1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* Copyright (c) 2020, Mellanox Technologies inc.  All rights reserved. */
3 
4 #include "fw_reset.h"
5 #include "diag/fw_tracer.h"
6 #include "lib/tout.h"
7 
8 enum {
9 	MLX5_FW_RESET_FLAGS_RESET_REQUESTED,
10 	MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST,
11 	MLX5_FW_RESET_FLAGS_PENDING_COMP,
12 	MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS
13 };
14 
15 struct mlx5_fw_reset {
16 	struct mlx5_core_dev *dev;
17 	struct mlx5_nb nb;
18 	struct workqueue_struct *wq;
19 	struct work_struct fw_live_patch_work;
20 	struct work_struct reset_request_work;
21 	struct work_struct reset_reload_work;
22 	struct work_struct reset_now_work;
23 	struct work_struct reset_abort_work;
24 	unsigned long reset_flags;
25 	struct timer_list timer;
26 	struct completion done;
27 	int ret;
28 };
29 
mlx5_fw_reset_enable_remote_dev_reset_set(struct mlx5_core_dev * dev,bool enable)30 void mlx5_fw_reset_enable_remote_dev_reset_set(struct mlx5_core_dev *dev, bool enable)
31 {
32 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
33 
34 	if (enable)
35 		clear_bit(MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST, &fw_reset->reset_flags);
36 	else
37 		set_bit(MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST, &fw_reset->reset_flags);
38 }
39 
mlx5_fw_reset_enable_remote_dev_reset_get(struct mlx5_core_dev * dev)40 bool mlx5_fw_reset_enable_remote_dev_reset_get(struct mlx5_core_dev *dev)
41 {
42 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
43 
44 	return !test_bit(MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST, &fw_reset->reset_flags);
45 }
46 
mlx5_reg_mfrl_set(struct mlx5_core_dev * dev,u8 reset_level,u8 reset_type_sel,u8 sync_resp,bool sync_start)47 static int mlx5_reg_mfrl_set(struct mlx5_core_dev *dev, u8 reset_level,
48 			     u8 reset_type_sel, u8 sync_resp, bool sync_start)
49 {
50 	u32 out[MLX5_ST_SZ_DW(mfrl_reg)] = {};
51 	u32 in[MLX5_ST_SZ_DW(mfrl_reg)] = {};
52 
53 	MLX5_SET(mfrl_reg, in, reset_level, reset_level);
54 	MLX5_SET(mfrl_reg, in, rst_type_sel, reset_type_sel);
55 	MLX5_SET(mfrl_reg, in, pci_sync_for_fw_update_resp, sync_resp);
56 	MLX5_SET(mfrl_reg, in, pci_sync_for_fw_update_start, sync_start);
57 
58 	return mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_REG_MFRL, 0, 1);
59 }
60 
mlx5_reg_mfrl_query(struct mlx5_core_dev * dev,u8 * reset_level,u8 * reset_type,u8 * reset_state)61 static int mlx5_reg_mfrl_query(struct mlx5_core_dev *dev, u8 *reset_level,
62 			       u8 *reset_type, u8 *reset_state)
63 {
64 	u32 out[MLX5_ST_SZ_DW(mfrl_reg)] = {};
65 	u32 in[MLX5_ST_SZ_DW(mfrl_reg)] = {};
66 	int err;
67 
68 	err = mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_REG_MFRL, 0, 0);
69 	if (err)
70 		return err;
71 
72 	if (reset_level)
73 		*reset_level = MLX5_GET(mfrl_reg, out, reset_level);
74 	if (reset_type)
75 		*reset_type = MLX5_GET(mfrl_reg, out, reset_type);
76 	if (reset_state)
77 		*reset_state = MLX5_GET(mfrl_reg, out, reset_state);
78 
79 	return 0;
80 }
81 
mlx5_fw_reset_query(struct mlx5_core_dev * dev,u8 * reset_level,u8 * reset_type)82 int mlx5_fw_reset_query(struct mlx5_core_dev *dev, u8 *reset_level, u8 *reset_type)
83 {
84 	return mlx5_reg_mfrl_query(dev, reset_level, reset_type, NULL);
85 }
86 
mlx5_fw_reset_get_reset_state_err(struct mlx5_core_dev * dev,struct netlink_ext_ack * extack)87 static int mlx5_fw_reset_get_reset_state_err(struct mlx5_core_dev *dev,
88 					     struct netlink_ext_ack *extack)
89 {
90 	u8 reset_state;
91 
92 	if (mlx5_reg_mfrl_query(dev, NULL, NULL, &reset_state))
93 		goto out;
94 
95 	switch (reset_state) {
96 	case MLX5_MFRL_REG_RESET_STATE_IN_NEGOTIATION:
97 	case MLX5_MFRL_REG_RESET_STATE_RESET_IN_PROGRESS:
98 		NL_SET_ERR_MSG_MOD(extack, "Sync reset was already triggered");
99 		return -EBUSY;
100 	case MLX5_MFRL_REG_RESET_STATE_TIMEOUT:
101 		NL_SET_ERR_MSG_MOD(extack, "Sync reset got timeout");
102 		return -ETIMEDOUT;
103 	case MLX5_MFRL_REG_RESET_STATE_NACK:
104 		NL_SET_ERR_MSG_MOD(extack, "One of the hosts disabled reset");
105 		return -EPERM;
106 	}
107 
108 out:
109 	NL_SET_ERR_MSG_MOD(extack, "Sync reset failed");
110 	return -EIO;
111 }
112 
mlx5_fw_reset_set_reset_sync(struct mlx5_core_dev * dev,u8 reset_type_sel,struct netlink_ext_ack * extack)113 int mlx5_fw_reset_set_reset_sync(struct mlx5_core_dev *dev, u8 reset_type_sel,
114 				 struct netlink_ext_ack *extack)
115 {
116 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
117 	u32 out[MLX5_ST_SZ_DW(mfrl_reg)] = {};
118 	u32 in[MLX5_ST_SZ_DW(mfrl_reg)] = {};
119 	int err;
120 
121 	set_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags);
122 
123 	MLX5_SET(mfrl_reg, in, reset_level, MLX5_MFRL_REG_RESET_LEVEL3);
124 	MLX5_SET(mfrl_reg, in, rst_type_sel, reset_type_sel);
125 	MLX5_SET(mfrl_reg, in, pci_sync_for_fw_update_start, 1);
126 	err = mlx5_access_reg(dev, in, sizeof(in), out, sizeof(out),
127 			      MLX5_REG_MFRL, 0, 1, false);
128 	if (!err)
129 		return 0;
130 
131 	clear_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags);
132 	if (err == -EREMOTEIO && MLX5_CAP_MCAM_FEATURE(dev, reset_state))
133 		return mlx5_fw_reset_get_reset_state_err(dev, extack);
134 
135 	NL_SET_ERR_MSG_MOD(extack, "Sync reset command failed");
136 	return mlx5_cmd_check(dev, err, in, out);
137 }
138 
mlx5_fw_reset_set_live_patch(struct mlx5_core_dev * dev)139 int mlx5_fw_reset_set_live_patch(struct mlx5_core_dev *dev)
140 {
141 	return mlx5_reg_mfrl_set(dev, MLX5_MFRL_REG_RESET_LEVEL0, 0, 0, false);
142 }
143 
mlx5_fw_reset_complete_reload(struct mlx5_core_dev * dev)144 static void mlx5_fw_reset_complete_reload(struct mlx5_core_dev *dev)
145 {
146 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
147 
148 	/* if this is the driver that initiated the fw reset, devlink completed the reload */
149 	if (test_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags)) {
150 		complete(&fw_reset->done);
151 	} else {
152 		mlx5_load_one(dev, false);
153 		devlink_remote_reload_actions_performed(priv_to_devlink(dev), 0,
154 							BIT(DEVLINK_RELOAD_ACTION_DRIVER_REINIT) |
155 							BIT(DEVLINK_RELOAD_ACTION_FW_ACTIVATE));
156 	}
157 }
158 
mlx5_stop_sync_reset_poll(struct mlx5_core_dev * dev)159 static void mlx5_stop_sync_reset_poll(struct mlx5_core_dev *dev)
160 {
161 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
162 
163 	del_timer_sync(&fw_reset->timer);
164 }
165 
mlx5_sync_reset_clear_reset_requested(struct mlx5_core_dev * dev,bool poll_health)166 static int mlx5_sync_reset_clear_reset_requested(struct mlx5_core_dev *dev, bool poll_health)
167 {
168 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
169 
170 	if (!test_and_clear_bit(MLX5_FW_RESET_FLAGS_RESET_REQUESTED, &fw_reset->reset_flags)) {
171 		mlx5_core_warn(dev, "Reset request was already cleared\n");
172 		return -EALREADY;
173 	}
174 
175 	mlx5_stop_sync_reset_poll(dev);
176 	if (poll_health)
177 		mlx5_start_health_poll(dev);
178 	return 0;
179 }
180 
mlx5_sync_reset_reload_work(struct work_struct * work)181 static void mlx5_sync_reset_reload_work(struct work_struct *work)
182 {
183 	struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
184 						      reset_reload_work);
185 	struct mlx5_core_dev *dev = fw_reset->dev;
186 	int err;
187 
188 	mlx5_sync_reset_clear_reset_requested(dev, false);
189 	mlx5_enter_error_state(dev, true);
190 	mlx5_unload_one(dev);
191 	err = mlx5_health_wait_pci_up(dev);
192 	if (err)
193 		mlx5_core_err(dev, "reset reload flow aborted, PCI reads still not working\n");
194 	fw_reset->ret = err;
195 	mlx5_fw_reset_complete_reload(dev);
196 }
197 
198 #define MLX5_RESET_POLL_INTERVAL	(HZ / 10)
poll_sync_reset(struct timer_list * t)199 static void poll_sync_reset(struct timer_list *t)
200 {
201 	struct mlx5_fw_reset *fw_reset = from_timer(fw_reset, t, timer);
202 	struct mlx5_core_dev *dev = fw_reset->dev;
203 	u32 fatal_error;
204 
205 	if (!test_bit(MLX5_FW_RESET_FLAGS_RESET_REQUESTED, &fw_reset->reset_flags))
206 		return;
207 
208 	fatal_error = mlx5_health_check_fatal_sensors(dev);
209 
210 	if (fatal_error) {
211 		mlx5_core_warn(dev, "Got Device Reset\n");
212 		if (!test_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags))
213 			queue_work(fw_reset->wq, &fw_reset->reset_reload_work);
214 		else
215 			mlx5_core_err(dev, "Device is being removed, Drop new reset work\n");
216 		return;
217 	}
218 
219 	mod_timer(&fw_reset->timer, round_jiffies(jiffies + MLX5_RESET_POLL_INTERVAL));
220 }
221 
mlx5_start_sync_reset_poll(struct mlx5_core_dev * dev)222 static void mlx5_start_sync_reset_poll(struct mlx5_core_dev *dev)
223 {
224 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
225 
226 	timer_setup(&fw_reset->timer, poll_sync_reset, 0);
227 	fw_reset->timer.expires = round_jiffies(jiffies + MLX5_RESET_POLL_INTERVAL);
228 	add_timer(&fw_reset->timer);
229 }
230 
mlx5_fw_reset_set_reset_sync_ack(struct mlx5_core_dev * dev)231 static int mlx5_fw_reset_set_reset_sync_ack(struct mlx5_core_dev *dev)
232 {
233 	return mlx5_reg_mfrl_set(dev, MLX5_MFRL_REG_RESET_LEVEL3, 0, 1, false);
234 }
235 
mlx5_fw_reset_set_reset_sync_nack(struct mlx5_core_dev * dev)236 static int mlx5_fw_reset_set_reset_sync_nack(struct mlx5_core_dev *dev)
237 {
238 	return mlx5_reg_mfrl_set(dev, MLX5_MFRL_REG_RESET_LEVEL3, 0, 2, false);
239 }
240 
mlx5_sync_reset_set_reset_requested(struct mlx5_core_dev * dev)241 static int mlx5_sync_reset_set_reset_requested(struct mlx5_core_dev *dev)
242 {
243 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
244 
245 	if (test_and_set_bit(MLX5_FW_RESET_FLAGS_RESET_REQUESTED, &fw_reset->reset_flags)) {
246 		mlx5_core_warn(dev, "Reset request was already set\n");
247 		return -EALREADY;
248 	}
249 	mlx5_stop_health_poll(dev, true);
250 	mlx5_start_sync_reset_poll(dev);
251 	return 0;
252 }
253 
mlx5_fw_live_patch_event(struct work_struct * work)254 static void mlx5_fw_live_patch_event(struct work_struct *work)
255 {
256 	struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
257 						      fw_live_patch_work);
258 	struct mlx5_core_dev *dev = fw_reset->dev;
259 
260 	mlx5_core_info(dev, "Live patch updated firmware version: %d.%d.%d\n", fw_rev_maj(dev),
261 		       fw_rev_min(dev), fw_rev_sub(dev));
262 
263 	if (mlx5_fw_tracer_reload(dev->tracer))
264 		mlx5_core_err(dev, "Failed to reload FW tracer\n");
265 }
266 
mlx5_sync_reset_request_event(struct work_struct * work)267 static void mlx5_sync_reset_request_event(struct work_struct *work)
268 {
269 	struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
270 						      reset_request_work);
271 	struct mlx5_core_dev *dev = fw_reset->dev;
272 	int err;
273 
274 	if (test_bit(MLX5_FW_RESET_FLAGS_NACK_RESET_REQUEST, &fw_reset->reset_flags)) {
275 		err = mlx5_fw_reset_set_reset_sync_nack(dev);
276 		mlx5_core_warn(dev, "PCI Sync FW Update Reset Nack %s",
277 			       err ? "Failed" : "Sent");
278 		return;
279 	}
280 	if (mlx5_sync_reset_set_reset_requested(dev))
281 		return;
282 
283 	err = mlx5_fw_reset_set_reset_sync_ack(dev);
284 	if (err)
285 		mlx5_core_warn(dev, "PCI Sync FW Update Reset Ack Failed. Error code: %d\n", err);
286 	else
287 		mlx5_core_warn(dev, "PCI Sync FW Update Reset Ack. Device reset is expected.\n");
288 }
289 
mlx5_pci_link_toggle(struct mlx5_core_dev * dev)290 static int mlx5_pci_link_toggle(struct mlx5_core_dev *dev)
291 {
292 	struct pci_bus *bridge_bus = dev->pdev->bus;
293 	struct pci_dev *bridge = bridge_bus->self;
294 	u16 reg16, dev_id, sdev_id;
295 	unsigned long timeout;
296 	struct pci_dev *sdev;
297 	int cap, err;
298 	u32 reg32;
299 
300 	/* Check that all functions under the pci bridge are PFs of
301 	 * this device otherwise fail this function.
302 	 */
303 	err = pci_read_config_word(dev->pdev, PCI_DEVICE_ID, &dev_id);
304 	if (err)
305 		return err;
306 	list_for_each_entry(sdev, &bridge_bus->devices, bus_list) {
307 		err = pci_read_config_word(sdev, PCI_DEVICE_ID, &sdev_id);
308 		if (err)
309 			return err;
310 		if (sdev_id != dev_id)
311 			return -EPERM;
312 	}
313 
314 	cap = pci_find_capability(bridge, PCI_CAP_ID_EXP);
315 	if (!cap)
316 		return -EOPNOTSUPP;
317 
318 	list_for_each_entry(sdev, &bridge_bus->devices, bus_list) {
319 		pci_save_state(sdev);
320 		pci_cfg_access_lock(sdev);
321 	}
322 	/* PCI link toggle */
323 	err = pci_read_config_word(bridge, cap + PCI_EXP_LNKCTL, &reg16);
324 	if (err)
325 		return err;
326 	reg16 |= PCI_EXP_LNKCTL_LD;
327 	err = pci_write_config_word(bridge, cap + PCI_EXP_LNKCTL, reg16);
328 	if (err)
329 		return err;
330 	msleep(500);
331 	reg16 &= ~PCI_EXP_LNKCTL_LD;
332 	err = pci_write_config_word(bridge, cap + PCI_EXP_LNKCTL, reg16);
333 	if (err)
334 		return err;
335 
336 	/* Check link */
337 	err = pci_read_config_dword(bridge, cap + PCI_EXP_LNKCAP, &reg32);
338 	if (err)
339 		return err;
340 	if (!(reg32 & PCI_EXP_LNKCAP_DLLLARC)) {
341 		mlx5_core_warn(dev, "No PCI link reporting capability (0x%08x)\n", reg32);
342 		msleep(1000);
343 		goto restore;
344 	}
345 
346 	timeout = jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, PCI_TOGGLE));
347 	do {
348 		err = pci_read_config_word(bridge, cap + PCI_EXP_LNKSTA, &reg16);
349 		if (err)
350 			return err;
351 		if (reg16 & PCI_EXP_LNKSTA_DLLLA)
352 			break;
353 		msleep(20);
354 	} while (!time_after(jiffies, timeout));
355 
356 	if (reg16 & PCI_EXP_LNKSTA_DLLLA) {
357 		mlx5_core_info(dev, "PCI Link up\n");
358 	} else {
359 		mlx5_core_err(dev, "PCI link not ready (0x%04x) after %llu ms\n",
360 			      reg16, mlx5_tout_ms(dev, PCI_TOGGLE));
361 		err = -ETIMEDOUT;
362 	}
363 
364 restore:
365 	list_for_each_entry(sdev, &bridge_bus->devices, bus_list) {
366 		pci_cfg_access_unlock(sdev);
367 		pci_restore_state(sdev);
368 	}
369 
370 	return err;
371 }
372 
mlx5_sync_reset_now_event(struct work_struct * work)373 static void mlx5_sync_reset_now_event(struct work_struct *work)
374 {
375 	struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
376 						      reset_now_work);
377 	struct mlx5_core_dev *dev = fw_reset->dev;
378 	int err;
379 
380 	if (mlx5_sync_reset_clear_reset_requested(dev, false))
381 		return;
382 
383 	mlx5_core_warn(dev, "Sync Reset now. Device is going to reset.\n");
384 
385 	err = mlx5_cmd_fast_teardown_hca(dev);
386 	if (err) {
387 		mlx5_core_warn(dev, "Fast teardown failed, no reset done, err %d\n", err);
388 		goto done;
389 	}
390 
391 	err = mlx5_pci_link_toggle(dev);
392 	if (err) {
393 		mlx5_core_warn(dev, "mlx5_pci_link_toggle failed, no reset done, err %d\n", err);
394 		goto done;
395 	}
396 
397 	mlx5_enter_error_state(dev, true);
398 	mlx5_unload_one(dev);
399 done:
400 	fw_reset->ret = err;
401 	mlx5_fw_reset_complete_reload(dev);
402 }
403 
mlx5_sync_reset_abort_event(struct work_struct * work)404 static void mlx5_sync_reset_abort_event(struct work_struct *work)
405 {
406 	struct mlx5_fw_reset *fw_reset = container_of(work, struct mlx5_fw_reset,
407 						      reset_abort_work);
408 	struct mlx5_core_dev *dev = fw_reset->dev;
409 
410 	if (mlx5_sync_reset_clear_reset_requested(dev, true))
411 		return;
412 	mlx5_core_warn(dev, "PCI Sync FW Update Reset Aborted.\n");
413 }
414 
mlx5_sync_reset_events_handle(struct mlx5_fw_reset * fw_reset,struct mlx5_eqe * eqe)415 static void mlx5_sync_reset_events_handle(struct mlx5_fw_reset *fw_reset, struct mlx5_eqe *eqe)
416 {
417 	struct mlx5_eqe_sync_fw_update *sync_fw_update_eqe;
418 	u8 sync_event_rst_type;
419 
420 	sync_fw_update_eqe = &eqe->data.sync_fw_update;
421 	sync_event_rst_type = sync_fw_update_eqe->sync_rst_state & SYNC_RST_STATE_MASK;
422 	switch (sync_event_rst_type) {
423 	case MLX5_SYNC_RST_STATE_RESET_REQUEST:
424 		queue_work(fw_reset->wq, &fw_reset->reset_request_work);
425 		break;
426 	case MLX5_SYNC_RST_STATE_RESET_NOW:
427 		queue_work(fw_reset->wq, &fw_reset->reset_now_work);
428 		break;
429 	case MLX5_SYNC_RST_STATE_RESET_ABORT:
430 		queue_work(fw_reset->wq, &fw_reset->reset_abort_work);
431 		break;
432 	}
433 }
434 
fw_reset_event_notifier(struct notifier_block * nb,unsigned long action,void * data)435 static int fw_reset_event_notifier(struct notifier_block *nb, unsigned long action, void *data)
436 {
437 	struct mlx5_fw_reset *fw_reset = mlx5_nb_cof(nb, struct mlx5_fw_reset, nb);
438 	struct mlx5_eqe *eqe = data;
439 
440 	if (test_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags))
441 		return NOTIFY_DONE;
442 
443 	switch (eqe->sub_type) {
444 	case MLX5_GENERAL_SUBTYPE_FW_LIVE_PATCH_EVENT:
445 		queue_work(fw_reset->wq, &fw_reset->fw_live_patch_work);
446 		break;
447 	case MLX5_GENERAL_SUBTYPE_PCI_SYNC_FOR_FW_UPDATE_EVENT:
448 		mlx5_sync_reset_events_handle(fw_reset, eqe);
449 		break;
450 	default:
451 		return NOTIFY_DONE;
452 	}
453 
454 	return NOTIFY_OK;
455 }
456 
mlx5_fw_reset_wait_reset_done(struct mlx5_core_dev * dev)457 int mlx5_fw_reset_wait_reset_done(struct mlx5_core_dev *dev)
458 {
459 	unsigned long pci_sync_update_timeout = mlx5_tout_ms(dev, PCI_SYNC_UPDATE);
460 	unsigned long timeout = msecs_to_jiffies(pci_sync_update_timeout);
461 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
462 	int err;
463 
464 	if (!wait_for_completion_timeout(&fw_reset->done, timeout)) {
465 		mlx5_core_warn(dev, "FW sync reset timeout after %lu seconds\n",
466 			       pci_sync_update_timeout / 1000);
467 		err = -ETIMEDOUT;
468 		goto out;
469 	}
470 	err = fw_reset->ret;
471 out:
472 	clear_bit(MLX5_FW_RESET_FLAGS_PENDING_COMP, &fw_reset->reset_flags);
473 	return err;
474 }
475 
mlx5_fw_reset_events_start(struct mlx5_core_dev * dev)476 void mlx5_fw_reset_events_start(struct mlx5_core_dev *dev)
477 {
478 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
479 
480 	MLX5_NB_INIT(&fw_reset->nb, fw_reset_event_notifier, GENERAL_EVENT);
481 	mlx5_eq_notifier_register(dev, &fw_reset->nb);
482 }
483 
mlx5_fw_reset_events_stop(struct mlx5_core_dev * dev)484 void mlx5_fw_reset_events_stop(struct mlx5_core_dev *dev)
485 {
486 	mlx5_eq_notifier_unregister(dev, &dev->priv.fw_reset->nb);
487 }
488 
mlx5_drain_fw_reset(struct mlx5_core_dev * dev)489 void mlx5_drain_fw_reset(struct mlx5_core_dev *dev)
490 {
491 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
492 
493 	set_bit(MLX5_FW_RESET_FLAGS_DROP_NEW_REQUESTS, &fw_reset->reset_flags);
494 	cancel_work_sync(&fw_reset->fw_live_patch_work);
495 	cancel_work_sync(&fw_reset->reset_request_work);
496 	cancel_work_sync(&fw_reset->reset_reload_work);
497 	cancel_work_sync(&fw_reset->reset_now_work);
498 	cancel_work_sync(&fw_reset->reset_abort_work);
499 }
500 
mlx5_fw_reset_init(struct mlx5_core_dev * dev)501 int mlx5_fw_reset_init(struct mlx5_core_dev *dev)
502 {
503 	struct mlx5_fw_reset *fw_reset = kzalloc(sizeof(*fw_reset), GFP_KERNEL);
504 
505 	if (!fw_reset)
506 		return -ENOMEM;
507 	fw_reset->wq = create_singlethread_workqueue("mlx5_fw_reset_events");
508 	if (!fw_reset->wq) {
509 		kfree(fw_reset);
510 		return -ENOMEM;
511 	}
512 
513 	fw_reset->dev = dev;
514 	dev->priv.fw_reset = fw_reset;
515 
516 	INIT_WORK(&fw_reset->fw_live_patch_work, mlx5_fw_live_patch_event);
517 	INIT_WORK(&fw_reset->reset_request_work, mlx5_sync_reset_request_event);
518 	INIT_WORK(&fw_reset->reset_reload_work, mlx5_sync_reset_reload_work);
519 	INIT_WORK(&fw_reset->reset_now_work, mlx5_sync_reset_now_event);
520 	INIT_WORK(&fw_reset->reset_abort_work, mlx5_sync_reset_abort_event);
521 
522 	init_completion(&fw_reset->done);
523 	return 0;
524 }
525 
mlx5_fw_reset_cleanup(struct mlx5_core_dev * dev)526 void mlx5_fw_reset_cleanup(struct mlx5_core_dev *dev)
527 {
528 	struct mlx5_fw_reset *fw_reset = dev->priv.fw_reset;
529 
530 	destroy_workqueue(fw_reset->wq);
531 	kfree(dev->priv.fw_reset);
532 }
533