]>
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> | |
47 | ||
48 | #include "t4vf_common.h" | |
49 | #include "t4vf_defs.h" | |
50 | ||
51 | #include "../cxgb4/t4_regs.h" | |
52 | #include "../cxgb4/t4_msg.h" | |
53 | ||
54 | /* | |
55 | * Generic information about the driver. | |
56 | */ | |
622c62b5 SR |
57 | #define DRV_VERSION "2.0.0-ko" |
58 | #define DRV_DESC "Chelsio T4/T5 Virtual Function (VF) Network Driver" | |
be839e39 CL |
59 | |
60 | /* | |
61 | * Module Parameters. | |
62 | * ================== | |
63 | */ | |
64 | ||
65 | /* | |
66 | * Default ethtool "message level" for adapters. | |
67 | */ | |
68 | #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \ | |
69 | NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\ | |
70 | NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR) | |
71 | ||
72 | static int dflt_msg_enable = DFLT_MSG_ENABLE; | |
73 | ||
74 | module_param(dflt_msg_enable, int, 0644); | |
75 | MODULE_PARM_DESC(dflt_msg_enable, | |
76 | "default adapter ethtool message level bitmap"); | |
77 | ||
78 | /* | |
79 | * The driver uses the best interrupt scheme available on a platform in the | |
80 | * order MSI-X then MSI. This parameter determines which of these schemes the | |
81 | * driver may consider as follows: | |
82 | * | |
83 | * msi = 2: choose from among MSI-X and MSI | |
84 | * msi = 1: only consider MSI interrupts | |
85 | * | |
86 | * Note that unlike the Physical Function driver, this Virtual Function driver | |
87 | * does _not_ support legacy INTx interrupts (this limitation is mandated by | |
88 | * the PCI-E SR-IOV standard). | |
89 | */ | |
90 | #define MSI_MSIX 2 | |
91 | #define MSI_MSI 1 | |
92 | #define MSI_DEFAULT MSI_MSIX | |
93 | ||
94 | static int msi = MSI_DEFAULT; | |
95 | ||
96 | module_param(msi, int, 0644); | |
97 | MODULE_PARM_DESC(msi, "whether to use MSI-X or MSI"); | |
98 | ||
99 | /* | |
100 | * Fundamental constants. | |
101 | * ====================== | |
102 | */ | |
103 | ||
104 | enum { | |
105 | MAX_TXQ_ENTRIES = 16384, | |
106 | MAX_RSPQ_ENTRIES = 16384, | |
107 | MAX_RX_BUFFERS = 16384, | |
108 | ||
109 | MIN_TXQ_ENTRIES = 32, | |
110 | MIN_RSPQ_ENTRIES = 128, | |
111 | MIN_FL_ENTRIES = 16, | |
112 | ||
113 | /* | |
114 | * For purposes of manipulating the Free List size we need to | |
115 | * recognize that Free Lists are actually Egress Queues (the host | |
116 | * produces free buffers which the hardware consumes), Egress Queues | |
117 | * indices are all in units of Egress Context Units bytes, and free | |
118 | * list entries are 64-bit PCI DMA addresses. And since the state of | |
119 | * the Producer Index == the Consumer Index implies an EMPTY list, we | |
120 | * always have at least one Egress Unit's worth of Free List entries | |
121 | * unused. See sge.c for more details ... | |
122 | */ | |
123 | EQ_UNIT = SGE_EQ_IDXSIZE, | |
124 | FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64), | |
125 | MIN_FL_RESID = FL_PER_EQ_UNIT, | |
126 | }; | |
127 | ||
128 | /* | |
129 | * Global driver state. | |
130 | * ==================== | |
131 | */ | |
132 | ||
133 | static struct dentry *cxgb4vf_debugfs_root; | |
134 | ||
135 | /* | |
136 | * OS "Callback" functions. | |
137 | * ======================== | |
138 | */ | |
139 | ||
140 | /* | |
141 | * The link status has changed on the indicated "port" (Virtual Interface). | |
142 | */ | |
143 | void t4vf_os_link_changed(struct adapter *adapter, int pidx, int link_ok) | |
144 | { | |
145 | struct net_device *dev = adapter->port[pidx]; | |
146 | ||
147 | /* | |
148 | * If the port is disabled or the current recorded "link up" | |
149 | * status matches the new status, just return. | |
150 | */ | |
151 | if (!netif_running(dev) || link_ok == netif_carrier_ok(dev)) | |
152 | return; | |
153 | ||
154 | /* | |
155 | * Tell the OS that the link status has changed and print a short | |
156 | * informative message on the console about the event. | |
157 | */ | |
158 | if (link_ok) { | |
159 | const char *s; | |
160 | const char *fc; | |
161 | const struct port_info *pi = netdev_priv(dev); | |
162 | ||
163 | netif_carrier_on(dev); | |
164 | ||
165 | switch (pi->link_cfg.speed) { | |
897d55df HS |
166 | case 40000: |
167 | s = "40Gbps"; | |
168 | break; | |
169 | ||
170 | case 10000: | |
be839e39 CL |
171 | s = "10Gbps"; |
172 | break; | |
173 | ||
897d55df | 174 | case 1000: |
be839e39 CL |
175 | s = "1000Mbps"; |
176 | break; | |
177 | ||
897d55df | 178 | case 100: |
be839e39 CL |
179 | s = "100Mbps"; |
180 | break; | |
181 | ||
182 | default: | |
183 | s = "unknown"; | |
184 | break; | |
185 | } | |
186 | ||
187 | switch (pi->link_cfg.fc) { | |
188 | case PAUSE_RX: | |
189 | fc = "RX"; | |
190 | break; | |
191 | ||
192 | case PAUSE_TX: | |
193 | fc = "TX"; | |
194 | break; | |
195 | ||
196 | case PAUSE_RX|PAUSE_TX: | |
197 | fc = "RX/TX"; | |
198 | break; | |
199 | ||
200 | default: | |
201 | fc = "no"; | |
202 | break; | |
203 | } | |
204 | ||
428ac43f | 205 | netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s, fc); |
be839e39 CL |
206 | } else { |
207 | netif_carrier_off(dev); | |
428ac43f | 208 | netdev_info(dev, "link down\n"); |
be839e39 CL |
209 | } |
210 | } | |
211 | ||
212 | /* | |
213 | * Net device operations. | |
214 | * ====================== | |
215 | */ | |
216 | ||
be839e39 | 217 | |
87737663 | 218 | |
be839e39 CL |
219 | |
220 | /* | |
221 | * Perform the MAC and PHY actions needed to enable a "port" (Virtual | |
222 | * Interface). | |
223 | */ | |
224 | static int link_start(struct net_device *dev) | |
225 | { | |
226 | int ret; | |
227 | struct port_info *pi = netdev_priv(dev); | |
228 | ||
229 | /* | |
230 | * We do not set address filters and promiscuity here, the stack does | |
87737663 | 231 | * that step explicitly. Enable vlan accel. |
be839e39 | 232 | */ |
87737663 | 233 | ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, 1, |
be839e39 CL |
234 | true); |
235 | if (ret == 0) { | |
236 | ret = t4vf_change_mac(pi->adapter, pi->viid, | |
237 | pi->xact_addr_filt, dev->dev_addr, true); | |
238 | if (ret >= 0) { | |
239 | pi->xact_addr_filt = ret; | |
240 | ret = 0; | |
241 | } | |
242 | } | |
243 | ||
244 | /* | |
245 | * We don't need to actually "start the link" itself since the | |
246 | * firmware will do that for us when the first Virtual Interface | |
247 | * is enabled on a port. | |
248 | */ | |
249 | if (ret == 0) | |
250 | ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true); | |
251 | return ret; | |
252 | } | |
253 | ||
254 | /* | |
255 | * Name the MSI-X interrupts. | |
256 | */ | |
257 | static void name_msix_vecs(struct adapter *adapter) | |
258 | { | |
259 | int namelen = sizeof(adapter->msix_info[0].desc) - 1; | |
260 | int pidx; | |
261 | ||
262 | /* | |
263 | * Firmware events. | |
264 | */ | |
265 | snprintf(adapter->msix_info[MSIX_FW].desc, namelen, | |
266 | "%s-FWeventq", adapter->name); | |
267 | adapter->msix_info[MSIX_FW].desc[namelen] = 0; | |
268 | ||
269 | /* | |
270 | * Ethernet queues. | |
271 | */ | |
272 | for_each_port(adapter, pidx) { | |
273 | struct net_device *dev = adapter->port[pidx]; | |
274 | const struct port_info *pi = netdev_priv(dev); | |
275 | int qs, msi; | |
276 | ||
caedda35 | 277 | for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) { |
be839e39 CL |
278 | snprintf(adapter->msix_info[msi].desc, namelen, |
279 | "%s-%d", dev->name, qs); | |
280 | adapter->msix_info[msi].desc[namelen] = 0; | |
281 | } | |
282 | } | |
283 | } | |
284 | ||
285 | /* | |
286 | * Request all of our MSI-X resources. | |
287 | */ | |
288 | static int request_msix_queue_irqs(struct adapter *adapter) | |
289 | { | |
290 | struct sge *s = &adapter->sge; | |
291 | int rxq, msi, err; | |
292 | ||
293 | /* | |
294 | * Firmware events. | |
295 | */ | |
296 | err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix, | |
297 | 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq); | |
298 | if (err) | |
299 | return err; | |
300 | ||
301 | /* | |
302 | * Ethernet queues. | |
303 | */ | |
caedda35 | 304 | msi = MSIX_IQFLINT; |
be839e39 CL |
305 | for_each_ethrxq(s, rxq) { |
306 | err = request_irq(adapter->msix_info[msi].vec, | |
307 | t4vf_sge_intr_msix, 0, | |
308 | adapter->msix_info[msi].desc, | |
309 | &s->ethrxq[rxq].rspq); | |
310 | if (err) | |
311 | goto err_free_irqs; | |
312 | msi++; | |
313 | } | |
314 | return 0; | |
315 | ||
316 | err_free_irqs: | |
317 | while (--rxq >= 0) | |
318 | free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq); | |
319 | free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq); | |
320 | return err; | |
321 | } | |
322 | ||
323 | /* | |
324 | * Free our MSI-X resources. | |
325 | */ | |
326 | static void free_msix_queue_irqs(struct adapter *adapter) | |
327 | { | |
328 | struct sge *s = &adapter->sge; | |
329 | int rxq, msi; | |
330 | ||
331 | free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq); | |
caedda35 | 332 | msi = MSIX_IQFLINT; |
be839e39 CL |
333 | for_each_ethrxq(s, rxq) |
334 | free_irq(adapter->msix_info[msi++].vec, | |
335 | &s->ethrxq[rxq].rspq); | |
336 | } | |
337 | ||
338 | /* | |
339 | * Turn on NAPI and start up interrupts on a response queue. | |
340 | */ | |
341 | static void qenable(struct sge_rspq *rspq) | |
342 | { | |
343 | napi_enable(&rspq->napi); | |
344 | ||
345 | /* | |
346 | * 0-increment the Going To Sleep register to start the timer and | |
347 | * enable interrupts. | |
348 | */ | |
349 | t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS, | |
350 | CIDXINC(0) | | |
351 | SEINTARM(rspq->intr_params) | | |
352 | INGRESSQID(rspq->cntxt_id)); | |
353 | } | |
354 | ||
355 | /* | |
356 | * Enable NAPI scheduling and interrupt generation for all Receive Queues. | |
357 | */ | |
358 | static void enable_rx(struct adapter *adapter) | |
359 | { | |
360 | int rxq; | |
361 | struct sge *s = &adapter->sge; | |
362 | ||
363 | for_each_ethrxq(s, rxq) | |
364 | qenable(&s->ethrxq[rxq].rspq); | |
365 | qenable(&s->fw_evtq); | |
366 | ||
367 | /* | |
368 | * The interrupt queue doesn't use NAPI so we do the 0-increment of | |
369 | * its Going To Sleep register here to get it started. | |
370 | */ | |
371 | if (adapter->flags & USING_MSI) | |
372 | t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS, | |
373 | CIDXINC(0) | | |
374 | SEINTARM(s->intrq.intr_params) | | |
375 | INGRESSQID(s->intrq.cntxt_id)); | |
376 | ||
377 | } | |
378 | ||
379 | /* | |
380 | * Wait until all NAPI handlers are descheduled. | |
381 | */ | |
382 | static void quiesce_rx(struct adapter *adapter) | |
383 | { | |
384 | struct sge *s = &adapter->sge; | |
385 | int rxq; | |
386 | ||
387 | for_each_ethrxq(s, rxq) | |
388 | napi_disable(&s->ethrxq[rxq].rspq.napi); | |
389 | napi_disable(&s->fw_evtq.napi); | |
390 | } | |
391 | ||
392 | /* | |
393 | * Response queue handler for the firmware event queue. | |
394 | */ | |
395 | static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp, | |
396 | const struct pkt_gl *gl) | |
397 | { | |
398 | /* | |
399 | * Extract response opcode and get pointer to CPL message body. | |
400 | */ | |
401 | struct adapter *adapter = rspq->adapter; | |
402 | u8 opcode = ((const struct rss_header *)rsp)->opcode; | |
403 | void *cpl = (void *)(rsp + 1); | |
404 | ||
405 | switch (opcode) { | |
406 | case CPL_FW6_MSG: { | |
407 | /* | |
408 | * We've received an asynchronous message from the firmware. | |
409 | */ | |
410 | const struct cpl_fw6_msg *fw_msg = cpl; | |
411 | if (fw_msg->type == FW6_TYPE_CMD_RPL) | |
412 | t4vf_handle_fw_rpl(adapter, fw_msg->data); | |
413 | break; | |
414 | } | |
415 | ||
94dace10 VP |
416 | case CPL_FW4_MSG: { |
417 | /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG. | |
418 | */ | |
419 | const struct cpl_sge_egr_update *p = (void *)(rsp + 3); | |
420 | opcode = G_CPL_OPCODE(ntohl(p->opcode_qid)); | |
421 | if (opcode != CPL_SGE_EGR_UPDATE) { | |
422 | dev_err(adapter->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n" | |
423 | , opcode); | |
424 | break; | |
425 | } | |
426 | cpl = (void *)p; | |
427 | /*FALLTHROUGH*/ | |
428 | } | |
429 | ||
be839e39 CL |
430 | case CPL_SGE_EGR_UPDATE: { |
431 | /* | |
7f9dd2fa CL |
432 | * We've received an Egress Queue Status Update message. We |
433 | * get these, if the SGE is configured to send these when the | |
434 | * firmware passes certain points in processing our TX | |
435 | * Ethernet Queue or if we make an explicit request for one. | |
436 | * We use these updates to determine when we may need to | |
437 | * restart a TX Ethernet Queue which was stopped for lack of | |
438 | * free TX Queue Descriptors ... | |
be839e39 | 439 | */ |
64699336 | 440 | const struct cpl_sge_egr_update *p = cpl; |
be839e39 CL |
441 | unsigned int qid = EGR_QID(be32_to_cpu(p->opcode_qid)); |
442 | struct sge *s = &adapter->sge; | |
443 | struct sge_txq *tq; | |
444 | struct sge_eth_txq *txq; | |
445 | unsigned int eq_idx; | |
be839e39 CL |
446 | |
447 | /* | |
448 | * Perform sanity checking on the Queue ID to make sure it | |
449 | * really refers to one of our TX Ethernet Egress Queues which | |
450 | * is active and matches the queue's ID. None of these error | |
451 | * conditions should ever happen so we may want to either make | |
452 | * them fatal and/or conditionalized under DEBUG. | |
453 | */ | |
454 | eq_idx = EQ_IDX(s, qid); | |
455 | if (unlikely(eq_idx >= MAX_EGRQ)) { | |
456 | dev_err(adapter->pdev_dev, | |
457 | "Egress Update QID %d out of range\n", qid); | |
458 | break; | |
459 | } | |
460 | tq = s->egr_map[eq_idx]; | |
461 | if (unlikely(tq == NULL)) { | |
462 | dev_err(adapter->pdev_dev, | |
463 | "Egress Update QID %d TXQ=NULL\n", qid); | |
464 | break; | |
465 | } | |
466 | txq = container_of(tq, struct sge_eth_txq, q); | |
467 | if (unlikely(tq->abs_id != qid)) { | |
468 | dev_err(adapter->pdev_dev, | |
469 | "Egress Update QID %d refers to TXQ %d\n", | |
470 | qid, tq->abs_id); | |
471 | break; | |
472 | } | |
473 | ||
be839e39 CL |
474 | /* |
475 | * Restart a stopped TX Queue which has less than half of its | |
476 | * TX ring in use ... | |
477 | */ | |
478 | txq->q.restarts++; | |
479 | netif_tx_wake_queue(txq->txq); | |
480 | break; | |
481 | } | |
482 | ||
483 | default: | |
484 | dev_err(adapter->pdev_dev, | |
485 | "unexpected CPL %#x on FW event queue\n", opcode); | |
486 | } | |
487 | ||
488 | return 0; | |
489 | } | |
490 | ||
491 | /* | |
492 | * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues | |
493 | * to use and initializes them. We support multiple "Queue Sets" per port if | |
494 | * we have MSI-X, otherwise just one queue set per port. | |
495 | */ | |
496 | static int setup_sge_queues(struct adapter *adapter) | |
497 | { | |
498 | struct sge *s = &adapter->sge; | |
499 | int err, pidx, msix; | |
500 | ||
501 | /* | |
502 | * Clear "Queue Set" Free List Starving and TX Queue Mapping Error | |
503 | * state. | |
504 | */ | |
505 | bitmap_zero(s->starving_fl, MAX_EGRQ); | |
506 | ||
507 | /* | |
508 | * If we're using MSI interrupt mode we need to set up a "forwarded | |
509 | * interrupt" queue which we'll set up with our MSI vector. The rest | |
510 | * of the ingress queues will be set up to forward their interrupts to | |
511 | * this queue ... This must be first since t4vf_sge_alloc_rxq() uses | |
512 | * the intrq's queue ID as the interrupt forwarding queue for the | |
513 | * subsequent calls ... | |
514 | */ | |
515 | if (adapter->flags & USING_MSI) { | |
516 | err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false, | |
517 | adapter->port[0], 0, NULL, NULL); | |
518 | if (err) | |
519 | goto err_free_queues; | |
520 | } | |
521 | ||
522 | /* | |
523 | * Allocate our ingress queue for asynchronous firmware messages. | |
524 | */ | |
525 | err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0], | |
526 | MSIX_FW, NULL, fwevtq_handler); | |
527 | if (err) | |
528 | goto err_free_queues; | |
529 | ||
530 | /* | |
531 | * Allocate each "port"'s initial Queue Sets. These can be changed | |
532 | * later on ... up to the point where any interface on the adapter is | |
533 | * brought up at which point lots of things get nailed down | |
534 | * permanently ... | |
535 | */ | |
caedda35 | 536 | msix = MSIX_IQFLINT; |
be839e39 CL |
537 | for_each_port(adapter, pidx) { |
538 | struct net_device *dev = adapter->port[pidx]; | |
539 | struct port_info *pi = netdev_priv(dev); | |
540 | struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset]; | |
541 | struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset]; | |
be839e39 CL |
542 | int qs; |
543 | ||
c8639a82 | 544 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { |
be839e39 CL |
545 | err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false, |
546 | dev, msix++, | |
547 | &rxq->fl, t4vf_ethrx_handler); | |
548 | if (err) | |
549 | goto err_free_queues; | |
550 | ||
551 | err = t4vf_sge_alloc_eth_txq(adapter, txq, dev, | |
552 | netdev_get_tx_queue(dev, qs), | |
553 | s->fw_evtq.cntxt_id); | |
554 | if (err) | |
555 | goto err_free_queues; | |
556 | ||
557 | rxq->rspq.idx = qs; | |
558 | memset(&rxq->stats, 0, sizeof(rxq->stats)); | |
559 | } | |
560 | } | |
561 | ||
562 | /* | |
563 | * Create the reverse mappings for the queues. | |
564 | */ | |
565 | s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id; | |
566 | s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id; | |
567 | IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq; | |
568 | for_each_port(adapter, pidx) { | |
569 | struct net_device *dev = adapter->port[pidx]; | |
570 | struct port_info *pi = netdev_priv(dev); | |
571 | struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset]; | |
572 | struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset]; | |
be839e39 CL |
573 | int qs; |
574 | ||
c8639a82 | 575 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { |
be839e39 CL |
576 | IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq; |
577 | EQ_MAP(s, txq->q.abs_id) = &txq->q; | |
578 | ||
579 | /* | |
580 | * The FW_IQ_CMD doesn't return the Absolute Queue IDs | |
581 | * for Free Lists but since all of the Egress Queues | |
582 | * (including Free Lists) have Relative Queue IDs | |
583 | * which are computed as Absolute - Base Queue ID, we | |
584 | * can synthesize the Absolute Queue IDs for the Free | |
585 | * Lists. This is useful for debugging purposes when | |
586 | * we want to dump Queue Contexts via the PF Driver. | |
587 | */ | |
588 | rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base; | |
589 | EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl; | |
590 | } | |
591 | } | |
592 | return 0; | |
593 | ||
594 | err_free_queues: | |
595 | t4vf_free_sge_resources(adapter); | |
596 | return err; | |
597 | } | |
598 | ||
599 | /* | |
600 | * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive | |
601 | * queues. We configure the RSS CPU lookup table to distribute to the number | |
602 | * of HW receive queues, and the response queue lookup table to narrow that | |
603 | * down to the response queues actually configured for each "port" (Virtual | |
604 | * Interface). We always configure the RSS mapping for all ports since the | |
605 | * mapping table has plenty of entries. | |
606 | */ | |
607 | static int setup_rss(struct adapter *adapter) | |
608 | { | |
609 | int pidx; | |
610 | ||
611 | for_each_port(adapter, pidx) { | |
612 | struct port_info *pi = adap2pinfo(adapter, pidx); | |
613 | struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset]; | |
614 | u16 rss[MAX_PORT_QSETS]; | |
615 | int qs, err; | |
616 | ||
617 | for (qs = 0; qs < pi->nqsets; qs++) | |
618 | rss[qs] = rxq[qs].rspq.abs_id; | |
619 | ||
620 | err = t4vf_config_rss_range(adapter, pi->viid, | |
621 | 0, pi->rss_size, rss, pi->nqsets); | |
622 | if (err) | |
623 | return err; | |
624 | ||
625 | /* | |
626 | * Perform Global RSS Mode-specific initialization. | |
627 | */ | |
628 | switch (adapter->params.rss.mode) { | |
629 | case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL: | |
630 | /* | |
631 | * If Tunnel All Lookup isn't specified in the global | |
632 | * RSS Configuration, then we need to specify a | |
633 | * default Ingress Queue for any ingress packets which | |
634 | * aren't hashed. We'll use our first ingress queue | |
635 | * ... | |
636 | */ | |
637 | if (!adapter->params.rss.u.basicvirtual.tnlalllookup) { | |
638 | union rss_vi_config config; | |
639 | err = t4vf_read_rss_vi_config(adapter, | |
640 | pi->viid, | |
641 | &config); | |
642 | if (err) | |
643 | return err; | |
644 | config.basicvirtual.defaultq = | |
645 | rxq[0].rspq.abs_id; | |
646 | err = t4vf_write_rss_vi_config(adapter, | |
647 | pi->viid, | |
648 | &config); | |
649 | if (err) | |
650 | return err; | |
651 | } | |
652 | break; | |
653 | } | |
654 | } | |
655 | ||
656 | return 0; | |
657 | } | |
658 | ||
659 | /* | |
660 | * Bring the adapter up. Called whenever we go from no "ports" open to having | |
661 | * one open. This function performs the actions necessary to make an adapter | |
662 | * operational, such as completing the initialization of HW modules, and | |
663 | * enabling interrupts. Must be called with the rtnl lock held. (Note that | |
664 | * this is called "cxgb_up" in the PF Driver.) | |
665 | */ | |
666 | static int adapter_up(struct adapter *adapter) | |
667 | { | |
668 | int err; | |
669 | ||
670 | /* | |
671 | * If this is the first time we've been called, perform basic | |
672 | * adapter setup. Once we've done this, many of our adapter | |
673 | * parameters can no longer be changed ... | |
674 | */ | |
675 | if ((adapter->flags & FULL_INIT_DONE) == 0) { | |
676 | err = setup_sge_queues(adapter); | |
677 | if (err) | |
678 | return err; | |
679 | err = setup_rss(adapter); | |
680 | if (err) { | |
681 | t4vf_free_sge_resources(adapter); | |
682 | return err; | |
683 | } | |
684 | ||
685 | if (adapter->flags & USING_MSIX) | |
686 | name_msix_vecs(adapter); | |
687 | adapter->flags |= FULL_INIT_DONE; | |
688 | } | |
689 | ||
690 | /* | |
691 | * Acquire our interrupt resources. We only support MSI-X and MSI. | |
692 | */ | |
693 | BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0); | |
694 | if (adapter->flags & USING_MSIX) | |
695 | err = request_msix_queue_irqs(adapter); | |
696 | else | |
697 | err = request_irq(adapter->pdev->irq, | |
698 | t4vf_intr_handler(adapter), 0, | |
699 | adapter->name, adapter); | |
700 | if (err) { | |
701 | dev_err(adapter->pdev_dev, "request_irq failed, err %d\n", | |
702 | err); | |
703 | return err; | |
704 | } | |
705 | ||
706 | /* | |
707 | * Enable NAPI ingress processing and return success. | |
708 | */ | |
709 | enable_rx(adapter); | |
710 | t4vf_sge_start(adapter); | |
711 | return 0; | |
712 | } | |
713 | ||
714 | /* | |
715 | * Bring the adapter down. Called whenever the last "port" (Virtual | |
716 | * Interface) closed. (Note that this routine is called "cxgb_down" in the PF | |
717 | * Driver.) | |
718 | */ | |
719 | static void adapter_down(struct adapter *adapter) | |
720 | { | |
721 | /* | |
722 | * Free interrupt resources. | |
723 | */ | |
724 | if (adapter->flags & USING_MSIX) | |
725 | free_msix_queue_irqs(adapter); | |
726 | else | |
727 | free_irq(adapter->pdev->irq, adapter); | |
728 | ||
729 | /* | |
730 | * Wait for NAPI handlers to finish. | |
731 | */ | |
732 | quiesce_rx(adapter); | |
733 | } | |
734 | ||
735 | /* | |
736 | * Start up a net device. | |
737 | */ | |
738 | static int cxgb4vf_open(struct net_device *dev) | |
739 | { | |
740 | int err; | |
741 | struct port_info *pi = netdev_priv(dev); | |
742 | struct adapter *adapter = pi->adapter; | |
743 | ||
744 | /* | |
745 | * If this is the first interface that we're opening on the "adapter", | |
746 | * bring the "adapter" up now. | |
747 | */ | |
748 | if (adapter->open_device_map == 0) { | |
749 | err = adapter_up(adapter); | |
750 | if (err) | |
751 | return err; | |
752 | } | |
753 | ||
754 | /* | |
755 | * Note that this interface is up and start everything up ... | |
756 | */ | |
003ab674 BH |
757 | netif_set_real_num_tx_queues(dev, pi->nqsets); |
758 | err = netif_set_real_num_rx_queues(dev, pi->nqsets); | |
759 | if (err) | |
343a8d13 | 760 | goto err_unwind; |
e7a3795f CL |
761 | err = link_start(dev); |
762 | if (err) | |
343a8d13 CL |
763 | goto err_unwind; |
764 | ||
be839e39 | 765 | netif_tx_start_all_queues(dev); |
343a8d13 | 766 | set_bit(pi->port_id, &adapter->open_device_map); |
be839e39 | 767 | return 0; |
343a8d13 CL |
768 | |
769 | err_unwind: | |
770 | if (adapter->open_device_map == 0) | |
771 | adapter_down(adapter); | |
772 | return err; | |
be839e39 CL |
773 | } |
774 | ||
775 | /* | |
776 | * Shut down a net device. This routine is called "cxgb_close" in the PF | |
777 | * Driver ... | |
778 | */ | |
779 | static int cxgb4vf_stop(struct net_device *dev) | |
780 | { | |
be839e39 CL |
781 | struct port_info *pi = netdev_priv(dev); |
782 | struct adapter *adapter = pi->adapter; | |
783 | ||
784 | netif_tx_stop_all_queues(dev); | |
785 | netif_carrier_off(dev); | |
343a8d13 | 786 | t4vf_enable_vi(adapter, pi->viid, false, false); |
be839e39 CL |
787 | pi->link_cfg.link_ok = 0; |
788 | ||
789 | clear_bit(pi->port_id, &adapter->open_device_map); | |
790 | if (adapter->open_device_map == 0) | |
791 | adapter_down(adapter); | |
792 | return 0; | |
793 | } | |
794 | ||
795 | /* | |
796 | * Translate our basic statistics into the standard "ifconfig" statistics. | |
797 | */ | |
798 | static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev) | |
799 | { | |
800 | struct t4vf_port_stats stats; | |
801 | struct port_info *pi = netdev2pinfo(dev); | |
802 | struct adapter *adapter = pi->adapter; | |
803 | struct net_device_stats *ns = &dev->stats; | |
804 | int err; | |
805 | ||
806 | spin_lock(&adapter->stats_lock); | |
807 | err = t4vf_get_port_stats(adapter, pi->pidx, &stats); | |
808 | spin_unlock(&adapter->stats_lock); | |
809 | ||
810 | memset(ns, 0, sizeof(*ns)); | |
811 | if (err) | |
812 | return ns; | |
813 | ||
814 | ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes + | |
815 | stats.tx_ucast_bytes + stats.tx_offload_bytes); | |
816 | ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames + | |
817 | stats.tx_ucast_frames + stats.tx_offload_frames); | |
818 | ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes + | |
819 | stats.rx_ucast_bytes); | |
820 | ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames + | |
821 | stats.rx_ucast_frames); | |
822 | ns->multicast = stats.rx_mcast_frames; | |
823 | ns->tx_errors = stats.tx_drop_frames; | |
824 | ns->rx_errors = stats.rx_err_frames; | |
825 | ||
826 | return ns; | |
827 | } | |
828 | ||
829 | /* | |
42eb59d3 CL |
830 | * Collect up to maxaddrs worth of a netdevice's unicast addresses, starting |
831 | * at a specified offset within the list, into an array of addrss pointers and | |
832 | * return the number collected. | |
be839e39 | 833 | */ |
42eb59d3 CL |
834 | static inline unsigned int collect_netdev_uc_list_addrs(const struct net_device *dev, |
835 | const u8 **addr, | |
836 | unsigned int offset, | |
837 | unsigned int maxaddrs) | |
be839e39 | 838 | { |
42eb59d3 | 839 | unsigned int index = 0; |
be839e39 CL |
840 | unsigned int naddr = 0; |
841 | const struct netdev_hw_addr *ha; | |
842 | ||
42eb59d3 CL |
843 | for_each_dev_addr(dev, ha) |
844 | if (index++ >= offset) { | |
845 | addr[naddr++] = ha->addr; | |
846 | if (naddr >= maxaddrs) | |
847 | break; | |
848 | } | |
be839e39 CL |
849 | return naddr; |
850 | } | |
851 | ||
852 | /* | |
42eb59d3 CL |
853 | * Collect up to maxaddrs worth of a netdevice's multicast addresses, starting |
854 | * at a specified offset within the list, into an array of addrss pointers and | |
855 | * return the number collected. | |
be839e39 | 856 | */ |
42eb59d3 CL |
857 | static inline unsigned int collect_netdev_mc_list_addrs(const struct net_device *dev, |
858 | const u8 **addr, | |
859 | unsigned int offset, | |
860 | unsigned int maxaddrs) | |
be839e39 | 861 | { |
42eb59d3 | 862 | unsigned int index = 0; |
be839e39 CL |
863 | unsigned int naddr = 0; |
864 | const struct netdev_hw_addr *ha; | |
865 | ||
42eb59d3 CL |
866 | netdev_for_each_mc_addr(ha, dev) |
867 | if (index++ >= offset) { | |
868 | addr[naddr++] = ha->addr; | |
869 | if (naddr >= maxaddrs) | |
870 | break; | |
871 | } | |
be839e39 CL |
872 | return naddr; |
873 | } | |
874 | ||
875 | /* | |
876 | * Configure the exact and hash address filters to handle a port's multicast | |
877 | * and secondary unicast MAC addresses. | |
878 | */ | |
879 | static int set_addr_filters(const struct net_device *dev, bool sleep) | |
880 | { | |
881 | u64 mhash = 0; | |
882 | u64 uhash = 0; | |
883 | bool free = true; | |
42eb59d3 | 884 | unsigned int offset, naddr; |
be839e39 | 885 | const u8 *addr[7]; |
42eb59d3 | 886 | int ret; |
be839e39 CL |
887 | const struct port_info *pi = netdev_priv(dev); |
888 | ||
889 | /* first do the secondary unicast addresses */ | |
42eb59d3 CL |
890 | for (offset = 0; ; offset += naddr) { |
891 | naddr = collect_netdev_uc_list_addrs(dev, addr, offset, | |
892 | ARRAY_SIZE(addr)); | |
893 | if (naddr == 0) | |
894 | break; | |
895 | ||
be839e39 | 896 | ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free, |
42eb59d3 | 897 | naddr, addr, NULL, &uhash, sleep); |
be839e39 CL |
898 | if (ret < 0) |
899 | return ret; | |
900 | ||
901 | free = false; | |
902 | } | |
903 | ||
904 | /* next set up the multicast addresses */ | |
42eb59d3 CL |
905 | for (offset = 0; ; offset += naddr) { |
906 | naddr = collect_netdev_mc_list_addrs(dev, addr, offset, | |
907 | ARRAY_SIZE(addr)); | |
908 | if (naddr == 0) | |
909 | break; | |
910 | ||
be839e39 | 911 | ret = t4vf_alloc_mac_filt(pi->adapter, pi->viid, free, |
42eb59d3 | 912 | naddr, addr, NULL, &mhash, sleep); |
be839e39 CL |
913 | if (ret < 0) |
914 | return ret; | |
42eb59d3 | 915 | free = false; |
be839e39 CL |
916 | } |
917 | ||
918 | return t4vf_set_addr_hash(pi->adapter, pi->viid, uhash != 0, | |
919 | uhash | mhash, sleep); | |
920 | } | |
921 | ||
922 | /* | |
923 | * Set RX properties of a port, such as promiscruity, address filters, and MTU. | |
924 | * If @mtu is -1 it is left unchanged. | |
925 | */ | |
926 | static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok) | |
927 | { | |
928 | int ret; | |
929 | struct port_info *pi = netdev_priv(dev); | |
930 | ||
931 | ret = set_addr_filters(dev, sleep_ok); | |
932 | if (ret == 0) | |
933 | ret = t4vf_set_rxmode(pi->adapter, pi->viid, -1, | |
934 | (dev->flags & IFF_PROMISC) != 0, | |
935 | (dev->flags & IFF_ALLMULTI) != 0, | |
936 | 1, -1, sleep_ok); | |
937 | return ret; | |
938 | } | |
939 | ||
940 | /* | |
941 | * Set the current receive modes on the device. | |
942 | */ | |
943 | static void cxgb4vf_set_rxmode(struct net_device *dev) | |
944 | { | |
945 | /* unfortunately we can't return errors to the stack */ | |
946 | set_rxmode(dev, -1, false); | |
947 | } | |
948 | ||
949 | /* | |
950 | * Find the entry in the interrupt holdoff timer value array which comes | |
951 | * closest to the specified interrupt holdoff value. | |
952 | */ | |
953 | static int closest_timer(const struct sge *s, int us) | |
954 | { | |
955 | int i, timer_idx = 0, min_delta = INT_MAX; | |
956 | ||
957 | for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) { | |
958 | int delta = us - s->timer_val[i]; | |
959 | if (delta < 0) | |
960 | delta = -delta; | |
961 | if (delta < min_delta) { | |
962 | min_delta = delta; | |
963 | timer_idx = i; | |
964 | } | |
965 | } | |
966 | return timer_idx; | |
967 | } | |
968 | ||
969 | static int closest_thres(const struct sge *s, int thres) | |
970 | { | |
971 | int i, delta, pktcnt_idx = 0, min_delta = INT_MAX; | |
972 | ||
973 | for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) { | |
974 | delta = thres - s->counter_val[i]; | |
975 | if (delta < 0) | |
976 | delta = -delta; | |
977 | if (delta < min_delta) { | |
978 | min_delta = delta; | |
979 | pktcnt_idx = i; | |
980 | } | |
981 | } | |
982 | return pktcnt_idx; | |
983 | } | |
984 | ||
985 | /* | |
986 | * Return a queue's interrupt hold-off time in us. 0 means no timer. | |
987 | */ | |
988 | static unsigned int qtimer_val(const struct adapter *adapter, | |
989 | const struct sge_rspq *rspq) | |
990 | { | |
991 | unsigned int timer_idx = QINTR_TIMER_IDX_GET(rspq->intr_params); | |
992 | ||
993 | return timer_idx < SGE_NTIMERS | |
994 | ? adapter->sge.timer_val[timer_idx] | |
995 | : 0; | |
996 | } | |
997 | ||
998 | /** | |
999 | * set_rxq_intr_params - set a queue's interrupt holdoff parameters | |
1000 | * @adapter: the adapter | |
1001 | * @rspq: the RX response queue | |
1002 | * @us: the hold-off time in us, or 0 to disable timer | |
1003 | * @cnt: the hold-off packet count, or 0 to disable counter | |
1004 | * | |
1005 | * Sets an RX response queue's interrupt hold-off time and packet count. | |
1006 | * At least one of the two needs to be enabled for the queue to generate | |
1007 | * interrupts. | |
1008 | */ | |
1009 | static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq, | |
1010 | unsigned int us, unsigned int cnt) | |
1011 | { | |
1012 | unsigned int timer_idx; | |
1013 | ||
1014 | /* | |
1015 | * If both the interrupt holdoff timer and count are specified as | |
1016 | * zero, default to a holdoff count of 1 ... | |
1017 | */ | |
1018 | if ((us | cnt) == 0) | |
1019 | cnt = 1; | |
1020 | ||
1021 | /* | |
1022 | * If an interrupt holdoff count has been specified, then find the | |
1023 | * closest configured holdoff count and use that. If the response | |
1024 | * queue has already been created, then update its queue context | |
1025 | * parameters ... | |
1026 | */ | |
1027 | if (cnt) { | |
1028 | int err; | |
1029 | u32 v, pktcnt_idx; | |
1030 | ||
1031 | pktcnt_idx = closest_thres(&adapter->sge, cnt); | |
1032 | if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) { | |
5167865a HS |
1033 | v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) | |
1034 | FW_PARAMS_PARAM_X_V( | |
be839e39 | 1035 | FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) | |
5167865a | 1036 | FW_PARAMS_PARAM_YZ_V(rspq->cntxt_id); |
be839e39 CL |
1037 | err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx); |
1038 | if (err) | |
1039 | return err; | |
1040 | } | |
1041 | rspq->pktcnt_idx = pktcnt_idx; | |
1042 | } | |
1043 | ||
1044 | /* | |
1045 | * Compute the closest holdoff timer index from the supplied holdoff | |
1046 | * timer value. | |
1047 | */ | |
1048 | timer_idx = (us == 0 | |
1049 | ? SGE_TIMER_RSTRT_CNTR | |
1050 | : closest_timer(&adapter->sge, us)); | |
1051 | ||
1052 | /* | |
1053 | * Update the response queue's interrupt coalescing parameters and | |
1054 | * return success. | |
1055 | */ | |
1056 | rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) | | |
1057 | (cnt > 0 ? QINTR_CNT_EN : 0)); | |
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Return a version number to identify the type of adapter. The scheme is: | |
1063 | * - bits 0..9: chip version | |
1064 | * - bits 10..15: chip revision | |
1065 | */ | |
1066 | static inline unsigned int mk_adap_vers(const struct adapter *adapter) | |
1067 | { | |
1068 | /* | |
1069 | * Chip version 4, revision 0x3f (cxgb4vf). | |
1070 | */ | |
70ee3666 | 1071 | return CHELSIO_CHIP_VERSION(adapter->params.chip) | (0x3f << 10); |
be839e39 CL |
1072 | } |
1073 | ||
1074 | /* | |
1075 | * Execute the specified ioctl command. | |
1076 | */ | |
1077 | static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd) | |
1078 | { | |
1079 | int ret = 0; | |
1080 | ||
1081 | switch (cmd) { | |
1082 | /* | |
1083 | * The VF Driver doesn't have access to any of the other | |
1084 | * common Ethernet device ioctl()'s (like reading/writing | |
1085 | * PHY registers, etc. | |
1086 | */ | |
1087 | ||
1088 | default: | |
1089 | ret = -EOPNOTSUPP; | |
1090 | break; | |
1091 | } | |
1092 | return ret; | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * Change the device's MTU. | |
1097 | */ | |
1098 | static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu) | |
1099 | { | |
1100 | int ret; | |
1101 | struct port_info *pi = netdev_priv(dev); | |
1102 | ||
1103 | /* accommodate SACK */ | |
1104 | if (new_mtu < 81) | |
1105 | return -EINVAL; | |
1106 | ||
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 | ||
1196 | /* | |
1197 | * Return current port link settings. | |
1198 | */ | |
1199 | static int cxgb4vf_get_settings(struct net_device *dev, | |
1200 | struct ethtool_cmd *cmd) | |
1201 | { | |
1202 | const struct port_info *pi = netdev_priv(dev); | |
1203 | ||
1204 | cmd->supported = pi->link_cfg.supported; | |
1205 | cmd->advertising = pi->link_cfg.advertising; | |
70739497 DD |
1206 | ethtool_cmd_speed_set(cmd, |
1207 | netif_carrier_ok(dev) ? pi->link_cfg.speed : -1); | |
be839e39 CL |
1208 | cmd->duplex = DUPLEX_FULL; |
1209 | ||
1210 | cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE; | |
1211 | cmd->phy_address = pi->port_id; | |
1212 | cmd->transceiver = XCVR_EXTERNAL; | |
1213 | cmd->autoneg = pi->link_cfg.autoneg; | |
1214 | cmd->maxtxpkt = 0; | |
1215 | cmd->maxrxpkt = 0; | |
1216 | return 0; | |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * Return our driver information. | |
1221 | */ | |
1222 | static void cxgb4vf_get_drvinfo(struct net_device *dev, | |
1223 | struct ethtool_drvinfo *drvinfo) | |
1224 | { | |
1225 | struct adapter *adapter = netdev2adap(dev); | |
1226 | ||
23020ab3 RJ |
1227 | strlcpy(drvinfo->driver, KBUILD_MODNAME, sizeof(drvinfo->driver)); |
1228 | strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); | |
1229 | strlcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)), | |
1230 | sizeof(drvinfo->bus_info)); | |
be839e39 CL |
1231 | snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), |
1232 | "%u.%u.%u.%u, TP %u.%u.%u.%u", | |
b2e1a3f0 HS |
1233 | FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.fwrev), |
1234 | FW_HDR_FW_VER_MINOR_G(adapter->params.dev.fwrev), | |
1235 | FW_HDR_FW_VER_MICRO_G(adapter->params.dev.fwrev), | |
1236 | FW_HDR_FW_VER_BUILD_G(adapter->params.dev.fwrev), | |
1237 | FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.tprev), | |
1238 | FW_HDR_FW_VER_MINOR_G(adapter->params.dev.tprev), | |
1239 | FW_HDR_FW_VER_MICRO_G(adapter->params.dev.tprev), | |
1240 | FW_HDR_FW_VER_BUILD_G(adapter->params.dev.tprev)); | |
be839e39 CL |
1241 | } |
1242 | ||
1243 | /* | |
1244 | * Return current adapter message level. | |
1245 | */ | |
1246 | static u32 cxgb4vf_get_msglevel(struct net_device *dev) | |
1247 | { | |
1248 | return netdev2adap(dev)->msg_enable; | |
1249 | } | |
1250 | ||
1251 | /* | |
1252 | * Set current adapter message level. | |
1253 | */ | |
1254 | static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel) | |
1255 | { | |
1256 | netdev2adap(dev)->msg_enable = msglevel; | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Return the device's current Queue Set ring size parameters along with the | |
1261 | * allowed maximum values. Since ethtool doesn't understand the concept of | |
1262 | * multi-queue devices, we just return the current values associated with the | |
1263 | * first Queue Set. | |
1264 | */ | |
1265 | static void cxgb4vf_get_ringparam(struct net_device *dev, | |
1266 | struct ethtool_ringparam *rp) | |
1267 | { | |
1268 | const struct port_info *pi = netdev_priv(dev); | |
1269 | const struct sge *s = &pi->adapter->sge; | |
1270 | ||
1271 | rp->rx_max_pending = MAX_RX_BUFFERS; | |
1272 | rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES; | |
1273 | rp->rx_jumbo_max_pending = 0; | |
1274 | rp->tx_max_pending = MAX_TXQ_ENTRIES; | |
1275 | ||
1276 | rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID; | |
1277 | rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size; | |
1278 | rp->rx_jumbo_pending = 0; | |
1279 | rp->tx_pending = s->ethtxq[pi->first_qset].q.size; | |
1280 | } | |
1281 | ||
1282 | /* | |
1283 | * Set the Queue Set ring size parameters for the device. Again, since | |
1284 | * ethtool doesn't allow for the concept of multiple queues per device, we'll | |
1285 | * apply these new values across all of the Queue Sets associated with the | |
1286 | * device -- after vetting them of course! | |
1287 | */ | |
1288 | static int cxgb4vf_set_ringparam(struct net_device *dev, | |
1289 | struct ethtool_ringparam *rp) | |
1290 | { | |
1291 | const struct port_info *pi = netdev_priv(dev); | |
1292 | struct adapter *adapter = pi->adapter; | |
1293 | struct sge *s = &adapter->sge; | |
1294 | int qs; | |
1295 | ||
1296 | if (rp->rx_pending > MAX_RX_BUFFERS || | |
1297 | rp->rx_jumbo_pending || | |
1298 | rp->tx_pending > MAX_TXQ_ENTRIES || | |
1299 | rp->rx_mini_pending > MAX_RSPQ_ENTRIES || | |
1300 | rp->rx_mini_pending < MIN_RSPQ_ENTRIES || | |
1301 | rp->rx_pending < MIN_FL_ENTRIES || | |
1302 | rp->tx_pending < MIN_TXQ_ENTRIES) | |
1303 | return -EINVAL; | |
1304 | ||
1305 | if (adapter->flags & FULL_INIT_DONE) | |
1306 | return -EBUSY; | |
1307 | ||
1308 | for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) { | |
1309 | s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID; | |
1310 | s->ethrxq[qs].rspq.size = rp->rx_mini_pending; | |
1311 | s->ethtxq[qs].q.size = rp->tx_pending; | |
1312 | } | |
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * Return the interrupt holdoff timer and count for the first Queue Set on the | |
1318 | * device. Our extension ioctl() (the cxgbtool interface) allows the | |
1319 | * interrupt holdoff timer to be read on all of the device's Queue Sets. | |
1320 | */ | |
1321 | static int cxgb4vf_get_coalesce(struct net_device *dev, | |
1322 | struct ethtool_coalesce *coalesce) | |
1323 | { | |
1324 | const struct port_info *pi = netdev_priv(dev); | |
1325 | const struct adapter *adapter = pi->adapter; | |
1326 | const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq; | |
1327 | ||
1328 | coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq); | |
1329 | coalesce->rx_max_coalesced_frames = | |
1330 | ((rspq->intr_params & QINTR_CNT_EN) | |
1331 | ? adapter->sge.counter_val[rspq->pktcnt_idx] | |
1332 | : 0); | |
1333 | return 0; | |
1334 | } | |
1335 | ||
1336 | /* | |
1337 | * Set the RX interrupt holdoff timer and count for the first Queue Set on the | |
1338 | * interface. Our extension ioctl() (the cxgbtool interface) allows us to set | |
1339 | * the interrupt holdoff timer on any of the device's Queue Sets. | |
1340 | */ | |
1341 | static int cxgb4vf_set_coalesce(struct net_device *dev, | |
1342 | struct ethtool_coalesce *coalesce) | |
1343 | { | |
1344 | const struct port_info *pi = netdev_priv(dev); | |
1345 | struct adapter *adapter = pi->adapter; | |
1346 | ||
1347 | return set_rxq_intr_params(adapter, | |
1348 | &adapter->sge.ethrxq[pi->first_qset].rspq, | |
1349 | coalesce->rx_coalesce_usecs, | |
1350 | coalesce->rx_max_coalesced_frames); | |
1351 | } | |
1352 | ||
1353 | /* | |
1354 | * Report current port link pause parameter settings. | |
1355 | */ | |
1356 | static void cxgb4vf_get_pauseparam(struct net_device *dev, | |
1357 | struct ethtool_pauseparam *pauseparam) | |
1358 | { | |
1359 | struct port_info *pi = netdev_priv(dev); | |
1360 | ||
1361 | pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0; | |
1362 | pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0; | |
1363 | pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0; | |
1364 | } | |
1365 | ||
be839e39 CL |
1366 | /* |
1367 | * Identify the port by blinking the port's LED. | |
1368 | */ | |
857a3d0f DM |
1369 | static int cxgb4vf_phys_id(struct net_device *dev, |
1370 | enum ethtool_phys_id_state state) | |
be839e39 | 1371 | { |
857a3d0f | 1372 | unsigned int val; |
be839e39 CL |
1373 | struct port_info *pi = netdev_priv(dev); |
1374 | ||
857a3d0f DM |
1375 | if (state == ETHTOOL_ID_ACTIVE) |
1376 | val = 0xffff; | |
1377 | else if (state == ETHTOOL_ID_INACTIVE) | |
1378 | val = 0; | |
1379 | else | |
1380 | return -EINVAL; | |
1381 | ||
1382 | return t4vf_identify_port(pi->adapter, pi->viid, val); | |
be839e39 CL |
1383 | } |
1384 | ||
1385 | /* | |
1386 | * Port stats maintained per queue of the port. | |
1387 | */ | |
1388 | struct queue_port_stats { | |
1389 | u64 tso; | |
1390 | u64 tx_csum; | |
1391 | u64 rx_csum; | |
1392 | u64 vlan_ex; | |
1393 | u64 vlan_ins; | |
f12fe353 CL |
1394 | u64 lro_pkts; |
1395 | u64 lro_merged; | |
be839e39 CL |
1396 | }; |
1397 | ||
1398 | /* | |
1399 | * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that | |
1400 | * these need to match the order of statistics returned by | |
1401 | * t4vf_get_port_stats(). | |
1402 | */ | |
1403 | static const char stats_strings[][ETH_GSTRING_LEN] = { | |
1404 | /* | |
1405 | * These must match the layout of the t4vf_port_stats structure. | |
1406 | */ | |
1407 | "TxBroadcastBytes ", | |
1408 | "TxBroadcastFrames ", | |
1409 | "TxMulticastBytes ", | |
1410 | "TxMulticastFrames ", | |
1411 | "TxUnicastBytes ", | |
1412 | "TxUnicastFrames ", | |
1413 | "TxDroppedFrames ", | |
1414 | "TxOffloadBytes ", | |
1415 | "TxOffloadFrames ", | |
1416 | "RxBroadcastBytes ", | |
1417 | "RxBroadcastFrames ", | |
1418 | "RxMulticastBytes ", | |
1419 | "RxMulticastFrames ", | |
1420 | "RxUnicastBytes ", | |
1421 | "RxUnicastFrames ", | |
1422 | "RxErrorFrames ", | |
1423 | ||
1424 | /* | |
1425 | * These are accumulated per-queue statistics and must match the | |
1426 | * order of the fields in the queue_port_stats structure. | |
1427 | */ | |
1428 | "TSO ", | |
1429 | "TxCsumOffload ", | |
1430 | "RxCsumGood ", | |
1431 | "VLANextractions ", | |
1432 | "VLANinsertions ", | |
f12fe353 CL |
1433 | "GROPackets ", |
1434 | "GROMerged ", | |
be839e39 CL |
1435 | }; |
1436 | ||
1437 | /* | |
1438 | * Return the number of statistics in the specified statistics set. | |
1439 | */ | |
1440 | static int cxgb4vf_get_sset_count(struct net_device *dev, int sset) | |
1441 | { | |
1442 | switch (sset) { | |
1443 | case ETH_SS_STATS: | |
1444 | return ARRAY_SIZE(stats_strings); | |
1445 | default: | |
1446 | return -EOPNOTSUPP; | |
1447 | } | |
1448 | /*NOTREACHED*/ | |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * Return the strings for the specified statistics set. | |
1453 | */ | |
1454 | static void cxgb4vf_get_strings(struct net_device *dev, | |
1455 | u32 sset, | |
1456 | u8 *data) | |
1457 | { | |
1458 | switch (sset) { | |
1459 | case ETH_SS_STATS: | |
1460 | memcpy(data, stats_strings, sizeof(stats_strings)); | |
1461 | break; | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * Small utility routine to accumulate queue statistics across the queues of | |
1467 | * a "port". | |
1468 | */ | |
1469 | static void collect_sge_port_stats(const struct adapter *adapter, | |
1470 | const struct port_info *pi, | |
1471 | struct queue_port_stats *stats) | |
1472 | { | |
1473 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset]; | |
1474 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset]; | |
1475 | int qs; | |
1476 | ||
1477 | memset(stats, 0, sizeof(*stats)); | |
1478 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { | |
1479 | stats->tso += txq->tso; | |
1480 | stats->tx_csum += txq->tx_cso; | |
1481 | stats->rx_csum += rxq->stats.rx_cso; | |
1482 | stats->vlan_ex += rxq->stats.vlan_ex; | |
1483 | stats->vlan_ins += txq->vlan_ins; | |
f12fe353 CL |
1484 | stats->lro_pkts += rxq->stats.lro_pkts; |
1485 | stats->lro_merged += rxq->stats.lro_merged; | |
be839e39 CL |
1486 | } |
1487 | } | |
1488 | ||
1489 | /* | |
1490 | * Return the ETH_SS_STATS statistics set. | |
1491 | */ | |
1492 | static void cxgb4vf_get_ethtool_stats(struct net_device *dev, | |
1493 | struct ethtool_stats *stats, | |
1494 | u64 *data) | |
1495 | { | |
1496 | struct port_info *pi = netdev2pinfo(dev); | |
1497 | struct adapter *adapter = pi->adapter; | |
1498 | int err = t4vf_get_port_stats(adapter, pi->pidx, | |
1499 | (struct t4vf_port_stats *)data); | |
1500 | if (err) | |
1501 | memset(data, 0, sizeof(struct t4vf_port_stats)); | |
1502 | ||
1503 | data += sizeof(struct t4vf_port_stats) / sizeof(u64); | |
1504 | collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data); | |
1505 | } | |
1506 | ||
1507 | /* | |
1508 | * Return the size of our register map. | |
1509 | */ | |
1510 | static int cxgb4vf_get_regs_len(struct net_device *dev) | |
1511 | { | |
1512 | return T4VF_REGMAP_SIZE; | |
1513 | } | |
1514 | ||
1515 | /* | |
1516 | * Dump a block of registers, start to end inclusive, into a buffer. | |
1517 | */ | |
1518 | static void reg_block_dump(struct adapter *adapter, void *regbuf, | |
1519 | unsigned int start, unsigned int end) | |
1520 | { | |
1521 | u32 *bp = regbuf + start - T4VF_REGMAP_START; | |
1522 | ||
1523 | for ( ; start <= end; start += sizeof(u32)) { | |
1524 | /* | |
1525 | * Avoid reading the Mailbox Control register since that | |
1526 | * can trigger a Mailbox Ownership Arbitration cycle and | |
1527 | * interfere with communication with the firmware. | |
1528 | */ | |
1529 | if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL) | |
1530 | *bp++ = 0xffff; | |
1531 | else | |
1532 | *bp++ = t4_read_reg(adapter, start); | |
1533 | } | |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * Copy our entire register map into the provided buffer. | |
1538 | */ | |
1539 | static void cxgb4vf_get_regs(struct net_device *dev, | |
1540 | struct ethtool_regs *regs, | |
1541 | void *regbuf) | |
1542 | { | |
1543 | struct adapter *adapter = netdev2adap(dev); | |
1544 | ||
1545 | regs->version = mk_adap_vers(adapter); | |
1546 | ||
1547 | /* | |
1548 | * Fill in register buffer with our register map. | |
1549 | */ | |
1550 | memset(regbuf, 0, T4VF_REGMAP_SIZE); | |
1551 | ||
1552 | reg_block_dump(adapter, regbuf, | |
1553 | T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST, | |
1554 | T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST); | |
1555 | reg_block_dump(adapter, regbuf, | |
1556 | T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST, | |
1557 | T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST); | |
70ee3666 HS |
1558 | |
1559 | /* T5 adds new registers in the PL Register map. | |
1560 | */ | |
be839e39 CL |
1561 | reg_block_dump(adapter, regbuf, |
1562 | T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST, | |
70ee3666 HS |
1563 | T4VF_PL_BASE_ADDR + (is_t4(adapter->params.chip) |
1564 | ? A_PL_VF_WHOAMI : A_PL_VF_REVISION)); | |
be839e39 CL |
1565 | reg_block_dump(adapter, regbuf, |
1566 | T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST, | |
1567 | T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST); | |
1568 | ||
1569 | reg_block_dump(adapter, regbuf, | |
1570 | T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST, | |
1571 | T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST); | |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * Report current Wake On LAN settings. | |
1576 | */ | |
1577 | static void cxgb4vf_get_wol(struct net_device *dev, | |
1578 | struct ethtool_wolinfo *wol) | |
1579 | { | |
1580 | wol->supported = 0; | |
1581 | wol->wolopts = 0; | |
1582 | memset(&wol->sopass, 0, sizeof(wol->sopass)); | |
1583 | } | |
1584 | ||
410989f6 CL |
1585 | /* |
1586 | * TCP Segmentation Offload flags which we support. | |
1587 | */ | |
1588 | #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN) | |
1589 | ||
9b07be4b | 1590 | static const struct ethtool_ops cxgb4vf_ethtool_ops = { |
be839e39 CL |
1591 | .get_settings = cxgb4vf_get_settings, |
1592 | .get_drvinfo = cxgb4vf_get_drvinfo, | |
1593 | .get_msglevel = cxgb4vf_get_msglevel, | |
1594 | .set_msglevel = cxgb4vf_set_msglevel, | |
1595 | .get_ringparam = cxgb4vf_get_ringparam, | |
1596 | .set_ringparam = cxgb4vf_set_ringparam, | |
1597 | .get_coalesce = cxgb4vf_get_coalesce, | |
1598 | .set_coalesce = cxgb4vf_set_coalesce, | |
1599 | .get_pauseparam = cxgb4vf_get_pauseparam, | |
be839e39 CL |
1600 | .get_link = ethtool_op_get_link, |
1601 | .get_strings = cxgb4vf_get_strings, | |
857a3d0f | 1602 | .set_phys_id = cxgb4vf_phys_id, |
be839e39 CL |
1603 | .get_sset_count = cxgb4vf_get_sset_count, |
1604 | .get_ethtool_stats = cxgb4vf_get_ethtool_stats, | |
1605 | .get_regs_len = cxgb4vf_get_regs_len, | |
1606 | .get_regs = cxgb4vf_get_regs, | |
1607 | .get_wol = cxgb4vf_get_wol, | |
be839e39 CL |
1608 | }; |
1609 | ||
1610 | /* | |
1611 | * /sys/kernel/debug/cxgb4vf support code and data. | |
1612 | * ================================================ | |
1613 | */ | |
1614 | ||
1615 | /* | |
1616 | * Show SGE Queue Set information. We display QPL Queues Sets per line. | |
1617 | */ | |
1618 | #define QPL 4 | |
1619 | ||
1620 | static int sge_qinfo_show(struct seq_file *seq, void *v) | |
1621 | { | |
1622 | struct adapter *adapter = seq->private; | |
1623 | int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL); | |
1624 | int qs, r = (uintptr_t)v - 1; | |
1625 | ||
1626 | if (r) | |
1627 | seq_putc(seq, '\n'); | |
1628 | ||
1629 | #define S3(fmt_spec, s, v) \ | |
1630 | do {\ | |
1631 | seq_printf(seq, "%-12s", s); \ | |
1632 | for (qs = 0; qs < n; ++qs) \ | |
1633 | seq_printf(seq, " %16" fmt_spec, v); \ | |
1634 | seq_putc(seq, '\n'); \ | |
1635 | } while (0) | |
1636 | #define S(s, v) S3("s", s, v) | |
1637 | #define T(s, v) S3("u", s, txq[qs].v) | |
1638 | #define R(s, v) S3("u", s, rxq[qs].v) | |
1639 | ||
1640 | if (r < eth_entries) { | |
1641 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL]; | |
1642 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL]; | |
1643 | int n = min(QPL, adapter->sge.ethqsets - QPL * r); | |
1644 | ||
1645 | S("QType:", "Ethernet"); | |
1646 | S("Interface:", | |
1647 | (rxq[qs].rspq.netdev | |
1648 | ? rxq[qs].rspq.netdev->name | |
1649 | : "N/A")); | |
1650 | S3("d", "Port:", | |
1651 | (rxq[qs].rspq.netdev | |
1652 | ? ((struct port_info *) | |
1653 | netdev_priv(rxq[qs].rspq.netdev))->port_id | |
1654 | : -1)); | |
1655 | T("TxQ ID:", q.abs_id); | |
1656 | T("TxQ size:", q.size); | |
1657 | T("TxQ inuse:", q.in_use); | |
1658 | T("TxQ PIdx:", q.pidx); | |
1659 | T("TxQ CIdx:", q.cidx); | |
1660 | R("RspQ ID:", rspq.abs_id); | |
1661 | R("RspQ size:", rspq.size); | |
1662 | R("RspQE size:", rspq.iqe_len); | |
1663 | S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq)); | |
1664 | S3("u", "Intr pktcnt:", | |
1665 | adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]); | |
1666 | R("RspQ CIdx:", rspq.cidx); | |
1667 | R("RspQ Gen:", rspq.gen); | |
1668 | R("FL ID:", fl.abs_id); | |
1669 | R("FL size:", fl.size - MIN_FL_RESID); | |
1670 | R("FL avail:", fl.avail); | |
1671 | R("FL PIdx:", fl.pidx); | |
1672 | R("FL CIdx:", fl.cidx); | |
1673 | return 0; | |
1674 | } | |
1675 | ||
1676 | r -= eth_entries; | |
1677 | if (r == 0) { | |
1678 | const struct sge_rspq *evtq = &adapter->sge.fw_evtq; | |
1679 | ||
1680 | seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue"); | |
1681 | seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id); | |
1682 | seq_printf(seq, "%-12s %16u\n", "Intr delay:", | |
1683 | qtimer_val(adapter, evtq)); | |
1684 | seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:", | |
1685 | adapter->sge.counter_val[evtq->pktcnt_idx]); | |
1686 | seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx); | |
1687 | seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen); | |
1688 | } else if (r == 1) { | |
1689 | const struct sge_rspq *intrq = &adapter->sge.intrq; | |
1690 | ||
1691 | seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue"); | |
1692 | seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id); | |
1693 | seq_printf(seq, "%-12s %16u\n", "Intr delay:", | |
1694 | qtimer_val(adapter, intrq)); | |
1695 | seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:", | |
1696 | adapter->sge.counter_val[intrq->pktcnt_idx]); | |
1697 | seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx); | |
1698 | seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen); | |
1699 | } | |
1700 | ||
1701 | #undef R | |
1702 | #undef T | |
1703 | #undef S | |
1704 | #undef S3 | |
1705 | ||
1706 | return 0; | |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * Return the number of "entries" in our "file". We group the multi-Queue | |
1711 | * sections with QPL Queue Sets per "entry". The sections of the output are: | |
1712 | * | |
1713 | * Ethernet RX/TX Queue Sets | |
1714 | * Firmware Event Queue | |
1715 | * Forwarded Interrupt Queue (if in MSI mode) | |
1716 | */ | |
1717 | static int sge_queue_entries(const struct adapter *adapter) | |
1718 | { | |
1719 | return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 + | |
1720 | ((adapter->flags & USING_MSI) != 0); | |
1721 | } | |
1722 | ||
1723 | static void *sge_queue_start(struct seq_file *seq, loff_t *pos) | |
1724 | { | |
1725 | int entries = sge_queue_entries(seq->private); | |
1726 | ||
1727 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
1728 | } | |
1729 | ||
1730 | static void sge_queue_stop(struct seq_file *seq, void *v) | |
1731 | { | |
1732 | } | |
1733 | ||
1734 | static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos) | |
1735 | { | |
1736 | int entries = sge_queue_entries(seq->private); | |
1737 | ||
1738 | ++*pos; | |
1739 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
1740 | } | |
1741 | ||
1742 | static const struct seq_operations sge_qinfo_seq_ops = { | |
1743 | .start = sge_queue_start, | |
1744 | .next = sge_queue_next, | |
1745 | .stop = sge_queue_stop, | |
1746 | .show = sge_qinfo_show | |
1747 | }; | |
1748 | ||
1749 | static int sge_qinfo_open(struct inode *inode, struct file *file) | |
1750 | { | |
1751 | int res = seq_open(file, &sge_qinfo_seq_ops); | |
1752 | ||
1753 | if (!res) { | |
1754 | struct seq_file *seq = file->private_data; | |
1755 | seq->private = inode->i_private; | |
1756 | } | |
1757 | return res; | |
1758 | } | |
1759 | ||
1760 | static const struct file_operations sge_qinfo_debugfs_fops = { | |
1761 | .owner = THIS_MODULE, | |
1762 | .open = sge_qinfo_open, | |
1763 | .read = seq_read, | |
1764 | .llseek = seq_lseek, | |
1765 | .release = seq_release, | |
1766 | }; | |
1767 | ||
1768 | /* | |
1769 | * Show SGE Queue Set statistics. We display QPL Queues Sets per line. | |
1770 | */ | |
1771 | #define QPL 4 | |
1772 | ||
1773 | static int sge_qstats_show(struct seq_file *seq, void *v) | |
1774 | { | |
1775 | struct adapter *adapter = seq->private; | |
1776 | int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL); | |
1777 | int qs, r = (uintptr_t)v - 1; | |
1778 | ||
1779 | if (r) | |
1780 | seq_putc(seq, '\n'); | |
1781 | ||
1782 | #define S3(fmt, s, v) \ | |
1783 | do { \ | |
1784 | seq_printf(seq, "%-16s", s); \ | |
1785 | for (qs = 0; qs < n; ++qs) \ | |
1786 | seq_printf(seq, " %8" fmt, v); \ | |
1787 | seq_putc(seq, '\n'); \ | |
1788 | } while (0) | |
1789 | #define S(s, v) S3("s", s, v) | |
1790 | ||
1791 | #define T3(fmt, s, v) S3(fmt, s, txq[qs].v) | |
1792 | #define T(s, v) T3("lu", s, v) | |
1793 | ||
1794 | #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v) | |
1795 | #define R(s, v) R3("lu", s, v) | |
1796 | ||
1797 | if (r < eth_entries) { | |
1798 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL]; | |
1799 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL]; | |
1800 | int n = min(QPL, adapter->sge.ethqsets - QPL * r); | |
1801 | ||
1802 | S("QType:", "Ethernet"); | |
1803 | S("Interface:", | |
1804 | (rxq[qs].rspq.netdev | |
1805 | ? rxq[qs].rspq.netdev->name | |
1806 | : "N/A")); | |
68dc9d36 | 1807 | R3("u", "RspQNullInts:", rspq.unhandled_irqs); |
be839e39 CL |
1808 | R("RxPackets:", stats.pkts); |
1809 | R("RxCSO:", stats.rx_cso); | |
1810 | R("VLANxtract:", stats.vlan_ex); | |
1811 | R("LROmerged:", stats.lro_merged); | |
1812 | R("LROpackets:", stats.lro_pkts); | |
1813 | R("RxDrops:", stats.rx_drops); | |
1814 | T("TSO:", tso); | |
1815 | T("TxCSO:", tx_cso); | |
1816 | T("VLANins:", vlan_ins); | |
1817 | T("TxQFull:", q.stops); | |
1818 | T("TxQRestarts:", q.restarts); | |
1819 | T("TxMapErr:", mapping_err); | |
1820 | R("FLAllocErr:", fl.alloc_failed); | |
1821 | R("FLLrgAlcErr:", fl.large_alloc_failed); | |
1822 | R("FLStarving:", fl.starving); | |
1823 | return 0; | |
1824 | } | |
1825 | ||
1826 | r -= eth_entries; | |
1827 | if (r == 0) { | |
1828 | const struct sge_rspq *evtq = &adapter->sge.fw_evtq; | |
1829 | ||
1830 | seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue"); | |
68dc9d36 CL |
1831 | seq_printf(seq, "%-16s %8u\n", "RspQNullInts:", |
1832 | evtq->unhandled_irqs); | |
be839e39 CL |
1833 | seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx); |
1834 | seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen); | |
1835 | } else if (r == 1) { | |
1836 | const struct sge_rspq *intrq = &adapter->sge.intrq; | |
1837 | ||
1838 | seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue"); | |
68dc9d36 CL |
1839 | seq_printf(seq, "%-16s %8u\n", "RspQNullInts:", |
1840 | intrq->unhandled_irqs); | |
be839e39 CL |
1841 | seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx); |
1842 | seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen); | |
1843 | } | |
1844 | ||
1845 | #undef R | |
1846 | #undef T | |
1847 | #undef S | |
1848 | #undef R3 | |
1849 | #undef T3 | |
1850 | #undef S3 | |
1851 | ||
1852 | return 0; | |
1853 | } | |
1854 | ||
1855 | /* | |
1856 | * Return the number of "entries" in our "file". We group the multi-Queue | |
1857 | * sections with QPL Queue Sets per "entry". The sections of the output are: | |
1858 | * | |
1859 | * Ethernet RX/TX Queue Sets | |
1860 | * Firmware Event Queue | |
1861 | * Forwarded Interrupt Queue (if in MSI mode) | |
1862 | */ | |
1863 | static int sge_qstats_entries(const struct adapter *adapter) | |
1864 | { | |
1865 | return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 + | |
1866 | ((adapter->flags & USING_MSI) != 0); | |
1867 | } | |
1868 | ||
1869 | static void *sge_qstats_start(struct seq_file *seq, loff_t *pos) | |
1870 | { | |
1871 | int entries = sge_qstats_entries(seq->private); | |
1872 | ||
1873 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
1874 | } | |
1875 | ||
1876 | static void sge_qstats_stop(struct seq_file *seq, void *v) | |
1877 | { | |
1878 | } | |
1879 | ||
1880 | static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos) | |
1881 | { | |
1882 | int entries = sge_qstats_entries(seq->private); | |
1883 | ||
1884 | (*pos)++; | |
1885 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
1886 | } | |
1887 | ||
1888 | static const struct seq_operations sge_qstats_seq_ops = { | |
1889 | .start = sge_qstats_start, | |
1890 | .next = sge_qstats_next, | |
1891 | .stop = sge_qstats_stop, | |
1892 | .show = sge_qstats_show | |
1893 | }; | |
1894 | ||
1895 | static int sge_qstats_open(struct inode *inode, struct file *file) | |
1896 | { | |
1897 | int res = seq_open(file, &sge_qstats_seq_ops); | |
1898 | ||
1899 | if (res == 0) { | |
1900 | struct seq_file *seq = file->private_data; | |
1901 | seq->private = inode->i_private; | |
1902 | } | |
1903 | return res; | |
1904 | } | |
1905 | ||
1906 | static const struct file_operations sge_qstats_proc_fops = { | |
1907 | .owner = THIS_MODULE, | |
1908 | .open = sge_qstats_open, | |
1909 | .read = seq_read, | |
1910 | .llseek = seq_lseek, | |
1911 | .release = seq_release, | |
1912 | }; | |
1913 | ||
1914 | /* | |
1915 | * Show PCI-E SR-IOV Virtual Function Resource Limits. | |
1916 | */ | |
1917 | static int resources_show(struct seq_file *seq, void *v) | |
1918 | { | |
1919 | struct adapter *adapter = seq->private; | |
1920 | struct vf_resources *vfres = &adapter->params.vfres; | |
1921 | ||
1922 | #define S(desc, fmt, var) \ | |
1923 | seq_printf(seq, "%-60s " fmt "\n", \ | |
1924 | desc " (" #var "):", vfres->var) | |
1925 | ||
1926 | S("Virtual Interfaces", "%d", nvi); | |
1927 | S("Egress Queues", "%d", neq); | |
1928 | S("Ethernet Control", "%d", nethctrl); | |
1929 | S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint); | |
1930 | S("Ingress Queues", "%d", niq); | |
1931 | S("Traffic Class", "%d", tc); | |
1932 | S("Port Access Rights Mask", "%#x", pmask); | |
1933 | S("MAC Address Filters", "%d", nexactf); | |
1934 | S("Firmware Command Read Capabilities", "%#x", r_caps); | |
1935 | S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps); | |
1936 | ||
1937 | #undef S | |
1938 | ||
1939 | return 0; | |
1940 | } | |
1941 | ||
1942 | static int resources_open(struct inode *inode, struct file *file) | |
1943 | { | |
1944 | return single_open(file, resources_show, inode->i_private); | |
1945 | } | |
1946 | ||
1947 | static const struct file_operations resources_proc_fops = { | |
1948 | .owner = THIS_MODULE, | |
1949 | .open = resources_open, | |
1950 | .read = seq_read, | |
1951 | .llseek = seq_lseek, | |
1952 | .release = single_release, | |
1953 | }; | |
1954 | ||
1955 | /* | |
1956 | * Show Virtual Interfaces. | |
1957 | */ | |
1958 | static int interfaces_show(struct seq_file *seq, void *v) | |
1959 | { | |
1960 | if (v == SEQ_START_TOKEN) { | |
1961 | seq_puts(seq, "Interface Port VIID\n"); | |
1962 | } else { | |
1963 | struct adapter *adapter = seq->private; | |
1964 | int pidx = (uintptr_t)v - 2; | |
1965 | struct net_device *dev = adapter->port[pidx]; | |
1966 | struct port_info *pi = netdev_priv(dev); | |
1967 | ||
1968 | seq_printf(seq, "%9s %4d %#5x\n", | |
1969 | dev->name, pi->port_id, pi->viid); | |
1970 | } | |
1971 | return 0; | |
1972 | } | |
1973 | ||
1974 | static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos) | |
1975 | { | |
1976 | return pos <= adapter->params.nports | |
1977 | ? (void *)(uintptr_t)(pos + 1) | |
1978 | : NULL; | |
1979 | } | |
1980 | ||
1981 | static void *interfaces_start(struct seq_file *seq, loff_t *pos) | |
1982 | { | |
1983 | return *pos | |
1984 | ? interfaces_get_idx(seq->private, *pos) | |
1985 | : SEQ_START_TOKEN; | |
1986 | } | |
1987 | ||
1988 | static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos) | |
1989 | { | |
1990 | (*pos)++; | |
1991 | return interfaces_get_idx(seq->private, *pos); | |
1992 | } | |
1993 | ||
1994 | static void interfaces_stop(struct seq_file *seq, void *v) | |
1995 | { | |
1996 | } | |
1997 | ||
1998 | static const struct seq_operations interfaces_seq_ops = { | |
1999 | .start = interfaces_start, | |
2000 | .next = interfaces_next, | |
2001 | .stop = interfaces_stop, | |
2002 | .show = interfaces_show | |
2003 | }; | |
2004 | ||
2005 | static int interfaces_open(struct inode *inode, struct file *file) | |
2006 | { | |
2007 | int res = seq_open(file, &interfaces_seq_ops); | |
2008 | ||
2009 | if (res == 0) { | |
2010 | struct seq_file *seq = file->private_data; | |
2011 | seq->private = inode->i_private; | |
2012 | } | |
2013 | return res; | |
2014 | } | |
2015 | ||
2016 | static const struct file_operations interfaces_proc_fops = { | |
2017 | .owner = THIS_MODULE, | |
2018 | .open = interfaces_open, | |
2019 | .read = seq_read, | |
2020 | .llseek = seq_lseek, | |
2021 | .release = seq_release, | |
2022 | }; | |
2023 | ||
2024 | /* | |
2025 | * /sys/kernel/debugfs/cxgb4vf/ files list. | |
2026 | */ | |
2027 | struct cxgb4vf_debugfs_entry { | |
2028 | const char *name; /* name of debugfs node */ | |
f4ae40a6 | 2029 | umode_t mode; /* file system mode */ |
be839e39 CL |
2030 | const struct file_operations *fops; |
2031 | }; | |
2032 | ||
2033 | static struct cxgb4vf_debugfs_entry debugfs_files[] = { | |
2034 | { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops }, | |
2035 | { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops }, | |
2036 | { "resources", S_IRUGO, &resources_proc_fops }, | |
2037 | { "interfaces", S_IRUGO, &interfaces_proc_fops }, | |
2038 | }; | |
2039 | ||
2040 | /* | |
2041 | * Module and device initialization and cleanup code. | |
2042 | * ================================================== | |
2043 | */ | |
2044 | ||
2045 | /* | |
2046 | * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the | |
2047 | * directory (debugfs_root) has already been set up. | |
2048 | */ | |
d289f864 | 2049 | static int setup_debugfs(struct adapter *adapter) |
be839e39 CL |
2050 | { |
2051 | int i; | |
2052 | ||
843635e0 | 2053 | BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root)); |
be839e39 CL |
2054 | |
2055 | /* | |
2056 | * Debugfs support is best effort. | |
2057 | */ | |
2058 | for (i = 0; i < ARRAY_SIZE(debugfs_files); i++) | |
2059 | (void)debugfs_create_file(debugfs_files[i].name, | |
2060 | debugfs_files[i].mode, | |
2061 | adapter->debugfs_root, | |
2062 | (void *)adapter, | |
2063 | debugfs_files[i].fops); | |
2064 | ||
2065 | return 0; | |
2066 | } | |
2067 | ||
2068 | /* | |
2069 | * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave | |
2070 | * it to our caller to tear down the directory (debugfs_root). | |
2071 | */ | |
4204875d | 2072 | static void cleanup_debugfs(struct adapter *adapter) |
be839e39 | 2073 | { |
843635e0 | 2074 | BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root)); |
be839e39 CL |
2075 | |
2076 | /* | |
2077 | * Unlike our sister routine cleanup_proc(), we don't need to remove | |
2078 | * individual entries because a call will be made to | |
2079 | * debugfs_remove_recursive(). We just need to clean up any ancillary | |
2080 | * persistent state. | |
2081 | */ | |
2082 | /* nothing to do */ | |
2083 | } | |
2084 | ||
2085 | /* | |
2086 | * Perform early "adapter" initialization. This is where we discover what | |
2087 | * adapter parameters we're going to be using and initialize basic adapter | |
2088 | * hardware support. | |
2089 | */ | |
d289f864 | 2090 | static int adap_init0(struct adapter *adapter) |
be839e39 CL |
2091 | { |
2092 | struct vf_resources *vfres = &adapter->params.vfres; | |
2093 | struct sge_params *sge_params = &adapter->params.sge; | |
2094 | struct sge *s = &adapter->sge; | |
2095 | unsigned int ethqsets; | |
2096 | int err; | |
94dace10 | 2097 | u32 param, val = 0; |
be839e39 CL |
2098 | |
2099 | /* | |
2100 | * Wait for the device to become ready before proceeding ... | |
2101 | */ | |
2102 | err = t4vf_wait_dev_ready(adapter); | |
2103 | if (err) { | |
2104 | dev_err(adapter->pdev_dev, "device didn't become ready:" | |
2105 | " err=%d\n", err); | |
2106 | return err; | |
2107 | } | |
2108 | ||
e68e6133 CL |
2109 | /* |
2110 | * Some environments do not properly handle PCIE FLRs -- e.g. in Linux | |
2111 | * 2.6.31 and later we can't call pci_reset_function() in order to | |
2112 | * issue an FLR because of a self- deadlock on the device semaphore. | |
2113 | * Meanwhile, the OS infrastructure doesn't issue FLRs in all the | |
2114 | * cases where they're needed -- for instance, some versions of KVM | |
2115 | * fail to reset "Assigned Devices" when the VM reboots. Therefore we | |
2116 | * use the firmware based reset in order to reset any per function | |
2117 | * state. | |
2118 | */ | |
2119 | err = t4vf_fw_reset(adapter); | |
2120 | if (err < 0) { | |
2121 | dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err); | |
2122 | return err; | |
2123 | } | |
2124 | ||
be839e39 CL |
2125 | /* |
2126 | * Grab basic operational parameters. These will predominantly have | |
2127 | * been set up by the Physical Function Driver or will be hard coded | |
2128 | * into the adapter. We just have to live with them ... Note that | |
2129 | * we _must_ get our VPD parameters before our SGE parameters because | |
2130 | * we need to know the adapter's core clock from the VPD in order to | |
2131 | * properly decode the SGE Timer Values. | |
2132 | */ | |
2133 | err = t4vf_get_dev_params(adapter); | |
2134 | if (err) { | |
2135 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2136 | " device parameters: err=%d\n", err); | |
2137 | return err; | |
2138 | } | |
2139 | err = t4vf_get_vpd_params(adapter); | |
2140 | if (err) { | |
2141 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2142 | " VPD parameters: err=%d\n", err); | |
2143 | return err; | |
2144 | } | |
2145 | err = t4vf_get_sge_params(adapter); | |
2146 | if (err) { | |
2147 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2148 | " SGE parameters: err=%d\n", err); | |
2149 | return err; | |
2150 | } | |
2151 | err = t4vf_get_rss_glb_config(adapter); | |
2152 | if (err) { | |
2153 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2154 | " RSS parameters: err=%d\n", err); | |
2155 | return err; | |
2156 | } | |
2157 | if (adapter->params.rss.mode != | |
2158 | FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) { | |
2159 | dev_err(adapter->pdev_dev, "unable to operate with global RSS" | |
2160 | " mode %d\n", adapter->params.rss.mode); | |
2161 | return -EINVAL; | |
2162 | } | |
2163 | err = t4vf_sge_init(adapter); | |
2164 | if (err) { | |
2165 | dev_err(adapter->pdev_dev, "unable to use adapter parameters:" | |
2166 | " err=%d\n", err); | |
2167 | return err; | |
2168 | } | |
2169 | ||
94dace10 VP |
2170 | /* If we're running on newer firmware, let it know that we're |
2171 | * prepared to deal with encapsulated CPL messages. Older | |
2172 | * firmware won't understand this and we'll just get | |
2173 | * unencapsulated messages ... | |
2174 | */ | |
5167865a HS |
2175 | param = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | |
2176 | FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_CPLFW4MSG_ENCAP); | |
94dace10 VP |
2177 | val = 1; |
2178 | (void) t4vf_set_params(adapter, 1, ¶m, &val); | |
2179 | ||
be839e39 CL |
2180 | /* |
2181 | * Retrieve our RX interrupt holdoff timer values and counter | |
2182 | * threshold values from the SGE parameters. | |
2183 | */ | |
2184 | s->timer_val[0] = core_ticks_to_us(adapter, | |
2185 | TIMERVALUE0_GET(sge_params->sge_timer_value_0_and_1)); | |
2186 | s->timer_val[1] = core_ticks_to_us(adapter, | |
2187 | TIMERVALUE1_GET(sge_params->sge_timer_value_0_and_1)); | |
2188 | s->timer_val[2] = core_ticks_to_us(adapter, | |
2189 | TIMERVALUE0_GET(sge_params->sge_timer_value_2_and_3)); | |
2190 | s->timer_val[3] = core_ticks_to_us(adapter, | |
2191 | TIMERVALUE1_GET(sge_params->sge_timer_value_2_and_3)); | |
2192 | s->timer_val[4] = core_ticks_to_us(adapter, | |
2193 | TIMERVALUE0_GET(sge_params->sge_timer_value_4_and_5)); | |
2194 | s->timer_val[5] = core_ticks_to_us(adapter, | |
2195 | TIMERVALUE1_GET(sge_params->sge_timer_value_4_and_5)); | |
2196 | ||
2197 | s->counter_val[0] = | |
2198 | THRESHOLD_0_GET(sge_params->sge_ingress_rx_threshold); | |
2199 | s->counter_val[1] = | |
2200 | THRESHOLD_1_GET(sge_params->sge_ingress_rx_threshold); | |
2201 | s->counter_val[2] = | |
2202 | THRESHOLD_2_GET(sge_params->sge_ingress_rx_threshold); | |
2203 | s->counter_val[3] = | |
2204 | THRESHOLD_3_GET(sge_params->sge_ingress_rx_threshold); | |
2205 | ||
2206 | /* | |
2207 | * Grab our Virtual Interface resource allocation, extract the | |
2208 | * features that we're interested in and do a bit of sanity testing on | |
2209 | * what we discover. | |
2210 | */ | |
2211 | err = t4vf_get_vfres(adapter); | |
2212 | if (err) { | |
2213 | dev_err(adapter->pdev_dev, "unable to get virtual interface" | |
2214 | " resources: err=%d\n", err); | |
2215 | return err; | |
2216 | } | |
2217 | ||
2218 | /* | |
2219 | * The number of "ports" which we support is equal to the number of | |
2220 | * Virtual Interfaces with which we've been provisioned. | |
2221 | */ | |
2222 | adapter->params.nports = vfres->nvi; | |
2223 | if (adapter->params.nports > MAX_NPORTS) { | |
2224 | dev_warn(adapter->pdev_dev, "only using %d of %d allowed" | |
2225 | " virtual interfaces\n", MAX_NPORTS, | |
2226 | adapter->params.nports); | |
2227 | adapter->params.nports = MAX_NPORTS; | |
2228 | } | |
2229 | ||
2230 | /* | |
2231 | * We need to reserve a number of the ingress queues with Free List | |
2232 | * and Interrupt capabilities for special interrupt purposes (like | |
2233 | * asynchronous firmware messages, or forwarded interrupts if we're | |
2234 | * using MSI). The rest of the FL/Intr-capable ingress queues will be | |
2235 | * matched up one-for-one with Ethernet/Control egress queues in order | |
2236 | * to form "Queue Sets" which will be aportioned between the "ports". | |
2237 | * For each Queue Set, we'll need the ability to allocate two Egress | |
2238 | * Contexts -- one for the Ingress Queue Free List and one for the TX | |
2239 | * Ethernet Queue. | |
2240 | */ | |
2241 | ethqsets = vfres->niqflint - INGQ_EXTRAS; | |
2242 | if (vfres->nethctrl != ethqsets) { | |
2243 | dev_warn(adapter->pdev_dev, "unequal number of [available]" | |
2244 | " ingress/egress queues (%d/%d); using minimum for" | |
2245 | " number of Queue Sets\n", ethqsets, vfres->nethctrl); | |
2246 | ethqsets = min(vfres->nethctrl, ethqsets); | |
2247 | } | |
2248 | if (vfres->neq < ethqsets*2) { | |
2249 | dev_warn(adapter->pdev_dev, "Not enough Egress Contexts (%d)" | |
2250 | " to support Queue Sets (%d); reducing allowed Queue" | |
2251 | " Sets\n", vfres->neq, ethqsets); | |
2252 | ethqsets = vfres->neq/2; | |
2253 | } | |
2254 | if (ethqsets > MAX_ETH_QSETS) { | |
2255 | dev_warn(adapter->pdev_dev, "only using %d of %d allowed Queue" | |
2256 | " Sets\n", MAX_ETH_QSETS, adapter->sge.max_ethqsets); | |
2257 | ethqsets = MAX_ETH_QSETS; | |
2258 | } | |
2259 | if (vfres->niq != 0 || vfres->neq > ethqsets*2) { | |
2260 | dev_warn(adapter->pdev_dev, "unused resources niq/neq (%d/%d)" | |
2261 | " ignored\n", vfres->niq, vfres->neq - ethqsets*2); | |
2262 | } | |
2263 | adapter->sge.max_ethqsets = ethqsets; | |
2264 | ||
2265 | /* | |
2266 | * Check for various parameter sanity issues. Most checks simply | |
2267 | * result in us using fewer resources than our provissioning but we | |
2268 | * do need at least one "port" with which to work ... | |
2269 | */ | |
2270 | if (adapter->sge.max_ethqsets < adapter->params.nports) { | |
2271 | dev_warn(adapter->pdev_dev, "only using %d of %d available" | |
2272 | " virtual interfaces (too few Queue Sets)\n", | |
2273 | adapter->sge.max_ethqsets, adapter->params.nports); | |
2274 | adapter->params.nports = adapter->sge.max_ethqsets; | |
2275 | } | |
2276 | if (adapter->params.nports == 0) { | |
2277 | dev_err(adapter->pdev_dev, "no virtual interfaces configured/" | |
2278 | "usable!\n"); | |
2279 | return -EINVAL; | |
2280 | } | |
2281 | return 0; | |
2282 | } | |
2283 | ||
2284 | static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx, | |
2285 | u8 pkt_cnt_idx, unsigned int size, | |
2286 | unsigned int iqe_size) | |
2287 | { | |
2288 | rspq->intr_params = (QINTR_TIMER_IDX(timer_idx) | | |
2289 | (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0)); | |
2290 | rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS | |
2291 | ? pkt_cnt_idx | |
2292 | : 0); | |
2293 | rspq->iqe_len = iqe_size; | |
2294 | rspq->size = size; | |
2295 | } | |
2296 | ||
2297 | /* | |
2298 | * Perform default configuration of DMA queues depending on the number and | |
2299 | * type of ports we found and the number of available CPUs. Most settings can | |
2300 | * be modified by the admin via ethtool and cxgbtool prior to the adapter | |
2301 | * being brought up for the first time. | |
2302 | */ | |
d289f864 | 2303 | static void cfg_queues(struct adapter *adapter) |
be839e39 CL |
2304 | { |
2305 | struct sge *s = &adapter->sge; | |
2306 | int q10g, n10g, qidx, pidx, qs; | |
c710245c | 2307 | size_t iqe_size; |
be839e39 CL |
2308 | |
2309 | /* | |
2310 | * We should not be called till we know how many Queue Sets we can | |
2311 | * support. In particular, this means that we need to know what kind | |
2312 | * of interrupts we'll be using ... | |
2313 | */ | |
2314 | BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0); | |
2315 | ||
2316 | /* | |
2317 | * Count the number of 10GbE Virtual Interfaces that we have. | |
2318 | */ | |
2319 | n10g = 0; | |
2320 | for_each_port(adapter, pidx) | |
2321 | n10g += is_10g_port(&adap2pinfo(adapter, pidx)->link_cfg); | |
2322 | ||
2323 | /* | |
2324 | * We default to 1 queue per non-10G port and up to # of cores queues | |
2325 | * per 10G port. | |
2326 | */ | |
2327 | if (n10g == 0) | |
2328 | q10g = 0; | |
2329 | else { | |
2330 | int n1g = (adapter->params.nports - n10g); | |
2331 | q10g = (adapter->sge.max_ethqsets - n1g) / n10g; | |
2332 | if (q10g > num_online_cpus()) | |
2333 | q10g = num_online_cpus(); | |
2334 | } | |
2335 | ||
2336 | /* | |
2337 | * Allocate the "Queue Sets" to the various Virtual Interfaces. | |
2338 | * The layout will be established in setup_sge_queues() when the | |
2339 | * adapter is brough up for the first time. | |
2340 | */ | |
2341 | qidx = 0; | |
2342 | for_each_port(adapter, pidx) { | |
2343 | struct port_info *pi = adap2pinfo(adapter, pidx); | |
2344 | ||
2345 | pi->first_qset = qidx; | |
897d55df | 2346 | pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1; |
be839e39 CL |
2347 | qidx += pi->nqsets; |
2348 | } | |
2349 | s->ethqsets = qidx; | |
2350 | ||
c710245c CL |
2351 | /* |
2352 | * The Ingress Queue Entry Size for our various Response Queues needs | |
2353 | * to be big enough to accommodate the largest message we can receive | |
2354 | * from the chip/firmware; which is 64 bytes ... | |
2355 | */ | |
2356 | iqe_size = 64; | |
2357 | ||
be839e39 CL |
2358 | /* |
2359 | * Set up default Queue Set parameters ... Start off with the | |
2360 | * shortest interrupt holdoff timer. | |
2361 | */ | |
2362 | for (qs = 0; qs < s->max_ethqsets; qs++) { | |
2363 | struct sge_eth_rxq *rxq = &s->ethrxq[qs]; | |
2364 | struct sge_eth_txq *txq = &s->ethtxq[qs]; | |
2365 | ||
c710245c | 2366 | init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size); |
be839e39 CL |
2367 | rxq->fl.size = 72; |
2368 | txq->q.size = 1024; | |
2369 | } | |
2370 | ||
2371 | /* | |
2372 | * The firmware event queue is used for link state changes and | |
2373 | * notifications of TX DMA completions. | |
2374 | */ | |
c710245c | 2375 | init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size); |
be839e39 CL |
2376 | |
2377 | /* | |
2378 | * The forwarded interrupt queue is used when we're in MSI interrupt | |
2379 | * mode. In this mode all interrupts associated with RX queues will | |
2380 | * be forwarded to a single queue which we'll associate with our MSI | |
2381 | * interrupt vector. The messages dropped in the forwarded interrupt | |
2382 | * queue will indicate which ingress queue needs servicing ... This | |
2383 | * queue needs to be large enough to accommodate all of the ingress | |
2384 | * queues which are forwarding their interrupt (+1 to prevent the PIDX | |
2385 | * from equalling the CIDX if every ingress queue has an outstanding | |
2386 | * interrupt). The queue doesn't need to be any larger because no | |
2387 | * ingress queue will ever have more than one outstanding interrupt at | |
2388 | * any time ... | |
2389 | */ | |
2390 | init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1, | |
c710245c | 2391 | iqe_size); |
be839e39 CL |
2392 | } |
2393 | ||
2394 | /* | |
2395 | * Reduce the number of Ethernet queues across all ports to at most n. | |
2396 | * n provides at least one queue per port. | |
2397 | */ | |
d289f864 | 2398 | static void reduce_ethqs(struct adapter *adapter, int n) |
be839e39 CL |
2399 | { |
2400 | int i; | |
2401 | struct port_info *pi; | |
2402 | ||
2403 | /* | |
2404 | * While we have too many active Ether Queue Sets, interate across the | |
2405 | * "ports" and reduce their individual Queue Set allocations. | |
2406 | */ | |
2407 | BUG_ON(n < adapter->params.nports); | |
2408 | while (n < adapter->sge.ethqsets) | |
2409 | for_each_port(adapter, i) { | |
2410 | pi = adap2pinfo(adapter, i); | |
2411 | if (pi->nqsets > 1) { | |
2412 | pi->nqsets--; | |
2413 | adapter->sge.ethqsets--; | |
2414 | if (adapter->sge.ethqsets <= n) | |
2415 | break; | |
2416 | } | |
2417 | } | |
2418 | ||
2419 | /* | |
2420 | * Reassign the starting Queue Sets for each of the "ports" ... | |
2421 | */ | |
2422 | n = 0; | |
2423 | for_each_port(adapter, i) { | |
2424 | pi = adap2pinfo(adapter, i); | |
2425 | pi->first_qset = n; | |
2426 | n += pi->nqsets; | |
2427 | } | |
2428 | } | |
2429 | ||
2430 | /* | |
2431 | * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally | |
2432 | * we get a separate MSI-X vector for every "Queue Set" plus any extras we | |
2433 | * need. Minimally we need one for every Virtual Interface plus those needed | |
2434 | * for our "extras". Note that this process may lower the maximum number of | |
2435 | * allowed Queue Sets ... | |
2436 | */ | |
d289f864 | 2437 | static int enable_msix(struct adapter *adapter) |
be839e39 | 2438 | { |
bd663689 | 2439 | int i, want, need, nqsets; |
be839e39 CL |
2440 | struct msix_entry entries[MSIX_ENTRIES]; |
2441 | struct sge *s = &adapter->sge; | |
2442 | ||
2443 | for (i = 0; i < MSIX_ENTRIES; ++i) | |
2444 | entries[i].entry = i; | |
2445 | ||
2446 | /* | |
2447 | * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets" | |
2448 | * plus those needed for our "extras" (for example, the firmware | |
2449 | * message queue). We _need_ at least one "Queue Set" per Virtual | |
2450 | * Interface plus those needed for our "extras". So now we get to see | |
2451 | * if the song is right ... | |
2452 | */ | |
2453 | want = s->max_ethqsets + MSIX_EXTRAS; | |
2454 | need = adapter->params.nports + MSIX_EXTRAS; | |
bd663689 AG |
2455 | |
2456 | want = pci_enable_msix_range(adapter->pdev, entries, need, want); | |
2457 | if (want < 0) | |
2458 | return want; | |
2459 | ||
2460 | nqsets = want - MSIX_EXTRAS; | |
2461 | if (nqsets < s->max_ethqsets) { | |
2462 | dev_warn(adapter->pdev_dev, "only enough MSI-X vectors" | |
2463 | " for %d Queue Sets\n", nqsets); | |
2464 | s->max_ethqsets = nqsets; | |
2465 | if (nqsets < s->ethqsets) | |
2466 | reduce_ethqs(adapter, nqsets); | |
be839e39 | 2467 | } |
bd663689 AG |
2468 | for (i = 0; i < want; ++i) |
2469 | adapter->msix_info[i].vec = entries[i].vector; | |
2470 | ||
2471 | return 0; | |
be839e39 CL |
2472 | } |
2473 | ||
be839e39 CL |
2474 | static const struct net_device_ops cxgb4vf_netdev_ops = { |
2475 | .ndo_open = cxgb4vf_open, | |
2476 | .ndo_stop = cxgb4vf_stop, | |
2477 | .ndo_start_xmit = t4vf_eth_xmit, | |
2478 | .ndo_get_stats = cxgb4vf_get_stats, | |
2479 | .ndo_set_rx_mode = cxgb4vf_set_rxmode, | |
2480 | .ndo_set_mac_address = cxgb4vf_set_mac_addr, | |
be839e39 CL |
2481 | .ndo_validate_addr = eth_validate_addr, |
2482 | .ndo_do_ioctl = cxgb4vf_do_ioctl, | |
2483 | .ndo_change_mtu = cxgb4vf_change_mtu, | |
87737663 JP |
2484 | .ndo_fix_features = cxgb4vf_fix_features, |
2485 | .ndo_set_features = cxgb4vf_set_features, | |
be839e39 CL |
2486 | #ifdef CONFIG_NET_POLL_CONTROLLER |
2487 | .ndo_poll_controller = cxgb4vf_poll_controller, | |
2488 | #endif | |
2489 | }; | |
be839e39 CL |
2490 | |
2491 | /* | |
2492 | * "Probe" a device: initialize a device and construct all kernel and driver | |
2493 | * state needed to manage the device. This routine is called "init_one" in | |
2494 | * the PF Driver ... | |
2495 | */ | |
d289f864 | 2496 | static int cxgb4vf_pci_probe(struct pci_dev *pdev, |
1dd06ae8 | 2497 | const struct pci_device_id *ent) |
be839e39 | 2498 | { |
be839e39 CL |
2499 | int pci_using_dac; |
2500 | int err, pidx; | |
2501 | unsigned int pmask; | |
2502 | struct adapter *adapter; | |
2503 | struct port_info *pi; | |
2504 | struct net_device *netdev; | |
2505 | ||
be839e39 CL |
2506 | /* |
2507 | * Print our driver banner the first time we're called to initialize a | |
2508 | * device. | |
2509 | */ | |
428ac43f | 2510 | pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION); |
be839e39 CL |
2511 | |
2512 | /* | |
7a0c2029 | 2513 | * Initialize generic PCI device state. |
be839e39 | 2514 | */ |
7a0c2029 | 2515 | err = pci_enable_device(pdev); |
be839e39 | 2516 | if (err) { |
7a0c2029 | 2517 | dev_err(&pdev->dev, "cannot enable PCI device\n"); |
be839e39 CL |
2518 | return err; |
2519 | } | |
2520 | ||
2521 | /* | |
7a0c2029 KV |
2522 | * Reserve PCI resources for the device. If we can't get them some |
2523 | * other driver may have already claimed the device ... | |
be839e39 | 2524 | */ |
7a0c2029 | 2525 | err = pci_request_regions(pdev, KBUILD_MODNAME); |
be839e39 | 2526 | if (err) { |
7a0c2029 KV |
2527 | dev_err(&pdev->dev, "cannot obtain PCI resources\n"); |
2528 | goto err_disable_device; | |
be839e39 CL |
2529 | } |
2530 | ||
2531 | /* | |
2532 | * Set up our DMA mask: try for 64-bit address masking first and | |
2533 | * fall back to 32-bit if we can't get 64 bits ... | |
2534 | */ | |
2535 | err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); | |
2536 | if (err == 0) { | |
2537 | err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); | |
2538 | if (err) { | |
2539 | dev_err(&pdev->dev, "unable to obtain 64-bit DMA for" | |
2540 | " coherent allocations\n"); | |
7a0c2029 | 2541 | goto err_release_regions; |
be839e39 CL |
2542 | } |
2543 | pci_using_dac = 1; | |
2544 | } else { | |
2545 | err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); | |
2546 | if (err != 0) { | |
2547 | dev_err(&pdev->dev, "no usable DMA configuration\n"); | |
7a0c2029 | 2548 | goto err_release_regions; |
be839e39 CL |
2549 | } |
2550 | pci_using_dac = 0; | |
2551 | } | |
2552 | ||
2553 | /* | |
2554 | * Enable bus mastering for the device ... | |
2555 | */ | |
2556 | pci_set_master(pdev); | |
2557 | ||
2558 | /* | |
2559 | * Allocate our adapter data structure and attach it to the device. | |
2560 | */ | |
2561 | adapter = kzalloc(sizeof(*adapter), GFP_KERNEL); | |
2562 | if (!adapter) { | |
2563 | err = -ENOMEM; | |
7a0c2029 | 2564 | goto err_release_regions; |
be839e39 CL |
2565 | } |
2566 | pci_set_drvdata(pdev, adapter); | |
2567 | adapter->pdev = pdev; | |
2568 | adapter->pdev_dev = &pdev->dev; | |
2569 | ||
2570 | /* | |
2571 | * Initialize SMP data synchronization resources. | |
2572 | */ | |
2573 | spin_lock_init(&adapter->stats_lock); | |
2574 | ||
2575 | /* | |
2576 | * Map our I/O registers in BAR0. | |
2577 | */ | |
2578 | adapter->regs = pci_ioremap_bar(pdev, 0); | |
2579 | if (!adapter->regs) { | |
2580 | dev_err(&pdev->dev, "cannot map device registers\n"); | |
2581 | err = -ENOMEM; | |
2582 | goto err_free_adapter; | |
2583 | } | |
2584 | ||
e0a8b34a HS |
2585 | /* Wait for the device to become ready before proceeding ... |
2586 | */ | |
2587 | err = t4vf_prep_adapter(adapter); | |
2588 | if (err) { | |
2589 | dev_err(adapter->pdev_dev, "device didn't become ready:" | |
2590 | " err=%d\n", err); | |
2591 | goto err_unmap_bar0; | |
2592 | } | |
2593 | ||
2594 | /* For T5 and later we want to use the new BAR-based User Doorbells, | |
2595 | * so we need to map BAR2 here ... | |
2596 | */ | |
2597 | if (!is_t4(adapter->params.chip)) { | |
2598 | adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2), | |
2599 | pci_resource_len(pdev, 2)); | |
2600 | if (!adapter->bar2) { | |
2601 | dev_err(adapter->pdev_dev, "cannot map BAR2 doorbells\n"); | |
2602 | err = -ENOMEM; | |
2603 | goto err_unmap_bar0; | |
2604 | } | |
2605 | } | |
be839e39 CL |
2606 | /* |
2607 | * Initialize adapter level features. | |
2608 | */ | |
2609 | adapter->name = pci_name(pdev); | |
2610 | adapter->msg_enable = dflt_msg_enable; | |
2611 | err = adap_init0(adapter); | |
2612 | if (err) | |
2613 | goto err_unmap_bar; | |
2614 | ||
2615 | /* | |
2616 | * Allocate our "adapter ports" and stitch everything together. | |
2617 | */ | |
2618 | pmask = adapter->params.vfres.pmask; | |
2619 | for_each_port(adapter, pidx) { | |
2620 | int port_id, viid; | |
2621 | ||
2622 | /* | |
2623 | * We simplistically allocate our virtual interfaces | |
2624 | * sequentially across the port numbers to which we have | |
2625 | * access rights. This should be configurable in some manner | |
2626 | * ... | |
2627 | */ | |
2628 | if (pmask == 0) | |
2629 | break; | |
2630 | port_id = ffs(pmask) - 1; | |
2631 | pmask &= ~(1 << port_id); | |
2632 | viid = t4vf_alloc_vi(adapter, port_id); | |
2633 | if (viid < 0) { | |
2634 | dev_err(&pdev->dev, "cannot allocate VI for port %d:" | |
2635 | " err=%d\n", port_id, viid); | |
2636 | err = viid; | |
2637 | goto err_free_dev; | |
2638 | } | |
2639 | ||
2640 | /* | |
2641 | * Allocate our network device and stitch things together. | |
2642 | */ | |
2643 | netdev = alloc_etherdev_mq(sizeof(struct port_info), | |
2644 | MAX_PORT_QSETS); | |
2645 | if (netdev == NULL) { | |
be839e39 CL |
2646 | t4vf_free_vi(adapter, viid); |
2647 | err = -ENOMEM; | |
2648 | goto err_free_dev; | |
2649 | } | |
2650 | adapter->port[pidx] = netdev; | |
2651 | SET_NETDEV_DEV(netdev, &pdev->dev); | |
2652 | pi = netdev_priv(netdev); | |
2653 | pi->adapter = adapter; | |
2654 | pi->pidx = pidx; | |
2655 | pi->port_id = port_id; | |
2656 | pi->viid = viid; | |
2657 | ||
2658 | /* | |
2659 | * Initialize the starting state of our "port" and register | |
2660 | * it. | |
2661 | */ | |
2662 | pi->xact_addr_filt = -1; | |
be839e39 | 2663 | netif_carrier_off(netdev); |
be839e39 CL |
2664 | netdev->irq = pdev->irq; |
2665 | ||
2ed28baa MM |
2666 | netdev->hw_features = NETIF_F_SG | TSO_FLAGS | |
2667 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
f646968f | 2668 | NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXCSUM; |
2ed28baa MM |
2669 | netdev->vlan_features = NETIF_F_SG | TSO_FLAGS | |
2670 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
2671 | NETIF_F_HIGHDMA; | |
f646968f PM |
2672 | netdev->features = netdev->hw_features | |
2673 | NETIF_F_HW_VLAN_CTAG_TX; | |
be839e39 CL |
2674 | if (pci_using_dac) |
2675 | netdev->features |= NETIF_F_HIGHDMA; | |
be839e39 | 2676 | |
01789349 JP |
2677 | netdev->priv_flags |= IFF_UNICAST_FLT; |
2678 | ||
be839e39 | 2679 | netdev->netdev_ops = &cxgb4vf_netdev_ops; |
7ad24ea4 | 2680 | netdev->ethtool_ops = &cxgb4vf_ethtool_ops; |
be839e39 CL |
2681 | |
2682 | /* | |
2683 | * Initialize the hardware/software state for the port. | |
2684 | */ | |
2685 | err = t4vf_port_init(adapter, pidx); | |
2686 | if (err) { | |
2687 | dev_err(&pdev->dev, "cannot initialize port %d\n", | |
2688 | pidx); | |
2689 | goto err_free_dev; | |
2690 | } | |
2691 | } | |
2692 | ||
2693 | /* | |
2694 | * The "card" is now ready to go. If any errors occur during device | |
2695 | * registration we do not fail the whole "card" but rather proceed | |
2696 | * only with the ports we manage to register successfully. However we | |
2697 | * must register at least one net device. | |
2698 | */ | |
2699 | for_each_port(adapter, pidx) { | |
2700 | netdev = adapter->port[pidx]; | |
2701 | if (netdev == NULL) | |
2702 | continue; | |
2703 | ||
2704 | err = register_netdev(netdev); | |
2705 | if (err) { | |
2706 | dev_warn(&pdev->dev, "cannot register net device %s," | |
2707 | " skipping\n", netdev->name); | |
2708 | continue; | |
2709 | } | |
2710 | ||
2711 | set_bit(pidx, &adapter->registered_device_map); | |
2712 | } | |
2713 | if (adapter->registered_device_map == 0) { | |
2714 | dev_err(&pdev->dev, "could not register any net devices\n"); | |
2715 | goto err_free_dev; | |
2716 | } | |
2717 | ||
2718 | /* | |
2719 | * Set up our debugfs entries. | |
2720 | */ | |
843635e0 | 2721 | if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) { |
be839e39 CL |
2722 | adapter->debugfs_root = |
2723 | debugfs_create_dir(pci_name(pdev), | |
2724 | cxgb4vf_debugfs_root); | |
843635e0 | 2725 | if (IS_ERR_OR_NULL(adapter->debugfs_root)) |
be839e39 CL |
2726 | dev_warn(&pdev->dev, "could not create debugfs" |
2727 | " directory"); | |
2728 | else | |
2729 | setup_debugfs(adapter); | |
2730 | } | |
2731 | ||
2732 | /* | |
2733 | * See what interrupts we'll be using. If we've been configured to | |
2734 | * use MSI-X interrupts, try to enable them but fall back to using | |
2735 | * MSI interrupts if we can't enable MSI-X interrupts. If we can't | |
2736 | * get MSI interrupts we bail with the error. | |
2737 | */ | |
2738 | if (msi == MSI_MSIX && enable_msix(adapter) == 0) | |
2739 | adapter->flags |= USING_MSIX; | |
2740 | else { | |
2741 | err = pci_enable_msi(pdev); | |
2742 | if (err) { | |
2743 | dev_err(&pdev->dev, "Unable to allocate %s interrupts;" | |
2744 | " err=%d\n", | |
2745 | msi == MSI_MSIX ? "MSI-X or MSI" : "MSI", err); | |
2746 | goto err_free_debugfs; | |
2747 | } | |
2748 | adapter->flags |= USING_MSI; | |
2749 | } | |
2750 | ||
2751 | /* | |
2752 | * Now that we know how many "ports" we have and what their types are, | |
2753 | * and how many Queue Sets we can support, we can configure our queue | |
2754 | * resources. | |
2755 | */ | |
2756 | cfg_queues(adapter); | |
2757 | ||
2758 | /* | |
25985edc | 2759 | * Print a short notice on the existence and configuration of the new |
be839e39 CL |
2760 | * VF network device ... |
2761 | */ | |
2762 | for_each_port(adapter, pidx) { | |
2763 | dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n", | |
2764 | adapter->port[pidx]->name, | |
2765 | (adapter->flags & USING_MSIX) ? "MSI-X" : | |
2766 | (adapter->flags & USING_MSI) ? "MSI" : ""); | |
2767 | } | |
2768 | ||
2769 | /* | |
2770 | * Return success! | |
2771 | */ | |
2772 | return 0; | |
2773 | ||
2774 | /* | |
2775 | * Error recovery and exit code. Unwind state that's been created | |
2776 | * so far and return the error. | |
2777 | */ | |
2778 | ||
2779 | err_free_debugfs: | |
843635e0 | 2780 | if (!IS_ERR_OR_NULL(adapter->debugfs_root)) { |
be839e39 CL |
2781 | cleanup_debugfs(adapter); |
2782 | debugfs_remove_recursive(adapter->debugfs_root); | |
2783 | } | |
2784 | ||
2785 | err_free_dev: | |
2786 | for_each_port(adapter, pidx) { | |
2787 | netdev = adapter->port[pidx]; | |
2788 | if (netdev == NULL) | |
2789 | continue; | |
2790 | pi = netdev_priv(netdev); | |
2791 | t4vf_free_vi(adapter, pi->viid); | |
2792 | if (test_bit(pidx, &adapter->registered_device_map)) | |
2793 | unregister_netdev(netdev); | |
2794 | free_netdev(netdev); | |
2795 | } | |
2796 | ||
2797 | err_unmap_bar: | |
e0a8b34a HS |
2798 | if (!is_t4(adapter->params.chip)) |
2799 | iounmap(adapter->bar2); | |
2800 | ||
2801 | err_unmap_bar0: | |
be839e39 CL |
2802 | iounmap(adapter->regs); |
2803 | ||
2804 | err_free_adapter: | |
2805 | kfree(adapter); | |
be839e39 | 2806 | |
be839e39 CL |
2807 | err_release_regions: |
2808 | pci_release_regions(pdev); | |
7a0c2029 KV |
2809 | pci_clear_master(pdev); |
2810 | ||
2811 | err_disable_device: | |
2812 | pci_disable_device(pdev); | |
be839e39 | 2813 | |
be839e39 CL |
2814 | return err; |
2815 | } | |
2816 | ||
2817 | /* | |
2818 | * "Remove" a device: tear down all kernel and driver state created in the | |
2819 | * "probe" routine and quiesce the device (disable interrupts, etc.). (Note | |
2820 | * that this is called "remove_one" in the PF Driver.) | |
2821 | */ | |
d289f864 | 2822 | static void cxgb4vf_pci_remove(struct pci_dev *pdev) |
be839e39 CL |
2823 | { |
2824 | struct adapter *adapter = pci_get_drvdata(pdev); | |
2825 | ||
2826 | /* | |
2827 | * Tear down driver state associated with device. | |
2828 | */ | |
2829 | if (adapter) { | |
2830 | int pidx; | |
2831 | ||
2832 | /* | |
2833 | * Stop all of our activity. Unregister network port, | |
2834 | * disable interrupts, etc. | |
2835 | */ | |
2836 | for_each_port(adapter, pidx) | |
2837 | if (test_bit(pidx, &adapter->registered_device_map)) | |
2838 | unregister_netdev(adapter->port[pidx]); | |
2839 | t4vf_sge_stop(adapter); | |
2840 | if (adapter->flags & USING_MSIX) { | |
2841 | pci_disable_msix(adapter->pdev); | |
2842 | adapter->flags &= ~USING_MSIX; | |
2843 | } else if (adapter->flags & USING_MSI) { | |
2844 | pci_disable_msi(adapter->pdev); | |
2845 | adapter->flags &= ~USING_MSI; | |
2846 | } | |
2847 | ||
2848 | /* | |
2849 | * Tear down our debugfs entries. | |
2850 | */ | |
843635e0 | 2851 | if (!IS_ERR_OR_NULL(adapter->debugfs_root)) { |
be839e39 CL |
2852 | cleanup_debugfs(adapter); |
2853 | debugfs_remove_recursive(adapter->debugfs_root); | |
2854 | } | |
2855 | ||
2856 | /* | |
2857 | * Free all of the various resources which we've acquired ... | |
2858 | */ | |
2859 | t4vf_free_sge_resources(adapter); | |
2860 | for_each_port(adapter, pidx) { | |
2861 | struct net_device *netdev = adapter->port[pidx]; | |
2862 | struct port_info *pi; | |
2863 | ||
2864 | if (netdev == NULL) | |
2865 | continue; | |
2866 | ||
2867 | pi = netdev_priv(netdev); | |
2868 | t4vf_free_vi(adapter, pi->viid); | |
2869 | free_netdev(netdev); | |
2870 | } | |
2871 | iounmap(adapter->regs); | |
e0a8b34a HS |
2872 | if (!is_t4(adapter->params.chip)) |
2873 | iounmap(adapter->bar2); | |
be839e39 | 2874 | kfree(adapter); |
be839e39 CL |
2875 | } |
2876 | ||
2877 | /* | |
2878 | * Disable the device and release its PCI resources. | |
2879 | */ | |
2880 | pci_disable_device(pdev); | |
2881 | pci_clear_master(pdev); | |
2882 | pci_release_regions(pdev); | |
2883 | } | |
2884 | ||
7e9c2629 CL |
2885 | /* |
2886 | * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt | |
2887 | * delivery. | |
2888 | */ | |
d289f864 | 2889 | static void cxgb4vf_pci_shutdown(struct pci_dev *pdev) |
7e9c2629 CL |
2890 | { |
2891 | struct adapter *adapter; | |
2892 | int pidx; | |
2893 | ||
2894 | adapter = pci_get_drvdata(pdev); | |
2895 | if (!adapter) | |
2896 | return; | |
2897 | ||
c2a19856 | 2898 | /* Disable all Virtual Interfaces. This will shut down the |
7e9c2629 CL |
2899 | * delivery of all ingress packets into the chip for these |
2900 | * Virtual Interfaces. | |
2901 | */ | |
c2a19856 HS |
2902 | for_each_port(adapter, pidx) |
2903 | if (test_bit(pidx, &adapter->registered_device_map)) | |
2904 | unregister_netdev(adapter->port[pidx]); | |
7e9c2629 | 2905 | |
c2a19856 HS |
2906 | /* Free up all Queues which will prevent further DMA and |
2907 | * Interrupts allowing various internal pathways to drain. | |
2908 | */ | |
2909 | t4vf_sge_stop(adapter); | |
2910 | if (adapter->flags & USING_MSIX) { | |
2911 | pci_disable_msix(adapter->pdev); | |
2912 | adapter->flags &= ~USING_MSIX; | |
2913 | } else if (adapter->flags & USING_MSI) { | |
2914 | pci_disable_msi(adapter->pdev); | |
2915 | adapter->flags &= ~USING_MSI; | |
7e9c2629 CL |
2916 | } |
2917 | ||
2918 | /* | |
2919 | * Free up all Queues which will prevent further DMA and | |
2920 | * Interrupts allowing various internal pathways to drain. | |
2921 | */ | |
2922 | t4vf_free_sge_resources(adapter); | |
c2a19856 | 2923 | pci_set_drvdata(pdev, NULL); |
7e9c2629 CL |
2924 | } |
2925 | ||
3fedeab1 HS |
2926 | /* Macros needed to support the PCI Device ID Table ... |
2927 | */ | |
2928 | #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \ | |
2929 | static struct pci_device_id cxgb4vf_pci_tbl[] = { | |
2930 | #define CH_PCI_DEVICE_ID_FUNCTION 0x8 | |
2931 | ||
2932 | #define CH_PCI_ID_TABLE_ENTRY(devid) \ | |
2933 | { PCI_VDEVICE(CHELSIO, (devid)), 0 } | |
2934 | ||
2935 | #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END { 0, } } | |
2936 | ||
2937 | #include "../cxgb4/t4_pci_id_tbl.h" | |
be839e39 CL |
2938 | |
2939 | MODULE_DESCRIPTION(DRV_DESC); | |
2940 | MODULE_AUTHOR("Chelsio Communications"); | |
2941 | MODULE_LICENSE("Dual BSD/GPL"); | |
2942 | MODULE_VERSION(DRV_VERSION); | |
2943 | MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl); | |
2944 | ||
2945 | static struct pci_driver cxgb4vf_driver = { | |
2946 | .name = KBUILD_MODNAME, | |
2947 | .id_table = cxgb4vf_pci_tbl, | |
2948 | .probe = cxgb4vf_pci_probe, | |
d289f864 BP |
2949 | .remove = cxgb4vf_pci_remove, |
2950 | .shutdown = cxgb4vf_pci_shutdown, | |
be839e39 CL |
2951 | }; |
2952 | ||
2953 | /* | |
2954 | * Initialize global driver state. | |
2955 | */ | |
2956 | static int __init cxgb4vf_module_init(void) | |
2957 | { | |
2958 | int ret; | |
2959 | ||
bb14a1af CL |
2960 | /* |
2961 | * Vet our module parameters. | |
2962 | */ | |
2963 | if (msi != MSI_MSIX && msi != MSI_MSI) { | |
428ac43f JP |
2964 | pr_warn("bad module parameter msi=%d; must be %d (MSI-X or MSI) or %d (MSI)\n", |
2965 | msi, MSI_MSIX, MSI_MSI); | |
bb14a1af CL |
2966 | return -EINVAL; |
2967 | } | |
2968 | ||
be839e39 CL |
2969 | /* Debugfs support is optional, just warn if this fails */ |
2970 | cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL); | |
843635e0 | 2971 | if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) |
428ac43f | 2972 | pr_warn("could not create debugfs entry, continuing\n"); |
be839e39 CL |
2973 | |
2974 | ret = pci_register_driver(&cxgb4vf_driver); | |
843635e0 | 2975 | if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) |
be839e39 CL |
2976 | debugfs_remove(cxgb4vf_debugfs_root); |
2977 | return ret; | |
2978 | } | |
2979 | ||
2980 | /* | |
2981 | * Tear down global driver state. | |
2982 | */ | |
2983 | static void __exit cxgb4vf_module_exit(void) | |
2984 | { | |
2985 | pci_unregister_driver(&cxgb4vf_driver); | |
2986 | debugfs_remove(cxgb4vf_debugfs_root); | |
2987 | } | |
2988 | ||
2989 | module_init(cxgb4vf_module_init); | |
2990 | module_exit(cxgb4vf_module_exit); |