]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/net/ethernet/mellanox/mlx5/core/main.c
iio: imu: inv_mpu6050: test whoami first and against all known values
[mirror_ubuntu-artful-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 #ifdef CONFIG_RFS_ACCEL
51 #include <linux/cpu_rmap.h>
52 #endif
53 #include <net/devlink.h>
54 #include "mlx5_core.h"
55 #include "fs_core.h"
56 #ifdef CONFIG_MLX5_CORE_EN
57 #include "eswitch.h"
58 #endif
59
60 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
61 MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
62 MODULE_LICENSE("Dual BSD/GPL");
63 MODULE_VERSION(DRIVER_VERSION);
64
65 unsigned int mlx5_core_debug_mask;
66 module_param_named(debug_mask, mlx5_core_debug_mask, uint, 0644);
67 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
68
69 #define MLX5_DEFAULT_PROF 2
70 static unsigned int prof_sel = MLX5_DEFAULT_PROF;
71 module_param_named(prof_sel, prof_sel, uint, 0444);
72 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
73
74 enum {
75 MLX5_ATOMIC_REQ_MODE_BE = 0x0,
76 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
77 };
78
79 static struct mlx5_profile profile[] = {
80 [0] = {
81 .mask = 0,
82 },
83 [1] = {
84 .mask = MLX5_PROF_MASK_QP_SIZE,
85 .log_max_qp = 12,
86 },
87 [2] = {
88 .mask = MLX5_PROF_MASK_QP_SIZE |
89 MLX5_PROF_MASK_MR_CACHE,
90 .log_max_qp = 18,
91 .mr_cache[0] = {
92 .size = 500,
93 .limit = 250
94 },
95 .mr_cache[1] = {
96 .size = 500,
97 .limit = 250
98 },
99 .mr_cache[2] = {
100 .size = 500,
101 .limit = 250
102 },
103 .mr_cache[3] = {
104 .size = 500,
105 .limit = 250
106 },
107 .mr_cache[4] = {
108 .size = 500,
109 .limit = 250
110 },
111 .mr_cache[5] = {
112 .size = 500,
113 .limit = 250
114 },
115 .mr_cache[6] = {
116 .size = 500,
117 .limit = 250
118 },
119 .mr_cache[7] = {
120 .size = 500,
121 .limit = 250
122 },
123 .mr_cache[8] = {
124 .size = 500,
125 .limit = 250
126 },
127 .mr_cache[9] = {
128 .size = 500,
129 .limit = 250
130 },
131 .mr_cache[10] = {
132 .size = 500,
133 .limit = 250
134 },
135 .mr_cache[11] = {
136 .size = 500,
137 .limit = 250
138 },
139 .mr_cache[12] = {
140 .size = 64,
141 .limit = 32
142 },
143 .mr_cache[13] = {
144 .size = 32,
145 .limit = 16
146 },
147 .mr_cache[14] = {
148 .size = 16,
149 .limit = 8
150 },
151 .mr_cache[15] = {
152 .size = 8,
153 .limit = 4
154 },
155 .mr_cache[16] = {
156 .size = 8,
157 .limit = 4
158 },
159 .mr_cache[17] = {
160 .size = 8,
161 .limit = 4
162 },
163 .mr_cache[18] = {
164 .size = 8,
165 .limit = 4
166 },
167 .mr_cache[19] = {
168 .size = 4,
169 .limit = 2
170 },
171 .mr_cache[20] = {
172 .size = 4,
173 .limit = 2
174 },
175 },
176 };
177
178 #define FW_INIT_TIMEOUT_MILI 2000
179 #define FW_INIT_WAIT_MS 2
180
181 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili)
182 {
183 unsigned long end = jiffies + msecs_to_jiffies(max_wait_mili);
184 int err = 0;
185
186 while (fw_initializing(dev)) {
187 if (time_after(jiffies, end)) {
188 err = -EBUSY;
189 break;
190 }
191 msleep(FW_INIT_WAIT_MS);
192 }
193
194 return err;
195 }
196
197 static void mlx5_set_driver_version(struct mlx5_core_dev *dev)
198 {
199 int driver_ver_sz = MLX5_FLD_SZ_BYTES(set_driver_version_in,
200 driver_version);
201 u8 in[MLX5_ST_SZ_BYTES(set_driver_version_in)] = {0};
202 u8 out[MLX5_ST_SZ_BYTES(set_driver_version_out)] = {0};
203 int remaining_size = driver_ver_sz;
204 char *string;
205
206 if (!MLX5_CAP_GEN(dev, driver_version))
207 return;
208
209 string = MLX5_ADDR_OF(set_driver_version_in, in, driver_version);
210
211 strncpy(string, "Linux", remaining_size);
212
213 remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
214 strncat(string, ",", remaining_size);
215
216 remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
217 strncat(string, DRIVER_NAME, remaining_size);
218
219 remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
220 strncat(string, ",", remaining_size);
221
222 remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
223 strncat(string, DRIVER_VERSION, remaining_size);
224
225 /*Send the command*/
226 MLX5_SET(set_driver_version_in, in, opcode,
227 MLX5_CMD_OP_SET_DRIVER_VERSION);
228
229 mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
230 }
231
232 static int set_dma_caps(struct pci_dev *pdev)
233 {
234 int err;
235
236 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
237 if (err) {
238 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
239 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
240 if (err) {
241 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
242 return err;
243 }
244 }
245
246 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
247 if (err) {
248 dev_warn(&pdev->dev,
249 "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
250 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
251 if (err) {
252 dev_err(&pdev->dev,
253 "Can't set consistent PCI DMA mask, aborting\n");
254 return err;
255 }
256 }
257
258 dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
259 return err;
260 }
261
262 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev)
263 {
264 struct pci_dev *pdev = dev->pdev;
265 int err = 0;
266
267 mutex_lock(&dev->pci_status_mutex);
268 if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) {
269 err = pci_enable_device(pdev);
270 if (!err)
271 dev->pci_status = MLX5_PCI_STATUS_ENABLED;
272 }
273 mutex_unlock(&dev->pci_status_mutex);
274
275 return err;
276 }
277
278 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev)
279 {
280 struct pci_dev *pdev = dev->pdev;
281
282 mutex_lock(&dev->pci_status_mutex);
283 if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) {
284 pci_disable_device(pdev);
285 dev->pci_status = MLX5_PCI_STATUS_DISABLED;
286 }
287 mutex_unlock(&dev->pci_status_mutex);
288 }
289
290 static int request_bar(struct pci_dev *pdev)
291 {
292 int err = 0;
293
294 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
295 dev_err(&pdev->dev, "Missing registers BAR, aborting\n");
296 return -ENODEV;
297 }
298
299 err = pci_request_regions(pdev, DRIVER_NAME);
300 if (err)
301 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
302
303 return err;
304 }
305
306 static void release_bar(struct pci_dev *pdev)
307 {
308 pci_release_regions(pdev);
309 }
310
311 static int mlx5_enable_msix(struct mlx5_core_dev *dev)
312 {
313 struct mlx5_priv *priv = &dev->priv;
314 struct mlx5_eq_table *table = &priv->eq_table;
315 int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq);
316 int nvec;
317 int i;
318
319 nvec = MLX5_CAP_GEN(dev, num_ports) * num_online_cpus() +
320 MLX5_EQ_VEC_COMP_BASE;
321 nvec = min_t(int, nvec, num_eqs);
322 if (nvec <= MLX5_EQ_VEC_COMP_BASE)
323 return -ENOMEM;
324
325 priv->msix_arr = kcalloc(nvec, sizeof(*priv->msix_arr), GFP_KERNEL);
326
327 priv->irq_info = kcalloc(nvec, sizeof(*priv->irq_info), GFP_KERNEL);
328 if (!priv->msix_arr || !priv->irq_info)
329 goto err_free_msix;
330
331 for (i = 0; i < nvec; i++)
332 priv->msix_arr[i].entry = i;
333
334 nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr,
335 MLX5_EQ_VEC_COMP_BASE + 1, nvec);
336 if (nvec < 0)
337 return nvec;
338
339 table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
340
341 return 0;
342
343 err_free_msix:
344 kfree(priv->irq_info);
345 kfree(priv->msix_arr);
346 return -ENOMEM;
347 }
348
349 static void mlx5_disable_msix(struct mlx5_core_dev *dev)
350 {
351 struct mlx5_priv *priv = &dev->priv;
352
353 pci_disable_msix(dev->pdev);
354 kfree(priv->irq_info);
355 kfree(priv->msix_arr);
356 }
357
358 struct mlx5_reg_host_endianess {
359 u8 he;
360 u8 rsvd[15];
361 };
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_endianess_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_endianess_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 /* If the HCA supports 4K UARs use it */
541 if (MLX5_CAP_GEN_MAX(dev, uar_4k))
542 MLX5_SET(cmd_hca_cap, set_hca_cap, uar_4k, 1);
543
544 MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
545
546 if (MLX5_CAP_GEN_MAX(dev, cache_line_128byte))
547 MLX5_SET(cmd_hca_cap,
548 set_hca_cap,
549 cache_line_128byte,
550 cache_line_size() == 128 ? 1 : 0);
551
552 err = set_caps(dev, set_ctx, set_sz,
553 MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
554
555 query_ex:
556 kfree(set_ctx);
557 return err;
558 }
559
560 static int set_hca_ctrl(struct mlx5_core_dev *dev)
561 {
562 struct mlx5_reg_host_endianess he_in;
563 struct mlx5_reg_host_endianess he_out;
564 int err;
565
566 if (!mlx5_core_is_pf(dev))
567 return 0;
568
569 memset(&he_in, 0, sizeof(he_in));
570 he_in.he = MLX5_SET_HOST_ENDIANNESS;
571 err = mlx5_core_access_reg(dev, &he_in, sizeof(he_in),
572 &he_out, sizeof(he_out),
573 MLX5_REG_HOST_ENDIANNESS, 0, 1);
574 return err;
575 }
576
577 int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
578 {
579 u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0};
580 u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0};
581
582 MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
583 MLX5_SET(enable_hca_in, in, function_id, func_id);
584 return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
585 }
586
587 int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
588 {
589 u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0};
590 u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0};
591
592 MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
593 MLX5_SET(disable_hca_in, in, function_id, func_id);
594 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
595 }
596
597 u64 mlx5_read_internal_timer(struct mlx5_core_dev *dev)
598 {
599 u32 timer_h, timer_h1, timer_l;
600
601 timer_h = ioread32be(&dev->iseg->internal_timer_h);
602 timer_l = ioread32be(&dev->iseg->internal_timer_l);
603 timer_h1 = ioread32be(&dev->iseg->internal_timer_h);
604 if (timer_h != timer_h1) /* wrap around */
605 timer_l = ioread32be(&dev->iseg->internal_timer_l);
606
607 return (u64)timer_l | (u64)timer_h1 << 32;
608 }
609
610 static int mlx5_irq_set_affinity_hint(struct mlx5_core_dev *mdev, int i)
611 {
612 struct mlx5_priv *priv = &mdev->priv;
613 struct msix_entry *msix = priv->msix_arr;
614 int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
615 int err;
616
617 if (!zalloc_cpumask_var(&priv->irq_info[i].mask, GFP_KERNEL)) {
618 mlx5_core_warn(mdev, "zalloc_cpumask_var failed");
619 return -ENOMEM;
620 }
621
622 cpumask_set_cpu(cpumask_local_spread(i, priv->numa_node),
623 priv->irq_info[i].mask);
624
625 err = irq_set_affinity_hint(irq, priv->irq_info[i].mask);
626 if (err) {
627 mlx5_core_warn(mdev, "irq_set_affinity_hint failed,irq 0x%.4x",
628 irq);
629 goto err_clear_mask;
630 }
631
632 return 0;
633
634 err_clear_mask:
635 free_cpumask_var(priv->irq_info[i].mask);
636 return err;
637 }
638
639 static void mlx5_irq_clear_affinity_hint(struct mlx5_core_dev *mdev, int i)
640 {
641 struct mlx5_priv *priv = &mdev->priv;
642 struct msix_entry *msix = priv->msix_arr;
643 int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
644
645 irq_set_affinity_hint(irq, NULL);
646 free_cpumask_var(priv->irq_info[i].mask);
647 }
648
649 static int mlx5_irq_set_affinity_hints(struct mlx5_core_dev *mdev)
650 {
651 int err;
652 int i;
653
654 for (i = 0; i < mdev->priv.eq_table.num_comp_vectors; i++) {
655 err = mlx5_irq_set_affinity_hint(mdev, i);
656 if (err)
657 goto err_out;
658 }
659
660 return 0;
661
662 err_out:
663 for (i--; i >= 0; i--)
664 mlx5_irq_clear_affinity_hint(mdev, i);
665
666 return err;
667 }
668
669 static void mlx5_irq_clear_affinity_hints(struct mlx5_core_dev *mdev)
670 {
671 int i;
672
673 for (i = 0; i < mdev->priv.eq_table.num_comp_vectors; i++)
674 mlx5_irq_clear_affinity_hint(mdev, i);
675 }
676
677 int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn,
678 unsigned int *irqn)
679 {
680 struct mlx5_eq_table *table = &dev->priv.eq_table;
681 struct mlx5_eq *eq, *n;
682 int err = -ENOENT;
683
684 spin_lock(&table->lock);
685 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
686 if (eq->index == vector) {
687 *eqn = eq->eqn;
688 *irqn = eq->irqn;
689 err = 0;
690 break;
691 }
692 }
693 spin_unlock(&table->lock);
694
695 return err;
696 }
697 EXPORT_SYMBOL(mlx5_vector2eqn);
698
699 struct mlx5_eq *mlx5_eqn2eq(struct mlx5_core_dev *dev, int eqn)
700 {
701 struct mlx5_eq_table *table = &dev->priv.eq_table;
702 struct mlx5_eq *eq;
703
704 spin_lock(&table->lock);
705 list_for_each_entry(eq, &table->comp_eqs_list, list)
706 if (eq->eqn == eqn) {
707 spin_unlock(&table->lock);
708 return eq;
709 }
710
711 spin_unlock(&table->lock);
712
713 return ERR_PTR(-ENOENT);
714 }
715
716 static void free_comp_eqs(struct mlx5_core_dev *dev)
717 {
718 struct mlx5_eq_table *table = &dev->priv.eq_table;
719 struct mlx5_eq *eq, *n;
720
721 #ifdef CONFIG_RFS_ACCEL
722 if (dev->rmap) {
723 free_irq_cpu_rmap(dev->rmap);
724 dev->rmap = NULL;
725 }
726 #endif
727 spin_lock(&table->lock);
728 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
729 list_del(&eq->list);
730 spin_unlock(&table->lock);
731 if (mlx5_destroy_unmap_eq(dev, eq))
732 mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n",
733 eq->eqn);
734 kfree(eq);
735 spin_lock(&table->lock);
736 }
737 spin_unlock(&table->lock);
738 }
739
740 static int alloc_comp_eqs(struct mlx5_core_dev *dev)
741 {
742 struct mlx5_eq_table *table = &dev->priv.eq_table;
743 char name[MLX5_MAX_IRQ_NAME];
744 struct mlx5_eq *eq;
745 int ncomp_vec;
746 int nent;
747 int err;
748 int i;
749
750 INIT_LIST_HEAD(&table->comp_eqs_list);
751 ncomp_vec = table->num_comp_vectors;
752 nent = MLX5_COMP_EQ_SIZE;
753 #ifdef CONFIG_RFS_ACCEL
754 dev->rmap = alloc_irq_cpu_rmap(ncomp_vec);
755 if (!dev->rmap)
756 return -ENOMEM;
757 #endif
758 for (i = 0; i < ncomp_vec; i++) {
759 eq = kzalloc(sizeof(*eq), GFP_KERNEL);
760 if (!eq) {
761 err = -ENOMEM;
762 goto clean;
763 }
764
765 #ifdef CONFIG_RFS_ACCEL
766 irq_cpu_rmap_add(dev->rmap,
767 dev->priv.msix_arr[i + MLX5_EQ_VEC_COMP_BASE].vector);
768 #endif
769 snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i);
770 err = mlx5_create_map_eq(dev, eq,
771 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
772 name, MLX5_EQ_TYPE_COMP);
773 if (err) {
774 kfree(eq);
775 goto clean;
776 }
777 mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
778 eq->index = i;
779 spin_lock(&table->lock);
780 list_add_tail(&eq->list, &table->comp_eqs_list);
781 spin_unlock(&table->lock);
782 }
783
784 return 0;
785
786 clean:
787 free_comp_eqs(dev);
788 return err;
789 }
790
791 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
792 {
793 u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0};
794 u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0};
795 u32 sup_issi;
796 int err;
797
798 MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
799 err = mlx5_cmd_exec(dev, query_in, sizeof(query_in),
800 query_out, sizeof(query_out));
801 if (err) {
802 u32 syndrome;
803 u8 status;
804
805 mlx5_cmd_mbox_status(query_out, &status, &syndrome);
806 if (!status || syndrome == MLX5_DRIVER_SYND) {
807 mlx5_core_err(dev, "Failed to query ISSI err(%d) status(%d) synd(%d)\n",
808 err, status, syndrome);
809 return err;
810 }
811
812 mlx5_core_warn(dev, "Query ISSI is not supported by FW, ISSI is 0\n");
813 dev->issi = 0;
814 return 0;
815 }
816
817 sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
818
819 if (sup_issi & (1 << 1)) {
820 u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {0};
821 u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0};
822
823 MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
824 MLX5_SET(set_issi_in, set_in, current_issi, 1);
825 err = mlx5_cmd_exec(dev, set_in, sizeof(set_in),
826 set_out, sizeof(set_out));
827 if (err) {
828 mlx5_core_err(dev, "Failed to set ISSI to 1 err(%d)\n",
829 err);
830 return err;
831 }
832
833 dev->issi = 1;
834
835 return 0;
836 } else if (sup_issi & (1 << 0) || !sup_issi) {
837 return 0;
838 }
839
840 return -EOPNOTSUPP;
841 }
842
843
844 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
845 {
846 struct pci_dev *pdev = dev->pdev;
847 int err = 0;
848
849 pci_set_drvdata(dev->pdev, dev);
850 strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
851 priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
852
853 mutex_init(&priv->pgdir_mutex);
854 INIT_LIST_HEAD(&priv->pgdir_list);
855 spin_lock_init(&priv->mkey_lock);
856
857 mutex_init(&priv->alloc_mutex);
858
859 priv->numa_node = dev_to_node(&dev->pdev->dev);
860
861 priv->dbg_root = debugfs_create_dir(dev_name(&pdev->dev), mlx5_debugfs_root);
862 if (!priv->dbg_root)
863 return -ENOMEM;
864
865 err = mlx5_pci_enable_device(dev);
866 if (err) {
867 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
868 goto err_dbg;
869 }
870
871 err = request_bar(pdev);
872 if (err) {
873 dev_err(&pdev->dev, "error requesting BARs, aborting\n");
874 goto err_disable;
875 }
876
877 pci_set_master(pdev);
878
879 err = set_dma_caps(pdev);
880 if (err) {
881 dev_err(&pdev->dev, "Failed setting DMA capabilities mask, aborting\n");
882 goto err_clr_master;
883 }
884
885 dev->iseg_base = pci_resource_start(dev->pdev, 0);
886 dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
887 if (!dev->iseg) {
888 err = -ENOMEM;
889 dev_err(&pdev->dev, "Failed mapping initialization segment, aborting\n");
890 goto err_clr_master;
891 }
892
893 return 0;
894
895 err_clr_master:
896 pci_clear_master(dev->pdev);
897 release_bar(dev->pdev);
898 err_disable:
899 mlx5_pci_disable_device(dev);
900
901 err_dbg:
902 debugfs_remove(priv->dbg_root);
903 return err;
904 }
905
906 static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
907 {
908 iounmap(dev->iseg);
909 pci_clear_master(dev->pdev);
910 release_bar(dev->pdev);
911 mlx5_pci_disable_device(dev);
912 debugfs_remove(priv->dbg_root);
913 }
914
915 static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
916 {
917 struct pci_dev *pdev = dev->pdev;
918 int err;
919
920 err = mlx5_query_board_id(dev);
921 if (err) {
922 dev_err(&pdev->dev, "query board id failed\n");
923 goto out;
924 }
925
926 err = mlx5_eq_init(dev);
927 if (err) {
928 dev_err(&pdev->dev, "failed to initialize eq\n");
929 goto out;
930 }
931
932 err = mlx5_init_cq_table(dev);
933 if (err) {
934 dev_err(&pdev->dev, "failed to initialize cq table\n");
935 goto err_eq_cleanup;
936 }
937
938 mlx5_init_qp_table(dev);
939
940 mlx5_init_srq_table(dev);
941
942 mlx5_init_mkey_table(dev);
943
944 err = mlx5_init_rl_table(dev);
945 if (err) {
946 dev_err(&pdev->dev, "Failed to init rate limiting\n");
947 goto err_tables_cleanup;
948 }
949
950 #ifdef CONFIG_MLX5_CORE_EN
951 err = mlx5_eswitch_init(dev);
952 if (err) {
953 dev_err(&pdev->dev, "Failed to init eswitch %d\n", err);
954 goto err_rl_cleanup;
955 }
956 #endif
957
958 err = mlx5_sriov_init(dev);
959 if (err) {
960 dev_err(&pdev->dev, "Failed to init sriov %d\n", err);
961 goto err_eswitch_cleanup;
962 }
963
964 return 0;
965
966 err_eswitch_cleanup:
967 #ifdef CONFIG_MLX5_CORE_EN
968 mlx5_eswitch_cleanup(dev->priv.eswitch);
969
970 err_rl_cleanup:
971 #endif
972 mlx5_cleanup_rl_table(dev);
973
974 err_tables_cleanup:
975 mlx5_cleanup_mkey_table(dev);
976 mlx5_cleanup_srq_table(dev);
977 mlx5_cleanup_qp_table(dev);
978 mlx5_cleanup_cq_table(dev);
979
980 err_eq_cleanup:
981 mlx5_eq_cleanup(dev);
982
983 out:
984 return err;
985 }
986
987 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
988 {
989 mlx5_sriov_cleanup(dev);
990 #ifdef CONFIG_MLX5_CORE_EN
991 mlx5_eswitch_cleanup(dev->priv.eswitch);
992 #endif
993 mlx5_cleanup_rl_table(dev);
994 mlx5_cleanup_mkey_table(dev);
995 mlx5_cleanup_srq_table(dev);
996 mlx5_cleanup_qp_table(dev);
997 mlx5_cleanup_cq_table(dev);
998 mlx5_eq_cleanup(dev);
999 }
1000
1001 static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
1002 bool boot)
1003 {
1004 struct pci_dev *pdev = dev->pdev;
1005 int err;
1006
1007 mutex_lock(&dev->intf_state_mutex);
1008 if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1009 dev_warn(&dev->pdev->dev, "%s: interface is up, NOP\n",
1010 __func__);
1011 goto out;
1012 }
1013
1014 dev_info(&pdev->dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1015 fw_rev_min(dev), fw_rev_sub(dev));
1016
1017 /* on load removing any previous indication of internal error, device is
1018 * up
1019 */
1020 dev->state = MLX5_DEVICE_STATE_UP;
1021
1022 err = mlx5_cmd_init(dev);
1023 if (err) {
1024 dev_err(&pdev->dev, "Failed initializing command interface, aborting\n");
1025 goto out_err;
1026 }
1027
1028 err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI);
1029 if (err) {
1030 dev_err(&dev->pdev->dev, "Firmware over %d MS in initializing state, aborting\n",
1031 FW_INIT_TIMEOUT_MILI);
1032 goto err_cmd_cleanup;
1033 }
1034
1035 err = mlx5_core_enable_hca(dev, 0);
1036 if (err) {
1037 dev_err(&pdev->dev, "enable hca failed\n");
1038 goto err_cmd_cleanup;
1039 }
1040
1041 err = mlx5_core_set_issi(dev);
1042 if (err) {
1043 dev_err(&pdev->dev, "failed to set issi\n");
1044 goto err_disable_hca;
1045 }
1046
1047 err = mlx5_satisfy_startup_pages(dev, 1);
1048 if (err) {
1049 dev_err(&pdev->dev, "failed to allocate boot pages\n");
1050 goto err_disable_hca;
1051 }
1052
1053 err = set_hca_ctrl(dev);
1054 if (err) {
1055 dev_err(&pdev->dev, "set_hca_ctrl failed\n");
1056 goto reclaim_boot_pages;
1057 }
1058
1059 err = handle_hca_cap(dev);
1060 if (err) {
1061 dev_err(&pdev->dev, "handle_hca_cap failed\n");
1062 goto reclaim_boot_pages;
1063 }
1064
1065 err = handle_hca_cap_atomic(dev);
1066 if (err) {
1067 dev_err(&pdev->dev, "handle_hca_cap_atomic failed\n");
1068 goto reclaim_boot_pages;
1069 }
1070
1071 err = mlx5_satisfy_startup_pages(dev, 0);
1072 if (err) {
1073 dev_err(&pdev->dev, "failed to allocate init pages\n");
1074 goto reclaim_boot_pages;
1075 }
1076
1077 err = mlx5_pagealloc_start(dev);
1078 if (err) {
1079 dev_err(&pdev->dev, "mlx5_pagealloc_start failed\n");
1080 goto reclaim_boot_pages;
1081 }
1082
1083 err = mlx5_cmd_init_hca(dev);
1084 if (err) {
1085 dev_err(&pdev->dev, "init hca failed\n");
1086 goto err_pagealloc_stop;
1087 }
1088
1089 mlx5_set_driver_version(dev);
1090
1091 mlx5_start_health_poll(dev);
1092
1093 err = mlx5_query_hca_caps(dev);
1094 if (err) {
1095 dev_err(&pdev->dev, "query hca failed\n");
1096 goto err_stop_poll;
1097 }
1098
1099 if (boot && mlx5_init_once(dev, priv)) {
1100 dev_err(&pdev->dev, "sw objs init failed\n");
1101 goto err_stop_poll;
1102 }
1103
1104 err = mlx5_enable_msix(dev);
1105 if (err) {
1106 dev_err(&pdev->dev, "enable msix failed\n");
1107 goto err_cleanup_once;
1108 }
1109
1110 dev->priv.uar = mlx5_get_uars_page(dev);
1111 if (!dev->priv.uar) {
1112 dev_err(&pdev->dev, "Failed allocating uar, aborting\n");
1113 goto err_disable_msix;
1114 }
1115
1116 err = mlx5_start_eqs(dev);
1117 if (err) {
1118 dev_err(&pdev->dev, "Failed to start pages and async EQs\n");
1119 goto err_put_uars;
1120 }
1121
1122 err = alloc_comp_eqs(dev);
1123 if (err) {
1124 dev_err(&pdev->dev, "Failed to alloc completion EQs\n");
1125 goto err_stop_eqs;
1126 }
1127
1128 err = mlx5_irq_set_affinity_hints(dev);
1129 if (err) {
1130 dev_err(&pdev->dev, "Failed to alloc affinity hint cpumask\n");
1131 goto err_affinity_hints;
1132 }
1133
1134 err = mlx5_init_fs(dev);
1135 if (err) {
1136 dev_err(&pdev->dev, "Failed to init flow steering\n");
1137 goto err_fs;
1138 }
1139
1140 #ifdef CONFIG_MLX5_CORE_EN
1141 mlx5_eswitch_attach(dev->priv.eswitch);
1142 #endif
1143
1144 err = mlx5_sriov_attach(dev);
1145 if (err) {
1146 dev_err(&pdev->dev, "sriov init failed %d\n", err);
1147 goto err_sriov;
1148 }
1149
1150 if (mlx5_device_registered(dev)) {
1151 mlx5_attach_device(dev);
1152 } else {
1153 err = mlx5_register_device(dev);
1154 if (err) {
1155 dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err);
1156 goto err_reg_dev;
1157 }
1158 }
1159
1160 clear_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state);
1161 set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1162 out:
1163 mutex_unlock(&dev->intf_state_mutex);
1164
1165 return 0;
1166
1167 err_reg_dev:
1168 mlx5_sriov_detach(dev);
1169
1170 err_sriov:
1171 #ifdef CONFIG_MLX5_CORE_EN
1172 mlx5_eswitch_detach(dev->priv.eswitch);
1173 #endif
1174 mlx5_cleanup_fs(dev);
1175
1176 err_fs:
1177 mlx5_irq_clear_affinity_hints(dev);
1178
1179 err_affinity_hints:
1180 free_comp_eqs(dev);
1181
1182 err_stop_eqs:
1183 mlx5_stop_eqs(dev);
1184
1185 err_put_uars:
1186 mlx5_put_uars_page(dev, priv->uar);
1187
1188 err_disable_msix:
1189 mlx5_disable_msix(dev);
1190
1191 err_cleanup_once:
1192 if (boot)
1193 mlx5_cleanup_once(dev);
1194
1195 err_stop_poll:
1196 mlx5_stop_health_poll(dev);
1197 if (mlx5_cmd_teardown_hca(dev)) {
1198 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1199 goto out_err;
1200 }
1201
1202 err_pagealloc_stop:
1203 mlx5_pagealloc_stop(dev);
1204
1205 reclaim_boot_pages:
1206 mlx5_reclaim_startup_pages(dev);
1207
1208 err_disable_hca:
1209 mlx5_core_disable_hca(dev, 0);
1210
1211 err_cmd_cleanup:
1212 mlx5_cmd_cleanup(dev);
1213
1214 out_err:
1215 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1216 mutex_unlock(&dev->intf_state_mutex);
1217
1218 return err;
1219 }
1220
1221 static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
1222 bool cleanup)
1223 {
1224 int err = 0;
1225
1226 if (cleanup)
1227 mlx5_drain_health_wq(dev);
1228
1229 mutex_lock(&dev->intf_state_mutex);
1230 if (test_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state)) {
1231 dev_warn(&dev->pdev->dev, "%s: interface is down, NOP\n",
1232 __func__);
1233 if (cleanup)
1234 mlx5_cleanup_once(dev);
1235 goto out;
1236 }
1237
1238 if (mlx5_device_registered(dev))
1239 mlx5_detach_device(dev);
1240
1241 mlx5_sriov_detach(dev);
1242 #ifdef CONFIG_MLX5_CORE_EN
1243 mlx5_eswitch_detach(dev->priv.eswitch);
1244 #endif
1245 mlx5_cleanup_fs(dev);
1246 mlx5_irq_clear_affinity_hints(dev);
1247 free_comp_eqs(dev);
1248 mlx5_stop_eqs(dev);
1249 mlx5_put_uars_page(dev, priv->uar);
1250 mlx5_disable_msix(dev);
1251 if (cleanup)
1252 mlx5_cleanup_once(dev);
1253 mlx5_stop_health_poll(dev);
1254 err = mlx5_cmd_teardown_hca(dev);
1255 if (err) {
1256 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1257 goto out;
1258 }
1259 mlx5_pagealloc_stop(dev);
1260 mlx5_reclaim_startup_pages(dev);
1261 mlx5_core_disable_hca(dev, 0);
1262 mlx5_cmd_cleanup(dev);
1263
1264 out:
1265 clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1266 set_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state);
1267 mutex_unlock(&dev->intf_state_mutex);
1268 return err;
1269 }
1270
1271 struct mlx5_core_event_handler {
1272 void (*event)(struct mlx5_core_dev *dev,
1273 enum mlx5_dev_event event,
1274 void *data);
1275 };
1276
1277 static const struct devlink_ops mlx5_devlink_ops = {
1278 #ifdef CONFIG_MLX5_CORE_EN
1279 .eswitch_mode_set = mlx5_devlink_eswitch_mode_set,
1280 .eswitch_mode_get = mlx5_devlink_eswitch_mode_get,
1281 .eswitch_inline_mode_set = mlx5_devlink_eswitch_inline_mode_set,
1282 .eswitch_inline_mode_get = mlx5_devlink_eswitch_inline_mode_get,
1283 .eswitch_encap_mode_set = mlx5_devlink_eswitch_encap_mode_set,
1284 .eswitch_encap_mode_get = mlx5_devlink_eswitch_encap_mode_get,
1285 #endif
1286 };
1287
1288 #define MLX5_IB_MOD "mlx5_ib"
1289 static int init_one(struct pci_dev *pdev,
1290 const struct pci_device_id *id)
1291 {
1292 struct mlx5_core_dev *dev;
1293 struct devlink *devlink;
1294 struct mlx5_priv *priv;
1295 int err;
1296
1297 devlink = devlink_alloc(&mlx5_devlink_ops, sizeof(*dev));
1298 if (!devlink) {
1299 dev_err(&pdev->dev, "kzalloc failed\n");
1300 return -ENOMEM;
1301 }
1302
1303 dev = devlink_priv(devlink);
1304 priv = &dev->priv;
1305 priv->pci_dev_data = id->driver_data;
1306
1307 pci_set_drvdata(pdev, dev);
1308
1309 dev->pdev = pdev;
1310 dev->event = mlx5_core_event;
1311 dev->profile = &profile[prof_sel];
1312
1313 INIT_LIST_HEAD(&priv->ctx_list);
1314 spin_lock_init(&priv->ctx_lock);
1315 mutex_init(&dev->pci_status_mutex);
1316 mutex_init(&dev->intf_state_mutex);
1317
1318 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1319 err = init_srcu_struct(&priv->pfault_srcu);
1320 if (err) {
1321 dev_err(&pdev->dev, "init_srcu_struct failed with error code %d\n",
1322 err);
1323 goto clean_dev;
1324 }
1325 #endif
1326 mutex_init(&priv->bfregs.reg_head.lock);
1327 mutex_init(&priv->bfregs.wc_head.lock);
1328 INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1329 INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1330
1331 err = mlx5_pci_init(dev, priv);
1332 if (err) {
1333 dev_err(&pdev->dev, "mlx5_pci_init failed with error code %d\n", err);
1334 goto clean_srcu;
1335 }
1336
1337 err = mlx5_health_init(dev);
1338 if (err) {
1339 dev_err(&pdev->dev, "mlx5_health_init failed with error code %d\n", err);
1340 goto close_pci;
1341 }
1342
1343 mlx5_pagealloc_init(dev);
1344
1345 err = mlx5_load_one(dev, priv, true);
1346 if (err) {
1347 dev_err(&pdev->dev, "mlx5_load_one failed with error code %d\n", err);
1348 goto clean_health;
1349 }
1350
1351 request_module_nowait(MLX5_IB_MOD);
1352
1353 err = devlink_register(devlink, &pdev->dev);
1354 if (err)
1355 goto clean_load;
1356
1357 pci_save_state(pdev);
1358 return 0;
1359
1360 clean_load:
1361 mlx5_unload_one(dev, priv, true);
1362 clean_health:
1363 mlx5_pagealloc_cleanup(dev);
1364 mlx5_health_cleanup(dev);
1365 close_pci:
1366 mlx5_pci_close(dev, priv);
1367 clean_srcu:
1368 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1369 cleanup_srcu_struct(&priv->pfault_srcu);
1370 clean_dev:
1371 #endif
1372 pci_set_drvdata(pdev, NULL);
1373 devlink_free(devlink);
1374
1375 return err;
1376 }
1377
1378 static void remove_one(struct pci_dev *pdev)
1379 {
1380 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1381 struct devlink *devlink = priv_to_devlink(dev);
1382 struct mlx5_priv *priv = &dev->priv;
1383
1384 devlink_unregister(devlink);
1385 mlx5_unregister_device(dev);
1386
1387 if (mlx5_unload_one(dev, priv, true)) {
1388 dev_err(&dev->pdev->dev, "mlx5_unload_one failed\n");
1389 mlx5_health_cleanup(dev);
1390 return;
1391 }
1392
1393 mlx5_pagealloc_cleanup(dev);
1394 mlx5_health_cleanup(dev);
1395 mlx5_pci_close(dev, priv);
1396 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1397 cleanup_srcu_struct(&priv->pfault_srcu);
1398 #endif
1399 pci_set_drvdata(pdev, NULL);
1400 devlink_free(devlink);
1401 }
1402
1403 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
1404 pci_channel_state_t state)
1405 {
1406 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1407 struct mlx5_priv *priv = &dev->priv;
1408
1409 dev_info(&pdev->dev, "%s was called\n", __func__);
1410
1411 mlx5_enter_error_state(dev);
1412 mlx5_unload_one(dev, priv, false);
1413 /* In case of kernel call drain the health wq */
1414 if (state) {
1415 mlx5_drain_health_wq(dev);
1416 mlx5_pci_disable_device(dev);
1417 }
1418
1419 return state == pci_channel_io_perm_failure ?
1420 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1421 }
1422
1423 /* wait for the device to show vital signs by waiting
1424 * for the health counter to start counting.
1425 */
1426 static int wait_vital(struct pci_dev *pdev)
1427 {
1428 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1429 struct mlx5_core_health *health = &dev->priv.health;
1430 const int niter = 100;
1431 u32 last_count = 0;
1432 u32 count;
1433 int i;
1434
1435 for (i = 0; i < niter; i++) {
1436 count = ioread32be(health->health_counter);
1437 if (count && count != 0xffffffff) {
1438 if (last_count && last_count != count) {
1439 dev_info(&pdev->dev, "Counter value 0x%x after %d iterations\n", count, i);
1440 return 0;
1441 }
1442 last_count = count;
1443 }
1444 msleep(50);
1445 }
1446
1447 return -ETIMEDOUT;
1448 }
1449
1450 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
1451 {
1452 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1453 int err;
1454
1455 dev_info(&pdev->dev, "%s was called\n", __func__);
1456
1457 err = mlx5_pci_enable_device(dev);
1458 if (err) {
1459 dev_err(&pdev->dev, "%s: mlx5_pci_enable_device failed with error code: %d\n"
1460 , __func__, err);
1461 return PCI_ERS_RESULT_DISCONNECT;
1462 }
1463
1464 pci_set_master(pdev);
1465 pci_restore_state(pdev);
1466 pci_save_state(pdev);
1467
1468 if (wait_vital(pdev)) {
1469 dev_err(&pdev->dev, "%s: wait_vital timed out\n", __func__);
1470 return PCI_ERS_RESULT_DISCONNECT;
1471 }
1472
1473 return PCI_ERS_RESULT_RECOVERED;
1474 }
1475
1476 static void mlx5_pci_resume(struct pci_dev *pdev)
1477 {
1478 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1479 struct mlx5_priv *priv = &dev->priv;
1480 int err;
1481
1482 dev_info(&pdev->dev, "%s was called\n", __func__);
1483
1484 err = mlx5_load_one(dev, priv, false);
1485 if (err)
1486 dev_err(&pdev->dev, "%s: mlx5_load_one failed with error code: %d\n"
1487 , __func__, err);
1488 else
1489 dev_info(&pdev->dev, "%s: device recovered\n", __func__);
1490 }
1491
1492 static const struct pci_error_handlers mlx5_err_handler = {
1493 .error_detected = mlx5_pci_err_detected,
1494 .slot_reset = mlx5_pci_slot_reset,
1495 .resume = mlx5_pci_resume
1496 };
1497
1498 static void shutdown(struct pci_dev *pdev)
1499 {
1500 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1501 struct mlx5_priv *priv = &dev->priv;
1502
1503 dev_info(&pdev->dev, "Shutdown was called\n");
1504 /* Notify mlx5 clients that the kernel is being shut down */
1505 set_bit(MLX5_INTERFACE_STATE_SHUTDOWN, &dev->intf_state);
1506 mlx5_unload_one(dev, priv, false);
1507 mlx5_pci_disable_device(dev);
1508 }
1509
1510 static const struct pci_device_id mlx5_core_pci_table[] = {
1511 { PCI_VDEVICE(MELLANOX, 0x1011) }, /* Connect-IB */
1512 { PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF}, /* Connect-IB VF */
1513 { PCI_VDEVICE(MELLANOX, 0x1013) }, /* ConnectX-4 */
1514 { PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF}, /* ConnectX-4 VF */
1515 { PCI_VDEVICE(MELLANOX, 0x1015) }, /* ConnectX-4LX */
1516 { PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF}, /* ConnectX-4LX VF */
1517 { PCI_VDEVICE(MELLANOX, 0x1017) }, /* ConnectX-5, PCIe 3.0 */
1518 { PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF}, /* ConnectX-5 VF */
1519 { PCI_VDEVICE(MELLANOX, 0x1019) }, /* ConnectX-5 Ex */
1520 { PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF}, /* ConnectX-5 Ex VF */
1521 { PCI_VDEVICE(MELLANOX, 0x101b) }, /* ConnectX-6 */
1522 { PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF}, /* ConnectX-6 VF */
1523 { 0, }
1524 };
1525
1526 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1527
1528 void mlx5_disable_device(struct mlx5_core_dev *dev)
1529 {
1530 mlx5_pci_err_detected(dev->pdev, 0);
1531 }
1532
1533 void mlx5_recover_device(struct mlx5_core_dev *dev)
1534 {
1535 mlx5_pci_disable_device(dev);
1536 if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED)
1537 mlx5_pci_resume(dev->pdev);
1538 }
1539
1540 static struct pci_driver mlx5_core_driver = {
1541 .name = DRIVER_NAME,
1542 .id_table = mlx5_core_pci_table,
1543 .probe = init_one,
1544 .remove = remove_one,
1545 .shutdown = shutdown,
1546 .err_handler = &mlx5_err_handler,
1547 .sriov_configure = mlx5_core_sriov_configure,
1548 };
1549
1550 static void mlx5_core_verify_params(void)
1551 {
1552 if (prof_sel >= ARRAY_SIZE(profile)) {
1553 pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
1554 prof_sel,
1555 ARRAY_SIZE(profile) - 1,
1556 MLX5_DEFAULT_PROF);
1557 prof_sel = MLX5_DEFAULT_PROF;
1558 }
1559 }
1560
1561 static int __init init(void)
1562 {
1563 int err;
1564
1565 mlx5_core_verify_params();
1566 mlx5_register_debugfs();
1567
1568 err = pci_register_driver(&mlx5_core_driver);
1569 if (err)
1570 goto err_debug;
1571
1572 #ifdef CONFIG_MLX5_CORE_EN
1573 mlx5e_init();
1574 #endif
1575
1576 return 0;
1577
1578 err_debug:
1579 mlx5_unregister_debugfs();
1580 return err;
1581 }
1582
1583 static void __exit cleanup(void)
1584 {
1585 #ifdef CONFIG_MLX5_CORE_EN
1586 mlx5e_cleanup();
1587 #endif
1588 pci_unregister_driver(&mlx5_core_driver);
1589 mlx5_unregister_debugfs();
1590 }
1591
1592 module_init(init);
1593 module_exit(cleanup);