1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* Copyright (c) 2020, Mellanox Technologies inc. All rights reserved. */
3
4 #include <devlink.h>
5
6 #include "fw_reset.h"
7 #include "diag/fw_tracer.h"
8 #include "lib/tout.h"
9
10 enum {
11 MLX5_FW_RESET_FLAGS_RESET_REQUESTED,
12 MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST,
13 MLX5_FW_RESET_FLAGS_PENDING_COMP,
14 MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS,
15 MLX5_FW_RESET_FLAGS_RELOAD_REQUIRED
16 };
17
18 struct mlx5_fw_reset {
19 struct mlx5_core_dev *dev;
20 struct mlx5_nb nb;
21 struct workqueue_struct *wq;
22 struct work_struct fw_live_patch_work;
23 struct work_struct reset_request_work;
24 struct work_struct reset_unload_work;
25 struct work_struct reset_reload_work;
26 struct work_struct reset_now_work;
27 struct work_struct reset_abort_work;
28 unsigned long reset_flags;
29 struct timer_list timer;
30 struct completion done;
31 int ret;
32 };
33
34 enum {
35 MLX5_FW_RST_STATE_IDLE = 0,
36 MLX5_FW_RST_STATE_TOGGLE_REQ = 4,
37 };
38
39 enum {
40 MLX5_RST_STATE_BIT_NUM = 12,
41 MLX5_RST_ACK_BIT_NUM = 22,
42 };
43
mlx5_get_fw_rst_state(struct mlx5_core_dev * dev)44 static u8 mlx5_get_fw_rst_state(struct mlx5_core_dev *dev)
45 {
46 return (ioread32be(&dev->iseg->initializing) >> MLX5_RST_STATE_BIT_NUM) & 0xF;
47 }
48
mlx5_set_fw_rst_ack(struct mlx5_core_dev * dev)49 static void mlx5_set_fw_rst_ack(struct mlx5_core_dev *dev)
50 {
51 iowrite32be(BIT(MLX5_RST_ACK_BIT_NUM), &dev->iseg->initializing);
52 }
53
mlx5_fw_reset_enable_remote_dev_reset_set(struct devlink * devlink,u32 id,struct devlink_param_gset_ctx * ctx)54 static int mlx5_fw_reset_enable_remote_dev_reset_set(struct devlink *devlink, u32 id,
55 struct devlink_param_gset_ctx *ctx)
56 {
57 struct mlx5_core_dev *dev = devlink_priv(devlink);
58 struct mlx5_fw_reset *fw_reset;
59
60 fw_reset = dev->priv.fw_reset;
61
62 if (ctx->val.vbool)
63 clear_bit(MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST, &fw_reset->reset_flags);
64 else
65 set_bit(MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST, &fw_reset->reset_flags);
66 return 0;
67 }
68
mlx5_fw_reset_enable_remote_dev_reset_get(struct devlink * devlink,u32 id,struct devlink_param_gset_ctx * ctx)69 static int mlx5_fw_reset_enable_remote_dev_reset_get(struct devlink *devlink, u32 id,
70 struct devlink_param_gset_ctx *ctx)
71 {
72 struct mlx5_core_dev *dev = devlink_priv(devlink);
73 struct mlx5_fw_reset *fw_reset;
74
75 fw_reset = dev->priv.fw_reset;
76
77 ctx->val.vbool = !test_bit(MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST,
78 &fw_reset->reset_flags);
79 return 0;
80 }
81
mlx5_reg_mfrl_set(struct mlx5_core_dev * dev,u8 reset_level,u8 reset_type_sel,u8 sync_resp,bool sync_start)82 static int mlx5_reg_mfrl_set(struct mlx5_core_dev *dev, u8 reset_level,
83 u8 reset_type_sel, u8 sync_resp, bool sync_start)
84 {
85 u32 out[MLX5_ST_SZ_DW(mfrl_reg)] = {};
86 u32 in[MLX5_ST_SZ_DW(mfrl_reg)] = {};
87
88 MLX5_SET(mfrl_reg, in, reset_level, reset_level);
89 MLX5_SET(mfrl_reg, in, rst_type_sel, reset_type_sel);
90 MLX5_SET(mfrl_reg, in, pci_sync_for_fw_update_resp, sync_resp);
91 MLX5_SET(mfrl_reg, in, pci_sync_for_fw_update_start, sync_start);
92
93 return mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_REG_MFRL, 0, 1);
94 }
95
mlx5_reg_mfrl_query(struct mlx5_core_dev * dev,u8 * reset_level,u8 * reset_type,u8 * reset_state)96 static int mlx5_reg_mfrl_query(struct mlx5_core_dev *dev, u8 *reset_level,
97 u8 *reset_type, u8 *reset_state)
98 {
99 u32 out[MLX5_ST_SZ_DW(mfrl_reg)] = {};
100 u32 in[MLX5_ST_SZ_DW(mfrl_reg)] = {};
101 int err;
102
103 err = mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_REG_MFRL, 0, 0);
104 if (err)
105 return err;
106
107 if (reset_level)
108 *reset_level = MLX5_GET(mfrl_reg, out, reset_level);
109 if (reset_type)
110 *reset_type = MLX5_GET(mfrl_reg, out, reset_type);
111 if (reset_state)
112 *reset_state = MLX5_GET(mfrl_reg, out, reset_state);
113
114 return 0;
115 }
116
mlx5_fw_reset_query(struct mlx5_core_dev * dev,u8 * reset_level,u8 * reset_type)117 int mlx5_fw_reset_query(struct mlx5_core_dev *dev, u8 *reset_level, u8 *reset_type)
118 {
119 return mlx5_reg_mfrl_query(dev, reset_level, reset_type, NULL);
120 }
121
mlx5_fw_reset_get_reset_state_err(struct mlx5_core_dev * dev,struct netlink_ext_ack * extack)122 static int mlx5_fw_reset_get_reset_state_err(struct mlx5_core_dev *dev,
123 struct netlink_ext_ack *extack)
124 {
125 u8 reset_state;
126
127 if (mlx5_reg_mfrl_query(dev, NULL, NULL, &reset_state))
128 goto out;
129
130 if (!reset_state)
131 return 0;
132
133 switch (reset_state) {
134 case MLX5_MFRL_REG_RESET_STATE_IN_NEGOTIATION:
135 case MLX5_MFRL_REG_RESET_STATE_RESET_IN_PROGRESS:
136 NL_SET_ERR_MSG_MOD(extack, "Sync reset still in progress");
137 return -EBUSY;
138 case MLX5_MFRL_REG_RESET_STATE_NEG_TIMEOUT:
139 NL_SET_ERR_MSG_MOD(extack, "Sync reset negotiation timeout");
140 return -ETIMEDOUT;
141 case MLX5_MFRL_REG_RESET_STATE_NACK:
142 NL_SET_ERR_MSG_MOD(extack, "One of the hosts disabled reset");
143 return -EPERM;
144 case MLX5_MFRL_REG_RESET_STATE_UNLOAD_TIMEOUT:
145 NL_SET_ERR_MSG_MOD(extack, "Sync reset unload timeout");
146 return -ETIMEDOUT;
147 }
148
149 out:
150 NL_SET_ERR_MSG_MOD(extack, "Sync reset failed");
151 return -EIO;
152 }
153
mlx5_fw_reset_set_reset_sync(struct mlx5_core_dev * dev,u8 reset_type_sel,struct netlink_ext_ack * extack)154 int mlx5_fw_reset_set_reset_sync(struct mlx5_core_dev *dev, u8 reset_type_sel,
155 struct netlink_ext_ack *extack)
156 {
157 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
158 u32 out[MLX5_ST_SZ_DW(mfrl_reg)] = {};
159 u32 in[MLX5_ST_SZ_DW(mfrl_reg)] = {};
160 int err, rst_res;
161
162 set_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags);
163
164 MLX5_SET(mfrl_reg, in, reset_level, MLX5_MFRL_REG_RESET_LEVEL3);
165 MLX5_SET(mfrl_reg, in, rst_type_sel, reset_type_sel);
166 MLX5_SET(mfrl_reg, in, pci_sync_for_fw_update_start, 1);
167 err = mlx5_access_reg(dev, in, sizeof(in), out, sizeof(out),
168 MLX5_REG_MFRL, 0, 1, false);
169 if (!err)
170 return 0;
171
172 clear_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags);
173 if (err == -EREMOTEIO && MLX5_CAP_MCAM_FEATURE(dev, reset_state)) {
174 rst_res = mlx5_fw_reset_get_reset_state_err(dev, extack);
175 return rst_res ? rst_res : err;
176 }
177
178 NL_SET_ERR_MSG_MOD(extack, "Sync reset command failed");
179 return mlx5_cmd_check(dev, err, in, out);
180 }
181
mlx5_fw_reset_verify_fw_complete(struct mlx5_core_dev * dev,struct netlink_ext_ack * extack)182 int mlx5_fw_reset_verify_fw_complete(struct mlx5_core_dev *dev,
183 struct netlink_ext_ack *extack)
184 {
185 u8 rst_state;
186 int err;
187
188 err = mlx5_fw_reset_get_reset_state_err(dev, extack);
189 if (err)
190 return err;
191
192 rst_state = mlx5_get_fw_rst_state(dev);
193 if (!rst_state)
194 return 0;
195
196 mlx5_core_err(dev, "Sync reset did not complete, state=%d\n", rst_state);
197 NL_SET_ERR_MSG_MOD(extack, "Sync reset did not complete successfully");
198 return rst_state;
199 }
200
mlx5_fw_reset_set_live_patch(struct mlx5_core_dev * dev)201 int mlx5_fw_reset_set_live_patch(struct mlx5_core_dev *dev)
202 {
203 return mlx5_reg_mfrl_set(dev, MLX5_MFRL_REG_RESET_LEVEL0, 0, 0, false);
204 }
205
mlx5_fw_reset_complete_reload(struct mlx5_core_dev * dev,bool unloaded)206 static void mlx5_fw_reset_complete_reload(struct mlx5_core_dev *dev, bool unloaded)
207 {
208 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
209
210 /* if this is the driver that initiated the fw reset, devlink completed the reload */
211 if (test_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags)) {
212 complete(&fw_reset->done);
213 } else {
214 if (!unloaded)
215 mlx5_unload_one(dev, false);
216 if (mlx5_health_wait_pci_up(dev))
217 mlx5_core_err(dev, "reset reload flow aborted, PCI reads still not working\n");
218 else
219 mlx5_load_one(dev, true);
220 devlink_remote_reload_actions_performed(priv_to_devlink(dev), 0,
221 BIT(DEVLINK_RELOAD_ACTION_DRIVER_REINIT) |
222 BIT(DEVLINK_RELOAD_ACTION_FW_ACTIVATE));
223 }
224 }
225
mlx5_stop_sync_reset_poll(struct mlx5_core_dev * dev)226 static void mlx5_stop_sync_reset_poll(struct mlx5_core_dev *dev)
227 {
228 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
229
230 del_timer_sync(&fw_reset->timer);
231 }
232
mlx5_sync_reset_clear_reset_requested(struct mlx5_core_dev * dev,bool poll_health)233 static int mlx5_sync_reset_clear_reset_requested(struct mlx5_core_dev *dev, bool poll_health)
234 {
235 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
236
237 if (!test_and_clear_bit(MLX5_FW_RESET_FLAGS_RESET_REQUESTED, &fw_reset->reset_flags)) {
238 mlx5_core_warn(dev, "Reset request was already cleared\n");
239 return -EALREADY;
240 }
241
242 mlx5_stop_sync_reset_poll(dev);
243 if (poll_health)
244 mlx5_start_health_poll(dev);
245 return 0;
246 }
247
mlx5_sync_reset_reload_work(struct work_struct * work)248 static void mlx5_sync_reset_reload_work(struct work_struct *work)
249 {
250 struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
251 reset_reload_work);
252 struct mlx5_core_dev *dev = fw_reset->dev;
253
254 mlx5_sync_reset_clear_reset_requested(dev, false);
255 mlx5_enter_error_state(dev, true);
256 mlx5_fw_reset_complete_reload(dev, false);
257 }
258
259 #define MLX5_RESET_POLL_INTERVAL (HZ / 10)
poll_sync_reset(struct timer_list * t)260 static void poll_sync_reset(struct timer_list *t)
261 {
262 struct mlx5_fw_reset *fw_reset = from_timer(fw_reset, t, timer);
263 struct mlx5_core_dev *dev = fw_reset->dev;
264 u32 fatal_error;
265
266 if (!test_bit(MLX5_FW_RESET_FLAGS_RESET_REQUESTED, &fw_reset->reset_flags))
267 return;
268
269 fatal_error = mlx5_health_check_fatal_sensors(dev);
270
271 if (fatal_error) {
272 mlx5_core_warn(dev, "Got Device Reset\n");
273 if (!test_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags))
274 queue_work(fw_reset->wq, &fw_reset->reset_reload_work);
275 else
276 mlx5_core_err(dev, "Device is being removed, Drop new reset work\n");
277 return;
278 }
279
280 mod_timer(&fw_reset->timer, round_jiffies(jiffies + MLX5_RESET_POLL_INTERVAL));
281 }
282
mlx5_start_sync_reset_poll(struct mlx5_core_dev * dev)283 static void mlx5_start_sync_reset_poll(struct mlx5_core_dev *dev)
284 {
285 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
286
287 timer_setup(&fw_reset->timer, poll_sync_reset, 0);
288 fw_reset->timer.expires = round_jiffies(jiffies + MLX5_RESET_POLL_INTERVAL);
289 add_timer(&fw_reset->timer);
290 }
291
mlx5_fw_reset_set_reset_sync_ack(struct mlx5_core_dev * dev)292 static int mlx5_fw_reset_set_reset_sync_ack(struct mlx5_core_dev *dev)
293 {
294 return mlx5_reg_mfrl_set(dev, MLX5_MFRL_REG_RESET_LEVEL3, 0, 1, false);
295 }
296
mlx5_fw_reset_set_reset_sync_nack(struct mlx5_core_dev * dev)297 static int mlx5_fw_reset_set_reset_sync_nack(struct mlx5_core_dev *dev)
298 {
299 return mlx5_reg_mfrl_set(dev, MLX5_MFRL_REG_RESET_LEVEL3, 0, 2, false);
300 }
301
mlx5_sync_reset_set_reset_requested(struct mlx5_core_dev * dev)302 static int mlx5_sync_reset_set_reset_requested(struct mlx5_core_dev *dev)
303 {
304 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
305
306 if (test_and_set_bit(MLX5_FW_RESET_FLAGS_RESET_REQUESTED, &fw_reset->reset_flags)) {
307 mlx5_core_warn(dev, "Reset request was already set\n");
308 return -EALREADY;
309 }
310 mlx5_stop_health_poll(dev, true);
311 mlx5_start_sync_reset_poll(dev);
312 return 0;
313 }
314
mlx5_fw_live_patch_event(struct work_struct * work)315 static void mlx5_fw_live_patch_event(struct work_struct *work)
316 {
317 struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
318 fw_live_patch_work);
319 struct mlx5_core_dev *dev = fw_reset->dev;
320
321 mlx5_core_info(dev, "Live patch updated firmware version: %d.%d.%d\n", fw_rev_maj(dev),
322 fw_rev_min(dev), fw_rev_sub(dev));
323
324 if (mlx5_fw_tracer_reload(dev->tracer))
325 mlx5_core_err(dev, "Failed to reload FW tracer\n");
326 }
327
mlx5_check_dev_ids(struct mlx5_core_dev * dev,u16 dev_id)328 static int mlx5_check_dev_ids(struct mlx5_core_dev *dev, u16 dev_id)
329 {
330 struct pci_bus *bridge_bus = dev->pdev->bus;
331 struct pci_dev *sdev;
332 u16 sdev_id;
333 int err;
334
335 /* Check that all functions under the pci bridge are PFs of
336 * this device otherwise fail this function.
337 */
338 list_for_each_entry(sdev, &bridge_bus->devices, bus_list) {
339 err = pci_read_config_word(sdev, PCI_DEVICE_ID, &sdev_id);
340 if (err)
341 return pcibios_err_to_errno(err);
342 if (sdev_id != dev_id) {
343 mlx5_core_warn(dev, "unrecognized dev_id (0x%x)\n", sdev_id);
344 return -EPERM;
345 }
346 }
347 return 0;
348 }
349
mlx5_is_reset_now_capable(struct mlx5_core_dev * dev)350 static bool mlx5_is_reset_now_capable(struct mlx5_core_dev *dev)
351 {
352 u16 dev_id;
353 int err;
354
355 if (!MLX5_CAP_GEN(dev, fast_teardown)) {
356 mlx5_core_warn(dev, "fast teardown is not supported by firmware\n");
357 return false;
358 }
359
360 err = pci_read_config_word(dev->pdev, PCI_DEVICE_ID, &dev_id);
361 if (err)
362 return false;
363 return (!mlx5_check_dev_ids(dev, dev_id));
364 }
365
mlx5_sync_reset_request_event(struct work_struct * work)366 static void mlx5_sync_reset_request_event(struct work_struct *work)
367 {
368 struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
369 reset_request_work);
370 struct mlx5_core_dev *dev = fw_reset->dev;
371 int err;
372
373 if (test_bit(MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST, &fw_reset->reset_flags) ||
374 !mlx5_is_reset_now_capable(dev)) {
375 err = mlx5_fw_reset_set_reset_sync_nack(dev);
376 mlx5_core_warn(dev, "PCI Sync FW Update Reset Nack %s",
377 err ? "Failed" : "Sent");
378 return;
379 }
380 if (mlx5_sync_reset_set_reset_requested(dev))
381 return;
382
383 err = mlx5_fw_reset_set_reset_sync_ack(dev);
384 if (err)
385 mlx5_core_warn(dev, "PCI Sync FW Update Reset Ack Failed. Error code: %d\n", err);
386 else
387 mlx5_core_warn(dev, "PCI Sync FW Update Reset Ack. Device reset is expected.\n");
388 }
389
mlx5_pci_link_toggle(struct mlx5_core_dev * dev)390 static int mlx5_pci_link_toggle(struct mlx5_core_dev *dev)
391 {
392 struct pci_bus *bridge_bus = dev->pdev->bus;
393 struct pci_dev *bridge = bridge_bus->self;
394 unsigned long timeout;
395 struct pci_dev *sdev;
396 u16 reg16, dev_id;
397 int cap, err;
398
399 err = pci_read_config_word(dev->pdev, PCI_DEVICE_ID, &dev_id);
400 if (err)
401 return pcibios_err_to_errno(err);
402 err = mlx5_check_dev_ids(dev, dev_id);
403 if (err)
404 return err;
405 cap = pci_find_capability(bridge, PCI_CAP_ID_EXP);
406 if (!cap)
407 return -EOPNOTSUPP;
408
409 list_for_each_entry(sdev, &bridge_bus->devices, bus_list) {
410 pci_save_state(sdev);
411 pci_cfg_access_lock(sdev);
412 }
413 /* PCI link toggle */
414 err = pcie_capability_set_word(bridge, PCI_EXP_LNKCTL, PCI_EXP_LNKCTL_LD);
415 if (err)
416 return pcibios_err_to_errno(err);
417 msleep(500);
418 err = pcie_capability_clear_word(bridge, PCI_EXP_LNKCTL, PCI_EXP_LNKCTL_LD);
419 if (err)
420 return pcibios_err_to_errno(err);
421
422 /* Check link */
423 if (!bridge->link_active_reporting) {
424 mlx5_core_warn(dev, "No PCI link reporting capability\n");
425 msleep(1000);
426 goto restore;
427 }
428
429 timeout = jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, PCI_TOGGLE));
430 do {
431 err = pci_read_config_word(bridge, cap + PCI_EXP_LNKSTA, ®16);
432 if (err)
433 return pcibios_err_to_errno(err);
434 if (reg16 & PCI_EXP_LNKSTA_DLLLA)
435 break;
436 msleep(20);
437 } while (!time_after(jiffies, timeout));
438
439 if (reg16 & PCI_EXP_LNKSTA_DLLLA) {
440 mlx5_core_info(dev, "PCI Link up\n");
441 } else {
442 mlx5_core_err(dev, "PCI link not ready (0x%04x) after %llu ms\n",
443 reg16, mlx5_tout_ms(dev, PCI_TOGGLE));
444 err = -ETIMEDOUT;
445 goto restore;
446 }
447
448 do {
449 err = pci_read_config_word(dev->pdev, PCI_DEVICE_ID, ®16);
450 if (err)
451 return pcibios_err_to_errno(err);
452 if (reg16 == dev_id)
453 break;
454 msleep(20);
455 } while (!time_after(jiffies, timeout));
456
457 if (reg16 == dev_id) {
458 mlx5_core_info(dev, "Firmware responds to PCI config cycles again\n");
459 } else {
460 mlx5_core_err(dev, "Firmware is not responsive (0x%04x) after %llu ms\n",
461 reg16, mlx5_tout_ms(dev, PCI_TOGGLE));
462 err = -ETIMEDOUT;
463 }
464
465 restore:
466 list_for_each_entry(sdev, &bridge_bus->devices, bus_list) {
467 pci_cfg_access_unlock(sdev);
468 pci_restore_state(sdev);
469 }
470
471 return err;
472 }
473
mlx5_sync_reset_now_event(struct work_struct * work)474 static void mlx5_sync_reset_now_event(struct work_struct *work)
475 {
476 struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
477 reset_now_work);
478 struct mlx5_core_dev *dev = fw_reset->dev;
479 int err;
480
481 if (mlx5_sync_reset_clear_reset_requested(dev, false))
482 return;
483
484 mlx5_core_warn(dev, "Sync Reset now. Device is going to reset.\n");
485
486 err = mlx5_cmd_fast_teardown_hca(dev);
487 if (err) {
488 mlx5_core_warn(dev, "Fast teardown failed, no reset done, err %d\n", err);
489 goto done;
490 }
491
492 err = mlx5_pci_link_toggle(dev);
493 if (err) {
494 mlx5_core_warn(dev, "mlx5_pci_link_toggle failed, no reset done, err %d\n", err);
495 set_bit(MLX5_FW_RESET_FLAGS_RELOAD_REQUIRED, &fw_reset->reset_flags);
496 }
497
498 mlx5_enter_error_state(dev, true);
499 done:
500 fw_reset->ret = err;
501 mlx5_fw_reset_complete_reload(dev, false);
502 }
503
mlx5_sync_reset_unload_event(struct work_struct * work)504 static void mlx5_sync_reset_unload_event(struct work_struct *work)
505 {
506 struct mlx5_fw_reset *fw_reset;
507 struct mlx5_core_dev *dev;
508 unsigned long timeout;
509 bool reset_action;
510 u8 rst_state;
511 int err;
512
513 fw_reset = container_of(work, struct mlx5_fw_reset, reset_unload_work);
514 dev = fw_reset->dev;
515
516 if (mlx5_sync_reset_clear_reset_requested(dev, false))
517 return;
518
519 mlx5_core_warn(dev, "Sync Reset Unload. Function is forced down.\n");
520
521 err = mlx5_cmd_fast_teardown_hca(dev);
522 if (err)
523 mlx5_core_warn(dev, "Fast teardown failed, unloading, err %d\n", err);
524 else
525 mlx5_enter_error_state(dev, true);
526
527 if (test_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags))
528 mlx5_unload_one_devl_locked(dev, false);
529 else
530 mlx5_unload_one(dev, false);
531
532 mlx5_set_fw_rst_ack(dev);
533 mlx5_core_warn(dev, "Sync Reset Unload done, device reset expected\n");
534
535 reset_action = false;
536 timeout = jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, RESET_UNLOAD));
537 do {
538 rst_state = mlx5_get_fw_rst_state(dev);
539 if (rst_state == MLX5_FW_RST_STATE_TOGGLE_REQ ||
540 rst_state == MLX5_FW_RST_STATE_IDLE) {
541 reset_action = true;
542 break;
543 }
544 msleep(20);
545 } while (!time_after(jiffies, timeout));
546
547 if (!reset_action) {
548 mlx5_core_err(dev, "Got timeout waiting for sync reset action, state = %u\n",
549 rst_state);
550 fw_reset->ret = -ETIMEDOUT;
551 goto done;
552 }
553
554 mlx5_core_warn(dev, "Sync Reset, got reset action. rst_state = %u\n", rst_state);
555 if (rst_state == MLX5_FW_RST_STATE_TOGGLE_REQ) {
556 err = mlx5_pci_link_toggle(dev);
557 if (err) {
558 mlx5_core_warn(dev, "mlx5_pci_link_toggle failed, err %d\n", err);
559 fw_reset->ret = err;
560 }
561 }
562
563 done:
564 mlx5_fw_reset_complete_reload(dev, true);
565 }
566
mlx5_sync_reset_abort_event(struct work_struct * work)567 static void mlx5_sync_reset_abort_event(struct work_struct *work)
568 {
569 struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
570 reset_abort_work);
571 struct mlx5_core_dev *dev = fw_reset->dev;
572
573 if (mlx5_sync_reset_clear_reset_requested(dev, true))
574 return;
575 mlx5_core_warn(dev, "PCI Sync FW Update Reset Aborted.\n");
576 }
577
mlx5_sync_reset_events_handle(struct mlx5_fw_reset * fw_reset,struct mlx5_eqe * eqe)578 static void mlx5_sync_reset_events_handle(struct mlx5_fw_reset *fw_reset, struct mlx5_eqe *eqe)
579 {
580 struct mlx5_eqe_sync_fw_update *sync_fw_update_eqe;
581 u8 sync_event_rst_type;
582
583 sync_fw_update_eqe = &eqe->data.sync_fw_update;
584 sync_event_rst_type = sync_fw_update_eqe->sync_rst_state & SYNC_RST_STATE_MASK;
585 switch (sync_event_rst_type) {
586 case MLX5_SYNC_RST_STATE_RESET_REQUEST:
587 queue_work(fw_reset->wq, &fw_reset->reset_request_work);
588 break;
589 case MLX5_SYNC_RST_STATE_RESET_UNLOAD:
590 queue_work(fw_reset->wq, &fw_reset->reset_unload_work);
591 break;
592 case MLX5_SYNC_RST_STATE_RESET_NOW:
593 queue_work(fw_reset->wq, &fw_reset->reset_now_work);
594 break;
595 case MLX5_SYNC_RST_STATE_RESET_ABORT:
596 queue_work(fw_reset->wq, &fw_reset->reset_abort_work);
597 break;
598 }
599 }
600
fw_reset_event_notifier(struct notifier_block * nb,unsigned long action,void * data)601 static int fw_reset_event_notifier(struct notifier_block *nb, unsigned long action, void *data)
602 {
603 struct mlx5_fw_reset *fw_reset = mlx5_nb_cof(nb, struct mlx5_fw_reset, nb);
604 struct mlx5_eqe *eqe = data;
605
606 if (test_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags))
607 return NOTIFY_DONE;
608
609 switch (eqe->sub_type) {
610 case MLX5_GENERAL_SUBTYPE_FW_LIVE_PATCH_EVENT:
611 queue_work(fw_reset->wq, &fw_reset->fw_live_patch_work);
612 break;
613 case MLX5_GENERAL_SUBTYPE_PCI_SYNC_FOR_FW_UPDATE_EVENT:
614 mlx5_sync_reset_events_handle(fw_reset, eqe);
615 break;
616 default:
617 return NOTIFY_DONE;
618 }
619
620 return NOTIFY_OK;
621 }
622
mlx5_fw_reset_wait_reset_done(struct mlx5_core_dev * dev)623 int mlx5_fw_reset_wait_reset_done(struct mlx5_core_dev *dev)
624 {
625 unsigned long pci_sync_update_timeout = mlx5_tout_ms(dev, PCI_SYNC_UPDATE);
626 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
627 unsigned long timeout;
628 int err;
629
630 if (MLX5_CAP_GEN(dev, pci_sync_for_fw_update_with_driver_unload))
631 pci_sync_update_timeout += mlx5_tout_ms(dev, RESET_UNLOAD);
632 timeout = msecs_to_jiffies(pci_sync_update_timeout);
633 if (!wait_for_completion_timeout(&fw_reset->done, timeout)) {
634 mlx5_core_warn(dev, "FW sync reset timeout after %lu seconds\n",
635 pci_sync_update_timeout / 1000);
636 err = -ETIMEDOUT;
637 goto out;
638 }
639 err = fw_reset->ret;
640 if (test_and_clear_bit(MLX5_FW_RESET_FLAGS_RELOAD_REQUIRED, &fw_reset->reset_flags)) {
641 mlx5_unload_one_devl_locked(dev, false);
642 mlx5_load_one_devl_locked(dev, true);
643 }
644 out:
645 clear_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags);
646 return err;
647 }
648
mlx5_fw_reset_events_start(struct mlx5_core_dev * dev)649 void mlx5_fw_reset_events_start(struct mlx5_core_dev *dev)
650 {
651 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
652
653 MLX5_NB_INIT(&fw_reset->nb, fw_reset_event_notifier, GENERAL_EVENT);
654 mlx5_eq_notifier_register(dev, &fw_reset->nb);
655 }
656
mlx5_fw_reset_events_stop(struct mlx5_core_dev * dev)657 void mlx5_fw_reset_events_stop(struct mlx5_core_dev *dev)
658 {
659 mlx5_eq_notifier_unregister(dev, &dev->priv.fw_reset->nb);
660 }
661
mlx5_drain_fw_reset(struct mlx5_core_dev * dev)662 void mlx5_drain_fw_reset(struct mlx5_core_dev *dev)
663 {
664 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
665
666 set_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags);
667 cancel_work_sync(&fw_reset->fw_live_patch_work);
668 cancel_work_sync(&fw_reset->reset_request_work);
669 cancel_work_sync(&fw_reset->reset_unload_work);
670 cancel_work_sync(&fw_reset->reset_reload_work);
671 cancel_work_sync(&fw_reset->reset_now_work);
672 cancel_work_sync(&fw_reset->reset_abort_work);
673 }
674
675 static const struct devlink_param mlx5_fw_reset_devlink_params[] = {
676 DEVLINK_PARAM_GENERIC(ENABLE_REMOTE_DEV_RESET, BIT(DEVLINK_PARAM_CMODE_RUNTIME),
677 mlx5_fw_reset_enable_remote_dev_reset_get,
678 mlx5_fw_reset_enable_remote_dev_reset_set, NULL),
679 };
680
mlx5_fw_reset_init(struct mlx5_core_dev * dev)681 int mlx5_fw_reset_init(struct mlx5_core_dev *dev)
682 {
683 struct mlx5_fw_reset *fw_reset = kzalloc(sizeof(*fw_reset), GFP_KERNEL);
684 int err;
685
686 if (!fw_reset)
687 return -ENOMEM;
688 fw_reset->wq = create_singlethread_workqueue("mlx5_fw_reset_events");
689 if (!fw_reset->wq) {
690 kfree(fw_reset);
691 return -ENOMEM;
692 }
693
694 fw_reset->dev = dev;
695 dev->priv.fw_reset = fw_reset;
696
697 err = devl_params_register(priv_to_devlink(dev),
698 mlx5_fw_reset_devlink_params,
699 ARRAY_SIZE(mlx5_fw_reset_devlink_params));
700 if (err) {
701 destroy_workqueue(fw_reset->wq);
702 kfree(fw_reset);
703 return err;
704 }
705
706 INIT_WORK(&fw_reset->fw_live_patch_work, mlx5_fw_live_patch_event);
707 INIT_WORK(&fw_reset->reset_request_work, mlx5_sync_reset_request_event);
708 INIT_WORK(&fw_reset->reset_unload_work, mlx5_sync_reset_unload_event);
709 INIT_WORK(&fw_reset->reset_reload_work, mlx5_sync_reset_reload_work);
710 INIT_WORK(&fw_reset->reset_now_work, mlx5_sync_reset_now_event);
711 INIT_WORK(&fw_reset->reset_abort_work, mlx5_sync_reset_abort_event);
712
713 init_completion(&fw_reset->done);
714 return 0;
715 }
716
mlx5_fw_reset_cleanup(struct mlx5_core_dev * dev)717 void mlx5_fw_reset_cleanup(struct mlx5_core_dev *dev)
718 {
719 struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
720
721 devl_params_unregister(priv_to_devlink(dev),
722 mlx5_fw_reset_devlink_params,
723 ARRAY_SIZE(mlx5_fw_reset_devlink_params));
724 destroy_workqueue(fw_reset->wq);
725 kfree(dev->priv.fw_reset);
726 }
727
