1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
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#include "lib/mlx5.h"
60#include "fpga/core.h"
61#include "accel/ipsec.h"
62
63MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
64MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
65MODULE_LICENSE("Dual BSD/GPL");
66MODULE_VERSION(DRIVER_VERSION);
67
68unsigned int mlx5_core_debug_mask;
69module_param_named(debug_mask, mlx5_core_debug_mask, uint, 0644);
70MODULE_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
73static unsigned int prof_sel = MLX5_DEFAULT_PROF;
74module_param_named(prof_sel, prof_sel, uint, 0444);
75MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
76
77enum {
78 MLX5_ATOMIC_REQ_MODE_BE = 0x0,
79 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
80};
81
82static 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
185static 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
201static 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
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
236static 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
266static 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
282static 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
294static 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
310static void release_bar(struct pci_dev *pdev)
311{
312 pci_release_regions(pdev);
313}
314
315static int mlx5_enable_msix(struct mlx5_core_dev *dev)
316{
317 struct mlx5_priv *priv = &dev->priv;
318 struct mlx5_eq_table *table = &priv->eq_table;
319 int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq);
320 int nvec;
321 int i;
322
323 nvec = MLX5_CAP_GEN(dev, num_ports) * num_online_cpus() +
324 MLX5_EQ_VEC_COMP_BASE;
325 nvec = min_t(int, nvec, num_eqs);
326 if (nvec <= MLX5_EQ_VEC_COMP_BASE)
327 return -ENOMEM;
328
329 priv->msix_arr = kcalloc(nvec, sizeof(*priv->msix_arr), GFP_KERNEL);
330
331 priv->irq_info = kcalloc(nvec, sizeof(*priv->irq_info), GFP_KERNEL);
332 if (!priv->msix_arr || !priv->irq_info)
333 goto err_free_msix;
334
335 for (i = 0; i < nvec; i++)
336 priv->msix_arr[i].entry = i;
337
338 nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr,
339 MLX5_EQ_VEC_COMP_BASE + 1, nvec);
340 if (nvec < 0)
341 return nvec;
342
343 table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
344
345 return 0;
346
347err_free_msix:
348 kfree(priv->irq_info);
349 kfree(priv->msix_arr);
350 return -ENOMEM;
351}
352
353static void mlx5_disable_msix(struct mlx5_core_dev *dev)
354{
355 struct mlx5_priv *priv = &dev->priv;
356
357 pci_disable_msix(dev->pdev);
358 kfree(priv->irq_info);
359 kfree(priv->msix_arr);
360}
361
362struct mlx5_reg_host_endianness {
363 u8 he;
364 u8 rsvd[15];
365};
366
367#define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos))
368
369enum {
370 MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) |
371 MLX5_DEV_CAP_FLAG_DCT,
372};
373
374static u16 to_fw_pkey_sz(struct mlx5_core_dev *dev, u32 size)
375{
376 switch (size) {
377 case 128:
378 return 0;
379 case 256:
380 return 1;
381 case 512:
382 return 2;
383 case 1024:
384 return 3;
385 case 2048:
386 return 4;
387 case 4096:
388 return 5;
389 default:
390 mlx5_core_warn(dev, "invalid pkey table size %d\n", size);
391 return 0;
392 }
393}
394
395static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev,
396 enum mlx5_cap_type cap_type,
397 enum mlx5_cap_mode cap_mode)
398{
399 u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
400 int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
401 void *out, *hca_caps;
402 u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
403 int err;
404
405 memset(in, 0, sizeof(in));
406 out = kzalloc(out_sz, GFP_KERNEL);
407 if (!out)
408 return -ENOMEM;
409
410 MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
411 MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
412 err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz);
413 if (err) {
414 mlx5_core_warn(dev,
415 "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
416 cap_type, cap_mode, err);
417 goto query_ex;
418 }
419
420 hca_caps = MLX5_ADDR_OF(query_hca_cap_out, out, capability);
421
422 switch (cap_mode) {
423 case HCA_CAP_OPMOD_GET_MAX:
424 memcpy(dev->caps.hca_max[cap_type], hca_caps,
425 MLX5_UN_SZ_BYTES(hca_cap_union));
426 break;
427 case HCA_CAP_OPMOD_GET_CUR:
428 memcpy(dev->caps.hca_cur[cap_type], hca_caps,
429 MLX5_UN_SZ_BYTES(hca_cap_union));
430 break;
431 default:
432 mlx5_core_warn(dev,
433 "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
434 cap_type, cap_mode);
435 err = -EINVAL;
436 break;
437 }
438query_ex:
439 kfree(out);
440 return err;
441}
442
443int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
444{
445 int ret;
446
447 ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
448 if (ret)
449 return ret;
450 return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
451}
452
453static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz, int opmod)
454{
455 u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)] = {0};
456
457 MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
458 MLX5_SET(set_hca_cap_in, in, op_mod, opmod << 1);
459 return mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out));
460}
461
462static int handle_hca_cap_atomic(struct mlx5_core_dev *dev)
463{
464 void *set_ctx;
465 void *set_hca_cap;
466 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
467 int req_endianness;
468 int err;
469
470 if (MLX5_CAP_GEN(dev, atomic)) {
471 err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
472 if (err)
473 return err;
474 } else {
475 return 0;
476 }
477
478 req_endianness =
479 MLX5_CAP_ATOMIC(dev,
480 supported_atomic_req_8B_endianness_mode_1);
481
482 if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
483 return 0;
484
485 set_ctx = kzalloc(set_sz, GFP_KERNEL);
486 if (!set_ctx)
487 return -ENOMEM;
488
489 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
490
491
492 MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianness_mode,
493 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
494
495 err = set_caps(dev, set_ctx, set_sz, MLX5_SET_HCA_CAP_OP_MOD_ATOMIC);
496
497 kfree(set_ctx);
498 return err;
499}
500
501static int handle_hca_cap(struct mlx5_core_dev *dev)
502{
503 void *set_ctx = NULL;
504 struct mlx5_profile *prof = dev->profile;
505 int err = -ENOMEM;
506 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
507 void *set_hca_cap;
508
509 set_ctx = kzalloc(set_sz, GFP_KERNEL);
510 if (!set_ctx)
511 goto query_ex;
512
513 err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
514 if (err)
515 goto query_ex;
516
517 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
518 capability);
519 memcpy(set_hca_cap, dev->caps.hca_cur[MLX5_CAP_GENERAL],
520 MLX5_ST_SZ_BYTES(cmd_hca_cap));
521
522 mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
523 mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
524 128);
525
526 MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
527 to_fw_pkey_sz(dev, 128));
528
529
530 if (MLX5_CAP_GEN_MAX(dev, log_max_qp) < profile[prof_sel].log_max_qp) {
531 mlx5_core_warn(dev, "log_max_qp value in current profile is %d, changing it to HCA capability limit (%d)\n",
532 profile[prof_sel].log_max_qp,
533 MLX5_CAP_GEN_MAX(dev, log_max_qp));
534 profile[prof_sel].log_max_qp = MLX5_CAP_GEN_MAX(dev, log_max_qp);
535 }
536 if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
537 MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
538 prof->log_max_qp);
539
540
541 MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
542
543
544
545
546 if (MLX5_CAP_GEN_MAX(dev, uar_4k) && PAGE_SIZE > 4096)
547 MLX5_SET(cmd_hca_cap, set_hca_cap, uar_4k, 1);
548
549 MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
550
551 if (MLX5_CAP_GEN_MAX(dev, cache_line_128byte))
552 MLX5_SET(cmd_hca_cap,
553 set_hca_cap,
554 cache_line_128byte,
555 cache_line_size() == 128 ? 1 : 0);
556
557 err = set_caps(dev, set_ctx, set_sz,
558 MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
559
560query_ex:
561 kfree(set_ctx);
562 return err;
563}
564
565static int set_hca_ctrl(struct mlx5_core_dev *dev)
566{
567 struct mlx5_reg_host_endianness he_in;
568 struct mlx5_reg_host_endianness he_out;
569 int err;
570
571 if (!mlx5_core_is_pf(dev))
572 return 0;
573
574 memset(&he_in, 0, sizeof(he_in));
575 he_in.he = MLX5_SET_HOST_ENDIANNESS;
576 err = mlx5_core_access_reg(dev, &he_in, sizeof(he_in),
577 &he_out, sizeof(he_out),
578 MLX5_REG_HOST_ENDIANNESS, 0, 1);
579 return err;
580}
581
582int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
583{
584 u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0};
585 u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0};
586
587 MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
588 MLX5_SET(enable_hca_in, in, function_id, func_id);
589 return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
590}
591
592int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
593{
594 u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0};
595 u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0};
596
597 MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
598 MLX5_SET(disable_hca_in, in, function_id, func_id);
599 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
600}
601
602u64 mlx5_read_internal_timer(struct mlx5_core_dev *dev)
603{
604 u32 timer_h, timer_h1, timer_l;
605
606 timer_h = ioread32be(&dev->iseg->internal_timer_h);
607 timer_l = ioread32be(&dev->iseg->internal_timer_l);
608 timer_h1 = ioread32be(&dev->iseg->internal_timer_h);
609 if (timer_h != timer_h1)
610 timer_l = ioread32be(&dev->iseg->internal_timer_l);
611
612 return (u64)timer_l | (u64)timer_h1 << 32;
613}
614
615static int mlx5_irq_set_affinity_hint(struct mlx5_core_dev *mdev, int i)
616{
617 struct mlx5_priv *priv = &mdev->priv;
618 struct msix_entry *msix = priv->msix_arr;
619 int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
620
621 if (!zalloc_cpumask_var(&priv->irq_info[i].mask, GFP_KERNEL)) {
622 mlx5_core_warn(mdev, "zalloc_cpumask_var failed");
623 return -ENOMEM;
624 }
625
626 cpumask_set_cpu(cpumask_local_spread(i, priv->numa_node),
627 priv->irq_info[i].mask);
628
629 if (IS_ENABLED(CONFIG_SMP) &&
630 irq_set_affinity_hint(irq, priv->irq_info[i].mask))
631 mlx5_core_warn(mdev, "irq_set_affinity_hint failed, irq 0x%.4x", irq);
632
633 return 0;
634}
635
636static void mlx5_irq_clear_affinity_hint(struct mlx5_core_dev *mdev, int i)
637{
638 struct mlx5_priv *priv = &mdev->priv;
639 struct msix_entry *msix = priv->msix_arr;
640 int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
641
642 irq_set_affinity_hint(irq, NULL);
643 free_cpumask_var(priv->irq_info[i].mask);
644}
645
646static int mlx5_irq_set_affinity_hints(struct mlx5_core_dev *mdev)
647{
648 int err;
649 int i;
650
651 for (i = 0; i < mdev->priv.eq_table.num_comp_vectors; i++) {
652 err = mlx5_irq_set_affinity_hint(mdev, i);
653 if (err)
654 goto err_out;
655 }
656
657 return 0;
658
659err_out:
660 for (i--; i >= 0; i--)
661 mlx5_irq_clear_affinity_hint(mdev, i);
662
663 return err;
664}
665
666static void mlx5_irq_clear_affinity_hints(struct mlx5_core_dev *mdev)
667{
668 int i;
669
670 for (i = 0; i < mdev->priv.eq_table.num_comp_vectors; i++)
671 mlx5_irq_clear_affinity_hint(mdev, i);
672}
673
674int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn,
675 unsigned int *irqn)
676{
677 struct mlx5_eq_table *table = &dev->priv.eq_table;
678 struct mlx5_eq *eq, *n;
679 int err = -ENOENT;
680
681 spin_lock(&table->lock);
682 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
683 if (eq->index == vector) {
684 *eqn = eq->eqn;
685 *irqn = eq->irqn;
686 err = 0;
687 break;
688 }
689 }
690 spin_unlock(&table->lock);
691
692 return err;
693}
694EXPORT_SYMBOL(mlx5_vector2eqn);
695
696struct mlx5_eq *mlx5_eqn2eq(struct mlx5_core_dev *dev, int eqn)
697{
698 struct mlx5_eq_table *table = &dev->priv.eq_table;
699 struct mlx5_eq *eq;
700
701 spin_lock(&table->lock);
702 list_for_each_entry(eq, &table->comp_eqs_list, list)
703 if (eq->eqn == eqn) {
704 spin_unlock(&table->lock);
705 return eq;
706 }
707
708 spin_unlock(&table->lock);
709
710 return ERR_PTR(-ENOENT);
711}
712
713static void free_comp_eqs(struct mlx5_core_dev *dev)
714{
715 struct mlx5_eq_table *table = &dev->priv.eq_table;
716 struct mlx5_eq *eq, *n;
717
718#ifdef CONFIG_RFS_ACCEL
719 if (dev->rmap) {
720 free_irq_cpu_rmap(dev->rmap);
721 dev->rmap = NULL;
722 }
723#endif
724 spin_lock(&table->lock);
725 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
726 list_del(&eq->list);
727 spin_unlock(&table->lock);
728 if (mlx5_destroy_unmap_eq(dev, eq))
729 mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n",
730 eq->eqn);
731 kfree(eq);
732 spin_lock(&table->lock);
733 }
734 spin_unlock(&table->lock);
735}
736
737static int alloc_comp_eqs(struct mlx5_core_dev *dev)
738{
739 struct mlx5_eq_table *table = &dev->priv.eq_table;
740 char name[MLX5_MAX_IRQ_NAME];
741 struct mlx5_eq *eq;
742 int ncomp_vec;
743 int nent;
744 int err;
745 int i;
746
747 INIT_LIST_HEAD(&table->comp_eqs_list);
748 ncomp_vec = table->num_comp_vectors;
749 nent = MLX5_COMP_EQ_SIZE;
750#ifdef CONFIG_RFS_ACCEL
751 dev->rmap = alloc_irq_cpu_rmap(ncomp_vec);
752 if (!dev->rmap)
753 return -ENOMEM;
754#endif
755 for (i = 0; i < ncomp_vec; i++) {
756 eq = kzalloc(sizeof(*eq), GFP_KERNEL);
757 if (!eq) {
758 err = -ENOMEM;
759 goto clean;
760 }
761
762#ifdef CONFIG_RFS_ACCEL
763 irq_cpu_rmap_add(dev->rmap,
764 dev->priv.msix_arr[i + MLX5_EQ_VEC_COMP_BASE].vector);
765#endif
766 snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i);
767 err = mlx5_create_map_eq(dev, eq,
768 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
769 name, MLX5_EQ_TYPE_COMP);
770 if (err) {
771 kfree(eq);
772 goto clean;
773 }
774 mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
775 eq->index = i;
776 spin_lock(&table->lock);
777 list_add_tail(&eq->list, &table->comp_eqs_list);
778 spin_unlock(&table->lock);
779 }
780
781 return 0;
782
783clean:
784 free_comp_eqs(dev);
785 return err;
786}
787
788static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
789{
790 u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0};
791 u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0};
792 u32 sup_issi;
793 int err;
794
795 MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
796 err = mlx5_cmd_exec(dev, query_in, sizeof(query_in),
797 query_out, sizeof(query_out));
798 if (err) {
799 u32 syndrome;
800 u8 status;
801
802 mlx5_cmd_mbox_status(query_out, &status, &syndrome);
803 if (!status || syndrome == MLX5_DRIVER_SYND) {
804 mlx5_core_err(dev, "Failed to query ISSI err(%d) status(%d) synd(%d)\n",
805 err, status, syndrome);
806 return err;
807 }
808
809 mlx5_core_warn(dev, "Query ISSI is not supported by FW, ISSI is 0\n");
810 dev->issi = 0;
811 return 0;
812 }
813
814 sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
815
816 if (sup_issi & (1 << 1)) {
817 u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {0};
818 u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0};
819
820 MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
821 MLX5_SET(set_issi_in, set_in, current_issi, 1);
822 err = mlx5_cmd_exec(dev, set_in, sizeof(set_in),
823 set_out, sizeof(set_out));
824 if (err) {
825 mlx5_core_err(dev, "Failed to set ISSI to 1 err(%d)\n",
826 err);
827 return err;
828 }
829
830 dev->issi = 1;
831
832 return 0;
833 } else if (sup_issi & (1 << 0) || !sup_issi) {
834 return 0;
835 }
836
837 return -EOPNOTSUPP;
838}
839
840
841static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
842{
843 struct pci_dev *pdev = dev->pdev;
844 int err = 0;
845
846 pci_set_drvdata(dev->pdev, dev);
847 strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
848 priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
849
850 mutex_init(&priv->pgdir_mutex);
851 INIT_LIST_HEAD(&priv->pgdir_list);
852 spin_lock_init(&priv->mkey_lock);
853
854 mutex_init(&priv->alloc_mutex);
855
856 priv->numa_node = dev_to_node(&dev->pdev->dev);
857
858 priv->dbg_root = debugfs_create_dir(dev_name(&pdev->dev), mlx5_debugfs_root);
859 if (!priv->dbg_root)
860 return -ENOMEM;
861
862 err = mlx5_pci_enable_device(dev);
863 if (err) {
864 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
865 goto err_dbg;
866 }
867
868 err = request_bar(pdev);
869 if (err) {
870 dev_err(&pdev->dev, "error requesting BARs, aborting\n");
871 goto err_disable;
872 }
873
874 pci_set_master(pdev);
875
876 err = set_dma_caps(pdev);
877 if (err) {
878 dev_err(&pdev->dev, "Failed setting DMA capabilities mask, aborting\n");
879 goto err_clr_master;
880 }
881
882 dev->iseg_base = pci_resource_start(dev->pdev, 0);
883 dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
884 if (!dev->iseg) {
885 err = -ENOMEM;
886 dev_err(&pdev->dev, "Failed mapping initialization segment, aborting\n");
887 goto err_clr_master;
888 }
889
890 return 0;
891
892err_clr_master:
893 pci_clear_master(dev->pdev);
894 release_bar(dev->pdev);
895err_disable:
896 mlx5_pci_disable_device(dev);
897
898err_dbg:
899 debugfs_remove(priv->dbg_root);
900 return err;
901}
902
903static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
904{
905 iounmap(dev->iseg);
906 pci_clear_master(dev->pdev);
907 release_bar(dev->pdev);
908 mlx5_pci_disable_device(dev);
909 debugfs_remove(priv->dbg_root);
910}
911
912static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
913{
914 struct pci_dev *pdev = dev->pdev;
915 int err;
916
917 err = mlx5_query_board_id(dev);
918 if (err) {
919 dev_err(&pdev->dev, "query board id failed\n");
920 goto out;
921 }
922
923 err = mlx5_eq_init(dev);
924 if (err) {
925 dev_err(&pdev->dev, "failed to initialize eq\n");
926 goto out;
927 }
928
929 err = mlx5_init_cq_table(dev);
930 if (err) {
931 dev_err(&pdev->dev, "failed to initialize cq table\n");
932 goto err_eq_cleanup;
933 }
934
935 mlx5_init_qp_table(dev);
936
937 mlx5_init_srq_table(dev);
938
939 mlx5_init_mkey_table(dev);
940
941 mlx5_init_reserved_gids(dev);
942
943 err = mlx5_init_rl_table(dev);
944 if (err) {
945 dev_err(&pdev->dev, "Failed to init rate limiting\n");
946 goto err_tables_cleanup;
947 }
948
949#ifdef CONFIG_MLX5_CORE_EN
950 err = mlx5_eswitch_init(dev);
951 if (err) {
952 dev_err(&pdev->dev, "Failed to init eswitch %d\n", err);
953 goto err_rl_cleanup;
954 }
955#endif
956
957 err = mlx5_sriov_init(dev);
958 if (err) {
959 dev_err(&pdev->dev, "Failed to init sriov %d\n", err);
960 goto err_eswitch_cleanup;
961 }
962
963 err = mlx5_fpga_init(dev);
964 if (err) {
965 dev_err(&pdev->dev, "Failed to init fpga device %d\n", err);
966 goto err_sriov_cleanup;
967 }
968
969 return 0;
970
971err_sriov_cleanup:
972 mlx5_sriov_cleanup(dev);
973err_eswitch_cleanup:
974#ifdef CONFIG_MLX5_CORE_EN
975 mlx5_eswitch_cleanup(dev->priv.eswitch);
976
977err_rl_cleanup:
978#endif
979 mlx5_cleanup_rl_table(dev);
980
981err_tables_cleanup:
982 mlx5_cleanup_mkey_table(dev);
983 mlx5_cleanup_srq_table(dev);
984 mlx5_cleanup_qp_table(dev);
985 mlx5_cleanup_cq_table(dev);
986
987err_eq_cleanup:
988 mlx5_eq_cleanup(dev);
989
990out:
991 return err;
992}
993
994static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
995{
996 mlx5_fpga_cleanup(dev);
997 mlx5_sriov_cleanup(dev);
998#ifdef CONFIG_MLX5_CORE_EN
999 mlx5_eswitch_cleanup(dev->priv.eswitch);
1000#endif
1001 mlx5_cleanup_rl_table(dev);
1002 mlx5_cleanup_reserved_gids(dev);
1003 mlx5_cleanup_mkey_table(dev);
1004 mlx5_cleanup_srq_table(dev);
1005 mlx5_cleanup_qp_table(dev);
1006 mlx5_cleanup_cq_table(dev);
1007 mlx5_eq_cleanup(dev);
1008}
1009
1010static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
1011 bool boot)
1012{
1013 struct pci_dev *pdev = dev->pdev;
1014 int err;
1015
1016 mutex_lock(&dev->intf_state_mutex);
1017 if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1018 dev_warn(&dev->pdev->dev, "%s: interface is up, NOP\n",
1019 __func__);
1020 goto out;
1021 }
1022
1023 dev_info(&pdev->dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1024 fw_rev_min(dev), fw_rev_sub(dev));
1025
1026
1027
1028
1029 dev->state = MLX5_DEVICE_STATE_UP;
1030
1031
1032
1033 err = wait_fw_init(dev, FW_PRE_INIT_TIMEOUT_MILI);
1034 if (err) {
1035 dev_err(&dev->pdev->dev, "Firmware over %d MS in pre-initializing state, aborting\n",
1036 FW_PRE_INIT_TIMEOUT_MILI);
1037 goto out_err;
1038 }
1039
1040 err = mlx5_cmd_init(dev);
1041 if (err) {
1042 dev_err(&pdev->dev, "Failed initializing command interface, aborting\n");
1043 goto out_err;
1044 }
1045
1046 err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI);
1047 if (err) {
1048 dev_err(&dev->pdev->dev, "Firmware over %d MS in initializing state, aborting\n",
1049 FW_INIT_TIMEOUT_MILI);
1050 goto err_cmd_cleanup;
1051 }
1052
1053 err = mlx5_core_enable_hca(dev, 0);
1054 if (err) {
1055 dev_err(&pdev->dev, "enable hca failed\n");
1056 goto err_cmd_cleanup;
1057 }
1058
1059 err = mlx5_core_set_issi(dev);
1060 if (err) {
1061 dev_err(&pdev->dev, "failed to set issi\n");
1062 goto err_disable_hca;
1063 }
1064
1065 err = mlx5_satisfy_startup_pages(dev, 1);
1066 if (err) {
1067 dev_err(&pdev->dev, "failed to allocate boot pages\n");
1068 goto err_disable_hca;
1069 }
1070
1071 err = set_hca_ctrl(dev);
1072 if (err) {
1073 dev_err(&pdev->dev, "set_hca_ctrl failed\n");
1074 goto reclaim_boot_pages;
1075 }
1076
1077 err = handle_hca_cap(dev);
1078 if (err) {
1079 dev_err(&pdev->dev, "handle_hca_cap failed\n");
1080 goto reclaim_boot_pages;
1081 }
1082
1083 err = handle_hca_cap_atomic(dev);
1084 if (err) {
1085 dev_err(&pdev->dev, "handle_hca_cap_atomic failed\n");
1086 goto reclaim_boot_pages;
1087 }
1088
1089 err = mlx5_satisfy_startup_pages(dev, 0);
1090 if (err) {
1091 dev_err(&pdev->dev, "failed to allocate init pages\n");
1092 goto reclaim_boot_pages;
1093 }
1094
1095 err = mlx5_pagealloc_start(dev);
1096 if (err) {
1097 dev_err(&pdev->dev, "mlx5_pagealloc_start failed\n");
1098 goto reclaim_boot_pages;
1099 }
1100
1101 err = mlx5_cmd_init_hca(dev);
1102 if (err) {
1103 dev_err(&pdev->dev, "init hca failed\n");
1104 goto err_pagealloc_stop;
1105 }
1106
1107 mlx5_set_driver_version(dev);
1108
1109 mlx5_start_health_poll(dev);
1110
1111 err = mlx5_query_hca_caps(dev);
1112 if (err) {
1113 dev_err(&pdev->dev, "query hca failed\n");
1114 goto err_stop_poll;
1115 }
1116
1117 if (boot && mlx5_init_once(dev, priv)) {
1118 dev_err(&pdev->dev, "sw objs init failed\n");
1119 goto err_stop_poll;
1120 }
1121
1122 err = mlx5_enable_msix(dev);
1123 if (err) {
1124 dev_err(&pdev->dev, "enable msix failed\n");
1125 goto err_cleanup_once;
1126 }
1127
1128 dev->priv.uar = mlx5_get_uars_page(dev);
1129 if (!dev->priv.uar) {
1130 dev_err(&pdev->dev, "Failed allocating uar, aborting\n");
1131 goto err_disable_msix;
1132 }
1133
1134 err = mlx5_start_eqs(dev);
1135 if (err) {
1136 dev_err(&pdev->dev, "Failed to start pages and async EQs\n");
1137 goto err_put_uars;
1138 }
1139
1140 err = alloc_comp_eqs(dev);
1141 if (err) {
1142 dev_err(&pdev->dev, "Failed to alloc completion EQs\n");
1143 goto err_stop_eqs;
1144 }
1145
1146 err = mlx5_irq_set_affinity_hints(dev);
1147 if (err) {
1148 dev_err(&pdev->dev, "Failed to alloc affinity hint cpumask\n");
1149 goto err_affinity_hints;
1150 }
1151
1152 err = mlx5_init_fs(dev);
1153 if (err) {
1154 dev_err(&pdev->dev, "Failed to init flow steering\n");
1155 goto err_fs;
1156 }
1157
1158#ifdef CONFIG_MLX5_CORE_EN
1159 mlx5_eswitch_attach(dev->priv.eswitch);
1160#endif
1161
1162 err = mlx5_sriov_attach(dev);
1163 if (err) {
1164 dev_err(&pdev->dev, "sriov init failed %d\n", err);
1165 goto err_sriov;
1166 }
1167
1168 err = mlx5_fpga_device_start(dev);
1169 if (err) {
1170 dev_err(&pdev->dev, "fpga device start failed %d\n", err);
1171 goto err_fpga_start;
1172 }
1173 err = mlx5_accel_ipsec_init(dev);
1174 if (err) {
1175 dev_err(&pdev->dev, "IPSec device start failed %d\n", err);
1176 goto err_ipsec_start;
1177 }
1178
1179 if (mlx5_device_registered(dev)) {
1180 mlx5_attach_device(dev);
1181 } else {
1182 err = mlx5_register_device(dev);
1183 if (err) {
1184 dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err);
1185 goto err_reg_dev;
1186 }
1187 }
1188
1189 set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1190out:
1191 mutex_unlock(&dev->intf_state_mutex);
1192
1193 return 0;
1194
1195err_reg_dev:
1196 mlx5_accel_ipsec_cleanup(dev);
1197err_ipsec_start:
1198 mlx5_fpga_device_stop(dev);
1199
1200err_fpga_start:
1201 mlx5_sriov_detach(dev);
1202
1203err_sriov:
1204#ifdef CONFIG_MLX5_CORE_EN
1205 mlx5_eswitch_detach(dev->priv.eswitch);
1206#endif
1207 mlx5_cleanup_fs(dev);
1208
1209err_fs:
1210 mlx5_irq_clear_affinity_hints(dev);
1211
1212err_affinity_hints:
1213 free_comp_eqs(dev);
1214
1215err_stop_eqs:
1216 mlx5_stop_eqs(dev);
1217
1218err_put_uars:
1219 mlx5_put_uars_page(dev, priv->uar);
1220
1221err_disable_msix:
1222 mlx5_disable_msix(dev);
1223
1224err_cleanup_once:
1225 if (boot)
1226 mlx5_cleanup_once(dev);
1227
1228err_stop_poll:
1229 mlx5_stop_health_poll(dev);
1230 if (mlx5_cmd_teardown_hca(dev)) {
1231 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1232 goto out_err;
1233 }
1234
1235err_pagealloc_stop:
1236 mlx5_pagealloc_stop(dev);
1237
1238reclaim_boot_pages:
1239 mlx5_reclaim_startup_pages(dev);
1240
1241err_disable_hca:
1242 mlx5_core_disable_hca(dev, 0);
1243
1244err_cmd_cleanup:
1245 mlx5_cmd_cleanup(dev);
1246
1247out_err:
1248 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1249 mutex_unlock(&dev->intf_state_mutex);
1250
1251 return err;
1252}
1253
1254static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
1255 bool cleanup)
1256{
1257 int err = 0;
1258
1259 if (cleanup)
1260 mlx5_drain_health_recovery(dev);
1261
1262 mutex_lock(&dev->intf_state_mutex);
1263 if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1264 dev_warn(&dev->pdev->dev, "%s: interface is down, NOP\n",
1265 __func__);
1266 if (cleanup)
1267 mlx5_cleanup_once(dev);
1268 goto out;
1269 }
1270
1271 clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1272
1273 if (mlx5_device_registered(dev))
1274 mlx5_detach_device(dev);
1275
1276 mlx5_accel_ipsec_cleanup(dev);
1277 mlx5_fpga_device_stop(dev);
1278
1279 mlx5_sriov_detach(dev);
1280#ifdef CONFIG_MLX5_CORE_EN
1281 mlx5_eswitch_detach(dev->priv.eswitch);
1282#endif
1283 mlx5_cleanup_fs(dev);
1284 mlx5_irq_clear_affinity_hints(dev);
1285 free_comp_eqs(dev);
1286 mlx5_stop_eqs(dev);
1287 mlx5_put_uars_page(dev, priv->uar);
1288 mlx5_disable_msix(dev);
1289 if (cleanup)
1290 mlx5_cleanup_once(dev);
1291 mlx5_stop_health_poll(dev);
1292 err = mlx5_cmd_teardown_hca(dev);
1293 if (err) {
1294 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1295 goto out;
1296 }
1297 mlx5_pagealloc_stop(dev);
1298 mlx5_reclaim_startup_pages(dev);
1299 mlx5_core_disable_hca(dev, 0);
1300 mlx5_cmd_cleanup(dev);
1301
1302out:
1303 mutex_unlock(&dev->intf_state_mutex);
1304 return err;
1305}
1306
1307struct mlx5_core_event_handler {
1308 void (*event)(struct mlx5_core_dev *dev,
1309 enum mlx5_dev_event event,
1310 void *data);
1311};
1312
1313static const struct devlink_ops mlx5_devlink_ops = {
1314#ifdef CONFIG_MLX5_CORE_EN
1315 .eswitch_mode_set = mlx5_devlink_eswitch_mode_set,
1316 .eswitch_mode_get = mlx5_devlink_eswitch_mode_get,
1317 .eswitch_inline_mode_set = mlx5_devlink_eswitch_inline_mode_set,
1318 .eswitch_inline_mode_get = mlx5_devlink_eswitch_inline_mode_get,
1319 .eswitch_encap_mode_set = mlx5_devlink_eswitch_encap_mode_set,
1320 .eswitch_encap_mode_get = mlx5_devlink_eswitch_encap_mode_get,
1321#endif
1322};
1323
1324#define MLX5_IB_MOD "mlx5_ib"
1325static int init_one(struct pci_dev *pdev,
1326 const struct pci_device_id *id)
1327{
1328 struct mlx5_core_dev *dev;
1329 struct devlink *devlink;
1330 struct mlx5_priv *priv;
1331 int err;
1332
1333 devlink = devlink_alloc(&mlx5_devlink_ops, sizeof(*dev));
1334 if (!devlink) {
1335 dev_err(&pdev->dev, "kzalloc failed\n");
1336 return -ENOMEM;
1337 }
1338
1339 dev = devlink_priv(devlink);
1340 priv = &dev->priv;
1341 priv->pci_dev_data = id->driver_data;
1342
1343 pci_set_drvdata(pdev, dev);
1344
1345 dev->pdev = pdev;
1346 dev->event = mlx5_core_event;
1347 dev->profile = &profile[prof_sel];
1348
1349 INIT_LIST_HEAD(&priv->ctx_list);
1350 spin_lock_init(&priv->ctx_lock);
1351 mutex_init(&dev->pci_status_mutex);
1352 mutex_init(&dev->intf_state_mutex);
1353
1354#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1355 err = init_srcu_struct(&priv->pfault_srcu);
1356 if (err) {
1357 dev_err(&pdev->dev, "init_srcu_struct failed with error code %d\n",
1358 err);
1359 goto clean_dev;
1360 }
1361#endif
1362 mutex_init(&priv->bfregs.reg_head.lock);
1363 mutex_init(&priv->bfregs.wc_head.lock);
1364 INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1365 INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1366
1367 err = mlx5_pci_init(dev, priv);
1368 if (err) {
1369 dev_err(&pdev->dev, "mlx5_pci_init failed with error code %d\n", err);
1370 goto clean_srcu;
1371 }
1372
1373 err = mlx5_health_init(dev);
1374 if (err) {
1375 dev_err(&pdev->dev, "mlx5_health_init failed with error code %d\n", err);
1376 goto close_pci;
1377 }
1378
1379 mlx5_pagealloc_init(dev);
1380
1381 err = mlx5_load_one(dev, priv, true);
1382 if (err) {
1383 dev_err(&pdev->dev, "mlx5_load_one failed with error code %d\n", err);
1384 goto clean_health;
1385 }
1386
1387 request_module_nowait(MLX5_IB_MOD);
1388
1389 err = devlink_register(devlink, &pdev->dev);
1390 if (err)
1391 goto clean_load;
1392
1393 pci_save_state(pdev);
1394 return 0;
1395
1396clean_load:
1397 mlx5_unload_one(dev, priv, true);
1398clean_health:
1399 mlx5_pagealloc_cleanup(dev);
1400 mlx5_health_cleanup(dev);
1401close_pci:
1402 mlx5_pci_close(dev, priv);
1403clean_srcu:
1404#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1405 cleanup_srcu_struct(&priv->pfault_srcu);
1406clean_dev:
1407#endif
1408 pci_set_drvdata(pdev, NULL);
1409 devlink_free(devlink);
1410
1411 return err;
1412}
1413
1414static void remove_one(struct pci_dev *pdev)
1415{
1416 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1417 struct devlink *devlink = priv_to_devlink(dev);
1418 struct mlx5_priv *priv = &dev->priv;
1419
1420 devlink_unregister(devlink);
1421 mlx5_unregister_device(dev);
1422
1423 if (mlx5_unload_one(dev, priv, true)) {
1424 dev_err(&dev->pdev->dev, "mlx5_unload_one failed\n");
1425 mlx5_health_cleanup(dev);
1426 return;
1427 }
1428
1429 mlx5_pagealloc_cleanup(dev);
1430 mlx5_health_cleanup(dev);
1431 mlx5_pci_close(dev, priv);
1432#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1433 cleanup_srcu_struct(&priv->pfault_srcu);
1434#endif
1435 pci_set_drvdata(pdev, NULL);
1436 devlink_free(devlink);
1437}
1438
1439static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
1440 pci_channel_state_t state)
1441{
1442 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1443 struct mlx5_priv *priv = &dev->priv;
1444
1445 dev_info(&pdev->dev, "%s was called\n", __func__);
1446
1447 mlx5_enter_error_state(dev, false);
1448 mlx5_unload_one(dev, priv, false);
1449
1450 if (state) {
1451 mlx5_drain_health_wq(dev);
1452 mlx5_pci_disable_device(dev);
1453 }
1454
1455 return state == pci_channel_io_perm_failure ?
1456 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1457}
1458
1459
1460
1461
1462static int wait_vital(struct pci_dev *pdev)
1463{
1464 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1465 struct mlx5_core_health *health = &dev->priv.health;
1466 const int niter = 100;
1467 u32 last_count = 0;
1468 u32 count;
1469 int i;
1470
1471 for (i = 0; i < niter; i++) {
1472 count = ioread32be(health->health_counter);
1473 if (count && count != 0xffffffff) {
1474 if (last_count && last_count != count) {
1475 dev_info(&pdev->dev, "Counter value 0x%x after %d iterations\n", count, i);
1476 return 0;
1477 }
1478 last_count = count;
1479 }
1480 msleep(50);
1481 }
1482
1483 return -ETIMEDOUT;
1484}
1485
1486static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
1487{
1488 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1489 int err;
1490
1491 dev_info(&pdev->dev, "%s was called\n", __func__);
1492
1493 err = mlx5_pci_enable_device(dev);
1494 if (err) {
1495 dev_err(&pdev->dev, "%s: mlx5_pci_enable_device failed with error code: %d\n"
1496 , __func__, err);
1497 return PCI_ERS_RESULT_DISCONNECT;
1498 }
1499
1500 pci_set_master(pdev);
1501 pci_restore_state(pdev);
1502 pci_save_state(pdev);
1503
1504 if (wait_vital(pdev)) {
1505 dev_err(&pdev->dev, "%s: wait_vital timed out\n", __func__);
1506 return PCI_ERS_RESULT_DISCONNECT;
1507 }
1508
1509 return PCI_ERS_RESULT_RECOVERED;
1510}
1511
1512static void mlx5_pci_resume(struct pci_dev *pdev)
1513{
1514 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1515 struct mlx5_priv *priv = &dev->priv;
1516 int err;
1517
1518 dev_info(&pdev->dev, "%s was called\n", __func__);
1519
1520 err = mlx5_load_one(dev, priv, false);
1521 if (err)
1522 dev_err(&pdev->dev, "%s: mlx5_load_one failed with error code: %d\n"
1523 , __func__, err);
1524 else
1525 dev_info(&pdev->dev, "%s: device recovered\n", __func__);
1526}
1527
1528static const struct pci_error_handlers mlx5_err_handler = {
1529 .error_detected = mlx5_pci_err_detected,
1530 .slot_reset = mlx5_pci_slot_reset,
1531 .resume = mlx5_pci_resume
1532};
1533
1534static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
1535{
1536 int ret;
1537
1538 if (!MLX5_CAP_GEN(dev, force_teardown)) {
1539 mlx5_core_dbg(dev, "force teardown is not supported in the firmware\n");
1540 return -EOPNOTSUPP;
1541 }
1542
1543 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1544 mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
1545 return -EAGAIN;
1546 }
1547
1548 ret = mlx5_cmd_force_teardown_hca(dev);
1549 if (ret) {
1550 mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", ret);
1551 return ret;
1552 }
1553
1554 mlx5_enter_error_state(dev, true);
1555
1556 return 0;
1557}
1558
1559static void shutdown(struct pci_dev *pdev)
1560{
1561 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1562 struct mlx5_priv *priv = &dev->priv;
1563 int err;
1564
1565 dev_info(&pdev->dev, "Shutdown was called\n");
1566 err = mlx5_try_fast_unload(dev);
1567 if (err)
1568 mlx5_unload_one(dev, priv, false);
1569 mlx5_pci_disable_device(dev);
1570}
1571
1572static const struct pci_device_id mlx5_core_pci_table[] = {
1573 { PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTIB) },
1574 { PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF},
1575 { PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4) },
1576 { PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF},
1577 { PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4_LX) },
1578 { PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF},
1579 { PCI_VDEVICE(MELLANOX, 0x1017) },
1580 { PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF},
1581 { PCI_VDEVICE(MELLANOX, 0x1019) },
1582 { PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF},
1583 { PCI_VDEVICE(MELLANOX, 0x101b) },
1584 { PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF},
1585 { PCI_VDEVICE(MELLANOX, 0xa2d2) },
1586 { PCI_VDEVICE(MELLANOX, 0xa2d3), MLX5_PCI_DEV_IS_VF},
1587 { 0, }
1588};
1589
1590MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1591
1592void mlx5_disable_device(struct mlx5_core_dev *dev)
1593{
1594 mlx5_pci_err_detected(dev->pdev, 0);
1595}
1596
1597void mlx5_recover_device(struct mlx5_core_dev *dev)
1598{
1599 mlx5_pci_disable_device(dev);
1600 if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED)
1601 mlx5_pci_resume(dev->pdev);
1602}
1603
1604static struct pci_driver mlx5_core_driver = {
1605 .name = DRIVER_NAME,
1606 .id_table = mlx5_core_pci_table,
1607 .probe = init_one,
1608 .remove = remove_one,
1609 .shutdown = shutdown,
1610 .err_handler = &mlx5_err_handler,
1611 .sriov_configure = mlx5_core_sriov_configure,
1612};
1613
1614static void mlx5_core_verify_params(void)
1615{
1616 if (prof_sel >= ARRAY_SIZE(profile)) {
1617 pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
1618 prof_sel,
1619 ARRAY_SIZE(profile) - 1,
1620 MLX5_DEFAULT_PROF);
1621 prof_sel = MLX5_DEFAULT_PROF;
1622 }
1623}
1624
1625static int __init init(void)
1626{
1627 int err;
1628
1629 mlx5_core_verify_params();
1630 mlx5_register_debugfs();
1631
1632 err = pci_register_driver(&mlx5_core_driver);
1633 if (err)
1634 goto err_debug;
1635
1636#ifdef CONFIG_MLX5_CORE_EN
1637 mlx5e_init();
1638#endif
1639
1640 return 0;
1641
1642err_debug:
1643 mlx5_unregister_debugfs();
1644 return err;
1645}
1646
1647static void __exit cleanup(void)
1648{
1649#ifdef CONFIG_MLX5_CORE_EN
1650 mlx5e_cleanup();
1651#endif
1652 pci_unregister_driver(&mlx5_core_driver);
1653 mlx5_unregister_debugfs();
1654}
1655
1656module_init(init);
1657module_exit(cleanup);
1658
