]>
Commit | Line | Data |
---|---|---|
be839e39 CL |
1 | /* |
2 | * This file is part of the Chelsio T4 PCI-E SR-IOV Virtual Function Ethernet | |
3 | * driver for Linux. | |
4 | * | |
5 | * Copyright (c) 2009-2010 Chelsio Communications, Inc. All rights reserved. | |
6 | * | |
7 | * This software is available to you under a choice of one of two | |
8 | * licenses. You may choose to be licensed under the terms of the GNU | |
9 | * General Public License (GPL) Version 2, available from the file | |
10 | * COPYING in the main directory of this source tree, or the | |
11 | * OpenIB.org BSD license below: | |
12 | * | |
13 | * Redistribution and use in source and binary forms, with or | |
14 | * without modification, are permitted provided that the following | |
15 | * conditions are met: | |
16 | * | |
17 | * - Redistributions of source code must retain the above | |
18 | * copyright notice, this list of conditions and the following | |
19 | * disclaimer. | |
20 | * | |
21 | * - Redistributions in binary form must reproduce the above | |
22 | * copyright notice, this list of conditions and the following | |
23 | * disclaimer in the documentation and/or other materials | |
24 | * provided with the distribution. | |
25 | * | |
26 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
27 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
28 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
29 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
30 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
31 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
32 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
33 | * SOFTWARE. | |
34 | */ | |
35 | ||
428ac43f JP |
36 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
37 | ||
be839e39 CL |
38 | #include <linux/module.h> |
39 | #include <linux/moduleparam.h> | |
40 | #include <linux/init.h> | |
41 | #include <linux/pci.h> | |
42 | #include <linux/dma-mapping.h> | |
43 | #include <linux/netdevice.h> | |
44 | #include <linux/etherdevice.h> | |
45 | #include <linux/debugfs.h> | |
46 | #include <linux/ethtool.h> | |
5ad24def | 47 | #include <linux/mdio.h> |
be839e39 CL |
48 | |
49 | #include "t4vf_common.h" | |
50 | #include "t4vf_defs.h" | |
51 | ||
52 | #include "../cxgb4/t4_regs.h" | |
53 | #include "../cxgb4/t4_msg.h" | |
54 | ||
55 | /* | |
56 | * Generic information about the driver. | |
57 | */ | |
622c62b5 | 58 | #define DRV_VERSION "2.0.0-ko" |
52a5f846 | 59 | #define DRV_DESC "Chelsio T4/T5/T6 Virtual Function (VF) Network Driver" |
be839e39 CL |
60 | |
61 | /* | |
62 | * Module Parameters. | |
63 | * ================== | |
64 | */ | |
65 | ||
66 | /* | |
67 | * Default ethtool "message level" for adapters. | |
68 | */ | |
69 | #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \ | |
70 | NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\ | |
71 | NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR) | |
72 | ||
73 | static int dflt_msg_enable = DFLT_MSG_ENABLE; | |
74 | ||
75 | module_param(dflt_msg_enable, int, 0644); | |
76 | MODULE_PARM_DESC(dflt_msg_enable, | |
8a21ec4e HS |
77 | "default adapter ethtool message level bitmap, " |
78 | "deprecated parameter"); | |
be839e39 CL |
79 | |
80 | /* | |
81 | * The driver uses the best interrupt scheme available on a platform in the | |
82 | * order MSI-X then MSI. This parameter determines which of these schemes the | |
83 | * driver may consider as follows: | |
84 | * | |
85 | * msi = 2: choose from among MSI-X and MSI | |
86 | * msi = 1: only consider MSI interrupts | |
87 | * | |
88 | * Note that unlike the Physical Function driver, this Virtual Function driver | |
89 | * does _not_ support legacy INTx interrupts (this limitation is mandated by | |
90 | * the PCI-E SR-IOV standard). | |
91 | */ | |
92 | #define MSI_MSIX 2 | |
93 | #define MSI_MSI 1 | |
94 | #define MSI_DEFAULT MSI_MSIX | |
95 | ||
96 | static int msi = MSI_DEFAULT; | |
97 | ||
98 | module_param(msi, int, 0644); | |
99 | MODULE_PARM_DESC(msi, "whether to use MSI-X or MSI"); | |
100 | ||
101 | /* | |
102 | * Fundamental constants. | |
103 | * ====================== | |
104 | */ | |
105 | ||
106 | enum { | |
107 | MAX_TXQ_ENTRIES = 16384, | |
108 | MAX_RSPQ_ENTRIES = 16384, | |
109 | MAX_RX_BUFFERS = 16384, | |
110 | ||
111 | MIN_TXQ_ENTRIES = 32, | |
112 | MIN_RSPQ_ENTRIES = 128, | |
113 | MIN_FL_ENTRIES = 16, | |
114 | ||
115 | /* | |
116 | * For purposes of manipulating the Free List size we need to | |
117 | * recognize that Free Lists are actually Egress Queues (the host | |
118 | * produces free buffers which the hardware consumes), Egress Queues | |
119 | * indices are all in units of Egress Context Units bytes, and free | |
120 | * list entries are 64-bit PCI DMA addresses. And since the state of | |
121 | * the Producer Index == the Consumer Index implies an EMPTY list, we | |
122 | * always have at least one Egress Unit's worth of Free List entries | |
123 | * unused. See sge.c for more details ... | |
124 | */ | |
125 | EQ_UNIT = SGE_EQ_IDXSIZE, | |
126 | FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64), | |
127 | MIN_FL_RESID = FL_PER_EQ_UNIT, | |
128 | }; | |
129 | ||
130 | /* | |
131 | * Global driver state. | |
132 | * ==================== | |
133 | */ | |
134 | ||
135 | static struct dentry *cxgb4vf_debugfs_root; | |
136 | ||
137 | /* | |
138 | * OS "Callback" functions. | |
139 | * ======================== | |
140 | */ | |
141 | ||
142 | /* | |
143 | * The link status has changed on the indicated "port" (Virtual Interface). | |
144 | */ | |
145 | void t4vf_os_link_changed(struct adapter *adapter, int pidx, int link_ok) | |
146 | { | |
147 | struct net_device *dev = adapter->port[pidx]; | |
148 | ||
149 | /* | |
150 | * If the port is disabled or the current recorded "link up" | |
151 | * status matches the new status, just return. | |
152 | */ | |
153 | if (!netif_running(dev) || link_ok == netif_carrier_ok(dev)) | |
154 | return; | |
155 | ||
156 | /* | |
157 | * Tell the OS that the link status has changed and print a short | |
158 | * informative message on the console about the event. | |
159 | */ | |
160 | if (link_ok) { | |
161 | const char *s; | |
162 | const char *fc; | |
163 | const struct port_info *pi = netdev_priv(dev); | |
164 | ||
165 | netif_carrier_on(dev); | |
166 | ||
167 | switch (pi->link_cfg.speed) { | |
897d55df HS |
168 | case 40000: |
169 | s = "40Gbps"; | |
170 | break; | |
171 | ||
172 | case 10000: | |
be839e39 CL |
173 | s = "10Gbps"; |
174 | break; | |
175 | ||
897d55df | 176 | case 1000: |
be839e39 CL |
177 | s = "1000Mbps"; |
178 | break; | |
179 | ||
897d55df | 180 | case 100: |
be839e39 CL |
181 | s = "100Mbps"; |
182 | break; | |
183 | ||
184 | default: | |
185 | s = "unknown"; | |
186 | break; | |
187 | } | |
188 | ||
189 | switch (pi->link_cfg.fc) { | |
190 | case PAUSE_RX: | |
191 | fc = "RX"; | |
192 | break; | |
193 | ||
194 | case PAUSE_TX: | |
195 | fc = "TX"; | |
196 | break; | |
197 | ||
198 | case PAUSE_RX|PAUSE_TX: | |
199 | fc = "RX/TX"; | |
200 | break; | |
201 | ||
202 | default: | |
203 | fc = "no"; | |
204 | break; | |
205 | } | |
206 | ||
428ac43f | 207 | netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s, fc); |
be839e39 CL |
208 | } else { |
209 | netif_carrier_off(dev); | |
428ac43f | 210 | netdev_info(dev, "link down\n"); |
be839e39 CL |
211 | } |
212 | } | |
213 | ||
5ad24def HS |
214 | /* |
215 | * THe port module type has changed on the indicated "port" (Virtual | |
216 | * Interface). | |
217 | */ | |
218 | void t4vf_os_portmod_changed(struct adapter *adapter, int pidx) | |
219 | { | |
220 | static const char * const mod_str[] = { | |
221 | NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM" | |
222 | }; | |
223 | const struct net_device *dev = adapter->port[pidx]; | |
224 | const struct port_info *pi = netdev_priv(dev); | |
225 | ||
226 | if (pi->mod_type == FW_PORT_MOD_TYPE_NONE) | |
227 | dev_info(adapter->pdev_dev, "%s: port module unplugged\n", | |
228 | dev->name); | |
229 | else if (pi->mod_type < ARRAY_SIZE(mod_str)) | |
230 | dev_info(adapter->pdev_dev, "%s: %s port module inserted\n", | |
231 | dev->name, mod_str[pi->mod_type]); | |
232 | else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED) | |
233 | dev_info(adapter->pdev_dev, "%s: unsupported optical port " | |
234 | "module inserted\n", dev->name); | |
235 | else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN) | |
236 | dev_info(adapter->pdev_dev, "%s: unknown port module inserted," | |
237 | "forcing TWINAX\n", dev->name); | |
238 | else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR) | |
239 | dev_info(adapter->pdev_dev, "%s: transceiver module error\n", | |
240 | dev->name); | |
241 | else | |
242 | dev_info(adapter->pdev_dev, "%s: unknown module type %d " | |
243 | "inserted\n", dev->name, pi->mod_type); | |
244 | } | |
245 | ||
be839e39 CL |
246 | /* |
247 | * Net device operations. | |
248 | * ====================== | |
249 | */ | |
250 | ||
be839e39 | 251 | |
87737663 | 252 | |
be839e39 CL |
253 | |
254 | /* | |
255 | * Perform the MAC and PHY actions needed to enable a "port" (Virtual | |
256 | * Interface). | |
257 | */ | |
258 | static int link_start(struct net_device *dev) | |
259 | { | |
260 | int ret; | |
261 | struct port_info *pi = netdev_priv(dev); | |
262 | ||
263 | /* | |
264 | * We do not set address filters and promiscuity here, the stack does | |
87737663 | 265 | * that step explicitly. Enable vlan accel. |
be839e39 | 266 | */ |
87737663 | 267 | ret = t4vf_set_rxmode(pi->adapter, pi->viid, dev->mtu, -1, -1, -1, 1, |
be839e39 CL |
268 | true); |
269 | if (ret == 0) { | |
270 | ret = t4vf_change_mac(pi->adapter, pi->viid, | |
271 | pi->xact_addr_filt, dev->dev_addr, true); | |
272 | if (ret >= 0) { | |
273 | pi->xact_addr_filt = ret; | |
274 | ret = 0; | |
275 | } | |
276 | } | |
277 | ||
278 | /* | |
279 | * We don't need to actually "start the link" itself since the | |
280 | * firmware will do that for us when the first Virtual Interface | |
281 | * is enabled on a port. | |
282 | */ | |
283 | if (ret == 0) | |
284 | ret = t4vf_enable_vi(pi->adapter, pi->viid, true, true); | |
285 | return ret; | |
286 | } | |
287 | ||
288 | /* | |
289 | * Name the MSI-X interrupts. | |
290 | */ | |
291 | static void name_msix_vecs(struct adapter *adapter) | |
292 | { | |
293 | int namelen = sizeof(adapter->msix_info[0].desc) - 1; | |
294 | int pidx; | |
295 | ||
296 | /* | |
297 | * Firmware events. | |
298 | */ | |
299 | snprintf(adapter->msix_info[MSIX_FW].desc, namelen, | |
300 | "%s-FWeventq", adapter->name); | |
301 | adapter->msix_info[MSIX_FW].desc[namelen] = 0; | |
302 | ||
303 | /* | |
304 | * Ethernet queues. | |
305 | */ | |
306 | for_each_port(adapter, pidx) { | |
307 | struct net_device *dev = adapter->port[pidx]; | |
308 | const struct port_info *pi = netdev_priv(dev); | |
309 | int qs, msi; | |
310 | ||
caedda35 | 311 | for (qs = 0, msi = MSIX_IQFLINT; qs < pi->nqsets; qs++, msi++) { |
be839e39 CL |
312 | snprintf(adapter->msix_info[msi].desc, namelen, |
313 | "%s-%d", dev->name, qs); | |
314 | adapter->msix_info[msi].desc[namelen] = 0; | |
315 | } | |
316 | } | |
317 | } | |
318 | ||
319 | /* | |
320 | * Request all of our MSI-X resources. | |
321 | */ | |
322 | static int request_msix_queue_irqs(struct adapter *adapter) | |
323 | { | |
324 | struct sge *s = &adapter->sge; | |
325 | int rxq, msi, err; | |
326 | ||
327 | /* | |
328 | * Firmware events. | |
329 | */ | |
330 | err = request_irq(adapter->msix_info[MSIX_FW].vec, t4vf_sge_intr_msix, | |
331 | 0, adapter->msix_info[MSIX_FW].desc, &s->fw_evtq); | |
332 | if (err) | |
333 | return err; | |
334 | ||
335 | /* | |
336 | * Ethernet queues. | |
337 | */ | |
caedda35 | 338 | msi = MSIX_IQFLINT; |
be839e39 CL |
339 | for_each_ethrxq(s, rxq) { |
340 | err = request_irq(adapter->msix_info[msi].vec, | |
341 | t4vf_sge_intr_msix, 0, | |
342 | adapter->msix_info[msi].desc, | |
343 | &s->ethrxq[rxq].rspq); | |
344 | if (err) | |
345 | goto err_free_irqs; | |
346 | msi++; | |
347 | } | |
348 | return 0; | |
349 | ||
350 | err_free_irqs: | |
351 | while (--rxq >= 0) | |
352 | free_irq(adapter->msix_info[--msi].vec, &s->ethrxq[rxq].rspq); | |
353 | free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq); | |
354 | return err; | |
355 | } | |
356 | ||
357 | /* | |
358 | * Free our MSI-X resources. | |
359 | */ | |
360 | static void free_msix_queue_irqs(struct adapter *adapter) | |
361 | { | |
362 | struct sge *s = &adapter->sge; | |
363 | int rxq, msi; | |
364 | ||
365 | free_irq(adapter->msix_info[MSIX_FW].vec, &s->fw_evtq); | |
caedda35 | 366 | msi = MSIX_IQFLINT; |
be839e39 CL |
367 | for_each_ethrxq(s, rxq) |
368 | free_irq(adapter->msix_info[msi++].vec, | |
369 | &s->ethrxq[rxq].rspq); | |
370 | } | |
371 | ||
372 | /* | |
373 | * Turn on NAPI and start up interrupts on a response queue. | |
374 | */ | |
375 | static void qenable(struct sge_rspq *rspq) | |
376 | { | |
377 | napi_enable(&rspq->napi); | |
378 | ||
379 | /* | |
380 | * 0-increment the Going To Sleep register to start the timer and | |
381 | * enable interrupts. | |
382 | */ | |
383 | t4_write_reg(rspq->adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS, | |
f612b815 HS |
384 | CIDXINC_V(0) | |
385 | SEINTARM_V(rspq->intr_params) | | |
386 | INGRESSQID_V(rspq->cntxt_id)); | |
be839e39 CL |
387 | } |
388 | ||
389 | /* | |
390 | * Enable NAPI scheduling and interrupt generation for all Receive Queues. | |
391 | */ | |
392 | static void enable_rx(struct adapter *adapter) | |
393 | { | |
394 | int rxq; | |
395 | struct sge *s = &adapter->sge; | |
396 | ||
397 | for_each_ethrxq(s, rxq) | |
398 | qenable(&s->ethrxq[rxq].rspq); | |
399 | qenable(&s->fw_evtq); | |
400 | ||
401 | /* | |
402 | * The interrupt queue doesn't use NAPI so we do the 0-increment of | |
403 | * its Going To Sleep register here to get it started. | |
404 | */ | |
405 | if (adapter->flags & USING_MSI) | |
406 | t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS, | |
f612b815 HS |
407 | CIDXINC_V(0) | |
408 | SEINTARM_V(s->intrq.intr_params) | | |
409 | INGRESSQID_V(s->intrq.cntxt_id)); | |
be839e39 CL |
410 | |
411 | } | |
412 | ||
413 | /* | |
414 | * Wait until all NAPI handlers are descheduled. | |
415 | */ | |
416 | static void quiesce_rx(struct adapter *adapter) | |
417 | { | |
418 | struct sge *s = &adapter->sge; | |
419 | int rxq; | |
420 | ||
421 | for_each_ethrxq(s, rxq) | |
422 | napi_disable(&s->ethrxq[rxq].rspq.napi); | |
423 | napi_disable(&s->fw_evtq.napi); | |
424 | } | |
425 | ||
426 | /* | |
427 | * Response queue handler for the firmware event queue. | |
428 | */ | |
429 | static int fwevtq_handler(struct sge_rspq *rspq, const __be64 *rsp, | |
430 | const struct pkt_gl *gl) | |
431 | { | |
432 | /* | |
433 | * Extract response opcode and get pointer to CPL message body. | |
434 | */ | |
435 | struct adapter *adapter = rspq->adapter; | |
436 | u8 opcode = ((const struct rss_header *)rsp)->opcode; | |
437 | void *cpl = (void *)(rsp + 1); | |
438 | ||
439 | switch (opcode) { | |
440 | case CPL_FW6_MSG: { | |
441 | /* | |
442 | * We've received an asynchronous message from the firmware. | |
443 | */ | |
444 | const struct cpl_fw6_msg *fw_msg = cpl; | |
445 | if (fw_msg->type == FW6_TYPE_CMD_RPL) | |
446 | t4vf_handle_fw_rpl(adapter, fw_msg->data); | |
447 | break; | |
448 | } | |
449 | ||
94dace10 VP |
450 | case CPL_FW4_MSG: { |
451 | /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG. | |
452 | */ | |
453 | const struct cpl_sge_egr_update *p = (void *)(rsp + 3); | |
6c53e938 | 454 | opcode = CPL_OPCODE_G(ntohl(p->opcode_qid)); |
94dace10 VP |
455 | if (opcode != CPL_SGE_EGR_UPDATE) { |
456 | dev_err(adapter->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n" | |
457 | , opcode); | |
458 | break; | |
459 | } | |
460 | cpl = (void *)p; | |
461 | /*FALLTHROUGH*/ | |
462 | } | |
463 | ||
be839e39 CL |
464 | case CPL_SGE_EGR_UPDATE: { |
465 | /* | |
7f9dd2fa CL |
466 | * We've received an Egress Queue Status Update message. We |
467 | * get these, if the SGE is configured to send these when the | |
468 | * firmware passes certain points in processing our TX | |
469 | * Ethernet Queue or if we make an explicit request for one. | |
470 | * We use these updates to determine when we may need to | |
471 | * restart a TX Ethernet Queue which was stopped for lack of | |
472 | * free TX Queue Descriptors ... | |
be839e39 | 473 | */ |
64699336 | 474 | const struct cpl_sge_egr_update *p = cpl; |
bdc590b9 | 475 | unsigned int qid = EGR_QID_G(be32_to_cpu(p->opcode_qid)); |
be839e39 CL |
476 | struct sge *s = &adapter->sge; |
477 | struct sge_txq *tq; | |
478 | struct sge_eth_txq *txq; | |
479 | unsigned int eq_idx; | |
be839e39 CL |
480 | |
481 | /* | |
482 | * Perform sanity checking on the Queue ID to make sure it | |
483 | * really refers to one of our TX Ethernet Egress Queues which | |
484 | * is active and matches the queue's ID. None of these error | |
485 | * conditions should ever happen so we may want to either make | |
486 | * them fatal and/or conditionalized under DEBUG. | |
487 | */ | |
488 | eq_idx = EQ_IDX(s, qid); | |
489 | if (unlikely(eq_idx >= MAX_EGRQ)) { | |
490 | dev_err(adapter->pdev_dev, | |
491 | "Egress Update QID %d out of range\n", qid); | |
492 | break; | |
493 | } | |
494 | tq = s->egr_map[eq_idx]; | |
495 | if (unlikely(tq == NULL)) { | |
496 | dev_err(adapter->pdev_dev, | |
497 | "Egress Update QID %d TXQ=NULL\n", qid); | |
498 | break; | |
499 | } | |
500 | txq = container_of(tq, struct sge_eth_txq, q); | |
501 | if (unlikely(tq->abs_id != qid)) { | |
502 | dev_err(adapter->pdev_dev, | |
503 | "Egress Update QID %d refers to TXQ %d\n", | |
504 | qid, tq->abs_id); | |
505 | break; | |
506 | } | |
507 | ||
be839e39 CL |
508 | /* |
509 | * Restart a stopped TX Queue which has less than half of its | |
510 | * TX ring in use ... | |
511 | */ | |
512 | txq->q.restarts++; | |
513 | netif_tx_wake_queue(txq->txq); | |
514 | break; | |
515 | } | |
516 | ||
517 | default: | |
518 | dev_err(adapter->pdev_dev, | |
519 | "unexpected CPL %#x on FW event queue\n", opcode); | |
520 | } | |
521 | ||
522 | return 0; | |
523 | } | |
524 | ||
525 | /* | |
526 | * Allocate SGE TX/RX response queues. Determine how many sets of SGE queues | |
527 | * to use and initializes them. We support multiple "Queue Sets" per port if | |
528 | * we have MSI-X, otherwise just one queue set per port. | |
529 | */ | |
530 | static int setup_sge_queues(struct adapter *adapter) | |
531 | { | |
532 | struct sge *s = &adapter->sge; | |
533 | int err, pidx, msix; | |
534 | ||
535 | /* | |
536 | * Clear "Queue Set" Free List Starving and TX Queue Mapping Error | |
537 | * state. | |
538 | */ | |
539 | bitmap_zero(s->starving_fl, MAX_EGRQ); | |
540 | ||
541 | /* | |
542 | * If we're using MSI interrupt mode we need to set up a "forwarded | |
543 | * interrupt" queue which we'll set up with our MSI vector. The rest | |
544 | * of the ingress queues will be set up to forward their interrupts to | |
545 | * this queue ... This must be first since t4vf_sge_alloc_rxq() uses | |
546 | * the intrq's queue ID as the interrupt forwarding queue for the | |
547 | * subsequent calls ... | |
548 | */ | |
549 | if (adapter->flags & USING_MSI) { | |
550 | err = t4vf_sge_alloc_rxq(adapter, &s->intrq, false, | |
551 | adapter->port[0], 0, NULL, NULL); | |
552 | if (err) | |
553 | goto err_free_queues; | |
554 | } | |
555 | ||
556 | /* | |
557 | * Allocate our ingress queue for asynchronous firmware messages. | |
558 | */ | |
559 | err = t4vf_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->port[0], | |
560 | MSIX_FW, NULL, fwevtq_handler); | |
561 | if (err) | |
562 | goto err_free_queues; | |
563 | ||
564 | /* | |
565 | * Allocate each "port"'s initial Queue Sets. These can be changed | |
566 | * later on ... up to the point where any interface on the adapter is | |
567 | * brought up at which point lots of things get nailed down | |
568 | * permanently ... | |
569 | */ | |
caedda35 | 570 | msix = MSIX_IQFLINT; |
be839e39 CL |
571 | for_each_port(adapter, pidx) { |
572 | struct net_device *dev = adapter->port[pidx]; | |
573 | struct port_info *pi = netdev_priv(dev); | |
574 | struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset]; | |
575 | struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset]; | |
be839e39 CL |
576 | int qs; |
577 | ||
c8639a82 | 578 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { |
be839e39 CL |
579 | err = t4vf_sge_alloc_rxq(adapter, &rxq->rspq, false, |
580 | dev, msix++, | |
581 | &rxq->fl, t4vf_ethrx_handler); | |
582 | if (err) | |
583 | goto err_free_queues; | |
584 | ||
585 | err = t4vf_sge_alloc_eth_txq(adapter, txq, dev, | |
586 | netdev_get_tx_queue(dev, qs), | |
587 | s->fw_evtq.cntxt_id); | |
588 | if (err) | |
589 | goto err_free_queues; | |
590 | ||
591 | rxq->rspq.idx = qs; | |
592 | memset(&rxq->stats, 0, sizeof(rxq->stats)); | |
593 | } | |
594 | } | |
595 | ||
596 | /* | |
597 | * Create the reverse mappings for the queues. | |
598 | */ | |
599 | s->egr_base = s->ethtxq[0].q.abs_id - s->ethtxq[0].q.cntxt_id; | |
600 | s->ingr_base = s->ethrxq[0].rspq.abs_id - s->ethrxq[0].rspq.cntxt_id; | |
601 | IQ_MAP(s, s->fw_evtq.abs_id) = &s->fw_evtq; | |
602 | for_each_port(adapter, pidx) { | |
603 | struct net_device *dev = adapter->port[pidx]; | |
604 | struct port_info *pi = netdev_priv(dev); | |
605 | struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset]; | |
606 | struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset]; | |
be839e39 CL |
607 | int qs; |
608 | ||
c8639a82 | 609 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { |
be839e39 CL |
610 | IQ_MAP(s, rxq->rspq.abs_id) = &rxq->rspq; |
611 | EQ_MAP(s, txq->q.abs_id) = &txq->q; | |
612 | ||
613 | /* | |
614 | * The FW_IQ_CMD doesn't return the Absolute Queue IDs | |
615 | * for Free Lists but since all of the Egress Queues | |
616 | * (including Free Lists) have Relative Queue IDs | |
617 | * which are computed as Absolute - Base Queue ID, we | |
618 | * can synthesize the Absolute Queue IDs for the Free | |
619 | * Lists. This is useful for debugging purposes when | |
620 | * we want to dump Queue Contexts via the PF Driver. | |
621 | */ | |
622 | rxq->fl.abs_id = rxq->fl.cntxt_id + s->egr_base; | |
623 | EQ_MAP(s, rxq->fl.abs_id) = &rxq->fl; | |
624 | } | |
625 | } | |
626 | return 0; | |
627 | ||
628 | err_free_queues: | |
629 | t4vf_free_sge_resources(adapter); | |
630 | return err; | |
631 | } | |
632 | ||
633 | /* | |
634 | * Set up Receive Side Scaling (RSS) to distribute packets to multiple receive | |
635 | * queues. We configure the RSS CPU lookup table to distribute to the number | |
636 | * of HW receive queues, and the response queue lookup table to narrow that | |
637 | * down to the response queues actually configured for each "port" (Virtual | |
638 | * Interface). We always configure the RSS mapping for all ports since the | |
639 | * mapping table has plenty of entries. | |
640 | */ | |
641 | static int setup_rss(struct adapter *adapter) | |
642 | { | |
643 | int pidx; | |
644 | ||
645 | for_each_port(adapter, pidx) { | |
646 | struct port_info *pi = adap2pinfo(adapter, pidx); | |
647 | struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset]; | |
648 | u16 rss[MAX_PORT_QSETS]; | |
649 | int qs, err; | |
650 | ||
651 | for (qs = 0; qs < pi->nqsets; qs++) | |
652 | rss[qs] = rxq[qs].rspq.abs_id; | |
653 | ||
654 | err = t4vf_config_rss_range(adapter, pi->viid, | |
655 | 0, pi->rss_size, rss, pi->nqsets); | |
656 | if (err) | |
657 | return err; | |
658 | ||
659 | /* | |
660 | * Perform Global RSS Mode-specific initialization. | |
661 | */ | |
662 | switch (adapter->params.rss.mode) { | |
663 | case FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL: | |
664 | /* | |
665 | * If Tunnel All Lookup isn't specified in the global | |
666 | * RSS Configuration, then we need to specify a | |
667 | * default Ingress Queue for any ingress packets which | |
668 | * aren't hashed. We'll use our first ingress queue | |
669 | * ... | |
670 | */ | |
671 | if (!adapter->params.rss.u.basicvirtual.tnlalllookup) { | |
672 | union rss_vi_config config; | |
673 | err = t4vf_read_rss_vi_config(adapter, | |
674 | pi->viid, | |
675 | &config); | |
676 | if (err) | |
677 | return err; | |
678 | config.basicvirtual.defaultq = | |
679 | rxq[0].rspq.abs_id; | |
680 | err = t4vf_write_rss_vi_config(adapter, | |
681 | pi->viid, | |
682 | &config); | |
683 | if (err) | |
684 | return err; | |
685 | } | |
686 | break; | |
687 | } | |
688 | } | |
689 | ||
690 | return 0; | |
691 | } | |
692 | ||
693 | /* | |
694 | * Bring the adapter up. Called whenever we go from no "ports" open to having | |
695 | * one open. This function performs the actions necessary to make an adapter | |
696 | * operational, such as completing the initialization of HW modules, and | |
697 | * enabling interrupts. Must be called with the rtnl lock held. (Note that | |
698 | * this is called "cxgb_up" in the PF Driver.) | |
699 | */ | |
700 | static int adapter_up(struct adapter *adapter) | |
701 | { | |
702 | int err; | |
703 | ||
704 | /* | |
705 | * If this is the first time we've been called, perform basic | |
706 | * adapter setup. Once we've done this, many of our adapter | |
707 | * parameters can no longer be changed ... | |
708 | */ | |
709 | if ((adapter->flags & FULL_INIT_DONE) == 0) { | |
710 | err = setup_sge_queues(adapter); | |
711 | if (err) | |
712 | return err; | |
713 | err = setup_rss(adapter); | |
714 | if (err) { | |
715 | t4vf_free_sge_resources(adapter); | |
716 | return err; | |
717 | } | |
718 | ||
719 | if (adapter->flags & USING_MSIX) | |
720 | name_msix_vecs(adapter); | |
721 | adapter->flags |= FULL_INIT_DONE; | |
722 | } | |
723 | ||
724 | /* | |
725 | * Acquire our interrupt resources. We only support MSI-X and MSI. | |
726 | */ | |
727 | BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0); | |
728 | if (adapter->flags & USING_MSIX) | |
729 | err = request_msix_queue_irqs(adapter); | |
730 | else | |
731 | err = request_irq(adapter->pdev->irq, | |
732 | t4vf_intr_handler(adapter), 0, | |
733 | adapter->name, adapter); | |
734 | if (err) { | |
735 | dev_err(adapter->pdev_dev, "request_irq failed, err %d\n", | |
736 | err); | |
737 | return err; | |
738 | } | |
739 | ||
740 | /* | |
741 | * Enable NAPI ingress processing and return success. | |
742 | */ | |
743 | enable_rx(adapter); | |
744 | t4vf_sge_start(adapter); | |
fe5d2709 HS |
745 | |
746 | /* Initialize hash mac addr list*/ | |
747 | INIT_LIST_HEAD(&adapter->mac_hlist); | |
be839e39 CL |
748 | return 0; |
749 | } | |
750 | ||
751 | /* | |
752 | * Bring the adapter down. Called whenever the last "port" (Virtual | |
753 | * Interface) closed. (Note that this routine is called "cxgb_down" in the PF | |
754 | * Driver.) | |
755 | */ | |
756 | static void adapter_down(struct adapter *adapter) | |
757 | { | |
758 | /* | |
759 | * Free interrupt resources. | |
760 | */ | |
761 | if (adapter->flags & USING_MSIX) | |
762 | free_msix_queue_irqs(adapter); | |
763 | else | |
764 | free_irq(adapter->pdev->irq, adapter); | |
765 | ||
766 | /* | |
767 | * Wait for NAPI handlers to finish. | |
768 | */ | |
769 | quiesce_rx(adapter); | |
770 | } | |
771 | ||
772 | /* | |
773 | * Start up a net device. | |
774 | */ | |
775 | static int cxgb4vf_open(struct net_device *dev) | |
776 | { | |
777 | int err; | |
778 | struct port_info *pi = netdev_priv(dev); | |
779 | struct adapter *adapter = pi->adapter; | |
780 | ||
781 | /* | |
782 | * If this is the first interface that we're opening on the "adapter", | |
783 | * bring the "adapter" up now. | |
784 | */ | |
785 | if (adapter->open_device_map == 0) { | |
786 | err = adapter_up(adapter); | |
787 | if (err) | |
788 | return err; | |
789 | } | |
790 | ||
791 | /* | |
792 | * Note that this interface is up and start everything up ... | |
793 | */ | |
e7a3795f CL |
794 | err = link_start(dev); |
795 | if (err) | |
343a8d13 CL |
796 | goto err_unwind; |
797 | ||
be839e39 | 798 | netif_tx_start_all_queues(dev); |
343a8d13 | 799 | set_bit(pi->port_id, &adapter->open_device_map); |
be839e39 | 800 | return 0; |
343a8d13 CL |
801 | |
802 | err_unwind: | |
803 | if (adapter->open_device_map == 0) | |
804 | adapter_down(adapter); | |
805 | return err; | |
be839e39 CL |
806 | } |
807 | ||
808 | /* | |
809 | * Shut down a net device. This routine is called "cxgb_close" in the PF | |
810 | * Driver ... | |
811 | */ | |
812 | static int cxgb4vf_stop(struct net_device *dev) | |
813 | { | |
be839e39 CL |
814 | struct port_info *pi = netdev_priv(dev); |
815 | struct adapter *adapter = pi->adapter; | |
816 | ||
817 | netif_tx_stop_all_queues(dev); | |
818 | netif_carrier_off(dev); | |
343a8d13 | 819 | t4vf_enable_vi(adapter, pi->viid, false, false); |
be839e39 CL |
820 | pi->link_cfg.link_ok = 0; |
821 | ||
822 | clear_bit(pi->port_id, &adapter->open_device_map); | |
823 | if (adapter->open_device_map == 0) | |
824 | adapter_down(adapter); | |
825 | return 0; | |
826 | } | |
827 | ||
828 | /* | |
829 | * Translate our basic statistics into the standard "ifconfig" statistics. | |
830 | */ | |
831 | static struct net_device_stats *cxgb4vf_get_stats(struct net_device *dev) | |
832 | { | |
833 | struct t4vf_port_stats stats; | |
834 | struct port_info *pi = netdev2pinfo(dev); | |
835 | struct adapter *adapter = pi->adapter; | |
836 | struct net_device_stats *ns = &dev->stats; | |
837 | int err; | |
838 | ||
839 | spin_lock(&adapter->stats_lock); | |
840 | err = t4vf_get_port_stats(adapter, pi->pidx, &stats); | |
841 | spin_unlock(&adapter->stats_lock); | |
842 | ||
843 | memset(ns, 0, sizeof(*ns)); | |
844 | if (err) | |
845 | return ns; | |
846 | ||
847 | ns->tx_bytes = (stats.tx_bcast_bytes + stats.tx_mcast_bytes + | |
848 | stats.tx_ucast_bytes + stats.tx_offload_bytes); | |
849 | ns->tx_packets = (stats.tx_bcast_frames + stats.tx_mcast_frames + | |
850 | stats.tx_ucast_frames + stats.tx_offload_frames); | |
851 | ns->rx_bytes = (stats.rx_bcast_bytes + stats.rx_mcast_bytes + | |
852 | stats.rx_ucast_bytes); | |
853 | ns->rx_packets = (stats.rx_bcast_frames + stats.rx_mcast_frames + | |
854 | stats.rx_ucast_frames); | |
855 | ns->multicast = stats.rx_mcast_frames; | |
856 | ns->tx_errors = stats.tx_drop_frames; | |
857 | ns->rx_errors = stats.rx_err_frames; | |
858 | ||
859 | return ns; | |
860 | } | |
861 | ||
fe5d2709 HS |
862 | static inline int cxgb4vf_set_addr_hash(struct port_info *pi) |
863 | { | |
864 | struct adapter *adapter = pi->adapter; | |
865 | u64 vec = 0; | |
866 | bool ucast = false; | |
867 | struct hash_mac_addr *entry; | |
868 | ||
869 | /* Calculate the hash vector for the updated list and program it */ | |
870 | list_for_each_entry(entry, &adapter->mac_hlist, list) { | |
871 | ucast |= is_unicast_ether_addr(entry->addr); | |
872 | vec |= (1ULL << hash_mac_addr(entry->addr)); | |
873 | } | |
874 | return t4vf_set_addr_hash(adapter, pi->viid, ucast, vec, false); | |
875 | } | |
876 | ||
877 | static int cxgb4vf_mac_sync(struct net_device *netdev, const u8 *mac_addr) | |
be839e39 | 878 | { |
fe5d2709 HS |
879 | struct port_info *pi = netdev_priv(netdev); |
880 | struct adapter *adapter = pi->adapter; | |
881 | int ret; | |
be839e39 CL |
882 | u64 mhash = 0; |
883 | u64 uhash = 0; | |
fe5d2709 HS |
884 | bool free = false; |
885 | bool ucast = is_unicast_ether_addr(mac_addr); | |
886 | const u8 *maclist[1] = {mac_addr}; | |
887 | struct hash_mac_addr *new_entry; | |
888 | ||
889 | ret = t4vf_alloc_mac_filt(adapter, pi->viid, free, 1, maclist, | |
890 | NULL, ucast ? &uhash : &mhash, false); | |
891 | if (ret < 0) | |
892 | goto out; | |
893 | /* if hash != 0, then add the addr to hash addr list | |
894 | * so on the end we will calculate the hash for the | |
895 | * list and program it | |
896 | */ | |
897 | if (uhash || mhash) { | |
898 | new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC); | |
899 | if (!new_entry) | |
900 | return -ENOMEM; | |
901 | ether_addr_copy(new_entry->addr, mac_addr); | |
902 | list_add_tail(&new_entry->list, &adapter->mac_hlist); | |
903 | ret = cxgb4vf_set_addr_hash(pi); | |
be839e39 | 904 | } |
fe5d2709 HS |
905 | out: |
906 | return ret < 0 ? ret : 0; | |
907 | } | |
be839e39 | 908 | |
fe5d2709 HS |
909 | static int cxgb4vf_mac_unsync(struct net_device *netdev, const u8 *mac_addr) |
910 | { | |
911 | struct port_info *pi = netdev_priv(netdev); | |
912 | struct adapter *adapter = pi->adapter; | |
913 | int ret; | |
914 | const u8 *maclist[1] = {mac_addr}; | |
915 | struct hash_mac_addr *entry, *tmp; | |
42eb59d3 | 916 | |
fe5d2709 HS |
917 | /* If the MAC address to be removed is in the hash addr |
918 | * list, delete it from the list and update hash vector | |
919 | */ | |
920 | list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist, list) { | |
921 | if (ether_addr_equal(entry->addr, mac_addr)) { | |
922 | list_del(&entry->list); | |
923 | kfree(entry); | |
924 | return cxgb4vf_set_addr_hash(pi); | |
925 | } | |
be839e39 CL |
926 | } |
927 | ||
fe5d2709 HS |
928 | ret = t4vf_free_mac_filt(adapter, pi->viid, 1, maclist, false); |
929 | return ret < 0 ? -EINVAL : 0; | |
be839e39 CL |
930 | } |
931 | ||
932 | /* | |
933 | * Set RX properties of a port, such as promiscruity, address filters, and MTU. | |
934 | * If @mtu is -1 it is left unchanged. | |
935 | */ | |
936 | static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok) | |
937 | { | |
be839e39 CL |
938 | struct port_info *pi = netdev_priv(dev); |
939 | ||
d01f7abc HS |
940 | __dev_uc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync); |
941 | __dev_mc_sync(dev, cxgb4vf_mac_sync, cxgb4vf_mac_unsync); | |
fe5d2709 HS |
942 | return t4vf_set_rxmode(pi->adapter, pi->viid, -1, |
943 | (dev->flags & IFF_PROMISC) != 0, | |
944 | (dev->flags & IFF_ALLMULTI) != 0, | |
945 | 1, -1, sleep_ok); | |
be839e39 CL |
946 | } |
947 | ||
948 | /* | |
949 | * Set the current receive modes on the device. | |
950 | */ | |
951 | static void cxgb4vf_set_rxmode(struct net_device *dev) | |
952 | { | |
953 | /* unfortunately we can't return errors to the stack */ | |
954 | set_rxmode(dev, -1, false); | |
955 | } | |
956 | ||
957 | /* | |
958 | * Find the entry in the interrupt holdoff timer value array which comes | |
959 | * closest to the specified interrupt holdoff value. | |
960 | */ | |
961 | static int closest_timer(const struct sge *s, int us) | |
962 | { | |
963 | int i, timer_idx = 0, min_delta = INT_MAX; | |
964 | ||
965 | for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) { | |
966 | int delta = us - s->timer_val[i]; | |
967 | if (delta < 0) | |
968 | delta = -delta; | |
969 | if (delta < min_delta) { | |
970 | min_delta = delta; | |
971 | timer_idx = i; | |
972 | } | |
973 | } | |
974 | return timer_idx; | |
975 | } | |
976 | ||
977 | static int closest_thres(const struct sge *s, int thres) | |
978 | { | |
979 | int i, delta, pktcnt_idx = 0, min_delta = INT_MAX; | |
980 | ||
981 | for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) { | |
982 | delta = thres - s->counter_val[i]; | |
983 | if (delta < 0) | |
984 | delta = -delta; | |
985 | if (delta < min_delta) { | |
986 | min_delta = delta; | |
987 | pktcnt_idx = i; | |
988 | } | |
989 | } | |
990 | return pktcnt_idx; | |
991 | } | |
992 | ||
993 | /* | |
994 | * Return a queue's interrupt hold-off time in us. 0 means no timer. | |
995 | */ | |
996 | static unsigned int qtimer_val(const struct adapter *adapter, | |
997 | const struct sge_rspq *rspq) | |
998 | { | |
1ecc7b7a | 999 | unsigned int timer_idx = QINTR_TIMER_IDX_G(rspq->intr_params); |
be839e39 CL |
1000 | |
1001 | return timer_idx < SGE_NTIMERS | |
1002 | ? adapter->sge.timer_val[timer_idx] | |
1003 | : 0; | |
1004 | } | |
1005 | ||
1006 | /** | |
1007 | * set_rxq_intr_params - set a queue's interrupt holdoff parameters | |
1008 | * @adapter: the adapter | |
1009 | * @rspq: the RX response queue | |
1010 | * @us: the hold-off time in us, or 0 to disable timer | |
1011 | * @cnt: the hold-off packet count, or 0 to disable counter | |
1012 | * | |
1013 | * Sets an RX response queue's interrupt hold-off time and packet count. | |
1014 | * At least one of the two needs to be enabled for the queue to generate | |
1015 | * interrupts. | |
1016 | */ | |
1017 | static int set_rxq_intr_params(struct adapter *adapter, struct sge_rspq *rspq, | |
1018 | unsigned int us, unsigned int cnt) | |
1019 | { | |
1020 | unsigned int timer_idx; | |
1021 | ||
1022 | /* | |
1023 | * If both the interrupt holdoff timer and count are specified as | |
1024 | * zero, default to a holdoff count of 1 ... | |
1025 | */ | |
1026 | if ((us | cnt) == 0) | |
1027 | cnt = 1; | |
1028 | ||
1029 | /* | |
1030 | * If an interrupt holdoff count has been specified, then find the | |
1031 | * closest configured holdoff count and use that. If the response | |
1032 | * queue has already been created, then update its queue context | |
1033 | * parameters ... | |
1034 | */ | |
1035 | if (cnt) { | |
1036 | int err; | |
1037 | u32 v, pktcnt_idx; | |
1038 | ||
1039 | pktcnt_idx = closest_thres(&adapter->sge, cnt); | |
1040 | if (rspq->desc && rspq->pktcnt_idx != pktcnt_idx) { | |
5167865a HS |
1041 | v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) | |
1042 | FW_PARAMS_PARAM_X_V( | |
be839e39 | 1043 | FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) | |
5167865a | 1044 | FW_PARAMS_PARAM_YZ_V(rspq->cntxt_id); |
be839e39 CL |
1045 | err = t4vf_set_params(adapter, 1, &v, &pktcnt_idx); |
1046 | if (err) | |
1047 | return err; | |
1048 | } | |
1049 | rspq->pktcnt_idx = pktcnt_idx; | |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * Compute the closest holdoff timer index from the supplied holdoff | |
1054 | * timer value. | |
1055 | */ | |
1056 | timer_idx = (us == 0 | |
1057 | ? SGE_TIMER_RSTRT_CNTR | |
1058 | : closest_timer(&adapter->sge, us)); | |
1059 | ||
1060 | /* | |
1061 | * Update the response queue's interrupt coalescing parameters and | |
1062 | * return success. | |
1063 | */ | |
1ecc7b7a HS |
1064 | rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) | |
1065 | QINTR_CNT_EN_V(cnt > 0)); | |
be839e39 CL |
1066 | return 0; |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * Return a version number to identify the type of adapter. The scheme is: | |
1071 | * - bits 0..9: chip version | |
1072 | * - bits 10..15: chip revision | |
1073 | */ | |
1074 | static inline unsigned int mk_adap_vers(const struct adapter *adapter) | |
1075 | { | |
1076 | /* | |
1077 | * Chip version 4, revision 0x3f (cxgb4vf). | |
1078 | */ | |
70ee3666 | 1079 | return CHELSIO_CHIP_VERSION(adapter->params.chip) | (0x3f << 10); |
be839e39 CL |
1080 | } |
1081 | ||
1082 | /* | |
1083 | * Execute the specified ioctl command. | |
1084 | */ | |
1085 | static int cxgb4vf_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd) | |
1086 | { | |
1087 | int ret = 0; | |
1088 | ||
1089 | switch (cmd) { | |
1090 | /* | |
1091 | * The VF Driver doesn't have access to any of the other | |
1092 | * common Ethernet device ioctl()'s (like reading/writing | |
1093 | * PHY registers, etc. | |
1094 | */ | |
1095 | ||
1096 | default: | |
1097 | ret = -EOPNOTSUPP; | |
1098 | break; | |
1099 | } | |
1100 | return ret; | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Change the device's MTU. | |
1105 | */ | |
1106 | static int cxgb4vf_change_mtu(struct net_device *dev, int new_mtu) | |
1107 | { | |
1108 | int ret; | |
1109 | struct port_info *pi = netdev_priv(dev); | |
1110 | ||
1111 | /* accommodate SACK */ | |
1112 | if (new_mtu < 81) | |
1113 | return -EINVAL; | |
1114 | ||
1115 | ret = t4vf_set_rxmode(pi->adapter, pi->viid, new_mtu, | |
1116 | -1, -1, -1, -1, true); | |
1117 | if (!ret) | |
1118 | dev->mtu = new_mtu; | |
1119 | return ret; | |
1120 | } | |
1121 | ||
c8f44aff MM |
1122 | static netdev_features_t cxgb4vf_fix_features(struct net_device *dev, |
1123 | netdev_features_t features) | |
87737663 JP |
1124 | { |
1125 | /* | |
1126 | * Since there is no support for separate rx/tx vlan accel | |
1127 | * enable/disable make sure tx flag is always in same state as rx. | |
1128 | */ | |
f646968f PM |
1129 | if (features & NETIF_F_HW_VLAN_CTAG_RX) |
1130 | features |= NETIF_F_HW_VLAN_CTAG_TX; | |
87737663 | 1131 | else |
f646968f | 1132 | features &= ~NETIF_F_HW_VLAN_CTAG_TX; |
87737663 JP |
1133 | |
1134 | return features; | |
1135 | } | |
1136 | ||
c8f44aff MM |
1137 | static int cxgb4vf_set_features(struct net_device *dev, |
1138 | netdev_features_t features) | |
87737663 JP |
1139 | { |
1140 | struct port_info *pi = netdev_priv(dev); | |
c8f44aff | 1141 | netdev_features_t changed = dev->features ^ features; |
87737663 | 1142 | |
f646968f | 1143 | if (changed & NETIF_F_HW_VLAN_CTAG_RX) |
87737663 | 1144 | t4vf_set_rxmode(pi->adapter, pi->viid, -1, -1, -1, -1, |
f646968f | 1145 | features & NETIF_F_HW_VLAN_CTAG_TX, 0); |
87737663 JP |
1146 | |
1147 | return 0; | |
1148 | } | |
1149 | ||
be839e39 CL |
1150 | /* |
1151 | * Change the devices MAC address. | |
1152 | */ | |
1153 | static int cxgb4vf_set_mac_addr(struct net_device *dev, void *_addr) | |
1154 | { | |
1155 | int ret; | |
1156 | struct sockaddr *addr = _addr; | |
1157 | struct port_info *pi = netdev_priv(dev); | |
1158 | ||
1159 | if (!is_valid_ether_addr(addr->sa_data)) | |
504f9b5a | 1160 | return -EADDRNOTAVAIL; |
be839e39 CL |
1161 | |
1162 | ret = t4vf_change_mac(pi->adapter, pi->viid, pi->xact_addr_filt, | |
1163 | addr->sa_data, true); | |
1164 | if (ret < 0) | |
1165 | return ret; | |
1166 | ||
1167 | memcpy(dev->dev_addr, addr->sa_data, dev->addr_len); | |
1168 | pi->xact_addr_filt = ret; | |
1169 | return 0; | |
1170 | } | |
1171 | ||
be839e39 CL |
1172 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1173 | /* | |
1174 | * Poll all of our receive queues. This is called outside of normal interrupt | |
1175 | * context. | |
1176 | */ | |
1177 | static void cxgb4vf_poll_controller(struct net_device *dev) | |
1178 | { | |
1179 | struct port_info *pi = netdev_priv(dev); | |
1180 | struct adapter *adapter = pi->adapter; | |
1181 | ||
1182 | if (adapter->flags & USING_MSIX) { | |
1183 | struct sge_eth_rxq *rxq; | |
1184 | int nqsets; | |
1185 | ||
1186 | rxq = &adapter->sge.ethrxq[pi->first_qset]; | |
1187 | for (nqsets = pi->nqsets; nqsets; nqsets--) { | |
1188 | t4vf_sge_intr_msix(0, &rxq->rspq); | |
1189 | rxq++; | |
1190 | } | |
1191 | } else | |
1192 | t4vf_intr_handler(adapter)(0, adapter); | |
1193 | } | |
1194 | #endif | |
1195 | ||
1196 | /* | |
1197 | * Ethtool operations. | |
1198 | * =================== | |
1199 | * | |
1200 | * Note that we don't support any ethtool operations which change the physical | |
1201 | * state of the port to which we're linked. | |
1202 | */ | |
1203 | ||
5ad24def HS |
1204 | static unsigned int t4vf_from_fw_linkcaps(enum fw_port_type type, |
1205 | unsigned int caps) | |
be839e39 | 1206 | { |
5ad24def HS |
1207 | unsigned int v = 0; |
1208 | ||
1209 | if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI || | |
1210 | type == FW_PORT_TYPE_BT_XAUI) { | |
1211 | v |= SUPPORTED_TP; | |
1212 | if (caps & FW_PORT_CAP_SPEED_100M) | |
1213 | v |= SUPPORTED_100baseT_Full; | |
1214 | if (caps & FW_PORT_CAP_SPEED_1G) | |
1215 | v |= SUPPORTED_1000baseT_Full; | |
1216 | if (caps & FW_PORT_CAP_SPEED_10G) | |
1217 | v |= SUPPORTED_10000baseT_Full; | |
1218 | } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) { | |
1219 | v |= SUPPORTED_Backplane; | |
1220 | if (caps & FW_PORT_CAP_SPEED_1G) | |
1221 | v |= SUPPORTED_1000baseKX_Full; | |
1222 | if (caps & FW_PORT_CAP_SPEED_10G) | |
1223 | v |= SUPPORTED_10000baseKX4_Full; | |
1224 | } else if (type == FW_PORT_TYPE_KR) | |
1225 | v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full; | |
1226 | else if (type == FW_PORT_TYPE_BP_AP) | |
1227 | v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC | | |
1228 | SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full; | |
1229 | else if (type == FW_PORT_TYPE_BP4_AP) | |
1230 | v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC | | |
1231 | SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full | | |
1232 | SUPPORTED_10000baseKX4_Full; | |
1233 | else if (type == FW_PORT_TYPE_FIBER_XFI || | |
1234 | type == FW_PORT_TYPE_FIBER_XAUI || | |
1235 | type == FW_PORT_TYPE_SFP || | |
1236 | type == FW_PORT_TYPE_QSFP_10G || | |
1237 | type == FW_PORT_TYPE_QSA) { | |
1238 | v |= SUPPORTED_FIBRE; | |
1239 | if (caps & FW_PORT_CAP_SPEED_1G) | |
1240 | v |= SUPPORTED_1000baseT_Full; | |
1241 | if (caps & FW_PORT_CAP_SPEED_10G) | |
1242 | v |= SUPPORTED_10000baseT_Full; | |
1243 | } else if (type == FW_PORT_TYPE_BP40_BA || | |
1244 | type == FW_PORT_TYPE_QSFP) { | |
1245 | v |= SUPPORTED_40000baseSR4_Full; | |
1246 | v |= SUPPORTED_FIBRE; | |
1247 | } | |
1248 | ||
1249 | if (caps & FW_PORT_CAP_ANEG) | |
1250 | v |= SUPPORTED_Autoneg; | |
1251 | return v; | |
1252 | } | |
be839e39 | 1253 | |
5ad24def HS |
1254 | static int cxgb4vf_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) |
1255 | { | |
1256 | const struct port_info *p = netdev_priv(dev); | |
1257 | ||
1258 | if (p->port_type == FW_PORT_TYPE_BT_SGMII || | |
1259 | p->port_type == FW_PORT_TYPE_BT_XFI || | |
1260 | p->port_type == FW_PORT_TYPE_BT_XAUI) | |
1261 | cmd->port = PORT_TP; | |
1262 | else if (p->port_type == FW_PORT_TYPE_FIBER_XFI || | |
1263 | p->port_type == FW_PORT_TYPE_FIBER_XAUI) | |
1264 | cmd->port = PORT_FIBRE; | |
1265 | else if (p->port_type == FW_PORT_TYPE_SFP || | |
1266 | p->port_type == FW_PORT_TYPE_QSFP_10G || | |
1267 | p->port_type == FW_PORT_TYPE_QSA || | |
1268 | p->port_type == FW_PORT_TYPE_QSFP) { | |
1269 | if (p->mod_type == FW_PORT_MOD_TYPE_LR || | |
1270 | p->mod_type == FW_PORT_MOD_TYPE_SR || | |
1271 | p->mod_type == FW_PORT_MOD_TYPE_ER || | |
1272 | p->mod_type == FW_PORT_MOD_TYPE_LRM) | |
1273 | cmd->port = PORT_FIBRE; | |
1274 | else if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE || | |
1275 | p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE) | |
1276 | cmd->port = PORT_DA; | |
1277 | else | |
1278 | cmd->port = PORT_OTHER; | |
1279 | } else | |
1280 | cmd->port = PORT_OTHER; | |
1281 | ||
1282 | if (p->mdio_addr >= 0) { | |
1283 | cmd->phy_address = p->mdio_addr; | |
1284 | cmd->transceiver = XCVR_EXTERNAL; | |
1285 | cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ? | |
1286 | MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45; | |
1287 | } else { | |
1288 | cmd->phy_address = 0; /* not really, but no better option */ | |
1289 | cmd->transceiver = XCVR_INTERNAL; | |
1290 | cmd->mdio_support = 0; | |
1291 | } | |
1292 | ||
1293 | cmd->supported = t4vf_from_fw_linkcaps(p->port_type, | |
1294 | p->link_cfg.supported); | |
1295 | cmd->advertising = t4vf_from_fw_linkcaps(p->port_type, | |
1296 | p->link_cfg.advertising); | |
70739497 | 1297 | ethtool_cmd_speed_set(cmd, |
5ad24def | 1298 | netif_carrier_ok(dev) ? p->link_cfg.speed : 0); |
be839e39 | 1299 | cmd->duplex = DUPLEX_FULL; |
5ad24def | 1300 | cmd->autoneg = p->link_cfg.autoneg; |
be839e39 CL |
1301 | cmd->maxtxpkt = 0; |
1302 | cmd->maxrxpkt = 0; | |
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * Return our driver information. | |
1308 | */ | |
1309 | static void cxgb4vf_get_drvinfo(struct net_device *dev, | |
1310 | struct ethtool_drvinfo *drvinfo) | |
1311 | { | |
1312 | struct adapter *adapter = netdev2adap(dev); | |
1313 | ||
23020ab3 RJ |
1314 | strlcpy(drvinfo->driver, KBUILD_MODNAME, sizeof(drvinfo->driver)); |
1315 | strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); | |
1316 | strlcpy(drvinfo->bus_info, pci_name(to_pci_dev(dev->dev.parent)), | |
1317 | sizeof(drvinfo->bus_info)); | |
be839e39 CL |
1318 | snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), |
1319 | "%u.%u.%u.%u, TP %u.%u.%u.%u", | |
b2e1a3f0 HS |
1320 | FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.fwrev), |
1321 | FW_HDR_FW_VER_MINOR_G(adapter->params.dev.fwrev), | |
1322 | FW_HDR_FW_VER_MICRO_G(adapter->params.dev.fwrev), | |
1323 | FW_HDR_FW_VER_BUILD_G(adapter->params.dev.fwrev), | |
1324 | FW_HDR_FW_VER_MAJOR_G(adapter->params.dev.tprev), | |
1325 | FW_HDR_FW_VER_MINOR_G(adapter->params.dev.tprev), | |
1326 | FW_HDR_FW_VER_MICRO_G(adapter->params.dev.tprev), | |
1327 | FW_HDR_FW_VER_BUILD_G(adapter->params.dev.tprev)); | |
be839e39 CL |
1328 | } |
1329 | ||
1330 | /* | |
1331 | * Return current adapter message level. | |
1332 | */ | |
1333 | static u32 cxgb4vf_get_msglevel(struct net_device *dev) | |
1334 | { | |
1335 | return netdev2adap(dev)->msg_enable; | |
1336 | } | |
1337 | ||
1338 | /* | |
1339 | * Set current adapter message level. | |
1340 | */ | |
1341 | static void cxgb4vf_set_msglevel(struct net_device *dev, u32 msglevel) | |
1342 | { | |
1343 | netdev2adap(dev)->msg_enable = msglevel; | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Return the device's current Queue Set ring size parameters along with the | |
1348 | * allowed maximum values. Since ethtool doesn't understand the concept of | |
1349 | * multi-queue devices, we just return the current values associated with the | |
1350 | * first Queue Set. | |
1351 | */ | |
1352 | static void cxgb4vf_get_ringparam(struct net_device *dev, | |
1353 | struct ethtool_ringparam *rp) | |
1354 | { | |
1355 | const struct port_info *pi = netdev_priv(dev); | |
1356 | const struct sge *s = &pi->adapter->sge; | |
1357 | ||
1358 | rp->rx_max_pending = MAX_RX_BUFFERS; | |
1359 | rp->rx_mini_max_pending = MAX_RSPQ_ENTRIES; | |
1360 | rp->rx_jumbo_max_pending = 0; | |
1361 | rp->tx_max_pending = MAX_TXQ_ENTRIES; | |
1362 | ||
1363 | rp->rx_pending = s->ethrxq[pi->first_qset].fl.size - MIN_FL_RESID; | |
1364 | rp->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size; | |
1365 | rp->rx_jumbo_pending = 0; | |
1366 | rp->tx_pending = s->ethtxq[pi->first_qset].q.size; | |
1367 | } | |
1368 | ||
1369 | /* | |
1370 | * Set the Queue Set ring size parameters for the device. Again, since | |
1371 | * ethtool doesn't allow for the concept of multiple queues per device, we'll | |
1372 | * apply these new values across all of the Queue Sets associated with the | |
1373 | * device -- after vetting them of course! | |
1374 | */ | |
1375 | static int cxgb4vf_set_ringparam(struct net_device *dev, | |
1376 | struct ethtool_ringparam *rp) | |
1377 | { | |
1378 | const struct port_info *pi = netdev_priv(dev); | |
1379 | struct adapter *adapter = pi->adapter; | |
1380 | struct sge *s = &adapter->sge; | |
1381 | int qs; | |
1382 | ||
1383 | if (rp->rx_pending > MAX_RX_BUFFERS || | |
1384 | rp->rx_jumbo_pending || | |
1385 | rp->tx_pending > MAX_TXQ_ENTRIES || | |
1386 | rp->rx_mini_pending > MAX_RSPQ_ENTRIES || | |
1387 | rp->rx_mini_pending < MIN_RSPQ_ENTRIES || | |
1388 | rp->rx_pending < MIN_FL_ENTRIES || | |
1389 | rp->tx_pending < MIN_TXQ_ENTRIES) | |
1390 | return -EINVAL; | |
1391 | ||
1392 | if (adapter->flags & FULL_INIT_DONE) | |
1393 | return -EBUSY; | |
1394 | ||
1395 | for (qs = pi->first_qset; qs < pi->first_qset + pi->nqsets; qs++) { | |
1396 | s->ethrxq[qs].fl.size = rp->rx_pending + MIN_FL_RESID; | |
1397 | s->ethrxq[qs].rspq.size = rp->rx_mini_pending; | |
1398 | s->ethtxq[qs].q.size = rp->tx_pending; | |
1399 | } | |
1400 | return 0; | |
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * Return the interrupt holdoff timer and count for the first Queue Set on the | |
1405 | * device. Our extension ioctl() (the cxgbtool interface) allows the | |
1406 | * interrupt holdoff timer to be read on all of the device's Queue Sets. | |
1407 | */ | |
1408 | static int cxgb4vf_get_coalesce(struct net_device *dev, | |
1409 | struct ethtool_coalesce *coalesce) | |
1410 | { | |
1411 | const struct port_info *pi = netdev_priv(dev); | |
1412 | const struct adapter *adapter = pi->adapter; | |
1413 | const struct sge_rspq *rspq = &adapter->sge.ethrxq[pi->first_qset].rspq; | |
1414 | ||
1415 | coalesce->rx_coalesce_usecs = qtimer_val(adapter, rspq); | |
1416 | coalesce->rx_max_coalesced_frames = | |
1ecc7b7a | 1417 | ((rspq->intr_params & QINTR_CNT_EN_F) |
be839e39 CL |
1418 | ? adapter->sge.counter_val[rspq->pktcnt_idx] |
1419 | : 0); | |
1420 | return 0; | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * Set the RX interrupt holdoff timer and count for the first Queue Set on the | |
1425 | * interface. Our extension ioctl() (the cxgbtool interface) allows us to set | |
1426 | * the interrupt holdoff timer on any of the device's Queue Sets. | |
1427 | */ | |
1428 | static int cxgb4vf_set_coalesce(struct net_device *dev, | |
1429 | struct ethtool_coalesce *coalesce) | |
1430 | { | |
1431 | const struct port_info *pi = netdev_priv(dev); | |
1432 | struct adapter *adapter = pi->adapter; | |
1433 | ||
1434 | return set_rxq_intr_params(adapter, | |
1435 | &adapter->sge.ethrxq[pi->first_qset].rspq, | |
1436 | coalesce->rx_coalesce_usecs, | |
1437 | coalesce->rx_max_coalesced_frames); | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * Report current port link pause parameter settings. | |
1442 | */ | |
1443 | static void cxgb4vf_get_pauseparam(struct net_device *dev, | |
1444 | struct ethtool_pauseparam *pauseparam) | |
1445 | { | |
1446 | struct port_info *pi = netdev_priv(dev); | |
1447 | ||
1448 | pauseparam->autoneg = (pi->link_cfg.requested_fc & PAUSE_AUTONEG) != 0; | |
1449 | pauseparam->rx_pause = (pi->link_cfg.fc & PAUSE_RX) != 0; | |
1450 | pauseparam->tx_pause = (pi->link_cfg.fc & PAUSE_TX) != 0; | |
1451 | } | |
1452 | ||
be839e39 CL |
1453 | /* |
1454 | * Identify the port by blinking the port's LED. | |
1455 | */ | |
857a3d0f DM |
1456 | static int cxgb4vf_phys_id(struct net_device *dev, |
1457 | enum ethtool_phys_id_state state) | |
be839e39 | 1458 | { |
857a3d0f | 1459 | unsigned int val; |
be839e39 CL |
1460 | struct port_info *pi = netdev_priv(dev); |
1461 | ||
857a3d0f DM |
1462 | if (state == ETHTOOL_ID_ACTIVE) |
1463 | val = 0xffff; | |
1464 | else if (state == ETHTOOL_ID_INACTIVE) | |
1465 | val = 0; | |
1466 | else | |
1467 | return -EINVAL; | |
1468 | ||
1469 | return t4vf_identify_port(pi->adapter, pi->viid, val); | |
be839e39 CL |
1470 | } |
1471 | ||
1472 | /* | |
1473 | * Port stats maintained per queue of the port. | |
1474 | */ | |
1475 | struct queue_port_stats { | |
1476 | u64 tso; | |
1477 | u64 tx_csum; | |
1478 | u64 rx_csum; | |
1479 | u64 vlan_ex; | |
1480 | u64 vlan_ins; | |
f12fe353 CL |
1481 | u64 lro_pkts; |
1482 | u64 lro_merged; | |
be839e39 CL |
1483 | }; |
1484 | ||
1485 | /* | |
1486 | * Strings for the ETH_SS_STATS statistics set ("ethtool -S"). Note that | |
1487 | * these need to match the order of statistics returned by | |
1488 | * t4vf_get_port_stats(). | |
1489 | */ | |
1490 | static const char stats_strings[][ETH_GSTRING_LEN] = { | |
1491 | /* | |
1492 | * These must match the layout of the t4vf_port_stats structure. | |
1493 | */ | |
1494 | "TxBroadcastBytes ", | |
1495 | "TxBroadcastFrames ", | |
1496 | "TxMulticastBytes ", | |
1497 | "TxMulticastFrames ", | |
1498 | "TxUnicastBytes ", | |
1499 | "TxUnicastFrames ", | |
1500 | "TxDroppedFrames ", | |
1501 | "TxOffloadBytes ", | |
1502 | "TxOffloadFrames ", | |
1503 | "RxBroadcastBytes ", | |
1504 | "RxBroadcastFrames ", | |
1505 | "RxMulticastBytes ", | |
1506 | "RxMulticastFrames ", | |
1507 | "RxUnicastBytes ", | |
1508 | "RxUnicastFrames ", | |
1509 | "RxErrorFrames ", | |
1510 | ||
1511 | /* | |
1512 | * These are accumulated per-queue statistics and must match the | |
1513 | * order of the fields in the queue_port_stats structure. | |
1514 | */ | |
1515 | "TSO ", | |
1516 | "TxCsumOffload ", | |
1517 | "RxCsumGood ", | |
1518 | "VLANextractions ", | |
1519 | "VLANinsertions ", | |
f12fe353 CL |
1520 | "GROPackets ", |
1521 | "GROMerged ", | |
be839e39 CL |
1522 | }; |
1523 | ||
1524 | /* | |
1525 | * Return the number of statistics in the specified statistics set. | |
1526 | */ | |
1527 | static int cxgb4vf_get_sset_count(struct net_device *dev, int sset) | |
1528 | { | |
1529 | switch (sset) { | |
1530 | case ETH_SS_STATS: | |
1531 | return ARRAY_SIZE(stats_strings); | |
1532 | default: | |
1533 | return -EOPNOTSUPP; | |
1534 | } | |
1535 | /*NOTREACHED*/ | |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * Return the strings for the specified statistics set. | |
1540 | */ | |
1541 | static void cxgb4vf_get_strings(struct net_device *dev, | |
1542 | u32 sset, | |
1543 | u8 *data) | |
1544 | { | |
1545 | switch (sset) { | |
1546 | case ETH_SS_STATS: | |
1547 | memcpy(data, stats_strings, sizeof(stats_strings)); | |
1548 | break; | |
1549 | } | |
1550 | } | |
1551 | ||
1552 | /* | |
1553 | * Small utility routine to accumulate queue statistics across the queues of | |
1554 | * a "port". | |
1555 | */ | |
1556 | static void collect_sge_port_stats(const struct adapter *adapter, | |
1557 | const struct port_info *pi, | |
1558 | struct queue_port_stats *stats) | |
1559 | { | |
1560 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[pi->first_qset]; | |
1561 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[pi->first_qset]; | |
1562 | int qs; | |
1563 | ||
1564 | memset(stats, 0, sizeof(*stats)); | |
1565 | for (qs = 0; qs < pi->nqsets; qs++, rxq++, txq++) { | |
1566 | stats->tso += txq->tso; | |
1567 | stats->tx_csum += txq->tx_cso; | |
1568 | stats->rx_csum += rxq->stats.rx_cso; | |
1569 | stats->vlan_ex += rxq->stats.vlan_ex; | |
1570 | stats->vlan_ins += txq->vlan_ins; | |
f12fe353 CL |
1571 | stats->lro_pkts += rxq->stats.lro_pkts; |
1572 | stats->lro_merged += rxq->stats.lro_merged; | |
be839e39 CL |
1573 | } |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * Return the ETH_SS_STATS statistics set. | |
1578 | */ | |
1579 | static void cxgb4vf_get_ethtool_stats(struct net_device *dev, | |
1580 | struct ethtool_stats *stats, | |
1581 | u64 *data) | |
1582 | { | |
1583 | struct port_info *pi = netdev2pinfo(dev); | |
1584 | struct adapter *adapter = pi->adapter; | |
1585 | int err = t4vf_get_port_stats(adapter, pi->pidx, | |
1586 | (struct t4vf_port_stats *)data); | |
1587 | if (err) | |
1588 | memset(data, 0, sizeof(struct t4vf_port_stats)); | |
1589 | ||
1590 | data += sizeof(struct t4vf_port_stats) / sizeof(u64); | |
1591 | collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data); | |
1592 | } | |
1593 | ||
1594 | /* | |
1595 | * Return the size of our register map. | |
1596 | */ | |
1597 | static int cxgb4vf_get_regs_len(struct net_device *dev) | |
1598 | { | |
1599 | return T4VF_REGMAP_SIZE; | |
1600 | } | |
1601 | ||
1602 | /* | |
1603 | * Dump a block of registers, start to end inclusive, into a buffer. | |
1604 | */ | |
1605 | static void reg_block_dump(struct adapter *adapter, void *regbuf, | |
1606 | unsigned int start, unsigned int end) | |
1607 | { | |
1608 | u32 *bp = regbuf + start - T4VF_REGMAP_START; | |
1609 | ||
1610 | for ( ; start <= end; start += sizeof(u32)) { | |
1611 | /* | |
1612 | * Avoid reading the Mailbox Control register since that | |
1613 | * can trigger a Mailbox Ownership Arbitration cycle and | |
1614 | * interfere with communication with the firmware. | |
1615 | */ | |
1616 | if (start == T4VF_CIM_BASE_ADDR + CIM_VF_EXT_MAILBOX_CTRL) | |
1617 | *bp++ = 0xffff; | |
1618 | else | |
1619 | *bp++ = t4_read_reg(adapter, start); | |
1620 | } | |
1621 | } | |
1622 | ||
1623 | /* | |
1624 | * Copy our entire register map into the provided buffer. | |
1625 | */ | |
1626 | static void cxgb4vf_get_regs(struct net_device *dev, | |
1627 | struct ethtool_regs *regs, | |
1628 | void *regbuf) | |
1629 | { | |
1630 | struct adapter *adapter = netdev2adap(dev); | |
1631 | ||
1632 | regs->version = mk_adap_vers(adapter); | |
1633 | ||
1634 | /* | |
1635 | * Fill in register buffer with our register map. | |
1636 | */ | |
1637 | memset(regbuf, 0, T4VF_REGMAP_SIZE); | |
1638 | ||
1639 | reg_block_dump(adapter, regbuf, | |
1640 | T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_FIRST, | |
1641 | T4VF_SGE_BASE_ADDR + T4VF_MOD_MAP_SGE_LAST); | |
1642 | reg_block_dump(adapter, regbuf, | |
1643 | T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_FIRST, | |
1644 | T4VF_MPS_BASE_ADDR + T4VF_MOD_MAP_MPS_LAST); | |
70ee3666 HS |
1645 | |
1646 | /* T5 adds new registers in the PL Register map. | |
1647 | */ | |
be839e39 CL |
1648 | reg_block_dump(adapter, regbuf, |
1649 | T4VF_PL_BASE_ADDR + T4VF_MOD_MAP_PL_FIRST, | |
70ee3666 | 1650 | T4VF_PL_BASE_ADDR + (is_t4(adapter->params.chip) |
0d804338 | 1651 | ? PL_VF_WHOAMI_A : PL_VF_REVISION_A)); |
be839e39 CL |
1652 | reg_block_dump(adapter, regbuf, |
1653 | T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_FIRST, | |
1654 | T4VF_CIM_BASE_ADDR + T4VF_MOD_MAP_CIM_LAST); | |
1655 | ||
1656 | reg_block_dump(adapter, regbuf, | |
1657 | T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_FIRST, | |
1658 | T4VF_MBDATA_BASE_ADDR + T4VF_MBDATA_LAST); | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * Report current Wake On LAN settings. | |
1663 | */ | |
1664 | static void cxgb4vf_get_wol(struct net_device *dev, | |
1665 | struct ethtool_wolinfo *wol) | |
1666 | { | |
1667 | wol->supported = 0; | |
1668 | wol->wolopts = 0; | |
1669 | memset(&wol->sopass, 0, sizeof(wol->sopass)); | |
1670 | } | |
1671 | ||
410989f6 CL |
1672 | /* |
1673 | * TCP Segmentation Offload flags which we support. | |
1674 | */ | |
1675 | #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN) | |
1676 | ||
9b07be4b | 1677 | static const struct ethtool_ops cxgb4vf_ethtool_ops = { |
be839e39 CL |
1678 | .get_settings = cxgb4vf_get_settings, |
1679 | .get_drvinfo = cxgb4vf_get_drvinfo, | |
1680 | .get_msglevel = cxgb4vf_get_msglevel, | |
1681 | .set_msglevel = cxgb4vf_set_msglevel, | |
1682 | .get_ringparam = cxgb4vf_get_ringparam, | |
1683 | .set_ringparam = cxgb4vf_set_ringparam, | |
1684 | .get_coalesce = cxgb4vf_get_coalesce, | |
1685 | .set_coalesce = cxgb4vf_set_coalesce, | |
1686 | .get_pauseparam = cxgb4vf_get_pauseparam, | |
be839e39 CL |
1687 | .get_link = ethtool_op_get_link, |
1688 | .get_strings = cxgb4vf_get_strings, | |
857a3d0f | 1689 | .set_phys_id = cxgb4vf_phys_id, |
be839e39 CL |
1690 | .get_sset_count = cxgb4vf_get_sset_count, |
1691 | .get_ethtool_stats = cxgb4vf_get_ethtool_stats, | |
1692 | .get_regs_len = cxgb4vf_get_regs_len, | |
1693 | .get_regs = cxgb4vf_get_regs, | |
1694 | .get_wol = cxgb4vf_get_wol, | |
be839e39 CL |
1695 | }; |
1696 | ||
1697 | /* | |
1698 | * /sys/kernel/debug/cxgb4vf support code and data. | |
1699 | * ================================================ | |
1700 | */ | |
1701 | ||
ae7b7576 HS |
1702 | /* |
1703 | * Show Firmware Mailbox Command/Reply Log | |
1704 | * | |
1705 | * Note that we don't do any locking when dumping the Firmware Mailbox Log so | |
1706 | * it's possible that we can catch things during a log update and therefore | |
1707 | * see partially corrupted log entries. But i9t's probably Good Enough(tm). | |
1708 | * If we ever decide that we want to make sure that we're dumping a coherent | |
1709 | * log, we'd need to perform locking in the mailbox logging and in | |
1710 | * mboxlog_open() where we'd need to grab the entire mailbox log in one go | |
1711 | * like we do for the Firmware Device Log. But as stated above, meh ... | |
1712 | */ | |
1713 | static int mboxlog_show(struct seq_file *seq, void *v) | |
1714 | { | |
1715 | struct adapter *adapter = seq->private; | |
1716 | struct mbox_cmd_log *log = adapter->mbox_log; | |
1717 | struct mbox_cmd *entry; | |
1718 | int entry_idx, i; | |
1719 | ||
1720 | if (v == SEQ_START_TOKEN) { | |
1721 | seq_printf(seq, | |
1722 | "%10s %15s %5s %5s %s\n", | |
1723 | "Seq#", "Tstamp", "Atime", "Etime", | |
1724 | "Command/Reply"); | |
1725 | return 0; | |
1726 | } | |
1727 | ||
1728 | entry_idx = log->cursor + ((uintptr_t)v - 2); | |
1729 | if (entry_idx >= log->size) | |
1730 | entry_idx -= log->size; | |
1731 | entry = mbox_cmd_log_entry(log, entry_idx); | |
1732 | ||
1733 | /* skip over unused entries */ | |
1734 | if (entry->timestamp == 0) | |
1735 | return 0; | |
1736 | ||
1737 | seq_printf(seq, "%10u %15llu %5d %5d", | |
1738 | entry->seqno, entry->timestamp, | |
1739 | entry->access, entry->execute); | |
1740 | for (i = 0; i < MBOX_LEN / 8; i++) { | |
1741 | u64 flit = entry->cmd[i]; | |
1742 | u32 hi = (u32)(flit >> 32); | |
1743 | u32 lo = (u32)flit; | |
1744 | ||
1745 | seq_printf(seq, " %08x %08x", hi, lo); | |
1746 | } | |
1747 | seq_puts(seq, "\n"); | |
1748 | return 0; | |
1749 | } | |
1750 | ||
1751 | static inline void *mboxlog_get_idx(struct seq_file *seq, loff_t pos) | |
1752 | { | |
1753 | struct adapter *adapter = seq->private; | |
1754 | struct mbox_cmd_log *log = adapter->mbox_log; | |
1755 | ||
1756 | return ((pos <= log->size) ? (void *)(uintptr_t)(pos + 1) : NULL); | |
1757 | } | |
1758 | ||
1759 | static void *mboxlog_start(struct seq_file *seq, loff_t *pos) | |
1760 | { | |
1761 | return *pos ? mboxlog_get_idx(seq, *pos) : SEQ_START_TOKEN; | |
1762 | } | |
1763 | ||
1764 | static void *mboxlog_next(struct seq_file *seq, void *v, loff_t *pos) | |
1765 | { | |
1766 | ++*pos; | |
1767 | return mboxlog_get_idx(seq, *pos); | |
1768 | } | |
1769 | ||
1770 | static void mboxlog_stop(struct seq_file *seq, void *v) | |
1771 | { | |
1772 | } | |
1773 | ||
1774 | static const struct seq_operations mboxlog_seq_ops = { | |
1775 | .start = mboxlog_start, | |
1776 | .next = mboxlog_next, | |
1777 | .stop = mboxlog_stop, | |
1778 | .show = mboxlog_show | |
1779 | }; | |
1780 | ||
1781 | static int mboxlog_open(struct inode *inode, struct file *file) | |
1782 | { | |
1783 | int res = seq_open(file, &mboxlog_seq_ops); | |
1784 | ||
1785 | if (!res) { | |
1786 | struct seq_file *seq = file->private_data; | |
1787 | ||
1788 | seq->private = inode->i_private; | |
1789 | } | |
1790 | return res; | |
1791 | } | |
1792 | ||
1793 | static const struct file_operations mboxlog_fops = { | |
1794 | .owner = THIS_MODULE, | |
1795 | .open = mboxlog_open, | |
1796 | .read = seq_read, | |
1797 | .llseek = seq_lseek, | |
1798 | .release = seq_release, | |
1799 | }; | |
1800 | ||
be839e39 CL |
1801 | /* |
1802 | * Show SGE Queue Set information. We display QPL Queues Sets per line. | |
1803 | */ | |
1804 | #define QPL 4 | |
1805 | ||
1806 | static int sge_qinfo_show(struct seq_file *seq, void *v) | |
1807 | { | |
1808 | struct adapter *adapter = seq->private; | |
1809 | int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL); | |
1810 | int qs, r = (uintptr_t)v - 1; | |
1811 | ||
1812 | if (r) | |
1813 | seq_putc(seq, '\n'); | |
1814 | ||
1815 | #define S3(fmt_spec, s, v) \ | |
1816 | do {\ | |
1817 | seq_printf(seq, "%-12s", s); \ | |
1818 | for (qs = 0; qs < n; ++qs) \ | |
1819 | seq_printf(seq, " %16" fmt_spec, v); \ | |
1820 | seq_putc(seq, '\n'); \ | |
1821 | } while (0) | |
1822 | #define S(s, v) S3("s", s, v) | |
1823 | #define T(s, v) S3("u", s, txq[qs].v) | |
1824 | #define R(s, v) S3("u", s, rxq[qs].v) | |
1825 | ||
1826 | if (r < eth_entries) { | |
1827 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL]; | |
1828 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL]; | |
1829 | int n = min(QPL, adapter->sge.ethqsets - QPL * r); | |
1830 | ||
1831 | S("QType:", "Ethernet"); | |
1832 | S("Interface:", | |
1833 | (rxq[qs].rspq.netdev | |
1834 | ? rxq[qs].rspq.netdev->name | |
1835 | : "N/A")); | |
1836 | S3("d", "Port:", | |
1837 | (rxq[qs].rspq.netdev | |
1838 | ? ((struct port_info *) | |
1839 | netdev_priv(rxq[qs].rspq.netdev))->port_id | |
1840 | : -1)); | |
1841 | T("TxQ ID:", q.abs_id); | |
1842 | T("TxQ size:", q.size); | |
1843 | T("TxQ inuse:", q.in_use); | |
1844 | T("TxQ PIdx:", q.pidx); | |
1845 | T("TxQ CIdx:", q.cidx); | |
1846 | R("RspQ ID:", rspq.abs_id); | |
1847 | R("RspQ size:", rspq.size); | |
1848 | R("RspQE size:", rspq.iqe_len); | |
1849 | S3("u", "Intr delay:", qtimer_val(adapter, &rxq[qs].rspq)); | |
1850 | S3("u", "Intr pktcnt:", | |
1851 | adapter->sge.counter_val[rxq[qs].rspq.pktcnt_idx]); | |
1852 | R("RspQ CIdx:", rspq.cidx); | |
1853 | R("RspQ Gen:", rspq.gen); | |
1854 | R("FL ID:", fl.abs_id); | |
1855 | R("FL size:", fl.size - MIN_FL_RESID); | |
1856 | R("FL avail:", fl.avail); | |
1857 | R("FL PIdx:", fl.pidx); | |
1858 | R("FL CIdx:", fl.cidx); | |
1859 | return 0; | |
1860 | } | |
1861 | ||
1862 | r -= eth_entries; | |
1863 | if (r == 0) { | |
1864 | const struct sge_rspq *evtq = &adapter->sge.fw_evtq; | |
1865 | ||
1866 | seq_printf(seq, "%-12s %16s\n", "QType:", "FW event queue"); | |
1867 | seq_printf(seq, "%-12s %16u\n", "RspQ ID:", evtq->abs_id); | |
1868 | seq_printf(seq, "%-12s %16u\n", "Intr delay:", | |
1869 | qtimer_val(adapter, evtq)); | |
1870 | seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:", | |
1871 | adapter->sge.counter_val[evtq->pktcnt_idx]); | |
1872 | seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", evtq->cidx); | |
1873 | seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", evtq->gen); | |
1874 | } else if (r == 1) { | |
1875 | const struct sge_rspq *intrq = &adapter->sge.intrq; | |
1876 | ||
1877 | seq_printf(seq, "%-12s %16s\n", "QType:", "Interrupt Queue"); | |
1878 | seq_printf(seq, "%-12s %16u\n", "RspQ ID:", intrq->abs_id); | |
1879 | seq_printf(seq, "%-12s %16u\n", "Intr delay:", | |
1880 | qtimer_val(adapter, intrq)); | |
1881 | seq_printf(seq, "%-12s %16u\n", "Intr pktcnt:", | |
1882 | adapter->sge.counter_val[intrq->pktcnt_idx]); | |
1883 | seq_printf(seq, "%-12s %16u\n", "RspQ Cidx:", intrq->cidx); | |
1884 | seq_printf(seq, "%-12s %16u\n", "RspQ Gen:", intrq->gen); | |
1885 | } | |
1886 | ||
1887 | #undef R | |
1888 | #undef T | |
1889 | #undef S | |
1890 | #undef S3 | |
1891 | ||
1892 | return 0; | |
1893 | } | |
1894 | ||
1895 | /* | |
1896 | * Return the number of "entries" in our "file". We group the multi-Queue | |
1897 | * sections with QPL Queue Sets per "entry". The sections of the output are: | |
1898 | * | |
1899 | * Ethernet RX/TX Queue Sets | |
1900 | * Firmware Event Queue | |
1901 | * Forwarded Interrupt Queue (if in MSI mode) | |
1902 | */ | |
1903 | static int sge_queue_entries(const struct adapter *adapter) | |
1904 | { | |
1905 | return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 + | |
1906 | ((adapter->flags & USING_MSI) != 0); | |
1907 | } | |
1908 | ||
1909 | static void *sge_queue_start(struct seq_file *seq, loff_t *pos) | |
1910 | { | |
1911 | int entries = sge_queue_entries(seq->private); | |
1912 | ||
1913 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
1914 | } | |
1915 | ||
1916 | static void sge_queue_stop(struct seq_file *seq, void *v) | |
1917 | { | |
1918 | } | |
1919 | ||
1920 | static void *sge_queue_next(struct seq_file *seq, void *v, loff_t *pos) | |
1921 | { | |
1922 | int entries = sge_queue_entries(seq->private); | |
1923 | ||
1924 | ++*pos; | |
1925 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
1926 | } | |
1927 | ||
1928 | static const struct seq_operations sge_qinfo_seq_ops = { | |
1929 | .start = sge_queue_start, | |
1930 | .next = sge_queue_next, | |
1931 | .stop = sge_queue_stop, | |
1932 | .show = sge_qinfo_show | |
1933 | }; | |
1934 | ||
1935 | static int sge_qinfo_open(struct inode *inode, struct file *file) | |
1936 | { | |
1937 | int res = seq_open(file, &sge_qinfo_seq_ops); | |
1938 | ||
1939 | if (!res) { | |
1940 | struct seq_file *seq = file->private_data; | |
1941 | seq->private = inode->i_private; | |
1942 | } | |
1943 | return res; | |
1944 | } | |
1945 | ||
1946 | static const struct file_operations sge_qinfo_debugfs_fops = { | |
1947 | .owner = THIS_MODULE, | |
1948 | .open = sge_qinfo_open, | |
1949 | .read = seq_read, | |
1950 | .llseek = seq_lseek, | |
1951 | .release = seq_release, | |
1952 | }; | |
1953 | ||
1954 | /* | |
1955 | * Show SGE Queue Set statistics. We display QPL Queues Sets per line. | |
1956 | */ | |
1957 | #define QPL 4 | |
1958 | ||
1959 | static int sge_qstats_show(struct seq_file *seq, void *v) | |
1960 | { | |
1961 | struct adapter *adapter = seq->private; | |
1962 | int eth_entries = DIV_ROUND_UP(adapter->sge.ethqsets, QPL); | |
1963 | int qs, r = (uintptr_t)v - 1; | |
1964 | ||
1965 | if (r) | |
1966 | seq_putc(seq, '\n'); | |
1967 | ||
1968 | #define S3(fmt, s, v) \ | |
1969 | do { \ | |
1970 | seq_printf(seq, "%-16s", s); \ | |
1971 | for (qs = 0; qs < n; ++qs) \ | |
1972 | seq_printf(seq, " %8" fmt, v); \ | |
1973 | seq_putc(seq, '\n'); \ | |
1974 | } while (0) | |
1975 | #define S(s, v) S3("s", s, v) | |
1976 | ||
1977 | #define T3(fmt, s, v) S3(fmt, s, txq[qs].v) | |
1978 | #define T(s, v) T3("lu", s, v) | |
1979 | ||
1980 | #define R3(fmt, s, v) S3(fmt, s, rxq[qs].v) | |
1981 | #define R(s, v) R3("lu", s, v) | |
1982 | ||
1983 | if (r < eth_entries) { | |
1984 | const struct sge_eth_rxq *rxq = &adapter->sge.ethrxq[r * QPL]; | |
1985 | const struct sge_eth_txq *txq = &adapter->sge.ethtxq[r * QPL]; | |
1986 | int n = min(QPL, adapter->sge.ethqsets - QPL * r); | |
1987 | ||
1988 | S("QType:", "Ethernet"); | |
1989 | S("Interface:", | |
1990 | (rxq[qs].rspq.netdev | |
1991 | ? rxq[qs].rspq.netdev->name | |
1992 | : "N/A")); | |
68dc9d36 | 1993 | R3("u", "RspQNullInts:", rspq.unhandled_irqs); |
be839e39 CL |
1994 | R("RxPackets:", stats.pkts); |
1995 | R("RxCSO:", stats.rx_cso); | |
1996 | R("VLANxtract:", stats.vlan_ex); | |
1997 | R("LROmerged:", stats.lro_merged); | |
1998 | R("LROpackets:", stats.lro_pkts); | |
1999 | R("RxDrops:", stats.rx_drops); | |
2000 | T("TSO:", tso); | |
2001 | T("TxCSO:", tx_cso); | |
2002 | T("VLANins:", vlan_ins); | |
2003 | T("TxQFull:", q.stops); | |
2004 | T("TxQRestarts:", q.restarts); | |
2005 | T("TxMapErr:", mapping_err); | |
2006 | R("FLAllocErr:", fl.alloc_failed); | |
2007 | R("FLLrgAlcErr:", fl.large_alloc_failed); | |
2008 | R("FLStarving:", fl.starving); | |
2009 | return 0; | |
2010 | } | |
2011 | ||
2012 | r -= eth_entries; | |
2013 | if (r == 0) { | |
2014 | const struct sge_rspq *evtq = &adapter->sge.fw_evtq; | |
2015 | ||
2016 | seq_printf(seq, "%-8s %16s\n", "QType:", "FW event queue"); | |
68dc9d36 CL |
2017 | seq_printf(seq, "%-16s %8u\n", "RspQNullInts:", |
2018 | evtq->unhandled_irqs); | |
be839e39 CL |
2019 | seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", evtq->cidx); |
2020 | seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", evtq->gen); | |
2021 | } else if (r == 1) { | |
2022 | const struct sge_rspq *intrq = &adapter->sge.intrq; | |
2023 | ||
2024 | seq_printf(seq, "%-8s %16s\n", "QType:", "Interrupt Queue"); | |
68dc9d36 CL |
2025 | seq_printf(seq, "%-16s %8u\n", "RspQNullInts:", |
2026 | intrq->unhandled_irqs); | |
be839e39 CL |
2027 | seq_printf(seq, "%-16s %8u\n", "RspQ CIdx:", intrq->cidx); |
2028 | seq_printf(seq, "%-16s %8u\n", "RspQ Gen:", intrq->gen); | |
2029 | } | |
2030 | ||
2031 | #undef R | |
2032 | #undef T | |
2033 | #undef S | |
2034 | #undef R3 | |
2035 | #undef T3 | |
2036 | #undef S3 | |
2037 | ||
2038 | return 0; | |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * Return the number of "entries" in our "file". We group the multi-Queue | |
2043 | * sections with QPL Queue Sets per "entry". The sections of the output are: | |
2044 | * | |
2045 | * Ethernet RX/TX Queue Sets | |
2046 | * Firmware Event Queue | |
2047 | * Forwarded Interrupt Queue (if in MSI mode) | |
2048 | */ | |
2049 | static int sge_qstats_entries(const struct adapter *adapter) | |
2050 | { | |
2051 | return DIV_ROUND_UP(adapter->sge.ethqsets, QPL) + 1 + | |
2052 | ((adapter->flags & USING_MSI) != 0); | |
2053 | } | |
2054 | ||
2055 | static void *sge_qstats_start(struct seq_file *seq, loff_t *pos) | |
2056 | { | |
2057 | int entries = sge_qstats_entries(seq->private); | |
2058 | ||
2059 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
2060 | } | |
2061 | ||
2062 | static void sge_qstats_stop(struct seq_file *seq, void *v) | |
2063 | { | |
2064 | } | |
2065 | ||
2066 | static void *sge_qstats_next(struct seq_file *seq, void *v, loff_t *pos) | |
2067 | { | |
2068 | int entries = sge_qstats_entries(seq->private); | |
2069 | ||
2070 | (*pos)++; | |
2071 | return *pos < entries ? (void *)((uintptr_t)*pos + 1) : NULL; | |
2072 | } | |
2073 | ||
2074 | static const struct seq_operations sge_qstats_seq_ops = { | |
2075 | .start = sge_qstats_start, | |
2076 | .next = sge_qstats_next, | |
2077 | .stop = sge_qstats_stop, | |
2078 | .show = sge_qstats_show | |
2079 | }; | |
2080 | ||
2081 | static int sge_qstats_open(struct inode *inode, struct file *file) | |
2082 | { | |
2083 | int res = seq_open(file, &sge_qstats_seq_ops); | |
2084 | ||
2085 | if (res == 0) { | |
2086 | struct seq_file *seq = file->private_data; | |
2087 | seq->private = inode->i_private; | |
2088 | } | |
2089 | return res; | |
2090 | } | |
2091 | ||
2092 | static const struct file_operations sge_qstats_proc_fops = { | |
2093 | .owner = THIS_MODULE, | |
2094 | .open = sge_qstats_open, | |
2095 | .read = seq_read, | |
2096 | .llseek = seq_lseek, | |
2097 | .release = seq_release, | |
2098 | }; | |
2099 | ||
2100 | /* | |
2101 | * Show PCI-E SR-IOV Virtual Function Resource Limits. | |
2102 | */ | |
2103 | static int resources_show(struct seq_file *seq, void *v) | |
2104 | { | |
2105 | struct adapter *adapter = seq->private; | |
2106 | struct vf_resources *vfres = &adapter->params.vfres; | |
2107 | ||
2108 | #define S(desc, fmt, var) \ | |
2109 | seq_printf(seq, "%-60s " fmt "\n", \ | |
2110 | desc " (" #var "):", vfres->var) | |
2111 | ||
2112 | S("Virtual Interfaces", "%d", nvi); | |
2113 | S("Egress Queues", "%d", neq); | |
2114 | S("Ethernet Control", "%d", nethctrl); | |
2115 | S("Ingress Queues/w Free Lists/Interrupts", "%d", niqflint); | |
2116 | S("Ingress Queues", "%d", niq); | |
2117 | S("Traffic Class", "%d", tc); | |
2118 | S("Port Access Rights Mask", "%#x", pmask); | |
2119 | S("MAC Address Filters", "%d", nexactf); | |
2120 | S("Firmware Command Read Capabilities", "%#x", r_caps); | |
2121 | S("Firmware Command Write/Execute Capabilities", "%#x", wx_caps); | |
2122 | ||
2123 | #undef S | |
2124 | ||
2125 | return 0; | |
2126 | } | |
2127 | ||
2128 | static int resources_open(struct inode *inode, struct file *file) | |
2129 | { | |
2130 | return single_open(file, resources_show, inode->i_private); | |
2131 | } | |
2132 | ||
2133 | static const struct file_operations resources_proc_fops = { | |
2134 | .owner = THIS_MODULE, | |
2135 | .open = resources_open, | |
2136 | .read = seq_read, | |
2137 | .llseek = seq_lseek, | |
2138 | .release = single_release, | |
2139 | }; | |
2140 | ||
2141 | /* | |
2142 | * Show Virtual Interfaces. | |
2143 | */ | |
2144 | static int interfaces_show(struct seq_file *seq, void *v) | |
2145 | { | |
2146 | if (v == SEQ_START_TOKEN) { | |
2147 | seq_puts(seq, "Interface Port VIID\n"); | |
2148 | } else { | |
2149 | struct adapter *adapter = seq->private; | |
2150 | int pidx = (uintptr_t)v - 2; | |
2151 | struct net_device *dev = adapter->port[pidx]; | |
2152 | struct port_info *pi = netdev_priv(dev); | |
2153 | ||
2154 | seq_printf(seq, "%9s %4d %#5x\n", | |
2155 | dev->name, pi->port_id, pi->viid); | |
2156 | } | |
2157 | return 0; | |
2158 | } | |
2159 | ||
2160 | static inline void *interfaces_get_idx(struct adapter *adapter, loff_t pos) | |
2161 | { | |
2162 | return pos <= adapter->params.nports | |
2163 | ? (void *)(uintptr_t)(pos + 1) | |
2164 | : NULL; | |
2165 | } | |
2166 | ||
2167 | static void *interfaces_start(struct seq_file *seq, loff_t *pos) | |
2168 | { | |
2169 | return *pos | |
2170 | ? interfaces_get_idx(seq->private, *pos) | |
2171 | : SEQ_START_TOKEN; | |
2172 | } | |
2173 | ||
2174 | static void *interfaces_next(struct seq_file *seq, void *v, loff_t *pos) | |
2175 | { | |
2176 | (*pos)++; | |
2177 | return interfaces_get_idx(seq->private, *pos); | |
2178 | } | |
2179 | ||
2180 | static void interfaces_stop(struct seq_file *seq, void *v) | |
2181 | { | |
2182 | } | |
2183 | ||
2184 | static const struct seq_operations interfaces_seq_ops = { | |
2185 | .start = interfaces_start, | |
2186 | .next = interfaces_next, | |
2187 | .stop = interfaces_stop, | |
2188 | .show = interfaces_show | |
2189 | }; | |
2190 | ||
2191 | static int interfaces_open(struct inode *inode, struct file *file) | |
2192 | { | |
2193 | int res = seq_open(file, &interfaces_seq_ops); | |
2194 | ||
2195 | if (res == 0) { | |
2196 | struct seq_file *seq = file->private_data; | |
2197 | seq->private = inode->i_private; | |
2198 | } | |
2199 | return res; | |
2200 | } | |
2201 | ||
2202 | static const struct file_operations interfaces_proc_fops = { | |
2203 | .owner = THIS_MODULE, | |
2204 | .open = interfaces_open, | |
2205 | .read = seq_read, | |
2206 | .llseek = seq_lseek, | |
2207 | .release = seq_release, | |
2208 | }; | |
2209 | ||
2210 | /* | |
2211 | * /sys/kernel/debugfs/cxgb4vf/ files list. | |
2212 | */ | |
2213 | struct cxgb4vf_debugfs_entry { | |
2214 | const char *name; /* name of debugfs node */ | |
f4ae40a6 | 2215 | umode_t mode; /* file system mode */ |
be839e39 CL |
2216 | const struct file_operations *fops; |
2217 | }; | |
2218 | ||
2219 | static struct cxgb4vf_debugfs_entry debugfs_files[] = { | |
ae7b7576 | 2220 | { "mboxlog", S_IRUGO, &mboxlog_fops }, |
be839e39 CL |
2221 | { "sge_qinfo", S_IRUGO, &sge_qinfo_debugfs_fops }, |
2222 | { "sge_qstats", S_IRUGO, &sge_qstats_proc_fops }, | |
2223 | { "resources", S_IRUGO, &resources_proc_fops }, | |
2224 | { "interfaces", S_IRUGO, &interfaces_proc_fops }, | |
2225 | }; | |
2226 | ||
2227 | /* | |
2228 | * Module and device initialization and cleanup code. | |
2229 | * ================================================== | |
2230 | */ | |
2231 | ||
2232 | /* | |
2233 | * Set up out /sys/kernel/debug/cxgb4vf sub-nodes. We assume that the | |
2234 | * directory (debugfs_root) has already been set up. | |
2235 | */ | |
d289f864 | 2236 | static int setup_debugfs(struct adapter *adapter) |
be839e39 CL |
2237 | { |
2238 | int i; | |
2239 | ||
843635e0 | 2240 | BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root)); |
be839e39 CL |
2241 | |
2242 | /* | |
2243 | * Debugfs support is best effort. | |
2244 | */ | |
2245 | for (i = 0; i < ARRAY_SIZE(debugfs_files); i++) | |
2246 | (void)debugfs_create_file(debugfs_files[i].name, | |
2247 | debugfs_files[i].mode, | |
2248 | adapter->debugfs_root, | |
2249 | (void *)adapter, | |
2250 | debugfs_files[i].fops); | |
2251 | ||
2252 | return 0; | |
2253 | } | |
2254 | ||
2255 | /* | |
2256 | * Tear down the /sys/kernel/debug/cxgb4vf sub-nodes created above. We leave | |
2257 | * it to our caller to tear down the directory (debugfs_root). | |
2258 | */ | |
4204875d | 2259 | static void cleanup_debugfs(struct adapter *adapter) |
be839e39 | 2260 | { |
843635e0 | 2261 | BUG_ON(IS_ERR_OR_NULL(adapter->debugfs_root)); |
be839e39 CL |
2262 | |
2263 | /* | |
2264 | * Unlike our sister routine cleanup_proc(), we don't need to remove | |
2265 | * individual entries because a call will be made to | |
2266 | * debugfs_remove_recursive(). We just need to clean up any ancillary | |
2267 | * persistent state. | |
2268 | */ | |
2269 | /* nothing to do */ | |
2270 | } | |
2271 | ||
495c22bb HS |
2272 | /* Figure out how many Ports and Queue Sets we can support. This depends on |
2273 | * knowing our Virtual Function Resources and may be called a second time if | |
2274 | * we fall back from MSI-X to MSI Interrupt Mode. | |
2275 | */ | |
2276 | static void size_nports_qsets(struct adapter *adapter) | |
2277 | { | |
2278 | struct vf_resources *vfres = &adapter->params.vfres; | |
2279 | unsigned int ethqsets, pmask_nports; | |
2280 | ||
2281 | /* The number of "ports" which we support is equal to the number of | |
2282 | * Virtual Interfaces with which we've been provisioned. | |
2283 | */ | |
2284 | adapter->params.nports = vfres->nvi; | |
2285 | if (adapter->params.nports > MAX_NPORTS) { | |
2286 | dev_warn(adapter->pdev_dev, "only using %d of %d maximum" | |
2287 | " allowed virtual interfaces\n", MAX_NPORTS, | |
2288 | adapter->params.nports); | |
2289 | adapter->params.nports = MAX_NPORTS; | |
2290 | } | |
2291 | ||
2292 | /* We may have been provisioned with more VIs than the number of | |
2293 | * ports we're allowed to access (our Port Access Rights Mask). | |
2294 | * This is obviously a configuration conflict but we don't want to | |
2295 | * crash the kernel or anything silly just because of that. | |
2296 | */ | |
2297 | pmask_nports = hweight32(adapter->params.vfres.pmask); | |
2298 | if (pmask_nports < adapter->params.nports) { | |
2299 | dev_warn(adapter->pdev_dev, "only using %d of %d provissioned" | |
2300 | " virtual interfaces; limited by Port Access Rights" | |
2301 | " mask %#x\n", pmask_nports, adapter->params.nports, | |
2302 | adapter->params.vfres.pmask); | |
2303 | adapter->params.nports = pmask_nports; | |
2304 | } | |
2305 | ||
2306 | /* We need to reserve an Ingress Queue for the Asynchronous Firmware | |
2307 | * Event Queue. And if we're using MSI Interrupts, we'll also need to | |
2308 | * reserve an Ingress Queue for a Forwarded Interrupts. | |
2309 | * | |
2310 | * The rest of the FL/Intr-capable ingress queues will be matched up | |
2311 | * one-for-one with Ethernet/Control egress queues in order to form | |
2312 | * "Queue Sets" which will be aportioned between the "ports". For | |
2313 | * each Queue Set, we'll need the ability to allocate two Egress | |
2314 | * Contexts -- one for the Ingress Queue Free List and one for the TX | |
2315 | * Ethernet Queue. | |
2316 | * | |
2317 | * Note that even if we're currently configured to use MSI-X | |
2318 | * Interrupts (module variable msi == MSI_MSIX) we may get downgraded | |
2319 | * to MSI Interrupts if we can't get enough MSI-X Interrupts. If that | |
2320 | * happens we'll need to adjust things later. | |
2321 | */ | |
2322 | ethqsets = vfres->niqflint - 1 - (msi == MSI_MSI); | |
2323 | if (vfres->nethctrl != ethqsets) | |
2324 | ethqsets = min(vfres->nethctrl, ethqsets); | |
2325 | if (vfres->neq < ethqsets*2) | |
2326 | ethqsets = vfres->neq/2; | |
2327 | if (ethqsets > MAX_ETH_QSETS) | |
2328 | ethqsets = MAX_ETH_QSETS; | |
2329 | adapter->sge.max_ethqsets = ethqsets; | |
2330 | ||
2331 | if (adapter->sge.max_ethqsets < adapter->params.nports) { | |
2332 | dev_warn(adapter->pdev_dev, "only using %d of %d available" | |
2333 | " virtual interfaces (too few Queue Sets)\n", | |
2334 | adapter->sge.max_ethqsets, adapter->params.nports); | |
2335 | adapter->params.nports = adapter->sge.max_ethqsets; | |
2336 | } | |
2337 | } | |
2338 | ||
be839e39 CL |
2339 | /* |
2340 | * Perform early "adapter" initialization. This is where we discover what | |
2341 | * adapter parameters we're going to be using and initialize basic adapter | |
2342 | * hardware support. | |
2343 | */ | |
d289f864 | 2344 | static int adap_init0(struct adapter *adapter) |
be839e39 | 2345 | { |
be839e39 CL |
2346 | struct sge_params *sge_params = &adapter->params.sge; |
2347 | struct sge *s = &adapter->sge; | |
be839e39 | 2348 | int err; |
94dace10 | 2349 | u32 param, val = 0; |
be839e39 | 2350 | |
e68e6133 CL |
2351 | /* |
2352 | * Some environments do not properly handle PCIE FLRs -- e.g. in Linux | |
2353 | * 2.6.31 and later we can't call pci_reset_function() in order to | |
2354 | * issue an FLR because of a self- deadlock on the device semaphore. | |
2355 | * Meanwhile, the OS infrastructure doesn't issue FLRs in all the | |
2356 | * cases where they're needed -- for instance, some versions of KVM | |
2357 | * fail to reset "Assigned Devices" when the VM reboots. Therefore we | |
2358 | * use the firmware based reset in order to reset any per function | |
2359 | * state. | |
2360 | */ | |
2361 | err = t4vf_fw_reset(adapter); | |
2362 | if (err < 0) { | |
2363 | dev_err(adapter->pdev_dev, "FW reset failed: err=%d\n", err); | |
2364 | return err; | |
2365 | } | |
2366 | ||
be839e39 CL |
2367 | /* |
2368 | * Grab basic operational parameters. These will predominantly have | |
2369 | * been set up by the Physical Function Driver or will be hard coded | |
2370 | * into the adapter. We just have to live with them ... Note that | |
2371 | * we _must_ get our VPD parameters before our SGE parameters because | |
2372 | * we need to know the adapter's core clock from the VPD in order to | |
2373 | * properly decode the SGE Timer Values. | |
2374 | */ | |
2375 | err = t4vf_get_dev_params(adapter); | |
2376 | if (err) { | |
2377 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2378 | " device parameters: err=%d\n", err); | |
2379 | return err; | |
2380 | } | |
2381 | err = t4vf_get_vpd_params(adapter); | |
2382 | if (err) { | |
2383 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2384 | " VPD parameters: err=%d\n", err); | |
2385 | return err; | |
2386 | } | |
2387 | err = t4vf_get_sge_params(adapter); | |
2388 | if (err) { | |
2389 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2390 | " SGE parameters: err=%d\n", err); | |
2391 | return err; | |
2392 | } | |
2393 | err = t4vf_get_rss_glb_config(adapter); | |
2394 | if (err) { | |
2395 | dev_err(adapter->pdev_dev, "unable to retrieve adapter" | |
2396 | " RSS parameters: err=%d\n", err); | |
2397 | return err; | |
2398 | } | |
2399 | if (adapter->params.rss.mode != | |
2400 | FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL) { | |
2401 | dev_err(adapter->pdev_dev, "unable to operate with global RSS" | |
2402 | " mode %d\n", adapter->params.rss.mode); | |
2403 | return -EINVAL; | |
2404 | } | |
2405 | err = t4vf_sge_init(adapter); | |
2406 | if (err) { | |
2407 | dev_err(adapter->pdev_dev, "unable to use adapter parameters:" | |
2408 | " err=%d\n", err); | |
2409 | return err; | |
2410 | } | |
2411 | ||
94dace10 VP |
2412 | /* If we're running on newer firmware, let it know that we're |
2413 | * prepared to deal with encapsulated CPL messages. Older | |
2414 | * firmware won't understand this and we'll just get | |
2415 | * unencapsulated messages ... | |
2416 | */ | |
5167865a HS |
2417 | param = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | |
2418 | FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_CPLFW4MSG_ENCAP); | |
94dace10 VP |
2419 | val = 1; |
2420 | (void) t4vf_set_params(adapter, 1, ¶m, &val); | |
2421 | ||
be839e39 CL |
2422 | /* |
2423 | * Retrieve our RX interrupt holdoff timer values and counter | |
2424 | * threshold values from the SGE parameters. | |
2425 | */ | |
2426 | s->timer_val[0] = core_ticks_to_us(adapter, | |
f061de42 | 2427 | TIMERVALUE0_G(sge_params->sge_timer_value_0_and_1)); |
be839e39 | 2428 | s->timer_val[1] = core_ticks_to_us(adapter, |
f061de42 | 2429 | TIMERVALUE1_G(sge_params->sge_timer_value_0_and_1)); |
be839e39 | 2430 | s->timer_val[2] = core_ticks_to_us(adapter, |
f061de42 | 2431 | TIMERVALUE0_G(sge_params->sge_timer_value_2_and_3)); |
be839e39 | 2432 | s->timer_val[3] = core_ticks_to_us(adapter, |
f061de42 | 2433 | TIMERVALUE1_G(sge_params->sge_timer_value_2_and_3)); |
be839e39 | 2434 | s->timer_val[4] = core_ticks_to_us(adapter, |
f061de42 | 2435 | TIMERVALUE0_G(sge_params->sge_timer_value_4_and_5)); |
be839e39 | 2436 | s->timer_val[5] = core_ticks_to_us(adapter, |
f061de42 | 2437 | TIMERVALUE1_G(sge_params->sge_timer_value_4_and_5)); |
be839e39 | 2438 | |
f612b815 HS |
2439 | s->counter_val[0] = THRESHOLD_0_G(sge_params->sge_ingress_rx_threshold); |
2440 | s->counter_val[1] = THRESHOLD_1_G(sge_params->sge_ingress_rx_threshold); | |
2441 | s->counter_val[2] = THRESHOLD_2_G(sge_params->sge_ingress_rx_threshold); | |
2442 | s->counter_val[3] = THRESHOLD_3_G(sge_params->sge_ingress_rx_threshold); | |
be839e39 CL |
2443 | |
2444 | /* | |
2445 | * Grab our Virtual Interface resource allocation, extract the | |
2446 | * features that we're interested in and do a bit of sanity testing on | |
2447 | * what we discover. | |
2448 | */ | |
2449 | err = t4vf_get_vfres(adapter); | |
2450 | if (err) { | |
2451 | dev_err(adapter->pdev_dev, "unable to get virtual interface" | |
2452 | " resources: err=%d\n", err); | |
2453 | return err; | |
2454 | } | |
2455 | ||
495c22bb | 2456 | /* Check for various parameter sanity issues */ |
28f71c6d HS |
2457 | if (adapter->params.vfres.pmask == 0) { |
2458 | dev_err(adapter->pdev_dev, "no port access configured\n" | |
2459 | "usable!\n"); | |
2460 | return -EINVAL; | |
2461 | } | |
495c22bb | 2462 | if (adapter->params.vfres.nvi == 0) { |
be839e39 CL |
2463 | dev_err(adapter->pdev_dev, "no virtual interfaces configured/" |
2464 | "usable!\n"); | |
2465 | return -EINVAL; | |
2466 | } | |
495c22bb HS |
2467 | |
2468 | /* Initialize nports and max_ethqsets now that we have our Virtual | |
2469 | * Function Resources. | |
2470 | */ | |
2471 | size_nports_qsets(adapter); | |
2472 | ||
be839e39 CL |
2473 | return 0; |
2474 | } | |
2475 | ||
2476 | static inline void init_rspq(struct sge_rspq *rspq, u8 timer_idx, | |
2477 | u8 pkt_cnt_idx, unsigned int size, | |
2478 | unsigned int iqe_size) | |
2479 | { | |
1ecc7b7a HS |
2480 | rspq->intr_params = (QINTR_TIMER_IDX_V(timer_idx) | |
2481 | (pkt_cnt_idx < SGE_NCOUNTERS ? | |
2482 | QINTR_CNT_EN_F : 0)); | |
be839e39 CL |
2483 | rspq->pktcnt_idx = (pkt_cnt_idx < SGE_NCOUNTERS |
2484 | ? pkt_cnt_idx | |
2485 | : 0); | |
2486 | rspq->iqe_len = iqe_size; | |
2487 | rspq->size = size; | |
2488 | } | |
2489 | ||
2490 | /* | |
2491 | * Perform default configuration of DMA queues depending on the number and | |
2492 | * type of ports we found and the number of available CPUs. Most settings can | |
2493 | * be modified by the admin via ethtool and cxgbtool prior to the adapter | |
2494 | * being brought up for the first time. | |
2495 | */ | |
d289f864 | 2496 | static void cfg_queues(struct adapter *adapter) |
be839e39 CL |
2497 | { |
2498 | struct sge *s = &adapter->sge; | |
2499 | int q10g, n10g, qidx, pidx, qs; | |
c710245c | 2500 | size_t iqe_size; |
be839e39 CL |
2501 | |
2502 | /* | |
2503 | * We should not be called till we know how many Queue Sets we can | |
2504 | * support. In particular, this means that we need to know what kind | |
2505 | * of interrupts we'll be using ... | |
2506 | */ | |
2507 | BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0); | |
2508 | ||
2509 | /* | |
2510 | * Count the number of 10GbE Virtual Interfaces that we have. | |
2511 | */ | |
2512 | n10g = 0; | |
2513 | for_each_port(adapter, pidx) | |
14b3812f | 2514 | n10g += is_x_10g_port(&adap2pinfo(adapter, pidx)->link_cfg); |
be839e39 CL |
2515 | |
2516 | /* | |
2517 | * We default to 1 queue per non-10G port and up to # of cores queues | |
2518 | * per 10G port. | |
2519 | */ | |
2520 | if (n10g == 0) | |
2521 | q10g = 0; | |
2522 | else { | |
2523 | int n1g = (adapter->params.nports - n10g); | |
2524 | q10g = (adapter->sge.max_ethqsets - n1g) / n10g; | |
2525 | if (q10g > num_online_cpus()) | |
2526 | q10g = num_online_cpus(); | |
2527 | } | |
2528 | ||
2529 | /* | |
2530 | * Allocate the "Queue Sets" to the various Virtual Interfaces. | |
2531 | * The layout will be established in setup_sge_queues() when the | |
2532 | * adapter is brough up for the first time. | |
2533 | */ | |
2534 | qidx = 0; | |
2535 | for_each_port(adapter, pidx) { | |
2536 | struct port_info *pi = adap2pinfo(adapter, pidx); | |
2537 | ||
2538 | pi->first_qset = qidx; | |
897d55df | 2539 | pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1; |
be839e39 CL |
2540 | qidx += pi->nqsets; |
2541 | } | |
2542 | s->ethqsets = qidx; | |
2543 | ||
c710245c CL |
2544 | /* |
2545 | * The Ingress Queue Entry Size for our various Response Queues needs | |
2546 | * to be big enough to accommodate the largest message we can receive | |
2547 | * from the chip/firmware; which is 64 bytes ... | |
2548 | */ | |
2549 | iqe_size = 64; | |
2550 | ||
be839e39 CL |
2551 | /* |
2552 | * Set up default Queue Set parameters ... Start off with the | |
2553 | * shortest interrupt holdoff timer. | |
2554 | */ | |
2555 | for (qs = 0; qs < s->max_ethqsets; qs++) { | |
2556 | struct sge_eth_rxq *rxq = &s->ethrxq[qs]; | |
2557 | struct sge_eth_txq *txq = &s->ethtxq[qs]; | |
2558 | ||
c710245c | 2559 | init_rspq(&rxq->rspq, 0, 0, 1024, iqe_size); |
be839e39 CL |
2560 | rxq->fl.size = 72; |
2561 | txq->q.size = 1024; | |
2562 | } | |
2563 | ||
2564 | /* | |
2565 | * The firmware event queue is used for link state changes and | |
2566 | * notifications of TX DMA completions. | |
2567 | */ | |
c710245c | 2568 | init_rspq(&s->fw_evtq, SGE_TIMER_RSTRT_CNTR, 0, 512, iqe_size); |
be839e39 CL |
2569 | |
2570 | /* | |
2571 | * The forwarded interrupt queue is used when we're in MSI interrupt | |
2572 | * mode. In this mode all interrupts associated with RX queues will | |
2573 | * be forwarded to a single queue which we'll associate with our MSI | |
2574 | * interrupt vector. The messages dropped in the forwarded interrupt | |
2575 | * queue will indicate which ingress queue needs servicing ... This | |
2576 | * queue needs to be large enough to accommodate all of the ingress | |
2577 | * queues which are forwarding their interrupt (+1 to prevent the PIDX | |
2578 | * from equalling the CIDX if every ingress queue has an outstanding | |
2579 | * interrupt). The queue doesn't need to be any larger because no | |
2580 | * ingress queue will ever have more than one outstanding interrupt at | |
2581 | * any time ... | |
2582 | */ | |
2583 | init_rspq(&s->intrq, SGE_TIMER_RSTRT_CNTR, 0, MSIX_ENTRIES + 1, | |
c710245c | 2584 | iqe_size); |
be839e39 CL |
2585 | } |
2586 | ||
2587 | /* | |
2588 | * Reduce the number of Ethernet queues across all ports to at most n. | |
2589 | * n provides at least one queue per port. | |
2590 | */ | |
d289f864 | 2591 | static void reduce_ethqs(struct adapter *adapter, int n) |
be839e39 CL |
2592 | { |
2593 | int i; | |
2594 | struct port_info *pi; | |
2595 | ||
2596 | /* | |
2597 | * While we have too many active Ether Queue Sets, interate across the | |
2598 | * "ports" and reduce their individual Queue Set allocations. | |
2599 | */ | |
2600 | BUG_ON(n < adapter->params.nports); | |
2601 | while (n < adapter->sge.ethqsets) | |
2602 | for_each_port(adapter, i) { | |
2603 | pi = adap2pinfo(adapter, i); | |
2604 | if (pi->nqsets > 1) { | |
2605 | pi->nqsets--; | |
2606 | adapter->sge.ethqsets--; | |
2607 | if (adapter->sge.ethqsets <= n) | |
2608 | break; | |
2609 | } | |
2610 | } | |
2611 | ||
2612 | /* | |
2613 | * Reassign the starting Queue Sets for each of the "ports" ... | |
2614 | */ | |
2615 | n = 0; | |
2616 | for_each_port(adapter, i) { | |
2617 | pi = adap2pinfo(adapter, i); | |
2618 | pi->first_qset = n; | |
2619 | n += pi->nqsets; | |
2620 | } | |
2621 | } | |
2622 | ||
2623 | /* | |
2624 | * We need to grab enough MSI-X vectors to cover our interrupt needs. Ideally | |
2625 | * we get a separate MSI-X vector for every "Queue Set" plus any extras we | |
2626 | * need. Minimally we need one for every Virtual Interface plus those needed | |
2627 | * for our "extras". Note that this process may lower the maximum number of | |
2628 | * allowed Queue Sets ... | |
2629 | */ | |
d289f864 | 2630 | static int enable_msix(struct adapter *adapter) |
be839e39 | 2631 | { |
bd663689 | 2632 | int i, want, need, nqsets; |
be839e39 CL |
2633 | struct msix_entry entries[MSIX_ENTRIES]; |
2634 | struct sge *s = &adapter->sge; | |
2635 | ||
2636 | for (i = 0; i < MSIX_ENTRIES; ++i) | |
2637 | entries[i].entry = i; | |
2638 | ||
2639 | /* | |
2640 | * We _want_ enough MSI-X interrupts to cover all of our "Queue Sets" | |
2641 | * plus those needed for our "extras" (for example, the firmware | |
2642 | * message queue). We _need_ at least one "Queue Set" per Virtual | |
2643 | * Interface plus those needed for our "extras". So now we get to see | |
2644 | * if the song is right ... | |
2645 | */ | |
2646 | want = s->max_ethqsets + MSIX_EXTRAS; | |
2647 | need = adapter->params.nports + MSIX_EXTRAS; | |
bd663689 AG |
2648 | |
2649 | want = pci_enable_msix_range(adapter->pdev, entries, need, want); | |
2650 | if (want < 0) | |
2651 | return want; | |
2652 | ||
2653 | nqsets = want - MSIX_EXTRAS; | |
2654 | if (nqsets < s->max_ethqsets) { | |
2655 | dev_warn(adapter->pdev_dev, "only enough MSI-X vectors" | |
2656 | " for %d Queue Sets\n", nqsets); | |
2657 | s->max_ethqsets = nqsets; | |
2658 | if (nqsets < s->ethqsets) | |
2659 | reduce_ethqs(adapter, nqsets); | |
be839e39 | 2660 | } |
bd663689 AG |
2661 | for (i = 0; i < want; ++i) |
2662 | adapter->msix_info[i].vec = entries[i].vector; | |
2663 | ||
2664 | return 0; | |
be839e39 CL |
2665 | } |
2666 | ||
be839e39 CL |
2667 | static const struct net_device_ops cxgb4vf_netdev_ops = { |
2668 | .ndo_open = cxgb4vf_open, | |
2669 | .ndo_stop = cxgb4vf_stop, | |
2670 | .ndo_start_xmit = t4vf_eth_xmit, | |
2671 | .ndo_get_stats = cxgb4vf_get_stats, | |
2672 | .ndo_set_rx_mode = cxgb4vf_set_rxmode, | |
2673 | .ndo_set_mac_address = cxgb4vf_set_mac_addr, | |
be839e39 CL |
2674 | .ndo_validate_addr = eth_validate_addr, |
2675 | .ndo_do_ioctl = cxgb4vf_do_ioctl, | |
2676 | .ndo_change_mtu = cxgb4vf_change_mtu, | |
87737663 JP |
2677 | .ndo_fix_features = cxgb4vf_fix_features, |
2678 | .ndo_set_features = cxgb4vf_set_features, | |
be839e39 CL |
2679 | #ifdef CONFIG_NET_POLL_CONTROLLER |
2680 | .ndo_poll_controller = cxgb4vf_poll_controller, | |
2681 | #endif | |
2682 | }; | |
be839e39 CL |
2683 | |
2684 | /* | |
2685 | * "Probe" a device: initialize a device and construct all kernel and driver | |
2686 | * state needed to manage the device. This routine is called "init_one" in | |
2687 | * the PF Driver ... | |
2688 | */ | |
d289f864 | 2689 | static int cxgb4vf_pci_probe(struct pci_dev *pdev, |
1dd06ae8 | 2690 | const struct pci_device_id *ent) |
be839e39 | 2691 | { |
be839e39 CL |
2692 | int pci_using_dac; |
2693 | int err, pidx; | |
2694 | unsigned int pmask; | |
2695 | struct adapter *adapter; | |
2696 | struct port_info *pi; | |
2697 | struct net_device *netdev; | |
2698 | ||
be839e39 CL |
2699 | /* |
2700 | * Print our driver banner the first time we're called to initialize a | |
2701 | * device. | |
2702 | */ | |
428ac43f | 2703 | pr_info_once("%s - version %s\n", DRV_DESC, DRV_VERSION); |
be839e39 CL |
2704 | |
2705 | /* | |
7a0c2029 | 2706 | * Initialize generic PCI device state. |
be839e39 | 2707 | */ |
7a0c2029 | 2708 | err = pci_enable_device(pdev); |
be839e39 | 2709 | if (err) { |
7a0c2029 | 2710 | dev_err(&pdev->dev, "cannot enable PCI device\n"); |
be839e39 CL |
2711 | return err; |
2712 | } | |
2713 | ||
2714 | /* | |
7a0c2029 KV |
2715 | * Reserve PCI resources for the device. If we can't get them some |
2716 | * other driver may have already claimed the device ... | |
be839e39 | 2717 | */ |
7a0c2029 | 2718 | err = pci_request_regions(pdev, KBUILD_MODNAME); |
be839e39 | 2719 | if (err) { |
7a0c2029 KV |
2720 | dev_err(&pdev->dev, "cannot obtain PCI resources\n"); |
2721 | goto err_disable_device; | |
be839e39 CL |
2722 | } |
2723 | ||
2724 | /* | |
2725 | * Set up our DMA mask: try for 64-bit address masking first and | |
2726 | * fall back to 32-bit if we can't get 64 bits ... | |
2727 | */ | |
2728 | err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); | |
2729 | if (err == 0) { | |
2730 | err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); | |
2731 | if (err) { | |
2732 | dev_err(&pdev->dev, "unable to obtain 64-bit DMA for" | |
2733 | " coherent allocations\n"); | |
7a0c2029 | 2734 | goto err_release_regions; |
be839e39 CL |
2735 | } |
2736 | pci_using_dac = 1; | |
2737 | } else { | |
2738 | err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); | |
2739 | if (err != 0) { | |
2740 | dev_err(&pdev->dev, "no usable DMA configuration\n"); | |
7a0c2029 | 2741 | goto err_release_regions; |
be839e39 CL |
2742 | } |
2743 | pci_using_dac = 0; | |
2744 | } | |
2745 | ||
2746 | /* | |
2747 | * Enable bus mastering for the device ... | |
2748 | */ | |
2749 | pci_set_master(pdev); | |
2750 | ||
2751 | /* | |
2752 | * Allocate our adapter data structure and attach it to the device. | |
2753 | */ | |
2754 | adapter = kzalloc(sizeof(*adapter), GFP_KERNEL); | |
2755 | if (!adapter) { | |
2756 | err = -ENOMEM; | |
7a0c2029 | 2757 | goto err_release_regions; |
be839e39 CL |
2758 | } |
2759 | pci_set_drvdata(pdev, adapter); | |
2760 | adapter->pdev = pdev; | |
2761 | adapter->pdev_dev = &pdev->dev; | |
2762 | ||
ae7b7576 HS |
2763 | adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) + |
2764 | (sizeof(struct mbox_cmd) * | |
2765 | T4VF_OS_LOG_MBOX_CMDS), | |
2766 | GFP_KERNEL); | |
2767 | if (!adapter->mbox_log) { | |
2768 | err = -ENOMEM; | |
2769 | goto err_free_adapter; | |
2770 | } | |
2771 | adapter->mbox_log->size = T4VF_OS_LOG_MBOX_CMDS; | |
2772 | ||
be839e39 CL |
2773 | /* |
2774 | * Initialize SMP data synchronization resources. | |
2775 | */ | |
2776 | spin_lock_init(&adapter->stats_lock); | |
b38066da HS |
2777 | spin_lock_init(&adapter->mbox_lock); |
2778 | INIT_LIST_HEAD(&adapter->mlist.list); | |
be839e39 CL |
2779 | |
2780 | /* | |
2781 | * Map our I/O registers in BAR0. | |
2782 | */ | |
2783 | adapter->regs = pci_ioremap_bar(pdev, 0); | |
2784 | if (!adapter->regs) { | |
2785 | dev_err(&pdev->dev, "cannot map device registers\n"); | |
2786 | err = -ENOMEM; | |
2787 | goto err_free_adapter; | |
2788 | } | |
2789 | ||
e0a8b34a HS |
2790 | /* Wait for the device to become ready before proceeding ... |
2791 | */ | |
2792 | err = t4vf_prep_adapter(adapter); | |
2793 | if (err) { | |
2794 | dev_err(adapter->pdev_dev, "device didn't become ready:" | |
2795 | " err=%d\n", err); | |
2796 | goto err_unmap_bar0; | |
2797 | } | |
2798 | ||
2799 | /* For T5 and later we want to use the new BAR-based User Doorbells, | |
2800 | * so we need to map BAR2 here ... | |
2801 | */ | |
2802 | if (!is_t4(adapter->params.chip)) { | |
2803 | adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2), | |
2804 | pci_resource_len(pdev, 2)); | |
2805 | if (!adapter->bar2) { | |
2806 | dev_err(adapter->pdev_dev, "cannot map BAR2 doorbells\n"); | |
2807 | err = -ENOMEM; | |
2808 | goto err_unmap_bar0; | |
2809 | } | |
2810 | } | |
be839e39 CL |
2811 | /* |
2812 | * Initialize adapter level features. | |
2813 | */ | |
2814 | adapter->name = pci_name(pdev); | |
2815 | adapter->msg_enable = dflt_msg_enable; | |
2816 | err = adap_init0(adapter); | |
2817 | if (err) | |
2818 | goto err_unmap_bar; | |
2819 | ||
2820 | /* | |
2821 | * Allocate our "adapter ports" and stitch everything together. | |
2822 | */ | |
2823 | pmask = adapter->params.vfres.pmask; | |
2824 | for_each_port(adapter, pidx) { | |
2825 | int port_id, viid; | |
2826 | ||
2827 | /* | |
2828 | * We simplistically allocate our virtual interfaces | |
2829 | * sequentially across the port numbers to which we have | |
2830 | * access rights. This should be configurable in some manner | |
2831 | * ... | |
2832 | */ | |
2833 | if (pmask == 0) | |
2834 | break; | |
2835 | port_id = ffs(pmask) - 1; | |
2836 | pmask &= ~(1 << port_id); | |
2837 | viid = t4vf_alloc_vi(adapter, port_id); | |
2838 | if (viid < 0) { | |
2839 | dev_err(&pdev->dev, "cannot allocate VI for port %d:" | |
2840 | " err=%d\n", port_id, viid); | |
2841 | err = viid; | |
2842 | goto err_free_dev; | |
2843 | } | |
2844 | ||
2845 | /* | |
2846 | * Allocate our network device and stitch things together. | |
2847 | */ | |
2848 | netdev = alloc_etherdev_mq(sizeof(struct port_info), | |
2849 | MAX_PORT_QSETS); | |
2850 | if (netdev == NULL) { | |
be839e39 CL |
2851 | t4vf_free_vi(adapter, viid); |
2852 | err = -ENOMEM; | |
2853 | goto err_free_dev; | |
2854 | } | |
2855 | adapter->port[pidx] = netdev; | |
2856 | SET_NETDEV_DEV(netdev, &pdev->dev); | |
2857 | pi = netdev_priv(netdev); | |
2858 | pi->adapter = adapter; | |
2859 | pi->pidx = pidx; | |
2860 | pi->port_id = port_id; | |
2861 | pi->viid = viid; | |
2862 | ||
2863 | /* | |
2864 | * Initialize the starting state of our "port" and register | |
2865 | * it. | |
2866 | */ | |
2867 | pi->xact_addr_filt = -1; | |
be839e39 | 2868 | netif_carrier_off(netdev); |
be839e39 CL |
2869 | netdev->irq = pdev->irq; |
2870 | ||
2ed28baa MM |
2871 | netdev->hw_features = NETIF_F_SG | TSO_FLAGS | |
2872 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
f646968f | 2873 | NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_RXCSUM; |
2ed28baa MM |
2874 | netdev->vlan_features = NETIF_F_SG | TSO_FLAGS | |
2875 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
2876 | NETIF_F_HIGHDMA; | |
f646968f PM |
2877 | netdev->features = netdev->hw_features | |
2878 | NETIF_F_HW_VLAN_CTAG_TX; | |
be839e39 CL |
2879 | if (pci_using_dac) |
2880 | netdev->features |= NETIF_F_HIGHDMA; | |
be839e39 | 2881 | |
01789349 JP |
2882 | netdev->priv_flags |= IFF_UNICAST_FLT; |
2883 | ||
be839e39 | 2884 | netdev->netdev_ops = &cxgb4vf_netdev_ops; |
7ad24ea4 | 2885 | netdev->ethtool_ops = &cxgb4vf_ethtool_ops; |
be839e39 CL |
2886 | |
2887 | /* | |
2888 | * Initialize the hardware/software state for the port. | |
2889 | */ | |
2890 | err = t4vf_port_init(adapter, pidx); | |
2891 | if (err) { | |
2892 | dev_err(&pdev->dev, "cannot initialize port %d\n", | |
2893 | pidx); | |
2894 | goto err_free_dev; | |
2895 | } | |
2896 | } | |
2897 | ||
84f67018 HS |
2898 | /* See what interrupts we'll be using. If we've been configured to |
2899 | * use MSI-X interrupts, try to enable them but fall back to using | |
2900 | * MSI interrupts if we can't enable MSI-X interrupts. If we can't | |
2901 | * get MSI interrupts we bail with the error. | |
2902 | */ | |
2903 | if (msi == MSI_MSIX && enable_msix(adapter) == 0) | |
2904 | adapter->flags |= USING_MSIX; | |
2905 | else { | |
495c22bb HS |
2906 | if (msi == MSI_MSIX) { |
2907 | dev_info(adapter->pdev_dev, | |
2908 | "Unable to use MSI-X Interrupts; falling " | |
2909 | "back to MSI Interrupts\n"); | |
2910 | ||
2911 | /* We're going to need a Forwarded Interrupt Queue so | |
2912 | * that may cut into how many Queue Sets we can | |
2913 | * support. | |
2914 | */ | |
2915 | msi = MSI_MSI; | |
2916 | size_nports_qsets(adapter); | |
2917 | } | |
84f67018 HS |
2918 | err = pci_enable_msi(pdev); |
2919 | if (err) { | |
495c22bb HS |
2920 | dev_err(&pdev->dev, "Unable to allocate MSI Interrupts;" |
2921 | " err=%d\n", err); | |
84f67018 HS |
2922 | goto err_free_dev; |
2923 | } | |
2924 | adapter->flags |= USING_MSI; | |
2925 | } | |
2926 | ||
495c22bb HS |
2927 | /* Now that we know how many "ports" we have and what interrupt |
2928 | * mechanism we're going to use, we can configure our queue resources. | |
2929 | */ | |
2930 | cfg_queues(adapter); | |
2931 | ||
be839e39 CL |
2932 | /* |
2933 | * The "card" is now ready to go. If any errors occur during device | |
2934 | * registration we do not fail the whole "card" but rather proceed | |
2935 | * only with the ports we manage to register successfully. However we | |
2936 | * must register at least one net device. | |
2937 | */ | |
2938 | for_each_port(adapter, pidx) { | |
a8d16d08 | 2939 | struct port_info *pi = netdev_priv(adapter->port[pidx]); |
be839e39 CL |
2940 | netdev = adapter->port[pidx]; |
2941 | if (netdev == NULL) | |
2942 | continue; | |
2943 | ||
a8d16d08 HS |
2944 | netif_set_real_num_tx_queues(netdev, pi->nqsets); |
2945 | netif_set_real_num_rx_queues(netdev, pi->nqsets); | |
2946 | ||
be839e39 CL |
2947 | err = register_netdev(netdev); |
2948 | if (err) { | |
2949 | dev_warn(&pdev->dev, "cannot register net device %s," | |
2950 | " skipping\n", netdev->name); | |
2951 | continue; | |
2952 | } | |
2953 | ||
2954 | set_bit(pidx, &adapter->registered_device_map); | |
2955 | } | |
2956 | if (adapter->registered_device_map == 0) { | |
2957 | dev_err(&pdev->dev, "could not register any net devices\n"); | |
84f67018 | 2958 | goto err_disable_interrupts; |
be839e39 CL |
2959 | } |
2960 | ||
2961 | /* | |
2962 | * Set up our debugfs entries. | |
2963 | */ | |
843635e0 | 2964 | if (!IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) { |
be839e39 CL |
2965 | adapter->debugfs_root = |
2966 | debugfs_create_dir(pci_name(pdev), | |
2967 | cxgb4vf_debugfs_root); | |
843635e0 | 2968 | if (IS_ERR_OR_NULL(adapter->debugfs_root)) |
be839e39 CL |
2969 | dev_warn(&pdev->dev, "could not create debugfs" |
2970 | " directory"); | |
2971 | else | |
2972 | setup_debugfs(adapter); | |
2973 | } | |
2974 | ||
be839e39 | 2975 | /* |
25985edc | 2976 | * Print a short notice on the existence and configuration of the new |
be839e39 CL |
2977 | * VF network device ... |
2978 | */ | |
2979 | for_each_port(adapter, pidx) { | |
2980 | dev_info(adapter->pdev_dev, "%s: Chelsio VF NIC PCIe %s\n", | |
2981 | adapter->port[pidx]->name, | |
2982 | (adapter->flags & USING_MSIX) ? "MSI-X" : | |
2983 | (adapter->flags & USING_MSI) ? "MSI" : ""); | |
2984 | } | |
2985 | ||
2986 | /* | |
2987 | * Return success! | |
2988 | */ | |
2989 | return 0; | |
2990 | ||
2991 | /* | |
2992 | * Error recovery and exit code. Unwind state that's been created | |
2993 | * so far and return the error. | |
2994 | */ | |
84f67018 HS |
2995 | err_disable_interrupts: |
2996 | if (adapter->flags & USING_MSIX) { | |
2997 | pci_disable_msix(adapter->pdev); | |
2998 | adapter->flags &= ~USING_MSIX; | |
2999 | } else if (adapter->flags & USING_MSI) { | |
3000 | pci_disable_msi(adapter->pdev); | |
3001 | adapter->flags &= ~USING_MSI; | |
be839e39 CL |
3002 | } |
3003 | ||
3004 | err_free_dev: | |
3005 | for_each_port(adapter, pidx) { | |
3006 | netdev = adapter->port[pidx]; | |
3007 | if (netdev == NULL) | |
3008 | continue; | |
3009 | pi = netdev_priv(netdev); | |
3010 | t4vf_free_vi(adapter, pi->viid); | |
3011 | if (test_bit(pidx, &adapter->registered_device_map)) | |
3012 | unregister_netdev(netdev); | |
3013 | free_netdev(netdev); | |
3014 | } | |
3015 | ||
3016 | err_unmap_bar: | |
e0a8b34a HS |
3017 | if (!is_t4(adapter->params.chip)) |
3018 | iounmap(adapter->bar2); | |
3019 | ||
3020 | err_unmap_bar0: | |
be839e39 CL |
3021 | iounmap(adapter->regs); |
3022 | ||
3023 | err_free_adapter: | |
ae7b7576 | 3024 | kfree(adapter->mbox_log); |
be839e39 | 3025 | kfree(adapter); |
be839e39 | 3026 | |
be839e39 CL |
3027 | err_release_regions: |
3028 | pci_release_regions(pdev); | |
7a0c2029 KV |
3029 | pci_clear_master(pdev); |
3030 | ||
3031 | err_disable_device: | |
3032 | pci_disable_device(pdev); | |
be839e39 | 3033 | |
be839e39 CL |
3034 | return err; |
3035 | } | |
3036 | ||
3037 | /* | |
3038 | * "Remove" a device: tear down all kernel and driver state created in the | |
3039 | * "probe" routine and quiesce the device (disable interrupts, etc.). (Note | |
3040 | * that this is called "remove_one" in the PF Driver.) | |
3041 | */ | |
d289f864 | 3042 | static void cxgb4vf_pci_remove(struct pci_dev *pdev) |
be839e39 CL |
3043 | { |
3044 | struct adapter *adapter = pci_get_drvdata(pdev); | |
3045 | ||
3046 | /* | |
3047 | * Tear down driver state associated with device. | |
3048 | */ | |
3049 | if (adapter) { | |
3050 | int pidx; | |
3051 | ||
3052 | /* | |
3053 | * Stop all of our activity. Unregister network port, | |
3054 | * disable interrupts, etc. | |
3055 | */ | |
3056 | for_each_port(adapter, pidx) | |
3057 | if (test_bit(pidx, &adapter->registered_device_map)) | |
3058 | unregister_netdev(adapter->port[pidx]); | |
3059 | t4vf_sge_stop(adapter); | |
3060 | if (adapter->flags & USING_MSIX) { | |
3061 | pci_disable_msix(adapter->pdev); | |
3062 | adapter->flags &= ~USING_MSIX; | |
3063 | } else if (adapter->flags & USING_MSI) { | |
3064 | pci_disable_msi(adapter->pdev); | |
3065 | adapter->flags &= ~USING_MSI; | |
3066 | } | |
3067 | ||
3068 | /* | |
3069 | * Tear down our debugfs entries. | |
3070 | */ | |
843635e0 | 3071 | if (!IS_ERR_OR_NULL(adapter->debugfs_root)) { |
be839e39 CL |
3072 | cleanup_debugfs(adapter); |
3073 | debugfs_remove_recursive(adapter->debugfs_root); | |
3074 | } | |
3075 | ||
3076 | /* | |
3077 | * Free all of the various resources which we've acquired ... | |
3078 | */ | |
3079 | t4vf_free_sge_resources(adapter); | |
3080 | for_each_port(adapter, pidx) { | |
3081 | struct net_device *netdev = adapter->port[pidx]; | |
3082 | struct port_info *pi; | |
3083 | ||
3084 | if (netdev == NULL) | |
3085 | continue; | |
3086 | ||
3087 | pi = netdev_priv(netdev); | |
3088 | t4vf_free_vi(adapter, pi->viid); | |
3089 | free_netdev(netdev); | |
3090 | } | |
3091 | iounmap(adapter->regs); | |
e0a8b34a HS |
3092 | if (!is_t4(adapter->params.chip)) |
3093 | iounmap(adapter->bar2); | |
ae7b7576 | 3094 | kfree(adapter->mbox_log); |
be839e39 | 3095 | kfree(adapter); |
be839e39 CL |
3096 | } |
3097 | ||
3098 | /* | |
3099 | * Disable the device and release its PCI resources. | |
3100 | */ | |
3101 | pci_disable_device(pdev); | |
3102 | pci_clear_master(pdev); | |
3103 | pci_release_regions(pdev); | |
3104 | } | |
3105 | ||
7e9c2629 CL |
3106 | /* |
3107 | * "Shutdown" quiesce the device, stopping Ingress Packet and Interrupt | |
3108 | * delivery. | |
3109 | */ | |
d289f864 | 3110 | static void cxgb4vf_pci_shutdown(struct pci_dev *pdev) |
7e9c2629 CL |
3111 | { |
3112 | struct adapter *adapter; | |
3113 | int pidx; | |
3114 | ||
3115 | adapter = pci_get_drvdata(pdev); | |
3116 | if (!adapter) | |
3117 | return; | |
3118 | ||
c2a19856 | 3119 | /* Disable all Virtual Interfaces. This will shut down the |
7e9c2629 CL |
3120 | * delivery of all ingress packets into the chip for these |
3121 | * Virtual Interfaces. | |
3122 | */ | |
c2a19856 HS |
3123 | for_each_port(adapter, pidx) |
3124 | if (test_bit(pidx, &adapter->registered_device_map)) | |
3125 | unregister_netdev(adapter->port[pidx]); | |
7e9c2629 | 3126 | |
c2a19856 HS |
3127 | /* Free up all Queues which will prevent further DMA and |
3128 | * Interrupts allowing various internal pathways to drain. | |
3129 | */ | |
3130 | t4vf_sge_stop(adapter); | |
3131 | if (adapter->flags & USING_MSIX) { | |
3132 | pci_disable_msix(adapter->pdev); | |
3133 | adapter->flags &= ~USING_MSIX; | |
3134 | } else if (adapter->flags & USING_MSI) { | |
3135 | pci_disable_msi(adapter->pdev); | |
3136 | adapter->flags &= ~USING_MSI; | |
7e9c2629 CL |
3137 | } |
3138 | ||
3139 | /* | |
3140 | * Free up all Queues which will prevent further DMA and | |
3141 | * Interrupts allowing various internal pathways to drain. | |
3142 | */ | |
3143 | t4vf_free_sge_resources(adapter); | |
c2a19856 | 3144 | pci_set_drvdata(pdev, NULL); |
7e9c2629 CL |
3145 | } |
3146 | ||
3fedeab1 HS |
3147 | /* Macros needed to support the PCI Device ID Table ... |
3148 | */ | |
3149 | #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \ | |
768ffc66 | 3150 | static const struct pci_device_id cxgb4vf_pci_tbl[] = { |
3fedeab1 HS |
3151 | #define CH_PCI_DEVICE_ID_FUNCTION 0x8 |
3152 | ||
3153 | #define CH_PCI_ID_TABLE_ENTRY(devid) \ | |
3154 | { PCI_VDEVICE(CHELSIO, (devid)), 0 } | |
3155 | ||
3156 | #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END { 0, } } | |
3157 | ||
3158 | #include "../cxgb4/t4_pci_id_tbl.h" | |
be839e39 CL |
3159 | |
3160 | MODULE_DESCRIPTION(DRV_DESC); | |
3161 | MODULE_AUTHOR("Chelsio Communications"); | |
3162 | MODULE_LICENSE("Dual BSD/GPL"); | |
3163 | MODULE_VERSION(DRV_VERSION); | |
3164 | MODULE_DEVICE_TABLE(pci, cxgb4vf_pci_tbl); | |
3165 | ||
3166 | static struct pci_driver cxgb4vf_driver = { | |
3167 | .name = KBUILD_MODNAME, | |
3168 | .id_table = cxgb4vf_pci_tbl, | |
3169 | .probe = cxgb4vf_pci_probe, | |
d289f864 BP |
3170 | .remove = cxgb4vf_pci_remove, |
3171 | .shutdown = cxgb4vf_pci_shutdown, | |
be839e39 CL |
3172 | }; |
3173 | ||
3174 | /* | |
3175 | * Initialize global driver state. | |
3176 | */ | |
3177 | static int __init cxgb4vf_module_init(void) | |
3178 | { | |
3179 | int ret; | |
3180 | ||
bb14a1af CL |
3181 | /* |
3182 | * Vet our module parameters. | |
3183 | */ | |
3184 | if (msi != MSI_MSIX && msi != MSI_MSI) { | |
428ac43f JP |
3185 | pr_warn("bad module parameter msi=%d; must be %d (MSI-X or MSI) or %d (MSI)\n", |
3186 | msi, MSI_MSIX, MSI_MSI); | |
bb14a1af CL |
3187 | return -EINVAL; |
3188 | } | |
3189 | ||
be839e39 CL |
3190 | /* Debugfs support is optional, just warn if this fails */ |
3191 | cxgb4vf_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL); | |
843635e0 | 3192 | if (IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) |
428ac43f | 3193 | pr_warn("could not create debugfs entry, continuing\n"); |
be839e39 CL |
3194 | |
3195 | ret = pci_register_driver(&cxgb4vf_driver); | |
843635e0 | 3196 | if (ret < 0 && !IS_ERR_OR_NULL(cxgb4vf_debugfs_root)) |
be839e39 CL |
3197 | debugfs_remove(cxgb4vf_debugfs_root); |
3198 | return ret; | |
3199 | } | |
3200 | ||
3201 | /* | |
3202 | * Tear down global driver state. | |
3203 | */ | |
3204 | static void __exit cxgb4vf_module_exit(void) | |
3205 | { | |
3206 | pci_unregister_driver(&cxgb4vf_driver); | |
3207 | debugfs_remove(cxgb4vf_debugfs_root); | |
3208 | } | |
3209 | ||
3210 | module_init(cxgb4vf_module_init); | |
3211 | module_exit(cxgb4vf_module_exit); |