]> git.ipfire.org Git - people/arne_f/kernel.git/blob - drivers/net/ethernet/mellanox/mlx5/core/main.c
projects / people / arne_f / kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net/mlx5: Cancel health poll before sending panic teardown command
[people/arne_f/kernel.git] / drivers / net / ethernet / mellanox / mlx5 / core / main.c
1 /*
2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32
33 #include <linux/highmem.h>
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/errno.h>
37 #include <linux/pci.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/slab.h>
40 #include <linux/io-mapping.h>
41 #include <linux/interrupt.h>
42 #include <linux/delay.h>
43 #include <linux/mlx5/driver.h>
44 #include <linux/mlx5/cq.h>
45 #include <linux/mlx5/qp.h>
46 #include <linux/mlx5/srq.h>
47 #include <linux/debugfs.h>
48 #include <linux/kmod.h>
49 #include <linux/mlx5/mlx5_ifc.h>
50 #include <linux/mlx5/vport.h>
51 #ifdef CONFIG_RFS_ACCEL
52 #include <linux/cpu_rmap.h>
53 #endif
54 #include <net/devlink.h>
55 #include "mlx5_core.h"
56 #include "fs_core.h"
57 #include "lib/mpfs.h"
58 #include "eswitch.h"
59 #include "lib/mlx5.h"
60 #include "fpga/core.h"
61 #include "accel/ipsec.h"
62
63 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
64 MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
65 MODULE_LICENSE("Dual BSD/GPL");
66 MODULE_VERSION(DRIVER_VERSION);
67
68 unsigned int mlx5_core_debug_mask;
69 module_param_named(debug_mask, mlx5_core_debug_mask, uint, 0644);
70 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
71
72 #define MLX5_DEFAULT_PROF 2
73 static unsigned int prof_sel = MLX5_DEFAULT_PROF;
74 module_param_named(prof_sel, prof_sel, uint, 0444);
75 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
76
77 enum {
78 MLX5_ATOMIC_REQ_MODE_BE = 0x0,
79 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
80 };
81
82 static struct mlx5_profile profile[] = {
83 [0] = {
84 .mask = 0,
85 },
86 [1] = {
87 .mask = MLX5_PROF_MASK_QP_SIZE,
88 .log_max_qp = 12,
89 },
90 [2] = {
91 .mask = MLX5_PROF_MASK_QP_SIZE |
92 MLX5_PROF_MASK_MR_CACHE,
93 .log_max_qp = 18,
94 .mr_cache[0] = {
95 .size = 500,
96 .limit = 250
97 },
98 .mr_cache[1] = {
99 .size = 500,
100 .limit = 250
101 },
102 .mr_cache[2] = {
103 .size = 500,
104 .limit = 250
105 },
106 .mr_cache[3] = {
107 .size = 500,
108 .limit = 250
109 },
110 .mr_cache[4] = {
111 .size = 500,
112 .limit = 250
113 },
114 .mr_cache[5] = {
115 .size = 500,
116 .limit = 250
117 },
118 .mr_cache[6] = {
119 .size = 500,
120 .limit = 250
121 },
122 .mr_cache[7] = {
123 .size = 500,
124 .limit = 250
125 },
126 .mr_cache[8] = {
127 .size = 500,
128 .limit = 250
129 },
130 .mr_cache[9] = {
131 .size = 500,
132 .limit = 250
133 },
134 .mr_cache[10] = {
135 .size = 500,
136 .limit = 250
137 },
138 .mr_cache[11] = {
139 .size = 500,
140 .limit = 250
141 },
142 .mr_cache[12] = {
143 .size = 64,
144 .limit = 32
145 },
146 .mr_cache[13] = {
147 .size = 32,
148 .limit = 16
149 },
150 .mr_cache[14] = {
151 .size = 16,
152 .limit = 8
153 },
154 .mr_cache[15] = {
155 .size = 8,
156 .limit = 4
157 },
158 .mr_cache[16] = {
159 .size = 8,
160 .limit = 4
161 },
162 .mr_cache[17] = {
163 .size = 8,
164 .limit = 4
165 },
166 .mr_cache[18] = {
167 .size = 8,
168 .limit = 4
169 },
170 .mr_cache[19] = {
171 .size = 4,
172 .limit = 2
173 },
174 .mr_cache[20] = {
175 .size = 4,
176 .limit = 2
177 },
178 },
179 };
180
181 #define FW_INIT_TIMEOUT_MILI 2000
182 #define FW_INIT_WAIT_MS 2
183 #define FW_PRE_INIT_TIMEOUT_MILI 10000
184
185 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili)
186 {
187 unsigned long end = jiffies + msecs_to_jiffies(max_wait_mili);
188 int err = 0;
189
190 while (fw_initializing(dev)) {
191 if (time_after(jiffies, end)) {
192 err = -EBUSY;
193 break;
194 }
195 msleep(FW_INIT_WAIT_MS);
196 }
197
198 return err;
199 }
200
201 static void mlx5_set_driver_version(struct mlx5_core_dev *dev)
202 {
203 int driver_ver_sz = MLX5_FLD_SZ_BYTES(set_driver_version_in,
204 driver_version);
205 u8 in[MLX5_ST_SZ_BYTES(set_driver_version_in)] = {0};
206 u8 out[MLX5_ST_SZ_BYTES(set_driver_version_out)] = {0};
207 int remaining_size = driver_ver_sz;
208 char *string;
209
210 if (!MLX5_CAP_GEN(dev, driver_version))
211 return;
212
213 string = MLX5_ADDR_OF(set_driver_version_in, in, driver_version);
214
215 strncpy(string, "Linux", remaining_size);
216
217 remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
218 strncat(string, ",", remaining_size);
219
220 remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
221 strncat(string, DRIVER_NAME, remaining_size);
222
223 remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
224 strncat(string, ",", remaining_size);
225
226 remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
227 strncat(string, DRIVER_VERSION, remaining_size);
228
229 /*Send the command*/
230 MLX5_SET(set_driver_version_in, in, opcode,
231 MLX5_CMD_OP_SET_DRIVER_VERSION);
232
233 mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
234 }
235
236 static int set_dma_caps(struct pci_dev *pdev)
237 {
238 int err;
239
240 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
241 if (err) {
242 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
243 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
244 if (err) {
245 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
246 return err;
247 }
248 }
249
250 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
251 if (err) {
252 dev_warn(&pdev->dev,
253 "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
254 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
255 if (err) {
256 dev_err(&pdev->dev,
257 "Can't set consistent PCI DMA mask, aborting\n");
258 return err;
259 }
260 }
261
262 dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
263 return err;
264 }
265
266 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev)
267 {
268 struct pci_dev *pdev = dev->pdev;
269 int err = 0;
270
271 mutex_lock(&dev->pci_status_mutex);
272 if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) {
273 err = pci_enable_device(pdev);
274 if (!err)
275 dev->pci_status = MLX5_PCI_STATUS_ENABLED;
276 }
277 mutex_unlock(&dev->pci_status_mutex);
278
279 return err;
280 }
281
282 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev)
283 {
284 struct pci_dev *pdev = dev->pdev;
285
286 mutex_lock(&dev->pci_status_mutex);
287 if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) {
288 pci_disable_device(pdev);
289 dev->pci_status = MLX5_PCI_STATUS_DISABLED;
290 }
291 mutex_unlock(&dev->pci_status_mutex);
292 }
293
294 static int request_bar(struct pci_dev *pdev)
295 {
296 int err = 0;
297
298 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
299 dev_err(&pdev->dev, "Missing registers BAR, aborting\n");
300 return -ENODEV;
301 }
302
303 err = pci_request_regions(pdev, DRIVER_NAME);
304 if (err)
305 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
306
307 return err;
308 }
309
310 static void release_bar(struct pci_dev *pdev)
311 {
312 pci_release_regions(pdev);
313 }
314
315 static int mlx5_alloc_irq_vectors(struct mlx5_core_dev *dev)
316 {
317 struct mlx5_priv *priv = &dev->priv;
318 struct mlx5_eq_table *table = &priv->eq_table;
319 struct irq_affinity irqdesc = {
320 .pre_vectors = MLX5_EQ_VEC_COMP_BASE,
321 };
322 int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq);
323 int nvec;
324
325 nvec = MLX5_CAP_GEN(dev, num_ports) * num_online_cpus() +
326 MLX5_EQ_VEC_COMP_BASE;
327 nvec = min_t(int, nvec, num_eqs);
328 if (nvec <= MLX5_EQ_VEC_COMP_BASE)
329 return -ENOMEM;
330
331 priv->irq_info = kcalloc(nvec, sizeof(*priv->irq_info), GFP_KERNEL);
332 if (!priv->irq_info)
333 goto err_free_msix;
334
335 nvec = pci_alloc_irq_vectors_affinity(dev->pdev,
336 MLX5_EQ_VEC_COMP_BASE + 1, nvec,
337 PCI_IRQ_MSIX | PCI_IRQ_AFFINITY,
338 &irqdesc);
339 if (nvec < 0)
340 return nvec;
341
342 table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
343
344 return 0;
345
346 err_free_msix:
347 kfree(priv->irq_info);
348 return -ENOMEM;
349 }
350
351 static void mlx5_free_irq_vectors(struct mlx5_core_dev *dev)
352 {
353 struct mlx5_priv *priv = &dev->priv;
354
355 pci_free_irq_vectors(dev->pdev);
356 kfree(priv->irq_info);
357 }
358
359 struct mlx5_reg_host_endianness {
360 u8 he;
361 u8 rsvd[15];
362 };
363
364 #define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos))
365
366 enum {
367 MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) |
368 MLX5_DEV_CAP_FLAG_DCT,
369 };
370
371 static u16 to_fw_pkey_sz(struct mlx5_core_dev *dev, u32 size)
372 {
373 switch (size) {
374 case 128:
375 return 0;
376 case 256:
377 return 1;
378 case 512:
379 return 2;
380 case 1024:
381 return 3;
382 case 2048:
383 return 4;
384 case 4096:
385 return 5;
386 default:
387 mlx5_core_warn(dev, "invalid pkey table size %d\n", size);
388 return 0;
389 }
390 }
391
392 static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev,
393 enum mlx5_cap_type cap_type,
394 enum mlx5_cap_mode cap_mode)
395 {
396 u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
397 int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
398 void *out, *hca_caps;
399 u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
400 int err;
401
402 memset(in, 0, sizeof(in));
403 out = kzalloc(out_sz, GFP_KERNEL);
404 if (!out)
405 return -ENOMEM;
406
407 MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
408 MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
409 err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz);
410 if (err) {
411 mlx5_core_warn(dev,
412 "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
413 cap_type, cap_mode, err);
414 goto query_ex;
415 }
416
417 hca_caps = MLX5_ADDR_OF(query_hca_cap_out, out, capability);
418
419 switch (cap_mode) {
420 case HCA_CAP_OPMOD_GET_MAX:
421 memcpy(dev->caps.hca_max[cap_type], hca_caps,
422 MLX5_UN_SZ_BYTES(hca_cap_union));
423 break;
424 case HCA_CAP_OPMOD_GET_CUR:
425 memcpy(dev->caps.hca_cur[cap_type], hca_caps,
426 MLX5_UN_SZ_BYTES(hca_cap_union));
427 break;
428 default:
429 mlx5_core_warn(dev,
430 "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
431 cap_type, cap_mode);
432 err = -EINVAL;
433 break;
434 }
435 query_ex:
436 kfree(out);
437 return err;
438 }
439
440 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
441 {
442 int ret;
443
444 ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
445 if (ret)
446 return ret;
447 return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
448 }
449
450 static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz, int opmod)
451 {
452 u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)] = {0};
453
454 MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
455 MLX5_SET(set_hca_cap_in, in, op_mod, opmod << 1);
456 return mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out));
457 }
458
459 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev)
460 {
461 void *set_ctx;
462 void *set_hca_cap;
463 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
464 int req_endianness;
465 int err;
466
467 if (MLX5_CAP_GEN(dev, atomic)) {
468 err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
469 if (err)
470 return err;
471 } else {
472 return 0;
473 }
474
475 req_endianness =
476 MLX5_CAP_ATOMIC(dev,
477 supported_atomic_req_8B_endianness_mode_1);
478
479 if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
480 return 0;
481
482 set_ctx = kzalloc(set_sz, GFP_KERNEL);
483 if (!set_ctx)
484 return -ENOMEM;
485
486 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
487
488 /* Set requestor to host endianness */
489 MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianness_mode,
490 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
491
492 err = set_caps(dev, set_ctx, set_sz, MLX5_SET_HCA_CAP_OP_MOD_ATOMIC);
493
494 kfree(set_ctx);
495 return err;
496 }
497
498 static int handle_hca_cap(struct mlx5_core_dev *dev)
499 {
500 void *set_ctx = NULL;
501 struct mlx5_profile *prof = dev->profile;
502 int err = -ENOMEM;
503 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
504 void *set_hca_cap;
505
506 set_ctx = kzalloc(set_sz, GFP_KERNEL);
507 if (!set_ctx)
508 goto query_ex;
509
510 err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
511 if (err)
512 goto query_ex;
513
514 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
515 capability);
516 memcpy(set_hca_cap, dev->caps.hca_cur[MLX5_CAP_GENERAL],
517 MLX5_ST_SZ_BYTES(cmd_hca_cap));
518
519 mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
520 mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
521 128);
522 /* we limit the size of the pkey table to 128 entries for now */
523 MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
524 to_fw_pkey_sz(dev, 128));
525
526 /* Check log_max_qp from HCA caps to set in current profile */
527 if (MLX5_CAP_GEN_MAX(dev, log_max_qp) < profile[prof_sel].log_max_qp) {
528 mlx5_core_warn(dev, "log_max_qp value in current profile is %d, changing it to HCA capability limit (%d)\n",
529 profile[prof_sel].log_max_qp,
530 MLX5_CAP_GEN_MAX(dev, log_max_qp));
531 profile[prof_sel].log_max_qp = MLX5_CAP_GEN_MAX(dev, log_max_qp);
532 }
533 if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
534 MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
535 prof->log_max_qp);
536
537 /* disable cmdif checksum */
538 MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
539
540 /* Enable 4K UAR only when HCA supports it and page size is bigger
541 * than 4K.
542 */
543 if (MLX5_CAP_GEN_MAX(dev, uar_4k) && PAGE_SIZE > 4096)
544 MLX5_SET(cmd_hca_cap, set_hca_cap, uar_4k, 1);
545
546 MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
547
548 if (MLX5_CAP_GEN_MAX(dev, cache_line_128byte))
549 MLX5_SET(cmd_hca_cap,
550 set_hca_cap,
551 cache_line_128byte,
552 cache_line_size() == 128 ? 1 : 0);
553
554 err = set_caps(dev, set_ctx, set_sz,
555 MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
556
557 query_ex:
558 kfree(set_ctx);
559 return err;
560 }
561
562 static int set_hca_ctrl(struct mlx5_core_dev *dev)
563 {
564 struct mlx5_reg_host_endianness he_in;
565 struct mlx5_reg_host_endianness he_out;
566 int err;
567
568 if (!mlx5_core_is_pf(dev))
569 return 0;
570
571 memset(&he_in, 0, sizeof(he_in));
572 he_in.he = MLX5_SET_HOST_ENDIANNESS;
573 err = mlx5_core_access_reg(dev, &he_in, sizeof(he_in),
574 &he_out, sizeof(he_out),
575 MLX5_REG_HOST_ENDIANNESS, 0, 1);
576 return err;
577 }
578
579 static int mlx5_core_set_hca_defaults(struct mlx5_core_dev *dev)
580 {
581 int ret = 0;
582
583 /* Disable local_lb by default */
584 if ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
585 MLX5_CAP_GEN(dev, disable_local_lb))
586 ret = mlx5_nic_vport_update_local_lb(dev, false);
587
588 return ret;
589 }
590
591 int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
592 {
593 u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0};
594 u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0};
595
596 MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
597 MLX5_SET(enable_hca_in, in, function_id, func_id);
598 return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
599 }
600
601 int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
602 {
603 u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0};
604 u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0};
605
606 MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
607 MLX5_SET(disable_hca_in, in, function_id, func_id);
608 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
609 }
610
611 u64 mlx5_read_internal_timer(struct mlx5_core_dev *dev)
612 {
613 u32 timer_h, timer_h1, timer_l;
614
615 timer_h = ioread32be(&dev->iseg->internal_timer_h);
616 timer_l = ioread32be(&dev->iseg->internal_timer_l);
617 timer_h1 = ioread32be(&dev->iseg->internal_timer_h);
618 if (timer_h != timer_h1) /* wrap around */
619 timer_l = ioread32be(&dev->iseg->internal_timer_l);
620
621 return (u64)timer_l | (u64)timer_h1 << 32;
622 }
623
624 int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn,
625 unsigned int *irqn)
626 {
627 struct mlx5_eq_table *table = &dev->priv.eq_table;
628 struct mlx5_eq *eq, *n;
629 int err = -ENOENT;
630
631 spin_lock(&table->lock);
632 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
633 if (eq->index == vector) {
634 *eqn = eq->eqn;
635 *irqn = eq->irqn;
636 err = 0;
637 break;
638 }
639 }
640 spin_unlock(&table->lock);
641
642 return err;
643 }
644 EXPORT_SYMBOL(mlx5_vector2eqn);
645
646 struct mlx5_eq *mlx5_eqn2eq(struct mlx5_core_dev *dev, int eqn)
647 {
648 struct mlx5_eq_table *table = &dev->priv.eq_table;
649 struct mlx5_eq *eq;
650
651 spin_lock(&table->lock);
652 list_for_each_entry(eq, &table->comp_eqs_list, list)
653 if (eq->eqn == eqn) {
654 spin_unlock(&table->lock);
655 return eq;
656 }
657
658 spin_unlock(&table->lock);
659
660 return ERR_PTR(-ENOENT);
661 }
662
663 static void free_comp_eqs(struct mlx5_core_dev *dev)
664 {
665 struct mlx5_eq_table *table = &dev->priv.eq_table;
666 struct mlx5_eq *eq, *n;
667
668 #ifdef CONFIG_RFS_ACCEL
669 if (dev->rmap) {
670 free_irq_cpu_rmap(dev->rmap);
671 dev->rmap = NULL;
672 }
673 #endif
674 spin_lock(&table->lock);
675 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
676 list_del(&eq->list);
677 spin_unlock(&table->lock);
678 if (mlx5_destroy_unmap_eq(dev, eq))
679 mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n",
680 eq->eqn);
681 kfree(eq);
682 spin_lock(&table->lock);
683 }
684 spin_unlock(&table->lock);
685 }
686
687 static int alloc_comp_eqs(struct mlx5_core_dev *dev)
688 {
689 struct mlx5_eq_table *table = &dev->priv.eq_table;
690 char name[MLX5_MAX_IRQ_NAME];
691 struct mlx5_eq *eq;
692 int ncomp_vec;
693 int nent;
694 int err;
695 int i;
696
697 INIT_LIST_HEAD(&table->comp_eqs_list);
698 ncomp_vec = table->num_comp_vectors;
699 nent = MLX5_COMP_EQ_SIZE;
700 #ifdef CONFIG_RFS_ACCEL
701 dev->rmap = alloc_irq_cpu_rmap(ncomp_vec);
702 if (!dev->rmap)
703 return -ENOMEM;
704 #endif
705 for (i = 0; i < ncomp_vec; i++) {
706 eq = kzalloc(sizeof(*eq), GFP_KERNEL);
707 if (!eq) {
708 err = -ENOMEM;
709 goto clean;
710 }
711
712 #ifdef CONFIG_RFS_ACCEL
713 irq_cpu_rmap_add(dev->rmap, pci_irq_vector(dev->pdev,
714 MLX5_EQ_VEC_COMP_BASE + i));
715 #endif
716 snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i);
717 err = mlx5_create_map_eq(dev, eq,
718 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
719 name, MLX5_EQ_TYPE_COMP);
720 if (err) {
721 kfree(eq);
722 goto clean;
723 }
724 mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
725 eq->index = i;
726 spin_lock(&table->lock);
727 list_add_tail(&eq->list, &table->comp_eqs_list);
728 spin_unlock(&table->lock);
729 }
730
731 return 0;
732
733 clean:
734 free_comp_eqs(dev);
735 return err;
736 }
737
738 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
739 {
740 u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0};
741 u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0};
742 u32 sup_issi;
743 int err;
744
745 MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
746 err = mlx5_cmd_exec(dev, query_in, sizeof(query_in),
747 query_out, sizeof(query_out));
748 if (err) {
749 u32 syndrome;
750 u8 status;
751
752 mlx5_cmd_mbox_status(query_out, &status, &syndrome);
753 if (!status || syndrome == MLX5_DRIVER_SYND) {
754 mlx5_core_err(dev, "Failed to query ISSI err(%d) status(%d) synd(%d)\n",
755 err, status, syndrome);
756 return err;
757 }
758
759 mlx5_core_warn(dev, "Query ISSI is not supported by FW, ISSI is 0\n");
760 dev->issi = 0;
761 return 0;
762 }
763
764 sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
765
766 if (sup_issi & (1 << 1)) {
767 u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {0};
768 u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0};
769
770 MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
771 MLX5_SET(set_issi_in, set_in, current_issi, 1);
772 err = mlx5_cmd_exec(dev, set_in, sizeof(set_in),
773 set_out, sizeof(set_out));
774 if (err) {
775 mlx5_core_err(dev, "Failed to set ISSI to 1 err(%d)\n",
776 err);
777 return err;
778 }
779
780 dev->issi = 1;
781
782 return 0;
783 } else if (sup_issi & (1 << 0) || !sup_issi) {
784 return 0;
785 }
786
787 return -EOPNOTSUPP;
788 }
789
790 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
791 {
792 struct pci_dev *pdev = dev->pdev;
793 int err = 0;
794
795 pci_set_drvdata(dev->pdev, dev);
796 strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
797 priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
798
799 mutex_init(&priv->pgdir_mutex);
800 INIT_LIST_HEAD(&priv->pgdir_list);
801 spin_lock_init(&priv->mkey_lock);
802
803 mutex_init(&priv->alloc_mutex);
804
805 priv->numa_node = dev_to_node(&dev->pdev->dev);
806
807 priv->dbg_root = debugfs_create_dir(dev_name(&pdev->dev), mlx5_debugfs_root);
808 if (!priv->dbg_root)
809 return -ENOMEM;
810
811 err = mlx5_pci_enable_device(dev);
812 if (err) {
813 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
814 goto err_dbg;
815 }
816
817 err = request_bar(pdev);
818 if (err) {
819 dev_err(&pdev->dev, "error requesting BARs, aborting\n");
820 goto err_disable;
821 }
822
823 pci_set_master(pdev);
824
825 err = set_dma_caps(pdev);
826 if (err) {
827 dev_err(&pdev->dev, "Failed setting DMA capabilities mask, aborting\n");
828 goto err_clr_master;
829 }
830
831 dev->iseg_base = pci_resource_start(dev->pdev, 0);
832 dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
833 if (!dev->iseg) {
834 err = -ENOMEM;
835 dev_err(&pdev->dev, "Failed mapping initialization segment, aborting\n");
836 goto err_clr_master;
837 }
838
839 return 0;
840
841 err_clr_master:
842 pci_clear_master(dev->pdev);
843 release_bar(dev->pdev);
844 err_disable:
845 mlx5_pci_disable_device(dev);
846
847 err_dbg:
848 debugfs_remove(priv->dbg_root);
849 return err;
850 }
851
852 static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
853 {
854 iounmap(dev->iseg);
855 pci_clear_master(dev->pdev);
856 release_bar(dev->pdev);
857 mlx5_pci_disable_device(dev);
858 debugfs_remove(priv->dbg_root);
859 }
860
861 static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
862 {
863 struct pci_dev *pdev = dev->pdev;
864 int err;
865
866 err = mlx5_query_board_id(dev);
867 if (err) {
868 dev_err(&pdev->dev, "query board id failed\n");
869 goto out;
870 }
871
872 err = mlx5_eq_init(dev);
873 if (err) {
874 dev_err(&pdev->dev, "failed to initialize eq\n");
875 goto out;
876 }
877
878 err = mlx5_init_cq_table(dev);
879 if (err) {
880 dev_err(&pdev->dev, "failed to initialize cq table\n");
881 goto err_eq_cleanup;
882 }
883
884 mlx5_init_qp_table(dev);
885
886 mlx5_init_srq_table(dev);
887
888 mlx5_init_mkey_table(dev);
889
890 mlx5_init_reserved_gids(dev);
891
892 err = mlx5_init_rl_table(dev);
893 if (err) {
894 dev_err(&pdev->dev, "Failed to init rate limiting\n");
895 goto err_tables_cleanup;
896 }
897
898 err = mlx5_mpfs_init(dev);
899 if (err) {
900 dev_err(&pdev->dev, "Failed to init l2 table %d\n", err);
901 goto err_rl_cleanup;
902 }
903
904 err = mlx5_eswitch_init(dev);
905 if (err) {
906 dev_err(&pdev->dev, "Failed to init eswitch %d\n", err);
907 goto err_mpfs_cleanup;
908 }
909
910 err = mlx5_sriov_init(dev);
911 if (err) {
912 dev_err(&pdev->dev, "Failed to init sriov %d\n", err);
913 goto err_eswitch_cleanup;
914 }
915
916 err = mlx5_fpga_init(dev);
917 if (err) {
918 dev_err(&pdev->dev, "Failed to init fpga device %d\n", err);
919 goto err_sriov_cleanup;
920 }
921
922 return 0;
923
924 err_sriov_cleanup:
925 mlx5_sriov_cleanup(dev);
926 err_eswitch_cleanup:
927 mlx5_eswitch_cleanup(dev->priv.eswitch);
928 err_mpfs_cleanup:
929 mlx5_mpfs_cleanup(dev);
930 err_rl_cleanup:
931 mlx5_cleanup_rl_table(dev);
932 err_tables_cleanup:
933 mlx5_cleanup_mkey_table(dev);
934 mlx5_cleanup_srq_table(dev);
935 mlx5_cleanup_qp_table(dev);
936 mlx5_cleanup_cq_table(dev);
937
938 err_eq_cleanup:
939 mlx5_eq_cleanup(dev);
940
941 out:
942 return err;
943 }
944
945 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
946 {
947 mlx5_fpga_cleanup(dev);
948 mlx5_sriov_cleanup(dev);
949 mlx5_eswitch_cleanup(dev->priv.eswitch);
950 mlx5_mpfs_cleanup(dev);
951 mlx5_cleanup_rl_table(dev);
952 mlx5_cleanup_reserved_gids(dev);
953 mlx5_cleanup_mkey_table(dev);
954 mlx5_cleanup_srq_table(dev);
955 mlx5_cleanup_qp_table(dev);
956 mlx5_cleanup_cq_table(dev);
957 mlx5_eq_cleanup(dev);
958 }
959
960 static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
961 bool boot)
962 {
963 struct pci_dev *pdev = dev->pdev;
964 int err;
965
966 mutex_lock(&dev->intf_state_mutex);
967 if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
968 dev_warn(&dev->pdev->dev, "%s: interface is up, NOP\n",
969 __func__);
970 goto out;
971 }
972
973 dev_info(&pdev->dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
974 fw_rev_min(dev), fw_rev_sub(dev));
975
976 /* on load removing any previous indication of internal error, device is
977 * up
978 */
979 dev->state = MLX5_DEVICE_STATE_UP;
980
981 /* wait for firmware to accept initialization segments configurations
982 */
983 err = wait_fw_init(dev, FW_PRE_INIT_TIMEOUT_MILI);
984 if (err) {
985 dev_err(&dev->pdev->dev, "Firmware over %d MS in pre-initializing state, aborting\n",
986 FW_PRE_INIT_TIMEOUT_MILI);
987 goto out_err;
988 }
989
990 err = mlx5_cmd_init(dev);
991 if (err) {
992 dev_err(&pdev->dev, "Failed initializing command interface, aborting\n");
993 goto out_err;
994 }
995
996 err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI);
997 if (err) {
998 dev_err(&dev->pdev->dev, "Firmware over %d MS in initializing state, aborting\n",
999 FW_INIT_TIMEOUT_MILI);
1000 goto err_cmd_cleanup;
1001 }
1002
1003 err = mlx5_core_enable_hca(dev, 0);
1004 if (err) {
1005 dev_err(&pdev->dev, "enable hca failed\n");
1006 goto err_cmd_cleanup;
1007 }
1008
1009 err = mlx5_core_set_issi(dev);
1010 if (err) {
1011 dev_err(&pdev->dev, "failed to set issi\n");
1012 goto err_disable_hca;
1013 }
1014
1015 err = mlx5_satisfy_startup_pages(dev, 1);
1016 if (err) {
1017 dev_err(&pdev->dev, "failed to allocate boot pages\n");
1018 goto err_disable_hca;
1019 }
1020
1021 err = set_hca_ctrl(dev);
1022 if (err) {
1023 dev_err(&pdev->dev, "set_hca_ctrl failed\n");
1024 goto reclaim_boot_pages;
1025 }
1026
1027 err = handle_hca_cap(dev);
1028 if (err) {
1029 dev_err(&pdev->dev, "handle_hca_cap failed\n");
1030 goto reclaim_boot_pages;
1031 }
1032
1033 err = handle_hca_cap_atomic(dev);
1034 if (err) {
1035 dev_err(&pdev->dev, "handle_hca_cap_atomic failed\n");
1036 goto reclaim_boot_pages;
1037 }
1038
1039 err = mlx5_satisfy_startup_pages(dev, 0);
1040 if (err) {
1041 dev_err(&pdev->dev, "failed to allocate init pages\n");
1042 goto reclaim_boot_pages;
1043 }
1044
1045 err = mlx5_pagealloc_start(dev);
1046 if (err) {
1047 dev_err(&pdev->dev, "mlx5_pagealloc_start failed\n");
1048 goto reclaim_boot_pages;
1049 }
1050
1051 err = mlx5_cmd_init_hca(dev);
1052 if (err) {
1053 dev_err(&pdev->dev, "init hca failed\n");
1054 goto err_pagealloc_stop;
1055 }
1056
1057 mlx5_set_driver_version(dev);
1058
1059 mlx5_start_health_poll(dev);
1060
1061 err = mlx5_query_hca_caps(dev);
1062 if (err) {
1063 dev_err(&pdev->dev, "query hca failed\n");
1064 goto err_stop_poll;
1065 }
1066
1067 if (boot && mlx5_init_once(dev, priv)) {
1068 dev_err(&pdev->dev, "sw objs init failed\n");
1069 goto err_stop_poll;
1070 }
1071
1072 err = mlx5_alloc_irq_vectors(dev);
1073 if (err) {
1074 dev_err(&pdev->dev, "alloc irq vectors failed\n");
1075 goto err_cleanup_once;
1076 }
1077
1078 dev->priv.uar = mlx5_get_uars_page(dev);
1079 if (!dev->priv.uar) {
1080 dev_err(&pdev->dev, "Failed allocating uar, aborting\n");
1081 goto err_disable_msix;
1082 }
1083
1084 err = mlx5_start_eqs(dev);
1085 if (err) {
1086 dev_err(&pdev->dev, "Failed to start pages and async EQs\n");
1087 goto err_put_uars;
1088 }
1089
1090 err = alloc_comp_eqs(dev);
1091 if (err) {
1092 dev_err(&pdev->dev, "Failed to alloc completion EQs\n");
1093 goto err_stop_eqs;
1094 }
1095
1096 err = mlx5_init_fs(dev);
1097 if (err) {
1098 dev_err(&pdev->dev, "Failed to init flow steering\n");
1099 goto err_fs;
1100 }
1101
1102 err = mlx5_core_set_hca_defaults(dev);
1103 if (err) {
1104 dev_err(&pdev->dev, "Failed to set hca defaults\n");
1105 goto err_fs;
1106 }
1107
1108 err = mlx5_sriov_attach(dev);
1109 if (err) {
1110 dev_err(&pdev->dev, "sriov init failed %d\n", err);
1111 goto err_sriov;
1112 }
1113
1114 err = mlx5_fpga_device_start(dev);
1115 if (err) {
1116 dev_err(&pdev->dev, "fpga device start failed %d\n", err);
1117 goto err_fpga_start;
1118 }
1119 err = mlx5_accel_ipsec_init(dev);
1120 if (err) {
1121 dev_err(&pdev->dev, "IPSec device start failed %d\n", err);
1122 goto err_ipsec_start;
1123 }
1124
1125 if (mlx5_device_registered(dev)) {
1126 mlx5_attach_device(dev);
1127 } else {
1128 err = mlx5_register_device(dev);
1129 if (err) {
1130 dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err);
1131 goto err_reg_dev;
1132 }
1133 }
1134
1135 set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1136 out:
1137 mutex_unlock(&dev->intf_state_mutex);
1138
1139 return 0;
1140
1141 err_reg_dev:
1142 mlx5_accel_ipsec_cleanup(dev);
1143 err_ipsec_start:
1144 mlx5_fpga_device_stop(dev);
1145
1146 err_fpga_start:
1147 mlx5_sriov_detach(dev);
1148
1149 err_sriov:
1150 mlx5_cleanup_fs(dev);
1151
1152 err_fs:
1153 free_comp_eqs(dev);
1154
1155 err_stop_eqs:
1156 mlx5_stop_eqs(dev);
1157
1158 err_put_uars:
1159 mlx5_put_uars_page(dev, priv->uar);
1160
1161 err_disable_msix:
1162 mlx5_free_irq_vectors(dev);
1163
1164 err_cleanup_once:
1165 if (boot)
1166 mlx5_cleanup_once(dev);
1167
1168 err_stop_poll:
1169 mlx5_stop_health_poll(dev);
1170 if (mlx5_cmd_teardown_hca(dev)) {
1171 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1172 goto out_err;
1173 }
1174
1175 err_pagealloc_stop:
1176 mlx5_pagealloc_stop(dev);
1177
1178 reclaim_boot_pages:
1179 mlx5_reclaim_startup_pages(dev);
1180
1181 err_disable_hca:
1182 mlx5_core_disable_hca(dev, 0);
1183
1184 err_cmd_cleanup:
1185 mlx5_cmd_cleanup(dev);
1186
1187 out_err:
1188 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1189 mutex_unlock(&dev->intf_state_mutex);
1190
1191 return err;
1192 }
1193
1194 static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
1195 bool cleanup)
1196 {
1197 int err = 0;
1198
1199 if (cleanup)
1200 mlx5_drain_health_recovery(dev);
1201
1202 mutex_lock(&dev->intf_state_mutex);
1203 if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1204 dev_warn(&dev->pdev->dev, "%s: interface is down, NOP\n",
1205 __func__);
1206 if (cleanup)
1207 mlx5_cleanup_once(dev);
1208 goto out;
1209 }
1210
1211 clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1212
1213 if (mlx5_device_registered(dev))
1214 mlx5_detach_device(dev);
1215
1216 mlx5_accel_ipsec_cleanup(dev);
1217 mlx5_fpga_device_stop(dev);
1218
1219 mlx5_sriov_detach(dev);
1220 mlx5_cleanup_fs(dev);
1221 free_comp_eqs(dev);
1222 mlx5_stop_eqs(dev);
1223 mlx5_put_uars_page(dev, priv->uar);
1224 mlx5_free_irq_vectors(dev);
1225 if (cleanup)
1226 mlx5_cleanup_once(dev);
1227 mlx5_stop_health_poll(dev);
1228 err = mlx5_cmd_teardown_hca(dev);
1229 if (err) {
1230 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1231 goto out;
1232 }
1233 mlx5_pagealloc_stop(dev);
1234 mlx5_reclaim_startup_pages(dev);
1235 mlx5_core_disable_hca(dev, 0);
1236 mlx5_cmd_cleanup(dev);
1237
1238 out:
1239 mutex_unlock(&dev->intf_state_mutex);
1240 return err;
1241 }
1242
1243 struct mlx5_core_event_handler {
1244 void (*event)(struct mlx5_core_dev *dev,
1245 enum mlx5_dev_event event,
1246 void *data);
1247 };
1248
1249 static const struct devlink_ops mlx5_devlink_ops = {
1250 #ifdef CONFIG_MLX5_ESWITCH
1251 .eswitch_mode_set = mlx5_devlink_eswitch_mode_set,
1252 .eswitch_mode_get = mlx5_devlink_eswitch_mode_get,
1253 .eswitch_inline_mode_set = mlx5_devlink_eswitch_inline_mode_set,
1254 .eswitch_inline_mode_get = mlx5_devlink_eswitch_inline_mode_get,
1255 .eswitch_encap_mode_set = mlx5_devlink_eswitch_encap_mode_set,
1256 .eswitch_encap_mode_get = mlx5_devlink_eswitch_encap_mode_get,
1257 #endif
1258 };
1259
1260 #define MLX5_IB_MOD "mlx5_ib"
1261 static int init_one(struct pci_dev *pdev,
1262 const struct pci_device_id *id)
1263 {
1264 struct mlx5_core_dev *dev;
1265 struct devlink *devlink;
1266 struct mlx5_priv *priv;
1267 int err;
1268
1269 devlink = devlink_alloc(&mlx5_devlink_ops, sizeof(*dev));
1270 if (!devlink) {
1271 dev_err(&pdev->dev, "kzalloc failed\n");
1272 return -ENOMEM;
1273 }
1274
1275 dev = devlink_priv(devlink);
1276 priv = &dev->priv;
1277 priv->pci_dev_data = id->driver_data;
1278
1279 pci_set_drvdata(pdev, dev);
1280
1281 dev->pdev = pdev;
1282 dev->event = mlx5_core_event;
1283 dev->profile = &profile[prof_sel];
1284
1285 INIT_LIST_HEAD(&priv->ctx_list);
1286 spin_lock_init(&priv->ctx_lock);
1287 mutex_init(&dev->pci_status_mutex);
1288 mutex_init(&dev->intf_state_mutex);
1289
1290 INIT_LIST_HEAD(&priv->waiting_events_list);
1291 priv->is_accum_events = false;
1292
1293 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1294 err = init_srcu_struct(&priv->pfault_srcu);
1295 if (err) {
1296 dev_err(&pdev->dev, "init_srcu_struct failed with error code %d\n",
1297 err);
1298 goto clean_dev;
1299 }
1300 #endif
1301 mutex_init(&priv->bfregs.reg_head.lock);
1302 mutex_init(&priv->bfregs.wc_head.lock);
1303 INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1304 INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1305
1306 err = mlx5_pci_init(dev, priv);
1307 if (err) {
1308 dev_err(&pdev->dev, "mlx5_pci_init failed with error code %d\n", err);
1309 goto clean_srcu;
1310 }
1311
1312 err = mlx5_health_init(dev);
1313 if (err) {
1314 dev_err(&pdev->dev, "mlx5_health_init failed with error code %d\n", err);
1315 goto close_pci;
1316 }
1317
1318 mlx5_pagealloc_init(dev);
1319
1320 err = mlx5_load_one(dev, priv, true);
1321 if (err) {
1322 dev_err(&pdev->dev, "mlx5_load_one failed with error code %d\n", err);
1323 goto clean_health;
1324 }
1325
1326 request_module_nowait(MLX5_IB_MOD);
1327
1328 err = devlink_register(devlink, &pdev->dev);
1329 if (err)
1330 goto clean_load;
1331
1332 pci_save_state(pdev);
1333 return 0;
1334
1335 clean_load:
1336 mlx5_unload_one(dev, priv, true);
1337 clean_health:
1338 mlx5_pagealloc_cleanup(dev);
1339 mlx5_health_cleanup(dev);
1340 close_pci:
1341 mlx5_pci_close(dev, priv);
1342 clean_srcu:
1343 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1344 cleanup_srcu_struct(&priv->pfault_srcu);
1345 clean_dev:
1346 #endif
1347 devlink_free(devlink);
1348
1349 return err;
1350 }
1351
1352 static void remove_one(struct pci_dev *pdev)
1353 {
1354 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1355 struct devlink *devlink = priv_to_devlink(dev);
1356 struct mlx5_priv *priv = &dev->priv;
1357
1358 devlink_unregister(devlink);
1359 mlx5_unregister_device(dev);
1360
1361 if (mlx5_unload_one(dev, priv, true)) {
1362 dev_err(&dev->pdev->dev, "mlx5_unload_one failed\n");
1363 mlx5_health_cleanup(dev);
1364 return;
1365 }
1366
1367 mlx5_pagealloc_cleanup(dev);
1368 mlx5_health_cleanup(dev);
1369 mlx5_pci_close(dev, priv);
1370 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1371 cleanup_srcu_struct(&priv->pfault_srcu);
1372 #endif
1373 devlink_free(devlink);
1374 }
1375
1376 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
1377 pci_channel_state_t state)
1378 {
1379 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1380 struct mlx5_priv *priv = &dev->priv;
1381
1382 dev_info(&pdev->dev, "%s was called\n", __func__);
1383
1384 mlx5_enter_error_state(dev, false);
1385 mlx5_unload_one(dev, priv, false);
1386 /* In case of kernel call drain the health wq */
1387 if (state) {
1388 mlx5_drain_health_wq(dev);
1389 mlx5_pci_disable_device(dev);
1390 }
1391
1392 return state == pci_channel_io_perm_failure ?
1393 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1394 }
1395
1396 /* wait for the device to show vital signs by waiting
1397 * for the health counter to start counting.
1398 */
1399 static int wait_vital(struct pci_dev *pdev)
1400 {
1401 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1402 struct mlx5_core_health *health = &dev->priv.health;
1403 const int niter = 100;
1404 u32 last_count = 0;
1405 u32 count;
1406 int i;
1407
1408 for (i = 0; i < niter; i++) {
1409 count = ioread32be(health->health_counter);
1410 if (count && count != 0xffffffff) {
1411 if (last_count && last_count != count) {
1412 dev_info(&pdev->dev, "Counter value 0x%x after %d iterations\n", count, i);
1413 return 0;
1414 }
1415 last_count = count;
1416 }
1417 msleep(50);
1418 }
1419
1420 return -ETIMEDOUT;
1421 }
1422
1423 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
1424 {
1425 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1426 int err;
1427
1428 dev_info(&pdev->dev, "%s was called\n", __func__);
1429
1430 err = mlx5_pci_enable_device(dev);
1431 if (err) {
1432 dev_err(&pdev->dev, "%s: mlx5_pci_enable_device failed with error code: %d\n"
1433 , __func__, err);
1434 return PCI_ERS_RESULT_DISCONNECT;
1435 }
1436
1437 pci_set_master(pdev);
1438 pci_restore_state(pdev);
1439 pci_save_state(pdev);
1440
1441 if (wait_vital(pdev)) {
1442 dev_err(&pdev->dev, "%s: wait_vital timed out\n", __func__);
1443 return PCI_ERS_RESULT_DISCONNECT;
1444 }
1445
1446 return PCI_ERS_RESULT_RECOVERED;
1447 }
1448
1449 static void mlx5_pci_resume(struct pci_dev *pdev)
1450 {
1451 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1452 struct mlx5_priv *priv = &dev->priv;
1453 int err;
1454
1455 dev_info(&pdev->dev, "%s was called\n", __func__);
1456
1457 err = mlx5_load_one(dev, priv, false);
1458 if (err)
1459 dev_err(&pdev->dev, "%s: mlx5_load_one failed with error code: %d\n"
1460 , __func__, err);
1461 else
1462 dev_info(&pdev->dev, "%s: device recovered\n", __func__);
1463 }
1464
1465 static const struct pci_error_handlers mlx5_err_handler = {
1466 .error_detected = mlx5_pci_err_detected,
1467 .slot_reset = mlx5_pci_slot_reset,
1468 .resume = mlx5_pci_resume
1469 };
1470
1471 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
1472 {
1473 int ret;
1474
1475 if (!MLX5_CAP_GEN(dev, force_teardown)) {
1476 mlx5_core_dbg(dev, "force teardown is not supported in the firmware\n");
1477 return -EOPNOTSUPP;
1478 }
1479
1480 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1481 mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
1482 return -EAGAIN;
1483 }
1484
1485 /* Panic tear down fw command will stop the PCI bus communication
1486 * with the HCA, so the health polll is no longer needed.
1487 */
1488 mlx5_drain_health_wq(dev);
1489 mlx5_stop_health_poll(dev);
1490
1491 ret = mlx5_cmd_force_teardown_hca(dev);
1492 if (ret) {
1493 mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", ret);
1494 mlx5_start_health_poll(dev);
1495 return ret;
1496 }
1497
1498 mlx5_enter_error_state(dev, true);
1499
1500 return 0;
1501 }
1502
1503 static void shutdown(struct pci_dev *pdev)
1504 {
1505 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1506 struct mlx5_priv *priv = &dev->priv;
1507 int err;
1508
1509 dev_info(&pdev->dev, "Shutdown was called\n");
1510 err = mlx5_try_fast_unload(dev);
1511 if (err)
1512 mlx5_unload_one(dev, priv, false);
1513 mlx5_pci_disable_device(dev);
1514 }
1515
1516 static const struct pci_device_id mlx5_core_pci_table[] = {
1517 { PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTIB) },
1518 { PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF}, /* Connect-IB VF */
1519 { PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4) },
1520 { PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF}, /* ConnectX-4 VF */
1521 { PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4_LX) },
1522 { PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF}, /* ConnectX-4LX VF */
1523 { PCI_VDEVICE(MELLANOX, 0x1017) }, /* ConnectX-5, PCIe 3.0 */
1524 { PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF}, /* ConnectX-5 VF */
1525 { PCI_VDEVICE(MELLANOX, 0x1019) }, /* ConnectX-5 Ex */
1526 { PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF}, /* ConnectX-5 Ex VF */
1527 { PCI_VDEVICE(MELLANOX, 0x101b) }, /* ConnectX-6 */
1528 { PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF}, /* ConnectX-6 VF */
1529 { PCI_VDEVICE(MELLANOX, 0xa2d2) }, /* BlueField integrated ConnectX-5 network controller */
1530 { PCI_VDEVICE(MELLANOX, 0xa2d3), MLX5_PCI_DEV_IS_VF}, /* BlueField integrated ConnectX-5 network controller VF */
1531 { 0, }
1532 };
1533
1534 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1535
1536 void mlx5_disable_device(struct mlx5_core_dev *dev)
1537 {
1538 mlx5_pci_err_detected(dev->pdev, 0);
1539 }
1540
1541 void mlx5_recover_device(struct mlx5_core_dev *dev)
1542 {
1543 mlx5_pci_disable_device(dev);
1544 if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED)
1545 mlx5_pci_resume(dev->pdev);
1546 }
1547
1548 static struct pci_driver mlx5_core_driver = {
1549 .name = DRIVER_NAME,
1550 .id_table = mlx5_core_pci_table,
1551 .probe = init_one,
1552 .remove = remove_one,
1553 .shutdown = shutdown,
1554 .err_handler = &mlx5_err_handler,
1555 .sriov_configure = mlx5_core_sriov_configure,
1556 };
1557
1558 static void mlx5_core_verify_params(void)
1559 {
1560 if (prof_sel >= ARRAY_SIZE(profile)) {
1561 pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
1562 prof_sel,
1563 ARRAY_SIZE(profile) - 1,
1564 MLX5_DEFAULT_PROF);
1565 prof_sel = MLX5_DEFAULT_PROF;
1566 }
1567 }
1568
1569 static int __init init(void)
1570 {
1571 int err;
1572
1573 mlx5_core_verify_params();
1574 mlx5_register_debugfs();
1575
1576 err = pci_register_driver(&mlx5_core_driver);
1577 if (err)
1578 goto err_debug;
1579
1580 #ifdef CONFIG_MLX5_CORE_EN
1581 mlx5e_init();
1582 #endif
1583
1584 return 0;
1585
1586 err_debug:
1587 mlx5_unregister_debugfs();
1588 return err;
1589 }
1590
1591 static void __exit cleanup(void)
1592 {
1593 #ifdef CONFIG_MLX5_CORE_EN
1594 mlx5e_cleanup();
1595 #endif
1596 pci_unregister_driver(&mlx5_core_driver);
1597 mlx5_unregister_debugfs();
1598 }
1599
1600 module_init(init);
1601 module_exit(cleanup);
Arne's tree of linux kernel.