1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 // Copyright (c) 2018 Mellanox Technologies
3 
4 #include <linux/mlx5/driver.h>
5 
6 #include "mlx5_core.h"
7 #include "lib/eq.h"
8 #include "lib/events.h"
9 
10 struct mlx5_event_nb {
11 	struct mlx5_nb  nb;
12 	void           *ctx;
13 };
14 
15 /* General events handlers for the low level mlx5_core driver
16  *
17  * Other Major feature specific events such as
18  * clock/eswitch/fpga/FW trace and many others, are handled elsewhere, with
19  * separate notifiers callbacks, specifically by those mlx5 components.
20  */
21 static int any_notifier(struct notifier_block *, unsigned long, void *);
22 static int temp_warn(struct notifier_block *, unsigned long, void *);
23 static int port_module(struct notifier_block *, unsigned long, void *);
24 static int pcie_core(struct notifier_block *, unsigned long, void *);
25 
26 /* handler which forwards the event to events->fw_nh, driver notifiers */
27 static int forward_event(struct notifier_block *, unsigned long, void *);
28 
29 static struct mlx5_nb events_nbs_ref[] = {
30 	/* Events to be processed by mlx5_core */
31 	{.nb.notifier_call = any_notifier,  .event_type = MLX5_EVENT_TYPE_NOTIFY_ANY },
32 	{.nb.notifier_call = temp_warn,     .event_type = MLX5_EVENT_TYPE_TEMP_WARN_EVENT },
33 	{.nb.notifier_call = port_module,   .event_type = MLX5_EVENT_TYPE_PORT_MODULE_EVENT },
34 	{.nb.notifier_call = pcie_core,     .event_type = MLX5_EVENT_TYPE_GENERAL_EVENT },
35 
36 	/* Events to be forwarded (as is) to mlx5 core interfaces (mlx5e/mlx5_ib) */
37 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_PORT_CHANGE },
38 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_GENERAL_EVENT },
39 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_OBJECT_CHANGE },
40 	/* QP/WQ resource events to forward */
41 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_DCT_DRAINED },
42 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_PATH_MIG },
43 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_COMM_EST },
44 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_SQ_DRAINED },
45 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_SRQ_LAST_WQE },
46 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_WQ_CATAS_ERROR },
47 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_PATH_MIG_FAILED },
48 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR },
49 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_WQ_ACCESS_ERROR },
50 	/* SRQ events */
51 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_SRQ_CATAS_ERROR },
52 	{.nb.notifier_call = forward_event,   .event_type = MLX5_EVENT_TYPE_SRQ_RQ_LIMIT },
53 };
54 
55 struct mlx5_events {
56 	struct mlx5_core_dev *dev;
57 	struct workqueue_struct *wq;
58 	struct mlx5_event_nb  notifiers[ARRAY_SIZE(events_nbs_ref)];
59 	/* driver notifier chain for fw events */
60 	struct atomic_notifier_head fw_nh;
61 	/* port module events stats */
62 	struct mlx5_pme_stats pme_stats;
63 	/*pcie_core*/
64 	struct work_struct pcie_core_work;
65 	/* driver notifier chain for sw events */
66 	struct blocking_notifier_head sw_nh;
67 };
68 
eqe_type_str(u8 type)69 static const char *eqe_type_str(u8 type)
70 {
71 	switch (type) {
72 	case MLX5_EVENT_TYPE_COMP:
73 		return "MLX5_EVENT_TYPE_COMP";
74 	case MLX5_EVENT_TYPE_PATH_MIG:
75 		return "MLX5_EVENT_TYPE_PATH_MIG";
76 	case MLX5_EVENT_TYPE_COMM_EST:
77 		return "MLX5_EVENT_TYPE_COMM_EST";
78 	case MLX5_EVENT_TYPE_SQ_DRAINED:
79 		return "MLX5_EVENT_TYPE_SQ_DRAINED";
80 	case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
81 		return "MLX5_EVENT_TYPE_SRQ_LAST_WQE";
82 	case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
83 		return "MLX5_EVENT_TYPE_SRQ_RQ_LIMIT";
84 	case MLX5_EVENT_TYPE_CQ_ERROR:
85 		return "MLX5_EVENT_TYPE_CQ_ERROR";
86 	case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
87 		return "MLX5_EVENT_TYPE_WQ_CATAS_ERROR";
88 	case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
89 		return "MLX5_EVENT_TYPE_PATH_MIG_FAILED";
90 	case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
91 		return "MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR";
92 	case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
93 		return "MLX5_EVENT_TYPE_WQ_ACCESS_ERROR";
94 	case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
95 		return "MLX5_EVENT_TYPE_SRQ_CATAS_ERROR";
96 	case MLX5_EVENT_TYPE_INTERNAL_ERROR:
97 		return "MLX5_EVENT_TYPE_INTERNAL_ERROR";
98 	case MLX5_EVENT_TYPE_PORT_CHANGE:
99 		return "MLX5_EVENT_TYPE_PORT_CHANGE";
100 	case MLX5_EVENT_TYPE_GPIO_EVENT:
101 		return "MLX5_EVENT_TYPE_GPIO_EVENT";
102 	case MLX5_EVENT_TYPE_PORT_MODULE_EVENT:
103 		return "MLX5_EVENT_TYPE_PORT_MODULE_EVENT";
104 	case MLX5_EVENT_TYPE_TEMP_WARN_EVENT:
105 		return "MLX5_EVENT_TYPE_TEMP_WARN_EVENT";
106 	case MLX5_EVENT_TYPE_REMOTE_CONFIG:
107 		return "MLX5_EVENT_TYPE_REMOTE_CONFIG";
108 	case MLX5_EVENT_TYPE_DB_BF_CONGESTION:
109 		return "MLX5_EVENT_TYPE_DB_BF_CONGESTION";
110 	case MLX5_EVENT_TYPE_STALL_EVENT:
111 		return "MLX5_EVENT_TYPE_STALL_EVENT";
112 	case MLX5_EVENT_TYPE_CMD:
113 		return "MLX5_EVENT_TYPE_CMD";
114 	case MLX5_EVENT_TYPE_ESW_FUNCTIONS_CHANGED:
115 		return "MLX5_EVENT_TYPE_ESW_FUNCTIONS_CHANGED";
116 	case MLX5_EVENT_TYPE_VHCA_STATE_CHANGE:
117 		return "MLX5_EVENT_TYPE_VHCA_STATE_CHANGE";
118 	case MLX5_EVENT_TYPE_PAGE_REQUEST:
119 		return "MLX5_EVENT_TYPE_PAGE_REQUEST";
120 	case MLX5_EVENT_TYPE_PAGE_FAULT:
121 		return "MLX5_EVENT_TYPE_PAGE_FAULT";
122 	case MLX5_EVENT_TYPE_PPS_EVENT:
123 		return "MLX5_EVENT_TYPE_PPS_EVENT";
124 	case MLX5_EVENT_TYPE_NIC_VPORT_CHANGE:
125 		return "MLX5_EVENT_TYPE_NIC_VPORT_CHANGE";
126 	case MLX5_EVENT_TYPE_FPGA_ERROR:
127 		return "MLX5_EVENT_TYPE_FPGA_ERROR";
128 	case MLX5_EVENT_TYPE_FPGA_QP_ERROR:
129 		return "MLX5_EVENT_TYPE_FPGA_QP_ERROR";
130 	case MLX5_EVENT_TYPE_GENERAL_EVENT:
131 		return "MLX5_EVENT_TYPE_GENERAL_EVENT";
132 	case MLX5_EVENT_TYPE_MONITOR_COUNTER:
133 		return "MLX5_EVENT_TYPE_MONITOR_COUNTER";
134 	case MLX5_EVENT_TYPE_DEVICE_TRACER:
135 		return "MLX5_EVENT_TYPE_DEVICE_TRACER";
136 	case MLX5_EVENT_TYPE_OBJECT_CHANGE:
137 		return "MLX5_EVENT_TYPE_OBJECT_CHANGE";
138 	default:
139 		return "Unrecognized event";
140 	}
141 }
142 
143 /* handles all FW events, type == eqe->type */
any_notifier(struct notifier_block * nb,unsigned long type,void * data)144 static int any_notifier(struct notifier_block *nb,
145 			unsigned long type, void *data)
146 {
147 	struct mlx5_event_nb *event_nb = mlx5_nb_cof(nb, struct mlx5_event_nb, nb);
148 	struct mlx5_events   *events   = event_nb->ctx;
149 	struct mlx5_eqe      *eqe      = data;
150 
151 	mlx5_core_dbg(events->dev, "Async eqe type %s, subtype (%d)\n",
152 		      eqe_type_str(eqe->type), eqe->sub_type);
153 	return NOTIFY_OK;
154 }
155 
156 /* type == MLX5_EVENT_TYPE_TEMP_WARN_EVENT */
temp_warn(struct notifier_block * nb,unsigned long type,void * data)157 static int temp_warn(struct notifier_block *nb, unsigned long type, void *data)
158 {
159 	struct mlx5_event_nb *event_nb = mlx5_nb_cof(nb, struct mlx5_event_nb, nb);
160 	struct mlx5_events   *events   = event_nb->ctx;
161 	struct mlx5_eqe      *eqe      = data;
162 	u64 value_lsb;
163 	u64 value_msb;
164 
165 	value_lsb = be64_to_cpu(eqe->data.temp_warning.sensor_warning_lsb);
166 	value_msb = be64_to_cpu(eqe->data.temp_warning.sensor_warning_msb);
167 
168 	mlx5_core_warn(events->dev,
169 		       "High temperature on sensors with bit set %llx %llx",
170 		       value_msb, value_lsb);
171 
172 	return NOTIFY_OK;
173 }
174 
175 /* MLX5_EVENT_TYPE_PORT_MODULE_EVENT */
mlx5_pme_status_to_string(enum port_module_event_status_type status)176 static const char *mlx5_pme_status_to_string(enum port_module_event_status_type status)
177 {
178 	switch (status) {
179 	case MLX5_MODULE_STATUS_PLUGGED:
180 		return "Cable plugged";
181 	case MLX5_MODULE_STATUS_UNPLUGGED:
182 		return "Cable unplugged";
183 	case MLX5_MODULE_STATUS_ERROR:
184 		return "Cable error";
185 	case MLX5_MODULE_STATUS_DISABLED:
186 		return "Cable disabled";
187 	default:
188 		return "Unknown status";
189 	}
190 }
191 
mlx5_pme_error_to_string(enum port_module_event_error_type error)192 static const char *mlx5_pme_error_to_string(enum port_module_event_error_type error)
193 {
194 	switch (error) {
195 	case MLX5_MODULE_EVENT_ERROR_POWER_BUDGET_EXCEEDED:
196 		return "Power budget exceeded";
197 	case MLX5_MODULE_EVENT_ERROR_LONG_RANGE_FOR_NON_MLNX:
198 		return "Long Range for non MLNX cable";
199 	case MLX5_MODULE_EVENT_ERROR_BUS_STUCK:
200 		return "Bus stuck (I2C or data shorted)";
201 	case MLX5_MODULE_EVENT_ERROR_NO_EEPROM_RETRY_TIMEOUT:
202 		return "No EEPROM/retry timeout";
203 	case MLX5_MODULE_EVENT_ERROR_ENFORCE_PART_NUMBER_LIST:
204 		return "Enforce part number list";
205 	case MLX5_MODULE_EVENT_ERROR_UNKNOWN_IDENTIFIER:
206 		return "Unknown identifier";
207 	case MLX5_MODULE_EVENT_ERROR_HIGH_TEMPERATURE:
208 		return "High Temperature";
209 	case MLX5_MODULE_EVENT_ERROR_BAD_CABLE:
210 		return "Bad or shorted cable/module";
211 	case MLX5_MODULE_EVENT_ERROR_PCIE_POWER_SLOT_EXCEEDED:
212 		return "One or more network ports have been powered down due to insufficient/unadvertised power on the PCIe slot";
213 	default:
214 		return "Unknown error";
215 	}
216 }
217 
218 /* type == MLX5_EVENT_TYPE_PORT_MODULE_EVENT */
port_module(struct notifier_block * nb,unsigned long type,void * data)219 static int port_module(struct notifier_block *nb, unsigned long type, void *data)
220 {
221 	struct mlx5_event_nb *event_nb = mlx5_nb_cof(nb, struct mlx5_event_nb, nb);
222 	struct mlx5_events   *events   = event_nb->ctx;
223 	struct mlx5_eqe      *eqe      = data;
224 
225 	enum port_module_event_status_type module_status;
226 	enum port_module_event_error_type error_type;
227 	struct mlx5_eqe_port_module *module_event_eqe;
228 	const char *status_str;
229 	u8 module_num;
230 
231 	module_event_eqe = &eqe->data.port_module;
232 	module_status = module_event_eqe->module_status &
233 			PORT_MODULE_EVENT_MODULE_STATUS_MASK;
234 	error_type = module_event_eqe->error_type &
235 		     PORT_MODULE_EVENT_ERROR_TYPE_MASK;
236 
237 	if (module_status < MLX5_MODULE_STATUS_NUM)
238 		events->pme_stats.status_counters[module_status]++;
239 
240 	if (module_status == MLX5_MODULE_STATUS_ERROR)
241 		if (error_type < MLX5_MODULE_EVENT_ERROR_NUM)
242 			events->pme_stats.error_counters[error_type]++;
243 
244 	if (!printk_ratelimit())
245 		return NOTIFY_OK;
246 
247 	module_num = module_event_eqe->module;
248 	status_str = mlx5_pme_status_to_string(module_status);
249 	if (module_status == MLX5_MODULE_STATUS_ERROR) {
250 		const char *error_str = mlx5_pme_error_to_string(error_type);
251 
252 		mlx5_core_err(events->dev,
253 			      "Port module event[error]: module %u, %s, %s\n",
254 			      module_num, status_str, error_str);
255 	} else {
256 		mlx5_core_info(events->dev,
257 			       "Port module event: module %u, %s\n",
258 			       module_num, status_str);
259 	}
260 
261 	return NOTIFY_OK;
262 }
263 
264 enum {
265 	MLX5_PCI_POWER_COULD_NOT_BE_READ = 0x0,
266 	MLX5_PCI_POWER_SUFFICIENT_REPORTED = 0x1,
267 	MLX5_PCI_POWER_INSUFFICIENT_REPORTED = 0x2,
268 };
269 
mlx5_pcie_event(struct work_struct * work)270 static void mlx5_pcie_event(struct work_struct *work)
271 {
272 	u32 out[MLX5_ST_SZ_DW(mpein_reg)] = {0};
273 	u32 in[MLX5_ST_SZ_DW(mpein_reg)] = {0};
274 	struct mlx5_events *events;
275 	struct mlx5_core_dev *dev;
276 	u8 power_status;
277 	u16 pci_power;
278 
279 	events = container_of(work, struct mlx5_events, pcie_core_work);
280 	dev  = events->dev;
281 
282 	if (!MLX5_CAP_MCAM_FEATURE(dev, pci_status_and_power))
283 		return;
284 
285 	mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out),
286 			     MLX5_REG_MPEIN, 0, 0);
287 	power_status = MLX5_GET(mpein_reg, out, pwr_status);
288 	pci_power = MLX5_GET(mpein_reg, out, pci_power);
289 
290 	switch (power_status) {
291 	case MLX5_PCI_POWER_COULD_NOT_BE_READ:
292 		mlx5_core_info_rl(dev,
293 				  "PCIe slot power capability was not advertised.\n");
294 		break;
295 	case MLX5_PCI_POWER_INSUFFICIENT_REPORTED:
296 		mlx5_core_warn_rl(dev,
297 				  "Detected insufficient power on the PCIe slot (%uW).\n",
298 				  pci_power);
299 		break;
300 	case MLX5_PCI_POWER_SUFFICIENT_REPORTED:
301 		mlx5_core_info_rl(dev,
302 				  "PCIe slot advertised sufficient power (%uW).\n",
303 				  pci_power);
304 		break;
305 	}
306 }
307 
pcie_core(struct notifier_block * nb,unsigned long type,void * data)308 static int pcie_core(struct notifier_block *nb, unsigned long type, void *data)
309 {
310 	struct mlx5_event_nb    *event_nb = mlx5_nb_cof(nb,
311 							struct mlx5_event_nb,
312 							nb);
313 	struct mlx5_events      *events   = event_nb->ctx;
314 	struct mlx5_eqe         *eqe      = data;
315 
316 	switch (eqe->sub_type) {
317 	case MLX5_GENERAL_SUBTYPE_PCI_POWER_CHANGE_EVENT:
318 			queue_work(events->wq, &events->pcie_core_work);
319 		break;
320 	default:
321 		return NOTIFY_DONE;
322 	}
323 
324 	return NOTIFY_OK;
325 }
326 
mlx5_get_pme_stats(struct mlx5_core_dev * dev,struct mlx5_pme_stats * stats)327 void mlx5_get_pme_stats(struct mlx5_core_dev *dev, struct mlx5_pme_stats *stats)
328 {
329 	*stats = dev->priv.events->pme_stats;
330 }
331 
332 /* forward event as is to registered interfaces (mlx5e/mlx5_ib) */
forward_event(struct notifier_block * nb,unsigned long event,void * data)333 static int forward_event(struct notifier_block *nb, unsigned long event, void *data)
334 {
335 	struct mlx5_event_nb *event_nb = mlx5_nb_cof(nb, struct mlx5_event_nb, nb);
336 	struct mlx5_events   *events   = event_nb->ctx;
337 	struct mlx5_eqe      *eqe      = data;
338 
339 	mlx5_core_dbg(events->dev, "Async eqe type %s, subtype (%d) forward to interfaces\n",
340 		      eqe_type_str(eqe->type), eqe->sub_type);
341 	atomic_notifier_call_chain(&events->fw_nh, event, data);
342 	return NOTIFY_OK;
343 }
344 
mlx5_events_init(struct mlx5_core_dev * dev)345 int mlx5_events_init(struct mlx5_core_dev *dev)
346 {
347 	struct mlx5_events *events = kzalloc(sizeof(*events), GFP_KERNEL);
348 
349 	if (!events)
350 		return -ENOMEM;
351 
352 	ATOMIC_INIT_NOTIFIER_HEAD(&events->fw_nh);
353 	events->dev = dev;
354 	dev->priv.events = events;
355 	events->wq = create_singlethread_workqueue("mlx5_events");
356 	if (!events->wq) {
357 		kfree(events);
358 		return -ENOMEM;
359 	}
360 	INIT_WORK(&events->pcie_core_work, mlx5_pcie_event);
361 	BLOCKING_INIT_NOTIFIER_HEAD(&events->sw_nh);
362 
363 	return 0;
364 }
365 
mlx5_events_cleanup(struct mlx5_core_dev * dev)366 void mlx5_events_cleanup(struct mlx5_core_dev *dev)
367 {
368 	destroy_workqueue(dev->priv.events->wq);
369 	kvfree(dev->priv.events);
370 }
371 
mlx5_events_start(struct mlx5_core_dev * dev)372 void mlx5_events_start(struct mlx5_core_dev *dev)
373 {
374 	struct mlx5_events *events = dev->priv.events;
375 	int i;
376 
377 	for (i = 0; i < ARRAY_SIZE(events_nbs_ref); i++) {
378 		events->notifiers[i].nb  = events_nbs_ref[i];
379 		events->notifiers[i].ctx = events;
380 		mlx5_eq_notifier_register(dev, &events->notifiers[i].nb);
381 	}
382 }
383 
mlx5_events_stop(struct mlx5_core_dev * dev)384 void mlx5_events_stop(struct mlx5_core_dev *dev)
385 {
386 	struct mlx5_events *events = dev->priv.events;
387 	int i;
388 
389 	for (i = ARRAY_SIZE(events_nbs_ref) - 1; i >= 0 ; i--)
390 		mlx5_eq_notifier_unregister(dev, &events->notifiers[i].nb);
391 	flush_workqueue(events->wq);
392 }
393 
394 /* This API is used only for processing and forwarding firmware
395  * events to mlx5 consumer.
396  */
mlx5_notifier_register(struct mlx5_core_dev * dev,struct notifier_block * nb)397 int mlx5_notifier_register(struct mlx5_core_dev *dev, struct notifier_block *nb)
398 {
399 	struct mlx5_events *events = dev->priv.events;
400 
401 	return atomic_notifier_chain_register(&events->fw_nh, nb);
402 }
403 EXPORT_SYMBOL(mlx5_notifier_register);
404 
mlx5_notifier_unregister(struct mlx5_core_dev * dev,struct notifier_block * nb)405 int mlx5_notifier_unregister(struct mlx5_core_dev *dev, struct notifier_block *nb)
406 {
407 	struct mlx5_events *events = dev->priv.events;
408 
409 	return atomic_notifier_chain_unregister(&events->fw_nh, nb);
410 }
411 EXPORT_SYMBOL(mlx5_notifier_unregister);
412 
mlx5_notifier_call_chain(struct mlx5_events * events,unsigned int event,void * data)413 int mlx5_notifier_call_chain(struct mlx5_events *events, unsigned int event, void *data)
414 {
415 	return atomic_notifier_call_chain(&events->fw_nh, event, data);
416 }
417 
418 /* This API is used only for processing and forwarding driver-specific
419  * events to mlx5 consumers.
420  */
mlx5_blocking_notifier_register(struct mlx5_core_dev * dev,struct notifier_block * nb)421 int mlx5_blocking_notifier_register(struct mlx5_core_dev *dev, struct notifier_block *nb)
422 {
423 	struct mlx5_events *events = dev->priv.events;
424 
425 	return blocking_notifier_chain_register(&events->sw_nh, nb);
426 }
427 EXPORT_SYMBOL(mlx5_blocking_notifier_register);
428 
mlx5_blocking_notifier_unregister(struct mlx5_core_dev * dev,struct notifier_block * nb)429 int mlx5_blocking_notifier_unregister(struct mlx5_core_dev *dev, struct notifier_block *nb)
430 {
431 	struct mlx5_events *events = dev->priv.events;
432 
433 	return blocking_notifier_chain_unregister(&events->sw_nh, nb);
434 }
435 EXPORT_SYMBOL(mlx5_blocking_notifier_unregister);
436 
mlx5_blocking_notifier_call_chain(struct mlx5_core_dev * dev,unsigned int event,void * data)437 int mlx5_blocking_notifier_call_chain(struct mlx5_core_dev *dev, unsigned int event,
438 				      void *data)
439 {
440 	struct mlx5_events *events = dev->priv.events;
441 
442 	return blocking_notifier_call_chain(&events->sw_nh, event, data);
443 }
444 
mlx5_events_work_enqueue(struct mlx5_core_dev * dev,struct work_struct * work)445 void mlx5_events_work_enqueue(struct mlx5_core_dev *dev, struct work_struct *work)
446 {
447 	queue_work(dev->priv.events->wq, work);
448 }
449