||
1598 | /* | |
1599 | * Return the number of statistics in the specified statistics set. | |
1600 | */ | |
1601 | static int cxgb4vf_get_sset_count(struct net_device *dev, int sset) | |
1602 | { | |
1603 | switch (sset) { | |
1604 | case ETH_SS_STATS: | |
1605 | return ARRAY_SIZE(stats_strings); | |
1606 | default: | |
1607 | return -EOPNOTSUPP; | |
1608 | } | |
1609 | /*NOTREACHED*/ | |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * Return the strings for the specified statistics set. | |
1614 | */ | |
1615 | static void cxgb4vf_get_strings(struct net_device *dev, | |
1616 | u32 sset, | |
1617 | u8 *data) | |
1618 | { | |
1619 | switch (sset) { | |
1620 | case ETH_SS_STATS: | |
1621 | memcpy(data, stats_strings, sizeof(stats_strings)); | |
1622 | break; | |
1623 | } | |
1624 | } | |
1625 | ||
1626 | /* | |
1627 | * Small utility routine to accumulate queue statistics across the queues of | |
1628 | * a "port". | |
1629 | */ | |
1630 | static void collect_sge_port_stats(const struct adapter *adapter, | |
1631 | const struct port_info *pi, | |
1632 | struct queue_port_stats *stats) | |
1633 | { | |
1634 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset]; | |
1635 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset]; | |
1636 | int qs; | |
1637 | ||
1638 | memset(stats, 0, sizeof(*stats)); | |
1639 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { | |
1640 | stats->tso += txq->tso; | |
1641 | stats->tx_csum += txq->tx_cso; | |
1642 | stats->rx_csum += rxq->stats.rx_cso; | |
1643 | stats->vlan_ex += rxq->stats.vlan_ex; | |
1644 | stats->vlan_ins += txq->vlan_ins; | |
f12fe353 CL |
1645 | stats->lro_pkts += rxq->stats.lro_pkts; |
1646 | stats->lro_merged += rxq->stats.lro_merged; | |
be839e39 CL |
1647 | } |
1648 | } | |
1649 | ||
1650 | /* | |
1651 | * Return the ETH_SS_STATS statistics set. | |
1652 | */ | |
1653 | static void cxgb4vf_get_ethtool_stats(struct net_device *dev, | |
1654 | struct ethtool_stats *stats, | |
1655 | u64 *data) | |
1656 | { | |
1657 | struct port_info *pi = netdev2pinfo(dev); | |
1658 | struct adapter *adapter = pi->adapter; | |
1659 | int err = t4vf_get_port_stats(adapter, pi->pidx, | |
1660 | (struct t4vf_port_stats *)data); | |
1661 | if (err) | |
1662 | memset(data, 0, sizeof(struct t4vf_port_stats)); | |
1663 | ||
1664 | data += sizeof(struct t4vf_port_stats) / sizeof(u64); | |
1665 | collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data); | |
1666 | } | |
1667 | ||
1668 | /* | |
1669 | * Return the size of our register map. | |
1670 | */ | |
1671 | static int cxgb4vf_get_regs_len(struct net_device *dev) | |
1672 | { | |
1673 | return T4VF_REGMAP_SIZE; | |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * Dump a block of registers, start to end inclusive, into a buffer. | |
1678 | */ | |
1679 | static void reg_block_dump(struct adapter *adapter, void *regbuf, | |
1680 | unsigned int start, unsigned int end) | |
1681 | { | |
1682 | u32 *bp = regbuf + start - T4VF_REGMAP_START; | |
1683 | ||
1684 | for ( ; start <= end; start += sizeof(u32)) { | |
1685 | /* | |
1686 | * Avoid reading the Mailbox Control register since that | |
1687 | * can trigger a Mailbox Ownership Arbitration cycle and | |
1688 | * interfere with communication with the firmware. | |
1689 | */ | |
1690 | if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL) | |
1691 | *bp++ = 0xffff; | |
1692 | else | |
1693 | *bp++ = t4_read_reg(adapter, start); | |
1694 | } | |
1695 | } | |
1696 | ||
1697 | /* | |
1698 | * Copy our entire register map into the provided buffer. | |
1699 | */ | |
1700 | static void cxgb4vf_get_regs(struct net_device *dev, | |
1701 | struct ethtool_regs *regs, | |
1702 | void *regbuf) | |
1703 | { | |
1704 | struct adapter *adapter = netdev2adap(dev); | |
1705 | ||
1706 | regs->version = mk_adap_vers(adapter); | |
1707 | ||
1708 | /* | |
1709 | * Fill in register buffer with our register map. | |
1710 | */ | |
1711 | memset(regbuf, 0, T4VF_REGMAP_SIZE); | |
1712 | ||
1713 | reg_block_dump(adapter, regbuf, | |
1714 | T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST, | |
1715 | T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST); | |
1716 | reg_block_dump(adapter, regbuf, | |
1717 | T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST, | |
1718 | T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST); | |
70ee3666 HS |
1719 | |
1720 | /* T5 adds new registers in the PL Register map. | |
1721 | */ | |
be839e39 CL |
1722 | reg_block_dump(adapter, regbuf, |
1723 | T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST, | |
70ee3666 | 1724 | T4VF_PL_BASE_ADDR + (is_t4(adapter->params.chip) |
0d804338 | 1725 | ? PL_VF_WHOAMI_A : PL_VF_REVISION_A)); |
be839e39 CL |
1726 | reg_block_dump(adapter, regbuf, |
1727 | T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST, | |
1728 | T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST); | |
1729 | ||
1730 | reg_block_dump(adapter, regbuf, | |
1731 | T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST, | |
1732 | T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST); | |
1733 | } | |
1734 | ||
1735 | /* | |
1736 | * Report current Wake On LAN settings. | |
1737 | */ | |
1738 | static void cxgb4vf_get_wol(struct net_device *dev, | |
1739 | struct ethtool_wolinfo *wol) | |
1740 | { | |
1741 | wol->supported = 0; | |
1742 | wol->wolopts = 0; | |
1743 | memset(&wol->sopass, 0, sizeof(wol->sopass)); | |
1744 | } | |
1745 | ||
410989f6 CL |
1746 | /* |
1747 | * TCP Segmentation Offload flags which we support. | |
1748 | */ | |
1749 | #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN) | |
1750 | ||
9b07be4b | 1751 | static const struct ethtool_ops cxgb4vf_ethtool_ops = { |
eb97ad99 | 1752 | .get_link_ksettings = cxgb4vf_get_link_ksettings, |
be839e39 CL |
1753 | .get_drvinfo = cxgb4vf_get_drvinfo, |
1754 | .get_msglevel = cxgb4vf_get_msglevel, | |
1755 | .set_msglevel = cxgb4vf_set_msglevel, | |
1756 | .get_ringparam = cxgb4vf_get_ringparam, | |
1757 | .set_ringparam = cxgb4vf_set_ringparam, | |
1758 | .get_coalesce = cxgb4vf_get_coalesce, | |
1759 | .set_coalesce = cxgb4vf_set_coalesce, | |
1760 | .get_pauseparam = cxgb4vf_get_pauseparam, | |
be839e39 CL |
1761 | .get_link = ethtool_op_get_link, |
1762 | .get_strings = cxgb4vf_get_strings, | |
857a3d0f | 1763 | .set_phys_id = cxgb4vf_phys_id, |
be839e39 CL |
1764 | .get_sset_count = cxgb4vf_get_sset_count, |
1765 | .get_ethtool_stats = cxgb4vf_get_ethtool_stats, | |
1766 | .get_regs_len = cxgb4vf_get_regs_len, | |
1767 | .get_regs = cxgb4vf_get_regs, | |
1768 | .get_wol = cxgb4vf_get_wol, | |
be839e39 CL |
1769 | }; |
1770 | ||
1771 | /* | |
1772 | * /sys/kernel/debug/cxgb4vf support code and data. | |
1773 | * ================================================ | |
1774 | */ | |
1775 | ||
ae7b7576 HS |
1776 | /* |
1777 | * Show Firmware Mailbox Command/Reply Log | |
1778 | * | |
1779 | * Note that we don't do any locking when dumping the Firmware Mailbox Log so | |
1780 | * it's possible that we can catch things during a log update and therefore | |
1781 | * see partially corrupted log entries. But i9t's probably Good Enough(tm). | |
1782 | * If we ever decide that we want to make sure that we're dumping a coherent | |
1783 | * log, we'd need to perform locking in the mailbox logging and in | |
1784 | * mboxlog_open() where we'd need to grab the entire mailbox log in one go | |
1785 | * like we do for the Firmware Device Log. But as stated above, meh ... | |
1786 | */ | |
1787 | static int mboxlog_show(struct seq_file *seq, void *v) | |
1788 | { | |
1789 | struct adapter *adapter = seq->private; | |
1790 | struct mbox_cmd_log *log = adapter->mbox_log; | |
1791 | struct mbox_cmd *entry; | |
1792 | int entry_idx, i; | |
1793 | ||
1794 | if (v == SEQ_START_TOKEN) { | |
1795 | seq_printf(seq, | |
1796 | "%10s %15s %5s %5s %s\n", | |
1797 | "Seq#", "Tstamp", "Atime", "Etime", | |
1798 | "Command/Reply"); | |
1799 | return 0; | |
1800 | } | |
1801 | ||
1802 | entry_idx = log->cursor + ((uintptr_t)v - 2); | |
1803 | if (entry_idx >= log->size) | |
1804 | entry_idx -= log->size; | |
1805 | entry = mbox_cmd_log_entry(log, entry_idx); | |
1806 | ||
1807 | /* skip over unused entries */ | |
1808 | if (entry->timestamp == 0) | |
1809 | return 0; | |
1810 | ||
1811 | seq_printf(seq, "%10u %15llu %5d %5d", | |
1812 | entry->seqno, entry->timestamp, | |
1813 | entry->access, entry->execute); | |
1814 | for (i = 0; i < MBOX_LEN / 8; i++) { | |
1815 | u64 flit = entry->cmd[i]; | |
1816 | u32 hi = (u32)(flit >> 32); | |
1817 | u32 lo = (u32)flit; | |
1818 | ||
1819 | seq_printf(seq, " %08x %08x", hi, lo); | |
1820 | } | |
1821 | seq_puts(seq, "\n"); | |
1822 | return 0; | |
1823 | } | |
1824 | ||
1825 | static inline void *mboxlog_get_idx(struct seq_file *seq, loff_t pos) | |
1826 | { | |
1827 | struct adapter *adapter = seq->private; | |
1828 | struct mbox_cmd_log *log = adapter->mbox_log; | |
1829 | ||
1830 | return ((pos <= log->size) ? (void *)(uintptr_t)(pos + 1) : NULL); | |
1831 | } | |
1832 | ||
1833 | static void *mboxlog_start(struct seq_file *seq, loff_t *pos) | |
1834 | { | |
1835 | return *pos ? mboxlog_get_idx(seq, *pos) : SEQ_START_TOKEN; | |
1836 | } | |
1837 | ||
1838 | static void *mboxlog_next(struct seq_file *seq, void *v, loff_t *pos) | |
1839 | { | |
1840 | ++*pos; | |
1841 | return mboxlog_get_idx(seq, *pos); | |
1842 | } | |
1843 | ||
1844 | static void mboxlog_stop(struct seq_file *seq, void *v) | |
1845 | { | |
1846 | } | |
1847 | ||
1848 | static const struct seq_operations mboxlog_seq_ops = { | |
1849 | .start = mboxlog_start, | |
1850 | .next = mboxlog_next, | |
1851 | .stop = mboxlog_stop, | |
1852 | .show = mboxlog_show | |
1853 | }; | |
1854 | ||
1855 | static int mboxlog_open(struct inode *inode, struct file *file) | |
1856 | { | |
1857 | int res = seq_open(file, &mboxlog_seq_ops); | |
1858 | ||
1859 | if (!res) { | |
1860 | struct seq_file *seq = file->private_data; | |
1861 | ||
1862 | seq->private = inode->i_private; | |
1863 | } | |
1864 | return res; | |
1865 | } | |
1866 | ||
1867 | static const struct file_operations mboxlog_fops = { | |
1868 | .owner = THIS_MODULE, | |
1869 | .open = mboxlog_open, | |
1870 | .read = seq_read, | |
1871 | .llseek = seq_lseek, | |
1872 | .release = seq_release, | |
1873 | }; | |
1874 | ||
be839e39 CL |
1875 | /* |
1876 | * Show SGE Queue Set information. We display QPL Queues Sets per line. | |
1877 | */ | |
1878 | #define QPL 4 | |
1879 | ||
1880 | static int sge_qinfo_show(struct seq_file *seq, void *v) | |
1881 | { | |
1882 | struct adapter *adapter = seq->private; | |
1883 | int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL); | |
1884 | int qs, r = (uintptr_t)v - 1; | |
1885 | ||
1886 | if (r) | |
1887 | seq_putc(seq, '\n'); | |
1888 | ||
1889 | #define S3(fmt_spec, s, v) \ | |
1890 | do {\ | |
1891 | seq_printf(seq, "%-12s", s); \ | |
1892 | for (qs = 0; qs < n; ++qs) \ | |
1893 | seq_printf(seq, " %16" fmt_spec, v); \ | |
1894 | seq_putc(seq, '\n'); \ | |
1895 | } while (0) | |
1896 | #define S(s, v) S3("s", s, v) | |
1897 | #define T(s, v) S3("u", s, txq[qs].v) | |
1898 | #define R(s, v) S3("u", s, rxq[qs].v) | |
1899 | ||
1900 | if (r < eth_entries) { | |
1901 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL]; | |
1902 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL]; | |
1903 | int n = min(QPL, adapter->sge.ethqsets - QPL * r); | |
1904 | ||
1905 | S("QType:", "Ethernet"); | |
1906 | S("Interface:", | |
1907 | (rxq[qs].rspq.netdev | |
1908 | ? rxq[qs].rspq.netdev->name | |
1909 | : "N/A")); | |
1910 | S3("d", "Port:", | |
1911 | (rxq[qs].rspq.netdev | |
1912 | ? ((struct port_info *) | |
1913 | netdev_priv(rxq[qs].rspq.netdev))->port_id | |
1914 | : -1)); | |
1915 | T("TxQ ID:", q.abs_id); | |
1916 | T("TxQ size:", q.size); | |
1917 | T("TxQ inuse:", q.in_use); | |
1918 | T("TxQ PIdx:", q.pidx); | |
1919 | T("TxQ CIdx:", q.cidx); | |
1920 | R("RspQ ID:", rspq.abs_id); | |
1921 | R("RspQ size:", rspq.size); | |
1922 | R("RspQE size:", rspq.iqe_len); | |
1923 | S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq)); | |
1924 | S3("u", "Intr pktcnt:", | |
1925 | adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]); | |
1926 | R("RspQ CIdx:", rspq.cidx); | |
1927 | R("RspQ Gen:", rspq.gen); | |
1928 | R("FL ID:", fl.abs_id); | |
1929 | R("FL size:", fl.size - MIN_FL_RESID); | |
1930 | R("FL avail:", fl.avail); | |
1931 | R("FL PIdx:", fl.pidx); | |
1932 | R("FL CIdx:", fl.cidx); | |
1933 | return 0; | |
1934 | } | |
1935 | ||
1936 | r -= eth_entries; | |
1937 | if (r == 0) { | |
1938 | const struct sge_rspq *evtq = &adapter->sge.fw_evtq; | |
1939 | ||
1940 | seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue"); | |
1941 | seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id); | |
1942 | seq_printf(seq, "%-12s %16u\n", "Intr delay:", | |
1943 | qtimer_val(adapter, evtq)); | |
1944 | seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:", | |
1945 | adapter->sge.counter_val[evtq->pktcnt_idx]); | |
1946 | seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx); | |
1947 | seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen); | |
1948 | } else if (r == 1) { | |
1949 | const struct sge_rspq *intrq = &adapter->sge.intrq; | |
1950 | ||
1951 | seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue"); | |
1952 | seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id); | |
1953 | seq_printf(seq, "%-12s %16u\n", "Intr delay:", | |
1954 | qtimer_val(adapter, intrq)); | |
1955 | seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:", | |
1956 | adapter->sge.counter_val[intrq->pktcnt_idx]); | |
1957 | seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx); | |
1958 | seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen); | |
1959 | } | |
1960 | ||
1961 | #undef R | |
1962 | #undef T | |
1963 | #undef S | |
1964 | #undef S3 | |
1965 | ||
1966 | return 0; | |
1967 | } | |
1968 | ||
1969 | /* | |
1970 | * Return the number of "entries" in our "file". We group the multi-Queue | |
1971 | * sections with QPL Queue Sets per "entry". The sections of the output are: | |
1972 | * | |
1973 | * Ethernet RX/TX Queue Sets | |
1974 | * Firmware Event Queue | |
1975 | * Forwarded Interrupt Queue (if in MSI mode) | |
1976 | */ | |
1977 | static int sge_queue_entries(const struct adapter *adapter) | |
1978 | { | |
1979 | return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 + | |
1980 | ((adapter->flags & USING_MSI) != 0); | |
1981 | } | |
1982 | ||
1983 | static void *sge_queue_start(struct seq_file *seq, loff_t *pos) | |
1984 | { | |
1985 | int entries = sge_queue_entries(seq->private); | |
1986 | ||
1987 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
1988 | } | |
1989 | ||
1990 | static void sge_queue_stop(struct seq_file *seq, void *v) | |
1991 | { | |
1992 | } | |
1993 | ||
1994 | static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos) | |
1995 | { | |
1996 | int entries = sge_queue_entries(seq->private); | |
1997 | ||
1998 | ++*pos; | |
1999 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
2000 | } | |
2001 | ||
2002 | static const struct seq_operations sge_qinfo_seq_ops = { | |
2003 | .start = sge_queue_start, | |
2004 | .next = sge_queue_next, | |
2005 | .stop = sge_queue_stop, | |
2006 | .show = sge_qinfo_show | |
2007 | }; | |
2008 | ||
2009 | static int sge_qinfo_open(struct inode *inode, struct file *file) | |
2010 | { | |
2011 | int res = seq_open(file, &sge_qinfo_seq_ops); | |
2012 | ||
2013 | if (!res) { | |
2014 | struct seq_file *seq = file->private_data; | |
2015 | seq->private = inode->i_private; | |
2016 | } | |
2017 | return res; | |
2018 | } | |
2019 | ||
2020 | static const struct file_operations sge_qinfo_debugfs_fops = { | |
2021 | .owner = THIS_MODULE, | |
2022 | .open = sge_qinfo_open, | |
2023 | .read = seq_read, | |
2024 | .llseek = seq_lseek, | |
2025 | .release = seq_release, | |
2026 | }; | |
2027 | ||
2028 | /* | |
2029 | * Show SGE Queue Set statistics. We display QPL Queues Sets per line. | |
2030 | */ | |
2031 | #define QPL 4 | |
2032 | ||
2033 | static int sge_qstats_show(struct seq_file *seq, void *v) | |
2034 | { | |
2035 | struct adapter *adapter = seq->private; | |
2036 | int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL); | |
2037 | int qs, r = (uintptr_t)v - 1; | |
2038 | ||
2039 | if (r) | |
2040 | seq_putc(seq, '\n'); | |
2041 | ||
2042 | #define S3(fmt, s, v) \ | |
2043 | do { \ | |
2044 | seq_printf(seq, "%-16s", s); \ | |
2045 | for (qs = 0; qs < n; ++qs) \ | |
2046 | seq_printf(seq, " %8" fmt, v); \ | |
2047 | seq_putc(seq, '\n'); \ | |
2048 | } while (0) | |
2049 | #define S(s, v) S3("s", s, v) | |
2050 | ||
2051 | #define T3(fmt, s, v) S3(fmt, s, txq[qs].v) | |
2052 | #define T(s, v) T3("lu", s, v) | |
2053 | ||
2054 | #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v) | |
2055 | #define R(s, v) R3("lu", s, v) | |
2056 | ||
2057 | if (r < eth_entries) { | |
2058 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL]; | |
2059 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL]; | |
2060 | int n = min(QPL, adapter->sge.ethqsets - QPL * r); | |
2061 | ||
2062 | S("QType:", "Ethernet"); | |
2063 | S("Interface:", | |
2064 | (rxq[qs].rspq.netdev | |
2065 | ? rxq[qs].rspq.netdev->name | |
2066 | : "N/A")); | |
68dc9d36 | 2067 | R3("u", "RspQNullInts:", rspq.unhandled_irqs); |
be839e39 CL |
2068 | R("RxPackets:", stats.pkts); |
2069 | R("RxCSO:", stats.rx_cso); | |
2070 | R("VLANxtract:", stats.vlan_ex); | |
2071 | R("LROmerged:", stats.lro_merged); | |
2072 | R("LROpackets:", stats.lro_pkts); | |
2073 | R("RxDrops:", stats.rx_drops); | |
2074 | T("TSO:", tso); | |
2075 | T("TxCSO:", tx_cso); | |
2076 | T("VLANins:", vlan_ins); | |
2077 | T("TxQFull:", q.stops); | |
2078 | T("TxQRestarts:", q.restarts); | |
2079 | T("TxMapErr:", mapping_err); | |
2080 | R("FLAllocErr:", fl.alloc_failed); | |
2081 | R("FLLrgAlcErr:", fl.large_alloc_failed); | |
2082 | R("FLStarving:", fl.starving); | |
2083 | return 0; | |
2084 | } | |
2085 | ||
2086 | r -= eth_entries; | |
2087 | if (r == 0) { | |
2088 | const struct sge_rspq *evtq = &adapter->sge.fw_evtq; | |
2089 | ||
2090 | seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue"); | |
68dc9d36 CL |
2091 | seq_printf(seq, "%-16s %8u\n", "RspQNullInts:", |
2092 | evtq->unhandled_irqs); | |
be839e39 CL |
2093 | seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx); |
2094 | seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen); | |
2095 | } else if (r == 1) { | |
2096 | const struct sge_rspq *intrq = &adapter->sge.intrq; | |
2097 | ||
2098 | seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue"); | |
68dc9d36 CL |
2099 | seq_printf(seq, "%-16s %8u\n", "RspQNullInts:", |
2100 | intrq->unhandled_irqs); | |
be839e39 CL |
2101 | seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx); |
2102 | seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen); | |
2103 | } | |
2104 | ||
2105 | #undef R | |
2106 | #undef T | |
2107 | #undef S | |
2108 | #undef R3 | |
2109 | #undef T3 | |
2110 | #undef S3 | |
2111 | ||
2112 | return 0; | |
2113 | } | |
2114 | ||
2115 | /* | |
2116 | * Return the number of "entries" in our "file". We group the multi-Queue | |
2117 | * sections with QPL Queue Sets per "entry". The sections of the output are: | |
2118 | * | |
2119 | * Ethernet RX/TX Queue Sets | |
2120 | * Firmware Event Queue | |
2121 | * Forwarded Interrupt Queue (if in MSI mode) | |
2122 | */ | |
2123 | static int sge_qstats_entries(const struct adapter *adapter) | |
2124 | { | |
2125 | return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 + | |
2126 | ((adapter->flags & USING_MSI) != 0); | |
2127 | } | |
2128 | ||
2129 | static void *sge_qstats_start(struct seq_file *seq, loff_t *pos) | |
2130 | { | |
2131 | int entries = sge_qstats_entries(seq->private); | |
2132 | ||
2133 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
2134 | } | |
2135 | ||
2136 | static void sge_qstats_stop(struct seq_file *seq, void *v) | |
2137 | { | |
2138 | } | |
2139 | ||
2140 | static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos) | |
2141 | { | |
2142 | int entries = sge_qstats_entries(seq->private); | |
2143 | ||
2144 | (*pos)++; | |
2145 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
2146 | } | |
2147 | ||
2148 | static const struct seq_operations sge_qstats_seq_ops = { | |
2149 | .start = sge_qstats_start, | |
2150 | .next = sge_qstats_next, | |
2151 | .stop = sge_qstats_stop, | |
2152 | .show = sge_qstats_show | |
2153 | }; | |
2154 | ||
2155 | static int sge_qstats_open(struct inode *inode, struct file *file) | |
2156 | { | |
2157 | int res = seq_open(file, &sge_qstats_seq_ops); | |
2158 | ||
2159 | if (res == 0) { | |
2160 | struct seq_file *seq = file->private_data; | |
2161 | seq->private = inode->i_private; | |
2162 | } | |
2163 | return res; | |
2164 | } | |
2165 | ||
2166 | static const struct file_operations sge_qstats_proc_fops = { | |
2167 | .owner = THIS_MODULE, | |
2168 | .open = sge_qstats_open, | |
2169 | .read = seq_read, | |
2170 | .llseek = seq_lseek, | |
2171 | .release = seq_release, | |
2172 | }; | |
2173 | ||
2174 | /* | |
2175 | * Show PCI-E SR-IOV Virtual Function Resource Limits. | |
2176 | */ | |
2177 | static int resources_show(struct seq_file *seq, void *v) | |
2178 | { | |
2179 | struct adapter *adapter = seq->private; | |
2180 | struct vf_resources *vfres = &adapter->params.vfres; | |
2181 | ||
2182 | #define S(desc, fmt, var) \ | |
2183 | seq_printf(seq, "%-60s " fmt "\n", \ | |
2184 | desc " (" #var "):", vfres->var) | |
2185 | ||
2186 | S("Virtual Interfaces", "%d", nvi); | |
2187 | S("Egress Queues", "%d", neq); | |
2188 | S("Ethernet Control", "%d", nethctrl); | |
2189 | S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint); | |
2190 | S("Ingress Queues", "%d", niq); | |
2191 | S("Traffic Class", "%d", tc); | |
2192 | S("Port Access Rights Mask", "%#x", pmask); | |
2193 | S("MAC Address Filters", "%d", nexactf); | |
2194 | S("Firmware Command Read Capabilities", "%#x", r_caps); | |
2195 | S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps); | |
2196 | ||
2197 | #undef S | |
2198 | ||
2199 | return 0; | |
2200 | } | |
2201 | ||
2202 | static int resources_open(struct inode *inode, struct file *file) | |
2203 | { | |
2204 | return single_open(file, resources_show, inode->i_private); | |
2205 | } | |
2206 | ||
2207 | static const struct file_operations resources_proc_fops = { | |
2208 | .owner = THIS_MODULE, | |
2209 | .open = resources_open, | |
2210 | .read = seq_read, | |
2211 | .llseek = seq_lseek, | |
2212 | .release = single_release, | |
2213 | }; | |
2214 | ||
2215 | /* | |
2216 | * Show Virtual Interfaces. | |
2217 | */ | |
2218 | static int interfaces_show(struct seq_file *seq, void *v) | |
2219 | { | |
2220 | if (v == SEQ_START_TOKEN) { | |
2221 | seq_puts(seq, "Interface Port VIID\n"); | |
2222 | } else { | |
2223 | struct adapter *adapter = seq->private; | |
2224 | int pidx = (uintptr_t)v - 2; | |
2225 | struct net_device *dev = adapter->port[pidx]; | |
2226 | struct port_info *pi = netdev_priv(dev); | |
2227 | ||
2228 | seq_printf(seq, "%9s %4d %#5x\n", | |
2229 | dev->name, pi->port_id, pi->viid); | |
2230 | } | |
2231 | return 0; | |
2232 | } | |
2233 | ||
2234 | static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos) | |
2235 | { | |
2236 | return pos <= adapter->params.nports | |
2237 | ? (void *)(uintptr_t)(pos + 1) | |
2238 | : NULL; | |
2239 | } | |
2240 | ||
2241 | static void *interfaces_start(struct seq_file *seq, loff_t *pos) | |
2242 | { | |
2243 | return *pos | |
2244 | ? interfaces_get_idx(seq->private, *pos) | |
2245 | : SEQ_START_TOKEN; | |
2246 | } | |
2247 | ||
2248 | static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos) | |
2249 | { | |
2250 | (*pos)++; | |
2251 | return interfaces_get_idx(seq->private, *pos); | |
2252 | } | |
2253 | ||
2254 | static void interfaces_stop(struct seq_file *seq, void *v) | |
2255 | { | |
2256 | } | |
2257 | ||
2258 | static const struct seq_operations interfaces_seq_ops = { | |
2259 | .start = interfaces_start, | |
2260 | .next = interfaces_next, | |
2261 | .stop = interfaces_stop, | |
2262 | .show = interfaces_show | |
2263 | }; | |
2264 | ||
2265 | static int interfaces_open(struct inode *inode, struct file *file) | |
2266 | { | |
2267 | int res = seq_open(file, &interfaces_seq_ops); | |
2268 | ||
2269 | if (res == 0) { | |
2270 | struct seq_file *seq = file->private_data; | |
2271 | seq->private = inode->i_private; | |
2272 | } | |
2273 | return res; | |
2274 | } | |
2275 | ||
2276 | static const struct file_operations interfaces_proc_fops = { | |
2277 | .owner = THIS_MODULE, | |
2278 | .open = interfaces_open, | |
2279 | .read = seq_read, | |
2280 | .llseek = seq_lseek, | |
2281 | .release = seq_release, | |
2282 | }; | |
2283 | ||
2284 | /* | |
2285 | * /sys/kernel/debugfs/cxgb4vf/ files list. | |
2286 | */ | |
2287 | struct cxgb4vf_debugfs_entry { | |
2288 | const char *name; /* name of debugfs node */ | |
f4ae40a6 | 2289 | umode_t mode; /* file system mode */ |
be839e39 CL |
2290 | const struct file_operations *fops; |
2291 | }; | |
2292 | ||
2293 | static struct cxgb4vf_debugfs_entry debugfs_files[] = { | |
ae7b7576 | 2294 | { "mboxlog", S_IRUGO, &mboxlog_fops }, |
be839e39 CL |
2295 | { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops }, |
2296 | { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops }, | |
2297 | { "resources", S_IRUGO, &resources_proc_fops }, | |
2298 | { "interfaces", S_IRUGO, &interfaces_proc_fops }, | |
2299 | }; | |
2300 | ||
2301 | /* | |
2302 | * Module and device initialization and cleanup code. | |
2303 | * ================================================== | |
2304 | */ | |
2305 | ||
2306 | /* | |
2307 | * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the | |
2308 | * directory (debugfs_root) has already been set up. | |
2309 | */ | |
d289f864 | 2310 | static int setup_debugfs(struct adapter *adapter) |
be839e39 CL |
2311 | { |
2312 | int i; | |
2313 | ||
843635e0 | 2314 | BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root)); |
be839e39 CL |
2315 | |
2316 | /* | |
2317 | * Debugfs support is best effort. | |
2318 | */ | |
2319 | for (i = 0; i < ARRAY_SIZE(debugfs_files); i++) | |
2320 | (void)debugfs_create_file(debugfs_files[i].name, | |
2321 | debugfs_files[i].mode, | |
2322 | adapter->debugfs_root, | |
2323 | (void *)adapter, | |
2324 | debugfs_files[i].fops); | |
2325 | ||
2326 | return 0; | |
2327 | } | |
2328 | ||
2329 | /* | |
2330 | * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave | |
2331 | * it to our caller to tear down the directory (debugfs_root). | |
2332 | */ | |
4204875d | 2333 | static void cleanup_debugfs(struct adapter *adapter) |
be839e39 | 2334 | { |
843635e0 | 2335 | BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root)); |
be839e39 CL |
2336 | |
2337 | /* | |
2338 | * Unlike our sister routine cleanup_proc(), we don't need to remove | |
2339 | * individual entries because a call will be made to | |
2340 | * debugfs_remove_recursive(). We just need to clean up any ancillary | |
2341 | * persistent state. | |
2342 | */ | |
2343 | /* nothing to do */ | |
2344 | } | |
2345 | ||
495c22bb HS |
2346 | /* Figure out how many Ports and Queue Sets we can support. This depends on |
2347 | * knowing our Virtual Function Resources and may be called a second time if | |
2348 | * we fall back from MSI-X to MSI Interrupt Mode. | |
2349 | */ | |
2350 | static void size_nports_qsets(struct adapter *adapter) | |
2351 | { | |
2352 | struct vf_resources *vfres = &adapter->params.vfres; | |
2353 | unsigned int ethqsets, pmask_nports; | |
2354 | ||
2355 | /* The number of "ports" which we support is equal to the number of | |
2356 | * Virtual Interfaces with which we've been provisioned. | |
2357 | */ | |
2358 | adapter->params.nports = vfres->nvi; | |
2359 | if (adapter->params.nports > MAX_NPORTS) { | |
2360 | dev_warn(adapter->pdev_dev, "only using %d of %d maximum" | |
2361 | " allowed virtual interfaces\n", MAX_NPORTS, | |
2362 | adapter->params.nports); | |
2363 | adapter->params.nports = MAX_NPORTS; | |
2364 | } | |
2365 | ||
2366 | /* We may have been provisioned with more VIs than the number of | |
2367 | * ports we're allowed to access (our Port Access Rights Mask). | |
2368 | * This is obviously a configuration conflict but we don't want to | |
2369 | * crash the kernel or anything silly just because of that. | |
2370 | */ | |
2371 | pmask_nports = hweight32(adapter->params.vfres.pmask); | |
2372 | if (pmask_nports < adapter->params.nports) { | |
1a8ff8f5 | 2373 | dev_warn(adapter->pdev_dev, "only using %d of %d provisioned" |
495c22bb HS |
2374 | " virtual interfaces; limited by Port Access Rights" |
2375 | " mask %#x\n", pmask_nports, adapter->params.nports, | |
2376 | adapter->params.vfres.pmask); | |
2377 | adapter->params.nports = pmask_nports; | |
2378 | } | |
2379 | ||
2380 | /* We need to reserve an Ingress Queue for the Asynchronous Firmware | |
2381 | * Event Queue. And if we're using MSI Interrupts, we'll also need to | |
2382 | * reserve an Ingress Queue for a Forwarded Interrupts. | |
2383 | * | |
2384 | * The rest of the FL/Intr-capable ingress queues will be matched up | |
2385 | * one-for-one with Ethernet/Control egress queues in order to form | |
2386 | * "Queue Sets" which will be aportioned between the "ports". For | |
2387 | * each Queue Set, we'll need the ability to allocate two Egress | |
2388 | * Contexts -- one for the Ingress Queue Free List and one for the TX | |
2389 | * Ethernet Queue. | |
2390 | * | |
2391 | * Note that even if we're currently configured to use MSI-X | |
2392 | * Interrupts (module variable msi == MSI_MSIX) we may get downgraded | |
2393 | * to MSI Interrupts if we can't get enough MSI-X Interrupts. If that | |
2394 | * happens we'll need to adjust things later. | |
2395 | */ | |
2396 | ethqsets = vfres->niqflint - 1 - (msi == MSI_MSI); | |
2397 | if (vfres->nethctrl != ethqsets) | |
2398 | ethqsets = min(vfres->nethctrl, ethqsets); | |
2399 | if (vfres->neq < ethqsets*2) | |
2400 | ethqsets = vfres->neq/2; | |
2401 | if (ethqsets > MAX_ETH_QSETS) | |
2402 | ethqsets = MAX_ETH_QSETS; | |
2403 | adapter->sge.max_ethqsets = ethqsets; | |
2404 | ||
2405 | if (adapter->sge.max_ethqsets < adapter->params.nports) { | |
2406 | dev_warn(adapter->pdev_dev, "only using %d of %d available" | |
2407 | " virtual interfaces (too few Queue Sets)\n", | |
2408 | adapter->sge.max_ethqsets, adapter->params.nports); | |
2409 | adapter->params.nports = adapter->sge.max_ethqsets; | |
2410 | } | |
2411 | } | |
2412 | ||
be839e39 CL |
2413 | /* |
2414 | * Perform early "adapter" initialization. This is where we discover what | |
2415 | * adapter parameters we're going to be using and initialize basic adapter | |
2416 | * hardware support. | |
2417 | */ | |
d289f864 | 2418 | static int adap_init0(struct adapter *adapter) |
be839e39 | 2419 | { |
be839e39 CL |
2420 | struct sge_params *sge_params = &adapter->params.sge; |
2421 | struct sge *s = &adapter->sge; | |
be839e39 | 2422 | int err; |
94dace10 | 2423 | u32 param, val = 0; |
be839e39 | 2424 | |
e68e6133 CL |
2425 | /* |
2426 | * Some environments do not properly handle PCIE FLRs -- e.g. in Linux | |
2427 | * 2.6.31 and later we can't call pci_reset_function() in order to | |
2428 | * issue an FLR because of a self- deadlock on the device semaphore. | |
2429 | * Meanwhile, the OS infrastructure doesn't issue FLRs in all the | |
2430 | * cases where they're needed -- for instance, some versions of KVM | |
2431 | * fail to reset "Assigned Devices" when the VM reboots. Therefore we | |
2432 | * use the firmware based reset in order to reset any per function | |
2433 | * state. | |
2434 | */ | |
2435 | err = t4vf_fw_reset(adapter); | |
2436 | if (err < 0) { | |
2437 | dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err); | |
2438 | return err; | |
2439 | } | |
2440 | ||
be839e39 CL |
2441 | /* |
2442 | * Grab basic operational parameters. These will predominantly have | |
2443 | * been set up by the Physical Function Driver or will be hard coded | |
2444 | * into the adapter. We just have to live with them ... Note that | |
2445 | * we _must_ get our VPD parameters before our SGE parameters because | |
2446 | * we need to know the adapter's core clock from the VPD in order to | |
2447 | * properly decode the SGE Timer Values. | |
2448 | */ | |
2449 | err = t4vf_get_dev_params(adapter); | |
2450 | if (err) { | |
2451 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2452 | " device parameters: err=%d\n", err); | |
2453 | return err; | |
2454 | } | |
2455 | err = t4vf_get_vpd_params(adapter); | |
2456 | if (err) { | |
2457 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2458 | " VPD parameters: err=%d\n", err); | |
2459 | return err; | |
2460 | } | |
2461 | err = t4vf_get_sge_params(adapter); | |
2462 | if (err) { | |
2463 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2464 | " SGE parameters: err=%d\n", err); | |
2465 | return err; | |
2466 | } | |
2467 | err = t4vf_get_rss_glb_config(adapter); | |
2468 | if (err) { | |
2469 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2470 | " RSS parameters: err=%d\n", err); | |
2471 | return err; | |
2472 | } | |
2473 | if (adapter->params.rss.mode != | |
2474 | FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) { | |
2475 | dev_err(adapter->pdev_dev, "unable to operate with global RSS" | |
2476 | " mode %d\n", adapter->params.rss.mode); | |
2477 | return -EINVAL; | |
2478 | } | |
2479 | err = t4vf_sge_init(adapter); | |
2480 | if (err) { | |
2481 | dev_err(adapter->pdev_dev, "unable to use adapter parameters:" | |
2482 | " err=%d\n", err); | |
2483 | return err; | |
2484 | } | |
2485 | ||
94dace10 VP |
2486 | /* If we're running on newer firmware, let it know that we're |
2487 | * prepared to deal with encapsulated CPL messages. Older | |
2488 | * firmware won't understand this and we'll just get | |
2489 | * unencapsulated messages ... | |
2490 | */ | |
5167865a HS |
2491 | param = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | |
2492 | FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_CPLFW4MSG_ENCAP); | |
94dace10 VP |
2493 | val = 1; |
2494 | (void) t4vf_set_params(adapter, 1, ¶m, &val); | |
2495 | ||
be839e39 CL |
2496 | /* |
2497 | * Retrieve our RX interrupt holdoff timer values and counter | |
2498 | * threshold values from the SGE parameters. | |
2499 | */ | |
2500 | s->timer_val[0] = core_ticks_to_us(adapter, | |
f061de42 | 2501 | TIMERVALUE0_G(sge_params->sge_timer_value_0_and_1)); |
be839e39 | 2502 | s->timer_val[1] = core_ticks_to_us(adapter, |
f061de42 | 2503 | TIMERVALUE1_G(sge_params->sge_timer_value_0_and_1)); |
be839e39 | 2504 | s->timer_val[2] = core_ticks_to_us(adapter, |
f061de42 | 2505 | TIMERVALUE0_G(sge_params->sge_timer_value_2_and_3)); |
be839e39 | 2506 | s->timer_val[3] = core_ticks_to_us(adapter, |
f061de42 | 2507 | TIMERVALUE1_G(sge_params->sge_timer_value_2_and_3)); |
be839e39 | 2508 | s->timer_val[4] = core_ticks_to_us(adapter, |
f061de42 | 2509 | TIMERVALUE0_G(sge_params->sge_timer_value_4_and_5)); |
be839e39 | 2510 | s->timer_val[5] = core_ticks_to_us(adapter, |
f061de42 | 2511 | TIMERVALUE1_G(sge_params->sge_timer_value_4_and_5)); |
be839e39 | 2512 | |
f612b815 HS |
2513 | s->counter_val[0] = THRESHOLD_0_G(sge_params->sge_ingress_rx_threshold); |
2514 | s->counter_val[1] = THRESHOLD_1_G(sge_params->sge_ingress_rx_threshold); | |
2515 | s->counter_val[2] = THRESHOLD_2_G(sge_params->sge_ingress_rx_threshold); | |
2516 | s->counter_val[3] = THRESHOLD_3_G(sge_params->sge_ingress_rx_threshold); | |
be839e39 CL |
2517 | |
2518 | /* | |
2519 | * Grab our Virtual Interface resource allocation, extract the | |
2520 | * features that we're interested in and do a bit of sanity testing on | |
2521 | * what we discover. | |
2522 | */ | |
2523 | err = t4vf_get_vfres(adapter); | |
2524 | if (err) { | |
2525 | dev_err(adapter->pdev_dev, "unable to get virtual interface" | |
2526 | " resources: err=%d\n", err); | |
2527 | return err; | |
2528 | } | |
2529 | ||
495c22bb | 2530 | /* Check for various parameter sanity issues */ |
28f71c6d HS |
2531 | if (adapter->params.vfres.pmask == 0) { |
2532 | dev_err(adapter->pdev_dev, "no port access configured\n" | |
2533 | "usable!\n"); | |
2534 | return -EINVAL; | |
2535 | } | |
495c22bb | 2536 | if (adapter->params.vfres.nvi == 0) { |
be839e39 CL |
2537 | dev_err(adapter->pdev_dev, "no virtual interfaces configured/" |
2538 | "usable!\n"); | |
2539 | return -EINVAL; | |
2540 | } | |
495c22bb HS |
2541 | |
2542 | /* Initialize nports and max_ethqsets now that we have our Virtual | |
2543 | * Function Resources. | |
2544 | */ | |
2545 | size_nports_qsets(adapter); | |
2546 | ||
be839e39 CL |
2547 | return 0; |
2548 | } | |
2549 | ||
2550 | static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx, | |
2551 | u8 pkt_cnt_idx, unsigned int size, | |
2552 | unsigned int iqe_size) | |
2553 | { | |
1ecc7b7a HS |
2554 | rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) | |
2555 | (pkt_cnt_idx < SGE_NCOUNTERS ? | |
2556 | QINTR_CNT_EN_F : 0)); | |
be839e39 CL |
2557 | rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS |
2558 | ? pkt_cnt_idx | |
2559 | : 0); | |
2560 | rspq->iqe_len = iqe_size; | |
2561 | rspq->size = size; | |
2562 | } | |
2563 | ||
2564 | /* | |
2565 | * Perform default configuration of DMA queues depending on the number and | |
2566 | * type of ports we found and the number of available CPUs. Most settings can | |
2567 | * be modified by the admin via ethtool and cxgbtool prior to the adapter | |
2568 | * being brought up for the first time. | |
2569 | */ | |
d289f864 | 2570 | static void cfg_queues(struct adapter *adapter) |
be839e39 CL |
2571 | { |
2572 | struct sge *s = &adapter->sge; | |
2573 | int q10g, n10g, qidx, pidx, qs; | |
c710245c | 2574 | size_t iqe_size; |
be839e39 CL |
2575 | |
2576 | /* | |
2577 | * We should not be called till we know how many Queue Sets we can | |
2578 | * support. In particular, this means that we need to know what kind | |
2579 | * of interrupts we'll be using ... | |
2580 | */ | |
2581 | BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0); | |
2582 | ||
2583 | /* | |
2584 | * Count the number of 10GbE Virtual Interfaces that we have. | |
2585 | */ | |
2586 | n10g = 0; | |
2587 | for_each_port(adapter, pidx) | |
14b3812f | 2588 | n10g += is_x_10g_port(&adap2pinfo(adapter, pidx)->link_cfg); |
be839e39 CL |
2589 | |
2590 | /* | |
2591 | * We default to 1 queue per non-10G port and up to # of cores queues | |
2592 | * per 10G port. | |
2593 | */ | |
2594 | if (n10g == 0) | |
2595 | q10g = 0; | |
2596 | else { | |
2597 | int n1g = (adapter->params.nports - n10g); | |
2598 | q10g = (adapter->sge.max_ethqsets - n1g) / n10g; | |
2599 | if (q10g > num_online_cpus()) | |
2600 | q10g = num_online_cpus(); | |
2601 | } | |
2602 | ||
2603 | /* | |
2604 | * Allocate the "Queue Sets" to the various Virtual Interfaces. | |
2605 | * The layout will be established in setup_sge_queues() when the | |
2606 | * adapter is brough up for the first time. | |
2607 | */ | |
2608 | qidx = 0; | |
2609 | for_each_port(adapter, pidx) { | |
2610 | struct port_info *pi = adap2pinfo(adapter, pidx); | |
2611 | ||
2612 | pi->first_qset = qidx; | |
897d55df | 2613 | pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1; |
be839e39 CL |
2614 | qidx += pi->nqsets; |
2615 | } | |
2616 | s->ethqsets = qidx; | |
2617 | ||
c710245c CL |
2618 | /* |
2619 | * The Ingress Queue Entry Size for our various Response Queues needs | |
2620 | * to be big enough to accommodate the largest message we can receive | |
2621 | * from the chip/firmware; which is 64 bytes ... | |
2622 | */ | |
2623 | iqe_size = 64; | |
2624 | ||
be839e39 CL |
2625 | /* |
2626 | * Set up default Queue Set parameters ... Start off with the | |
2627 | * shortest interrupt holdoff timer. | |
2628 | */ | |
2629 | for (qs = 0; qs < s->max_ethqsets; qs++) { | |
2630 | struct sge_eth_rxq *rxq = &s->ethrxq[qs]; | |
2631 | struct sge_eth_txq *txq = &s->ethtxq[qs]; | |
2632 | ||
c710245c | 2633 | init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size); |
be839e39 CL |
2634 | rxq->fl.size = 72; |
2635 | txq->q.size = 1024; | |
2636 | } | |
2637 | ||
2638 | /* | |
2639 | * The firmware event queue is used for link state changes and | |
2640 | * notifications of TX DMA completions. | |
2641 | */ | |
c710245c | 2642 | init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size); |
be839e39 CL |
2643 | |
2644 | /* | |
2645 | * The forwarded interrupt queue is used when we're in MSI interrupt | |
2646 | * mode. In this mode all interrupts associated with RX queues will | |
2647 | * be forwarded to a single queue which we'll associate with our MSI | |
2648 | * interrupt vector. The messages dropped in the forwarded interrupt | |
2649 | * queue will indicate which ingress queue needs servicing ... This | |
2650 | * queue needs to be large enough to accommodate all of the ingress | |
2651 | * queues which are forwarding their interrupt (+1 to prevent the PIDX | |
2652 | * from equalling the CIDX if every ingress queue has an outstanding | |
2653 | * interrupt). The queue doesn't need to be any larger because no | |
2654 | * ingress queue will ever have more than one outstanding interrupt at | |
2655 | * any time ... | |
2656 | */ | |
2657 | init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1, | |
c710245c | 2658 | iqe_size); |
be839e39 CL |
2659 | } |
2660 | ||
2661 | /* | |
2662 | * Reduce the number of Ethernet queues across all ports to at most n. | |
2663 | * n provides at least one queue per port. | |
2664 | */ | |
d289f864 | 2665 | static void reduce_ethqs(struct adapter *adapter, int n) |
be839e39 CL |
2666 | { |
2667 | int i; | |
2668 | struct port_info *pi; | |
2669 | ||
2670 | /* | |
2671 | * While we have too many active Ether Queue Sets, interate across the | |
2672 | * "ports" and reduce their individual Queue Set allocations. | |
2673 | */ | |
2674 | BUG_ON(n < adapter->params.nports); | |
2675 | while (n < adapter->sge.ethqsets) | |
2676 | for_each_port(adapter, i) { | |
2677 | pi = adap2pinfo(adapter, i); | |
2678 | if (pi->nqsets > 1) { | |
2679 | pi->nqsets--; | |
2680 | adapter->sge.ethqsets--; | |
2681 | if (adapter->sge.ethqsets <= n) | |
2682 | break; | |
2683 | } | |
2684 | } | |
2685 | ||
2686 | /* | |
2687 | * Reassign the starting Queue Sets for each of the "ports" ... | |
2688 | */ | |
2689 | n = 0; | |
2690 | for_each_port(adapter, i) { | |
2691 | pi = adap2pinfo(adapter, i); | |
2692 | pi->first_qset = n; | |
2693 | n += pi->nqsets; | |
2694 | } | |
2695 | } | |
2696 | ||
2697 | /* | |
2698 | * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally | |
2699 | * we get a separate MSI-X vector for every "Queue Set" plus any extras we | |
2700 | * need. Minimally we need one for every Virtual Interface plus those needed | |
2701 | * for our "extras". Note that this process may lower the maximum number of | |
2702 | * allowed Queue Sets ... | |
2703 | */ | |
d289f864 | 2704 | static int enable_msix(struct adapter *adapter) |
be839e39 | 2705 | { |
bd663689 | 2706 | int i, want, need, nqsets; |
be839e39 CL |
2707 | struct msix_entry entries[MSIX_ENTRIES]; |
2708 | struct sge *s = &adapter->sge; | |
2709 | ||
2710 | for (i = 0; i < MSIX_ENTRIES; ++i) | |
2711 | entries[i].entry = i; | |
2712 | ||
2713 | /* | |
2714 | * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets" | |
2715 | * plus those needed for our "extras" (for example, the firmware | |
2716 | * message queue). We _need_ at least one "Queue Set" per Virtual | |
2717 | * Interface plus those needed for our "extras". So now we get to see | |
2718 | * if the song is right ... | |
2719 | */ | |
2720 | want = s->max_ethqsets + MSIX_EXTRAS; | |
2721 | need = adapter->params.nports + MSIX_EXTRAS; | |
bd663689 AG |
2722 | |
2723 | want = pci_enable_msix_range(adapter->pdev, entries, need, want); | |
2724 | if (want < 0) | |
2725 | return want; | |
2726 | ||
2727 | nqsets = want - MSIX_EXTRAS; | |
2728 | if (nqsets < s->max_ethqsets) { | |
2729 | dev_warn(adapter->pdev_dev, "only enough MSI-X vectors" | |
2730 | " for %d Queue Sets\n", nqsets); | |
2731 | s->max_ethqsets = nqsets; | |
2732 | if (nqsets < s->ethqsets) | |
2733 | reduce_ethqs(adapter, nqsets); | |
be839e39 | 2734 | } |
bd663689 AG |
2735 | for (i = 0; i < want; ++i) |
2736 | adapter->msix_info[i].vec = entries[i].vector; | |
2737 | ||
2738 | return 0; | |
be839e39 CL |
2739 | } |
2740 | ||
be839e39 CL |
2741 | static const struct net_device_ops cxgb4vf_netdev_ops = { |
2742 | .ndo_open = cxgb4vf_open, | |
2743 | .ndo_stop = cxgb4vf_stop, | |
2744 | .ndo_start_xmit = t4vf_eth_xmit, | |
2745 | .ndo_get_stats = cxgb4vf_get_stats, | |
2746 | .ndo_set_rx_mode = cxgb4vf_set_rxmode, | |
2747 | .ndo_set_mac_address = cxgb4vf_set_mac_addr, | |
be839e39 CL |
2748 | .ndo_validate_addr = eth_validate_addr, |
2749 | .ndo_do_ioctl = cxgb4vf_do_ioctl, | |
2750 | .ndo_change_mtu = cxgb4vf_change_mtu, | |
87737663 JP |
2751 | .ndo_fix_features = cxgb4vf_fix_features, |
2752 | .ndo_set_features = cxgb4vf_set_features, | |
be839e39 CL |
2753 | #ifdef CONFIG_NET_POLL_CONTROLLER |
2754 | .ndo_poll_controller = cxgb4vf_poll_controller, | |
2755 | #endif | |
2756 | }; | |
be839e39 CL |
2757 | |
2758 | /* | |
2759 | * "Probe" a device: initialize a device and construct all kernel and driver | |
2760 | * state needed to manage the device. This routine is called "init_one" in | |
2761 | * the PF Driver ... | |
2762 | */ | |
d289f864 | 2763 | static int cxgb4vf_pci_probe(struct pci_dev *pdev, |
1dd06ae8 | 2764 | const struct pci_device_id *ent) |
be839e39 | 2765 | { |
be839e39 CL |
2766 | int pci_using_dac; |
2767 | int err, pidx; | |
2768 | unsigned int pmask; | |
2769 | struct adapter *adapter; | |
2770 | struct port_info *pi; | |
2771 | struct net_device *netdev; | |
858aa65c | 2772 | unsigned int pf; |
be839e39 | 2773 | |
be839e39 CL |
2774 | /* |
2775 | * Print our driver banner the first time we're called to initialize a | |
2776 | * device. | |
2777 | */ | |
428ac43f | 2778 | pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION); |
be839e39 CL |
2779 | |
2780 | /* | |
7a0c2029 | 2781 | * Initialize generic PCI device state. |
be839e39 | 2782 | */ |
7a0c2029 | 2783 | err = pci_enable_device(pdev); |
be839e39 | 2784 | if (err) { |
7a0c2029 | 2785 | dev_err(&pdev->dev, "cannot enable PCI device\n"); |
be839e39 CL |
2786 | return err; |
2787 | } | |
2788 | ||
2789 | /* | |
7a0c2029 KV |
2790 | * Reserve PCI resources for the device. If we can't get them some |
2791 | * other driver may have already claimed the device ... | |
be839e39 | 2792 | */ |
7a0c2029 | 2793 | err = pci_request_regions(pdev, KBUILD_MODNAME); |
be839e39 | 2794 | if (err) { |
7a0c2029 KV |
2795 | dev_err(&pdev->dev, "cannot obtain PCI resources\n"); |
2796 | goto err_disable_device; | |
be839e39 CL |
2797 | } |
2798 | ||
2799 | /* | |
2800 | * Set up our DMA mask: try for 64-bit address masking first and | |
2801 | * fall back to 32-bit if we can't get 64 bits ... | |
2802 | */ | |
2803 | err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); | |
2804 | if (err == 0) { | |
2805 | err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); | |
2806 | if (err) { | |
2807 | dev_err(&pdev->dev, "unable to obtain 64-bit DMA for" | |
2808 | " coherent allocations\n"); | |
7a0c2029 | 2809 | goto err_release_regions; |
be839e39 CL |
2810 | } |
2811 | pci_using_dac = 1; | |
2812 | } else { | |
2813 | err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); | |
2814 | if (err != 0) { | |
2815 | dev_err(&pdev->dev, "no usable DMA configuration\n"); | |
7a0c2029 | 2816 | goto err_release_regions; |
be839e39 CL |
2817 | } |
2818 | pci_using_dac = 0; | |
2819 | } | |
2820 | ||
2821 | /* | |
2822 | * Enable bus mastering for the device ... | |
2823 | */ | |
2824 | pci_set_master(pdev); | |
2825 | ||
2826 | /* | |
2827 | * Allocate our adapter data structure and attach it to the device. | |
2828 | */ | |
2829 | adapter = kzalloc(sizeof(*adapter), GFP_KERNEL); | |
2830 | if (!adapter) { | |
2831 | err = -ENOMEM; | |
7a0c2029 | 2832 | goto err_release_regions; |
be839e39 CL |
2833 | } |
2834 | pci_set_drvdata(pdev, adapter); | |
2835 | adapter->pdev = pdev; | |
2836 | adapter->pdev_dev = &pdev->dev; | |
2837 | ||
ae7b7576 HS |
2838 | adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) + |
2839 | (sizeof(struct mbox_cmd) * | |
2840 | T4VF_OS_LOG_MBOX_CMDS), | |
2841 | GFP_KERNEL); | |
2842 | if (!adapter->mbox_log) { | |
2843 | err = -ENOMEM; | |
2844 | goto err_free_adapter; | |
2845 | } | |
2846 | adapter->mbox_log->size = T4VF_OS_LOG_MBOX_CMDS; | |
2847 | ||
be839e39 CL |
2848 | /* |
2849 | * Initialize SMP data synchronization resources. | |
2850 | */ | |
2851 | spin_lock_init(&adapter->stats_lock); | |
b38066da HS |
2852 | spin_lock_init(&adapter->mbox_lock); |
2853 | INIT_LIST_HEAD(&adapter->mlist.list); | |
be839e39 CL |
2854 | |
2855 | /* | |
2856 | * Map our I/O registers in BAR0. | |
2857 | */ | |
2858 | adapter->regs = pci_ioremap_bar(pdev, 0); | |
2859 | if (!adapter->regs) { | |
2860 | dev_err(&pdev->dev, "cannot map device registers\n"); | |
2861 | err = -ENOMEM; | |
2862 | goto err_free_adapter; | |
2863 | } | |
2864 | ||
e0a8b34a HS |
2865 | /* Wait for the device to become ready before proceeding ... |
2866 | */ | |
2867 | err = t4vf_prep_adapter(adapter); | |
2868 | if (err) { | |
2869 | dev_err(adapter->pdev_dev, "device didn't become ready:" | |
2870 | " err=%d\n", err); | |
2871 | goto err_unmap_bar0; | |
2872 | } | |
2873 | ||
2874 | /* For T5 and later we want to use the new BAR-based User Doorbells, | |
2875 | * so we need to map BAR2 here ... | |
2876 | */ | |
2877 | if (!is_t4(adapter->params.chip)) { | |
2878 | adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2), | |
2879 | pci_resource_len(pdev, 2)); | |
2880 | if (!adapter->bar2) { | |
2881 | dev_err(adapter->pdev_dev, "cannot map BAR2 doorbells\n"); | |
2882 | err = -ENOMEM; | |
2883 | goto err_unmap_bar0; | |
2884 | } | |
2885 | } | |
be839e39 CL |
2886 | /* |
2887 | * Initialize adapter level features. | |
2888 | */ | |
2889 | adapter->name = pci_name(pdev); | |
ea1e76f7 | 2890 | adapter->msg_enable = DFLT_MSG_ENABLE; |
b629276d CL |
2891 | |
2892 | /* If possible, we use PCIe Relaxed Ordering Attribute to deliver | |
2893 | * Ingress Packet Data to Free List Buffers in order to allow for | |
2894 | * chipset performance optimizations between the Root Complex and | |
2895 | * Memory Controllers. (Messages to the associated Ingress Queue | |
2896 | * notifying new Packet Placement in the Free Lists Buffers will be | |
2897 | * send without the Relaxed Ordering Attribute thus guaranteeing that | |
2898 | * all preceding PCIe Transaction Layer Packets will be processed | |
2899 | * first.) But some Root Complexes have various issues with Upstream | |
2900 | * Transaction Layer Packets with the Relaxed Ordering Attribute set. | |
2901 | * The PCIe devices which under the Root Complexes will be cleared the | |
2902 | * Relaxed Ordering bit in the configuration space, So we check our | |
2903 | * PCIe configuration space to see if it's flagged with advice against | |
2904 | * using Relaxed Ordering. | |
2905 | */ | |
2906 | if (!pcie_relaxed_ordering_enabled(pdev)) | |
2907 | adapter->flags |= ROOT_NO_RELAXED_ORDERING; | |
2908 | ||
be839e39 CL |
2909 | err = adap_init0(adapter); |
2910 | if (err) | |
2911 | goto err_unmap_bar; | |
2912 | ||
2913 | /* | |
2914 | * Allocate our "adapter ports" and stitch everything together. | |
2915 | */ | |
2916 | pmask = adapter->params.vfres.pmask; | |
858aa65c | 2917 | pf = t4vf_get_pf_from_vf(adapter); |
be839e39 CL |
2918 | for_each_port(adapter, pidx) { |
2919 | int port_id, viid; | |
858aa65c HS |
2920 | u8 mac[ETH_ALEN]; |
2921 | unsigned int naddr = 1; | |
be839e39 CL |
2922 | |
2923 | /* | |
2924 | * We simplistically allocate our virtual interfaces | |
2925 | * sequentially across the port numbers to which we have | |
2926 | * access rights. This should be configurable in some manner | |
2927 | * ... | |
2928 | */ | |
2929 | if (pmask == 0) | |
2930 | break; | |
2931 | port_id = ffs(pmask) - 1; | |
2932 | pmask &= ~(1 << port_id); | |
2933 | viid = t4vf_alloc_vi(adapter, port_id); | |
2934 | if (viid < 0) { | |
2935 | dev_err(&pdev->dev, "cannot allocate VI for port %d:" | |
2936 | " err=%d\n", port_id, viid); | |
2937 | err = viid; | |
2938 | goto err_free_dev; | |
2939 | } | |
2940 | ||
2941 | /* | |
2942 | * Allocate our network device and stitch things together. | |
2943 | */ | |
2944 | netdev = alloc_etherdev_mq(sizeof(struct port_info), | |
2945 | MAX_PORT_QSETS); | |
2946 | if (netdev == NULL) { | |
be839e39 CL |
2947 | t4vf_free_vi(adapter, viid); |
2948 | err = -ENOMEM; | |
2949 | goto err_free_dev; | |
2950 | } | |
2951 | adapter->port[pidx] = netdev; | |
2952 | SET_NETDEV_DEV(netdev, &pdev->dev); | |
2953 | pi = netdev_priv(netdev); | |
2954 | pi->adapter = adapter; | |
2955 | pi->pidx = pidx; | |
2956 | pi->port_id = port_id; | |
2957 | pi->viid = viid; | |
2958 | ||
2959 | /* | |
2960 | * Initialize the starting state of our "port" and register | |
2961 | * it. | |
2962 | */ | |
2963 | pi->xact_addr_filt = -1; | |
be839e39 | 2964 | netif_carrier_off(netdev); |
be839e39 CL |
2965 | netdev->irq = pdev->irq; |
2966 | ||
2ed28baa MM |
2967 | netdev->hw_features = NETIF_F_SG | TSO_FLAGS | |
2968 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
f646968f | 2969 | NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXCSUM; |
2ed28baa MM |
2970 | netdev->vlan_features = NETIF_F_SG | TSO_FLAGS | |
2971 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
2972 | NETIF_F_HIGHDMA; | |
f646968f PM |
2973 | netdev->features = netdev->hw_features | |
2974 | NETIF_F_HW_VLAN_CTAG_TX; | |
be839e39 CL |
2975 | if (pci_using_dac) |
2976 | netdev->features |= NETIF_F_HIGHDMA; | |
be839e39 | 2977 | |
01789349 | 2978 | netdev->priv_flags |= IFF_UNICAST_FLT; |
d894be57 JW |
2979 | netdev->min_mtu = 81; |
2980 | netdev->max_mtu = ETH_MAX_MTU; | |
01789349 | 2981 | |
be839e39 | 2982 | netdev->netdev_ops = &cxgb4vf_netdev_ops; |
7ad24ea4 | 2983 | netdev->ethtool_ops = &cxgb4vf_ethtool_ops; |
d2a007ab | 2984 | netdev->dev_port = pi->port_id; |
be839e39 CL |
2985 | |
2986 | /* | |
2987 | * Initialize the hardware/software state for the port. | |
2988 | */ | |
2989 | err = t4vf_port_init(adapter, pidx); | |
2990 | if (err) { | |
2991 | dev_err(&pdev->dev, "cannot initialize port %d\n", | |
2992 | pidx); | |
2993 | goto err_free_dev; | |
2994 | } | |
858aa65c HS |
2995 | |
2996 | err = t4vf_get_vf_mac_acl(adapter, pf, &naddr, mac); | |
2997 | if (err) { | |
2998 | dev_err(&pdev->dev, | |
2999 | "unable to determine MAC ACL address, " | |
3000 | "continuing anyway.. (status %d)\n", err); | |
3001 | } else if (naddr && adapter->params.vfres.nvi == 1) { | |
3002 | struct sockaddr addr; | |
3003 | ||
3004 | ether_addr_copy(addr.sa_data, mac); | |
3005 | err = cxgb4vf_set_mac_addr(netdev, &addr); | |
3006 | if (err) { | |
3007 | dev_err(&pdev->dev, | |
3008 | "unable to set MAC address %pM\n", | |
3009 | mac); | |
3010 | goto err_free_dev; | |
3011 | } | |
3012 | dev_info(&pdev->dev, | |
3013 | "Using assigned MAC ACL: %pM\n", mac); | |
3014 | } | |
be839e39 CL |
3015 | } |
3016 | ||
84f67018 HS |
3017 | /* See what interrupts we'll be using. If we've been configured to |
3018 | * use MSI-X interrupts, try to enable them but fall back to using | |
3019 | * MSI interrupts if we can't enable MSI-X interrupts. If we can't | |
3020 | * get MSI interrupts we bail with the error. | |
3021 | */ | |
3022 | if (msi == MSI_MSIX && enable_msix(adapter) == 0) | |
3023 | adapter->flags |= USING_MSIX; | |
3024 | else { | |
495c22bb HS |
3025 | if (msi == MSI_MSIX) { |
3026 | dev_info(adapter->pdev_dev, | |
3027 | "Unable to use MSI-X Interrupts; falling " | |
3028 | "back to MSI Interrupts\n"); | |
3029 | ||
3030 | /* We're going to need a Forwarded Interrupt Queue so | |
3031 | * that may cut into how many Queue Sets we can | |
3032 | * support. | |
3033 | */ | |
3034 | msi = MSI_MSI; | |
3035 | size_nports_qsets(adapter); | |
3036 | } | |
84f67018 HS |
3037 | err = pci_enable_msi(pdev); |
3038 | if (err) { | |
495c22bb HS |
3039 | dev_err(&pdev->dev, "Unable to allocate MSI Interrupts;" |
3040 | " err=%d\n", err); | |
84f67018 HS |
3041 | goto err_free_dev; |
3042 | } | |
3043 | adapter->flags |= USING_MSI; | |
3044 | } | |
3045 | ||
495c22bb HS |
3046 | /* Now that we know how many "ports" we have and what interrupt |
3047 | * mechanism we're going to use, we can configure our queue resources. | |
3048 | */ | |
3049 | cfg_queues(adapter); | |
3050 | ||
be839e39 CL |
3051 | /* |
3052 | * The "card" is now ready to go. If any errors occur during device | |
3053 | * registration we do not fail the whole "card" but rather proceed | |
3054 | * only with the ports we manage to register successfully. However we | |
3055 | * must register at least one net device. | |
3056 | */ | |
3057 | for_each_port(adapter, pidx) { | |
a8d16d08 | 3058 | struct port_info *pi = netdev_priv(adapter->port[pidx]); |
be839e39 CL |
3059 | netdev = adapter->port[pidx]; |
3060 | if (netdev == NULL) | |
3061 | continue; | |
3062 | ||
a8d16d08 HS |
3063 | netif_set_real_num_tx_queues(netdev, pi->nqsets); |
3064 | netif_set_real_num_rx_queues(netdev, pi->nqsets); | |
3065 | ||
be839e39 CL |
3066 | err = register_netdev(netdev); |
3067 | if (err) { | |
3068 | dev_warn(&pdev->dev, "cannot register net device %s," | |
3069 | " skipping\n", netdev->name); | |
3070 | continue; | |
3071 | } | |
3072 | ||
3073 | set_bit(pidx, &adapter->registered_device_map); | |
3074 | } | |
3075 | if (adapter->registered_device_map == 0) { | |
3076 | dev_err(&pdev->dev, "could not register any net devices\n"); | |
84f67018 | 3077 | goto err_disable_interrupts; |
be839e39 CL |
3078 | } |
3079 | ||
3080 | /* | |
3081 | * Set up our debugfs entries. | |
3082 | */ | |
843635e0 | 3083 | if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) { |
be839e39 CL |
3084 | adapter->debugfs_root = |
3085 | debugfs_create_dir(pci_name(pdev), | |
3086 | cxgb4vf_debugfs_root); | |
843635e0 | 3087 | if (IS_ERR_OR_NULL(adapter->debugfs_root)) |
be839e39 CL |
3088 | dev_warn(&pdev->dev, "could not create debugfs" |
3089 | " directory"); | |
3090 | else | |
3091 | setup_debugfs(adapter); | |
3092 | } | |
3093 | ||
be839e39 | 3094 | /* |
25985edc | 3095 | * Print a short notice on the existence and configuration of the new |
be839e39 CL |
3096 | * VF network device ... |
3097 | */ | |
3098 | for_each_port(adapter, pidx) { | |
3099 | dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n", | |
3100 | adapter->port[pidx]->name, | |
3101 | (adapter->flags & USING_MSIX) ? "MSI-X" : | |
3102 | (adapter->flags & USING_MSI) ? "MSI" : ""); | |
3103 | } | |
3104 | ||
3105 | /* | |
3106 | * Return success! | |
3107 | */ | |
3108 | return 0; | |
3109 | ||
3110 | /* | |
3111 | * Error recovery and exit code. Unwind state that's been created | |
3112 | * so far and return the error. | |
3113 | */ | |
84f67018 HS |
3114 | err_disable_interrupts: |
3115 | if (adapter->flags & USING_MSIX) { | |
3116 | pci_disable_msix(adapter->pdev); | |
3117 | adapter->flags &= ~USING_MSIX; | |
3118 | } else if (adapter->flags & USING_MSI) { | |
3119 | pci_disable_msi(adapter->pdev); | |
3120 | adapter->flags &= ~USING_MSI; | |
be839e39 CL |
3121 | } |
3122 | ||
3123 | err_free_dev: | |
3124 | for_each_port(adapter, pidx) { | |
3125 | netdev = adapter->port[pidx]; | |
3126 | if (netdev == NULL) | |
3127 | continue; | |
3128 | pi = netdev_priv(netdev); | |
3129 | t4vf_free_vi(adapter, pi->viid); | |
3130 | if (test_bit(pidx, &adapter->registered_device_map)) | |
3131 | unregister_netdev(netdev); | |
3132 | free_netdev(netdev); | |
3133 | } | |
3134 | ||
3135 | err_unmap_bar: | |
e0a8b34a HS |
3136 | if (!is_t4(adapter->params.chip)) |
3137 | iounmap(adapter->bar2); | |
3138 | ||
3139 | err_unmap_bar0: | |
be839e39 CL |
3140 | iounmap(adapter->regs); |
3141 | ||
3142 | err_free_adapter: | |
ae7b7576 | 3143 | kfree(adapter->mbox_log); |
be839e39 | 3144 | kfree(adapter); |
be839e39 | 3145 | |
be839e39 CL |
3146 | err_release_regions: |
3147 | pci_release_regions(pdev); | |
7a0c2029 KV |
3148 | pci_clear_master(pdev); |
3149 | ||
3150 | err_disable_device: | |
3151 | pci_disable_device(pdev); | |
be839e39 | 3152 | |
be839e39 CL |
3153 | return err; |
3154 | } | |
3155 | ||
3156 | /* | |
3157 | * "Remove" a device: tear down all kernel and driver state created in the | |
3158 | * "probe" routine and quiesce the device (disable interrupts, etc.). (Note | |
3159 | * that this is called "remove_one" in the PF Driver.) | |
3160 | */ | |
d289f864 | 3161 | static void cxgb4vf_pci_remove(struct pci_dev *pdev) |
be839e39 CL |
3162 | { |
3163 | struct adapter *adapter = pci_get_drvdata(pdev); | |
3164 | ||
3165 | /* | |
3166 | * Tear down driver state associated with device. | |
3167 | */ | |
3168 | if (adapter) { | |
3169 | int pidx; | |
3170 | ||
3171 | /* | |
3172 | * Stop all of our activity. Unregister network port, | |
3173 | * disable interrupts, etc. | |
3174 | */ | |
3175 | for_each_port(adapter, pidx) | |
3176 | if (test_bit(pidx, &adapter->registered_device_map)) | |
3177 | unregister_netdev(adapter->port[pidx]); | |
3178 | t4vf_sge_stop(adapter); | |
3179 | if (adapter->flags & USING_MSIX) { | |
3180 | pci_disable_msix(adapter->pdev); | |
3181 | adapter->flags &= ~USING_MSIX; | |
3182 | } else if (adapter->flags & USING_MSI) { | |
3183 | pci_disable_msi(adapter->pdev); | |
3184 | adapter->flags &= ~USING_MSI; | |
3185 | } | |
3186 | ||
3187 | /* | |
3188 | * Tear down our debugfs entries. | |
3189 | */ | |
843635e0 | 3190 | if (!IS_ERR_OR_NULL(adapter->debugfs_root)) { |
be839e39 CL |
3191 | cleanup_debugfs(adapter); |
3192 | debugfs_remove_recursive(adapter->debugfs_root); | |
3193 | } | |
3194 | ||
3195 | /* | |
3196 | * Free all of the various resources which we've acquired ... | |
3197 | */ | |
3198 | t4vf_free_sge_resources(adapter); | |
3199 | for_each_port(adapter, pidx) { | |
3200 | struct net_device *netdev = adapter->port[pidx]; | |
3201 | struct port_info *pi; | |
3202 | ||
3203 | if (netdev == NULL) | |
3204 | continue; | |
3205 | ||
3206 | pi = netdev_priv(netdev); | |
3207 | t4vf_free_vi(adapter, pi->viid); | |
3208 | free_netdev(netdev); | |
3209 | } | |
3210 | iounmap(adapter->regs); | |
e0a8b34a HS |
3211 | if (!is_t4(adapter->params.chip)) |
3212 | iounmap(adapter->bar2); | |
ae7b7576 | 3213 | kfree(adapter->mbox_log); |
be839e39 | 3214 | kfree(adapter); |
be839e39 CL |
3215 | } |
3216 | ||
3217 | /* | |
3218 | * Disable the device and release its PCI resources. | |
3219 | */ | |
3220 | pci_disable_device(pdev); | |
3221 | pci_clear_master(pdev); | |
3222 | pci_release_regions(pdev); | |
3223 | } | |
3224 | ||
7e9c2629 CL |
3225 | /* |
3226 | * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt | |
3227 | * delivery. | |
3228 | */ | |
d289f864 | 3229 | static void cxgb4vf_pci_shutdown(struct pci_dev *pdev) |
7e9c2629 CL |
3230 | { |
3231 | struct adapter *adapter; | |
3232 | int pidx; | |
3233 | ||
3234 | adapter = pci_get_drvdata(pdev); | |
3235 | if (!adapter) | |
3236 | return; | |
3237 | ||
c2a19856 | 3238 | /* Disable all Virtual Interfaces. This will shut down the |
7e9c2629 CL |
3239 | * delivery of all ingress packets into the chip for these |
3240 | * Virtual Interfaces. | |
3241 | */ | |
c2a19856 HS |
3242 | for_each_port(adapter, pidx) |
3243 | if (test_bit(pidx, &adapter->registered_device_map)) | |
3244 | unregister_netdev(adapter->port[pidx]); | |
7e9c2629 | 3245 | |
c2a19856 HS |
3246 | /* Free up all Queues which will prevent further DMA and |
3247 | * Interrupts allowing various internal pathways to drain. | |
3248 | */ | |
3249 | t4vf_sge_stop(adapter); | |
3250 | if (adapter->flags & USING_MSIX) { | |
3251 | pci_disable_msix(adapter->pdev); | |
3252 | adapter->flags &= ~USING_MSIX; | |
3253 | } else if (adapter->flags & USING_MSI) { | |
3254 | pci_disable_msi(adapter->pdev); | |
3255 | adapter->flags &= ~USING_MSI; | |
7e9c2629 CL |
3256 | } |
3257 | ||
3258 | /* | |
3259 | * Free up all Queues which will prevent further DMA and | |
3260 | * Interrupts allowing various internal pathways to drain. | |
3261 | */ | |
3262 | t4vf_free_sge_resources(adapter); | |
c2a19856 | 3263 | pci_set_drvdata(pdev, NULL); |
7e9c2629 CL |
3264 | } |
3265 | ||
3fedeab1 HS |
3266 | /* Macros needed to support the PCI Device ID Table ... |
3267 | */ | |
3268 | #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \ | |
768ffc66 | 3269 | static const struct pci_device_id cxgb4vf_pci_tbl[] = { |
3fedeab1 HS |
3270 | #define CH_PCI_DEVICE_ID_FUNCTION 0x8 |
3271 | ||
3272 | #define CH_PCI_ID_TABLE_ENTRY(devid) \ | |
3273 | { PCI_VDEVICE(CHELSIO, (devid)), 0 } | |
3274 | ||
3275 | #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END { 0, } } | |
3276 | ||
3277 | #include "../cxgb4/t4_pci_id_tbl.h" | |
be839e39 CL |
3278 | |
3279 | MODULE_DESCRIPTION(DRV_DESC); | |
3280 | MODULE_AUTHOR("Chelsio Communications"); | |
3281 | MODULE_LICENSE("Dual BSD/GPL"); | |
3282 | MODULE_VERSION(DRV_VERSION); | |
3283 | MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl); | |
3284 | ||
3285 | static struct pci_driver cxgb4vf_driver = { | |
3286 | .name = KBUILD_MODNAME, | |
3287 | .id_table = cxgb4vf_pci_tbl, | |
3288 | .probe = cxgb4vf_pci_probe, | |
d289f864 BP |
3289 | .remove = cxgb4vf_pci_remove, |
3290 | .shutdown = cxgb4vf_pci_shutdown, | |
be839e39 CL |
3291 | }; |
3292 | ||
3293 | /* | |
3294 | * Initialize global driver state. | |
3295 | */ | |
3296 | static int __init cxgb4vf_module_init(void) | |
3297 | { | |
3298 | int ret; | |
3299 | ||
bb14a1af CL |
3300 | /* |
3301 | * Vet our module parameters. | |
3302 | */ | |
3303 | if (msi != MSI_MSIX && msi != MSI_MSI) { | |
428ac43f JP |
3304 | pr_warn("bad module parameter msi=%d; must be %d (MSI-X or MSI) or %d (MSI)\n", |
3305 | msi, MSI_MSIX, MSI_MSI); | |
bb14a1af CL |
3306 | return -EINVAL; |
3307 | } | |
3308 | ||
be839e39 CL |
3309 | /* Debugfs support is optional, just warn if this fails */ |
3310 | cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL); | |
843635e0 | 3311 | if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) |
428ac43f | 3312 | pr_warn("could not create debugfs entry, continuing\n"); |
be839e39 CL |
3313 | |
3314 | ret = pci_register_driver(&cxgb4vf_driver); | |
843635e0 | 3315 | if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) |
be839e39 CL |
3316 | debugfs_remove(cxgb4vf_debugfs_root); |
3317 | return ret; | |
3318 | } | |
3319 | ||
3320 | /* | |
3321 | * Tear down global driver state. | |
3322 | */ | |
3323 | static void __exit cxgb4vf_module_exit(void) | |
3324 | { | |
3325 | pci_unregister_driver(&cxgb4vf_driver); | |
3326 | debugfs_remove(cxgb4vf_debugfs_root); | |
3327 | } | |
3328 | ||
3329 | module_init(cxgb4vf_module_init); | |
3330 | module_exit(cxgb4vf_module_exit); |