1 /*
2 * Copyright (c) 2005, 2006, 2007, 2008, 2014 Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34 #include <linux/interrupt.h>
35 #include <linux/slab.h>
36 #include <linux/module.h>
37 #include <linux/mm.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/hardirq.h>
40
41 #include <dev/mlx4/cmd.h>
42
43 #include "mlx4.h"
44 #include "fw.h"
45
46 enum {
47 MLX4_IRQNAME_SIZE = 32
48 };
49
50 enum {
51 MLX4_NUM_ASYNC_EQE = 0x100,
52 MLX4_NUM_SPARE_EQE = 0x80,
53 MLX4_EQ_ENTRY_SIZE = 0x20
54 };
55
56 #define MLX4_EQ_STATUS_OK ( 0 << 28)
57 #define MLX4_EQ_STATUS_WRITE_FAIL (10 << 28)
58 #define MLX4_EQ_OWNER_SW ( 0 << 24)
59 #define MLX4_EQ_OWNER_HW ( 1 << 24)
60 #define MLX4_EQ_FLAG_EC ( 1 << 18)
61 #define MLX4_EQ_FLAG_OI ( 1 << 17)
62 #define MLX4_EQ_STATE_ARMED ( 9 << 8)
63 #define MLX4_EQ_STATE_FIRED (10 << 8)
64 #define MLX4_EQ_STATE_ALWAYS_ARMED (11 << 8)
65
66 #define MLX4_ASYNC_EVENT_MASK ((1ull << MLX4_EVENT_TYPE_PATH_MIG) | \
67 (1ull << MLX4_EVENT_TYPE_COMM_EST) | \
68 (1ull << MLX4_EVENT_TYPE_SQ_DRAINED) | \
69 (1ull << MLX4_EVENT_TYPE_CQ_ERROR) | \
70 (1ull << MLX4_EVENT_TYPE_WQ_CATAS_ERROR) | \
71 (1ull << MLX4_EVENT_TYPE_EEC_CATAS_ERROR) | \
72 (1ull << MLX4_EVENT_TYPE_PATH_MIG_FAILED) | \
73 (1ull << MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR) | \
74 (1ull << MLX4_EVENT_TYPE_WQ_ACCESS_ERROR) | \
75 (1ull << MLX4_EVENT_TYPE_PORT_CHANGE) | \
76 (1ull << MLX4_EVENT_TYPE_ECC_DETECT) | \
77 (1ull << MLX4_EVENT_TYPE_SRQ_CATAS_ERROR) | \
78 (1ull << MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE) | \
79 (1ull << MLX4_EVENT_TYPE_SRQ_LIMIT) | \
80 (1ull << MLX4_EVENT_TYPE_CMD) | \
81 (1ull << MLX4_EVENT_TYPE_OP_REQUIRED) | \
82 (1ull << MLX4_EVENT_TYPE_COMM_CHANNEL) | \
83 (1ull << MLX4_EVENT_TYPE_FLR_EVENT) | \
84 (1ull << MLX4_EVENT_TYPE_FATAL_WARNING))
85
get_async_ev_mask(struct mlx4_dev * dev)86 static u64 get_async_ev_mask(struct mlx4_dev *dev)
87 {
88 u64 async_ev_mask = MLX4_ASYNC_EVENT_MASK;
89 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV)
90 async_ev_mask |= (1ull << MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT);
91 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RECOVERABLE_ERROR_EVENT)
92 async_ev_mask |= (1ull << MLX4_EVENT_TYPE_RECOVERABLE_ERROR_EVENT);
93
94 return async_ev_mask;
95 }
96
eq_set_ci(struct mlx4_eq * eq,int req_not)97 static void eq_set_ci(struct mlx4_eq *eq, int req_not)
98 {
99 __raw_writel((__force u32) cpu_to_be32((eq->cons_index & 0xffffff) |
100 req_not << 31),
101 eq->doorbell);
102 /* We still want ordering, just not swabbing, so add a barrier */
103 mb();
104 }
105
get_eqe(struct mlx4_eq * eq,u32 entry,u8 eqe_factor,u8 eqe_size)106 static struct mlx4_eqe *get_eqe(struct mlx4_eq *eq, u32 entry, u8 eqe_factor,
107 u8 eqe_size)
108 {
109 /* (entry & (eq->nent - 1)) gives us a cyclic array */
110 unsigned long offset = (entry & (eq->nent - 1)) * eqe_size;
111 /* CX3 is capable of extending the EQE from 32 to 64 bytes with
112 * strides of 64B,128B and 256B.
113 * When 64B EQE is used, the first (in the lower addresses)
114 * 32 bytes in the 64 byte EQE are reserved and the next 32 bytes
115 * contain the legacy EQE information.
116 * In all other cases, the first 32B contains the legacy EQE info.
117 */
118 return eq->page_list[offset / PAGE_SIZE].buf + (offset + (eqe_factor ? MLX4_EQ_ENTRY_SIZE : 0)) % PAGE_SIZE;
119 }
120
next_eqe_sw(struct mlx4_eq * eq,u8 eqe_factor,u8 size)121 static struct mlx4_eqe *next_eqe_sw(struct mlx4_eq *eq, u8 eqe_factor, u8 size)
122 {
123 struct mlx4_eqe *eqe = get_eqe(eq, eq->cons_index, eqe_factor, size);
124 return !!(eqe->owner & 0x80) ^ !!(eq->cons_index & eq->nent) ? NULL : eqe;
125 }
126
next_slave_event_eqe(struct mlx4_slave_event_eq * slave_eq)127 static struct mlx4_eqe *next_slave_event_eqe(struct mlx4_slave_event_eq *slave_eq)
128 {
129 struct mlx4_eqe *eqe =
130 &slave_eq->event_eqe[slave_eq->cons & (SLAVE_EVENT_EQ_SIZE - 1)];
131 return (!!(eqe->owner & 0x80) ^
132 !!(slave_eq->cons & SLAVE_EVENT_EQ_SIZE)) ?
133 eqe : NULL;
134 }
135
mlx4_gen_slave_eqe(struct work_struct * work)136 void mlx4_gen_slave_eqe(struct work_struct *work)
137 {
138 struct mlx4_mfunc_master_ctx *master =
139 container_of(work, struct mlx4_mfunc_master_ctx,
140 slave_event_work);
141 struct mlx4_mfunc *mfunc =
142 container_of(master, struct mlx4_mfunc, master);
143 struct mlx4_priv *priv = container_of(mfunc, struct mlx4_priv, mfunc);
144 struct mlx4_dev *dev = &priv->dev;
145 struct mlx4_slave_event_eq *slave_eq = &mfunc->master.slave_eq;
146 struct mlx4_eqe *eqe;
147 u8 slave;
148 int i, phys_port, slave_port;
149
150 for (eqe = next_slave_event_eqe(slave_eq); eqe;
151 eqe = next_slave_event_eqe(slave_eq)) {
152 slave = eqe->slave_id;
153
154 if (eqe->type == MLX4_EVENT_TYPE_PORT_CHANGE &&
155 eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN &&
156 mlx4_is_bonded(dev)) {
157 struct mlx4_port_cap port_cap;
158
159 if (!mlx4_QUERY_PORT(dev, 1, &port_cap) && port_cap.link_state)
160 goto consume;
161
162 if (!mlx4_QUERY_PORT(dev, 2, &port_cap) && port_cap.link_state)
163 goto consume;
164 }
165 /* All active slaves need to receive the event */
166 if (slave == ALL_SLAVES) {
167 for (i = 0; i <= dev->persist->num_vfs; i++) {
168 phys_port = 0;
169 if (eqe->type == MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT &&
170 eqe->subtype == MLX4_DEV_PMC_SUBTYPE_PORT_INFO) {
171 phys_port = eqe->event.port_mgmt_change.port;
172 slave_port = mlx4_phys_to_slave_port(dev, i, phys_port);
173 if (slave_port < 0) /* VF doesn't have this port */
174 continue;
175 eqe->event.port_mgmt_change.port = slave_port;
176 }
177 if (mlx4_GEN_EQE(dev, i, eqe))
178 mlx4_warn(dev, "Failed to generate event for slave %d\n",
179 i);
180 if (phys_port)
181 eqe->event.port_mgmt_change.port = phys_port;
182 }
183 } else {
184 if (mlx4_GEN_EQE(dev, slave, eqe))
185 mlx4_warn(dev, "Failed to generate event for slave %d\n",
186 slave);
187 }
188 consume:
189 ++slave_eq->cons;
190 }
191 }
192
193
slave_event(struct mlx4_dev * dev,u8 slave,struct mlx4_eqe * eqe)194 static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
195 {
196 struct mlx4_priv *priv = mlx4_priv(dev);
197 struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq;
198 struct mlx4_eqe *s_eqe;
199 unsigned long flags;
200
201 spin_lock_irqsave(&slave_eq->event_lock, flags);
202 s_eqe = &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
203 if ((!!(s_eqe->owner & 0x80)) ^
204 (!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) {
205 mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. No free EQE on slave events queue\n",
206 slave);
207 spin_unlock_irqrestore(&slave_eq->event_lock, flags);
208 return;
209 }
210
211 memcpy(s_eqe, eqe, sizeof(struct mlx4_eqe) - 1);
212 s_eqe->slave_id = slave;
213 /* ensure all information is written before setting the ownersip bit */
214 wmb();
215 s_eqe->owner = !!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE) ? 0x0 : 0x80;
216 ++slave_eq->prod;
217
218 queue_work(priv->mfunc.master.comm_wq,
219 &priv->mfunc.master.slave_event_work);
220 spin_unlock_irqrestore(&slave_eq->event_lock, flags);
221 }
222
mlx4_slave_event(struct mlx4_dev * dev,int slave,struct mlx4_eqe * eqe)223 static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
224 struct mlx4_eqe *eqe)
225 {
226 struct mlx4_priv *priv = mlx4_priv(dev);
227
228 if (slave < 0 || slave > dev->persist->num_vfs ||
229 slave == dev->caps.function ||
230 !priv->mfunc.master.slave_state[slave].active)
231 return;
232
233 slave_event(dev, slave, eqe);
234 }
235
mlx4_set_eq_affinity_hint(struct mlx4_priv * priv,int vec)236 static void mlx4_set_eq_affinity_hint(struct mlx4_priv *priv, int vec)
237 {
238 int hint_err;
239 struct mlx4_dev *dev = &priv->dev;
240 struct mlx4_eq *eq = &priv->eq_table.eq[vec];
241
242 hint_err = bind_irq_to_cpu(eq->irq, eq->affinity_cpu_id);
243
244 if (hint_err)
245 mlx4_warn(dev, "bind_irq_to_cpu failed, err %d\n", hint_err);
246 }
247
mlx4_gen_pkey_eqe(struct mlx4_dev * dev,int slave,u8 port)248 int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port)
249 {
250 struct mlx4_eqe eqe;
251
252 struct mlx4_priv *priv = mlx4_priv(dev);
253 struct mlx4_slave_state *s_slave = &priv->mfunc.master.slave_state[slave];
254
255 if (!s_slave->active)
256 return 0;
257
258 memset(&eqe, 0, sizeof eqe);
259
260 eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
261 eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE;
262 eqe.event.port_mgmt_change.port = mlx4_phys_to_slave_port(dev, slave, port);
263
264 return mlx4_GEN_EQE(dev, slave, &eqe);
265 }
266 EXPORT_SYMBOL(mlx4_gen_pkey_eqe);
267
mlx4_gen_guid_change_eqe(struct mlx4_dev * dev,int slave,u8 port)268 int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port)
269 {
270 struct mlx4_eqe eqe;
271
272 /*don't send if we don't have the that slave */
273 if (dev->persist->num_vfs < slave)
274 return 0;
275 memset(&eqe, 0, sizeof eqe);
276
277 eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
278 eqe.subtype = MLX4_DEV_PMC_SUBTYPE_GUID_INFO;
279 eqe.event.port_mgmt_change.port = mlx4_phys_to_slave_port(dev, slave, port);
280
281 return mlx4_GEN_EQE(dev, slave, &eqe);
282 }
283 EXPORT_SYMBOL(mlx4_gen_guid_change_eqe);
284
mlx4_gen_port_state_change_eqe(struct mlx4_dev * dev,int slave,u8 port,u8 port_subtype_change)285 int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port,
286 u8 port_subtype_change)
287 {
288 struct mlx4_eqe eqe;
289 u8 slave_port = mlx4_phys_to_slave_port(dev, slave, port);
290
291 /*don't send if we don't have the that slave */
292 if (dev->persist->num_vfs < slave)
293 return 0;
294 memset(&eqe, 0, sizeof eqe);
295
296 eqe.type = MLX4_EVENT_TYPE_PORT_CHANGE;
297 eqe.subtype = port_subtype_change;
298 eqe.event.port_change.port = cpu_to_be32(slave_port << 28);
299
300 mlx4_dbg(dev, "%s: sending: %d to slave: %d on port: %d\n", __func__,
301 port_subtype_change, slave, port);
302 return mlx4_GEN_EQE(dev, slave, &eqe);
303 }
304 EXPORT_SYMBOL(mlx4_gen_port_state_change_eqe);
305
mlx4_get_slave_port_state(struct mlx4_dev * dev,int slave,u8 port)306 enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port)
307 {
308 struct mlx4_priv *priv = mlx4_priv(dev);
309 struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
310 struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
311
312 if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
313 port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
314 pr_err("%s: Error: asking for slave:%d, port:%d\n",
315 __func__, slave, port);
316 return SLAVE_PORT_DOWN;
317 }
318 return s_state[slave].port_state[port];
319 }
320 EXPORT_SYMBOL(mlx4_get_slave_port_state);
321
mlx4_set_slave_port_state(struct mlx4_dev * dev,int slave,u8 port,enum slave_port_state state)322 static int mlx4_set_slave_port_state(struct mlx4_dev *dev, int slave, u8 port,
323 enum slave_port_state state)
324 {
325 struct mlx4_priv *priv = mlx4_priv(dev);
326 struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
327 struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
328
329 if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
330 port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
331 pr_err("%s: Error: asking for slave:%d, port:%d\n",
332 __func__, slave, port);
333 return -1;
334 }
335 s_state[slave].port_state[port] = state;
336
337 return 0;
338 }
339
set_all_slave_state(struct mlx4_dev * dev,u8 port,int event)340 static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event)
341 {
342 int i;
343 enum slave_port_gen_event gen_event;
344 struct mlx4_slaves_pport slaves_pport = mlx4_phys_to_slaves_pport(dev,
345 port);
346
347 for (i = 0; i < dev->persist->num_vfs + 1; i++)
348 if (test_bit(i, slaves_pport.slaves))
349 set_and_calc_slave_port_state(dev, i, port,
350 event, &gen_event);
351 }
352 /**************************************************************************
353 The function get as input the new event to that port,
354 and according to the prev state change the slave's port state.
355 The events are:
356 MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
357 MLX4_PORT_STATE_DEV_EVENT_PORT_UP
358 MLX4_PORT_STATE_IB_EVENT_GID_VALID
359 MLX4_PORT_STATE_IB_EVENT_GID_INVALID
360 ***************************************************************************/
set_and_calc_slave_port_state(struct mlx4_dev * dev,int slave,u8 port,int event,enum slave_port_gen_event * gen_event)361 int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave,
362 u8 port, int event,
363 enum slave_port_gen_event *gen_event)
364 {
365 struct mlx4_priv *priv = mlx4_priv(dev);
366 struct mlx4_slave_state *ctx = NULL;
367 unsigned long flags;
368 int ret = -1;
369 struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
370 enum slave_port_state cur_state =
371 mlx4_get_slave_port_state(dev, slave, port);
372
373 *gen_event = SLAVE_PORT_GEN_EVENT_NONE;
374
375 if (slave >= dev->num_slaves || port > dev->caps.num_ports ||
376 port <= 0 || !test_bit(port - 1, actv_ports.ports)) {
377 pr_err("%s: Error: asking for slave:%d, port:%d\n",
378 __func__, slave, port);
379 return ret;
380 }
381
382 ctx = &priv->mfunc.master.slave_state[slave];
383 spin_lock_irqsave(&ctx->lock, flags);
384
385 switch (cur_state) {
386 case SLAVE_PORT_DOWN:
387 if (MLX4_PORT_STATE_DEV_EVENT_PORT_UP == event)
388 mlx4_set_slave_port_state(dev, slave, port,
389 SLAVE_PENDING_UP);
390 break;
391 case SLAVE_PENDING_UP:
392 if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event)
393 mlx4_set_slave_port_state(dev, slave, port,
394 SLAVE_PORT_DOWN);
395 else if (MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID == event) {
396 mlx4_set_slave_port_state(dev, slave, port,
397 SLAVE_PORT_UP);
398 *gen_event = SLAVE_PORT_GEN_EVENT_UP;
399 }
400 break;
401 case SLAVE_PORT_UP:
402 if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) {
403 mlx4_set_slave_port_state(dev, slave, port,
404 SLAVE_PORT_DOWN);
405 *gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
406 } else if (MLX4_PORT_STATE_IB_EVENT_GID_INVALID ==
407 event) {
408 mlx4_set_slave_port_state(dev, slave, port,
409 SLAVE_PENDING_UP);
410 *gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
411 }
412 break;
413 default:
414 pr_err("%s: BUG!!! UNKNOWN state: slave:%d, port:%d\n",
415 __func__, slave, port);
416 goto out;
417 }
418 ret = mlx4_get_slave_port_state(dev, slave, port);
419
420 out:
421 spin_unlock_irqrestore(&ctx->lock, flags);
422 return ret;
423 }
424
425 EXPORT_SYMBOL(set_and_calc_slave_port_state);
426
mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev * dev,u8 port,int attr)427 int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr)
428 {
429 struct mlx4_eqe eqe;
430
431 memset(&eqe, 0, sizeof eqe);
432
433 eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
434 eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PORT_INFO;
435 eqe.event.port_mgmt_change.port = port;
436 eqe.event.port_mgmt_change.params.port_info.changed_attr =
437 cpu_to_be32((u32) attr);
438
439 slave_event(dev, ALL_SLAVES, &eqe);
440 return 0;
441 }
442 EXPORT_SYMBOL(mlx4_gen_slaves_port_mgt_ev);
443
mlx4_master_handle_slave_flr(struct work_struct * work)444 void mlx4_master_handle_slave_flr(struct work_struct *work)
445 {
446 struct mlx4_mfunc_master_ctx *master =
447 container_of(work, struct mlx4_mfunc_master_ctx,
448 slave_flr_event_work);
449 struct mlx4_mfunc *mfunc =
450 container_of(master, struct mlx4_mfunc, master);
451 struct mlx4_priv *priv =
452 container_of(mfunc, struct mlx4_priv, mfunc);
453 struct mlx4_dev *dev = &priv->dev;
454 struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
455 int i;
456 int err;
457 unsigned long flags;
458
459 mlx4_dbg(dev, "mlx4_handle_slave_flr\n");
460
461 for (i = 0 ; i < dev->num_slaves; i++) {
462
463 if (MLX4_COMM_CMD_FLR == slave_state[i].last_cmd) {
464 mlx4_dbg(dev, "mlx4_handle_slave_flr: clean slave: %d\n",
465 i);
466 /* In case of 'Reset flow' FLR can be generated for
467 * a slave before mlx4_load_one is done.
468 * make sure interface is up before trying to delete
469 * slave resources which weren't allocated yet.
470 */
471 if (dev->persist->interface_state &
472 MLX4_INTERFACE_STATE_UP)
473 mlx4_delete_all_resources_for_slave(dev, i);
474 /*return the slave to running mode*/
475 spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
476 slave_state[i].last_cmd = MLX4_COMM_CMD_RESET;
477 slave_state[i].is_slave_going_down = 0;
478 spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
479 /*notify the FW:*/
480 err = mlx4_cmd(dev, 0, i, 0, MLX4_CMD_INFORM_FLR_DONE,
481 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
482 if (err)
483 mlx4_warn(dev, "Failed to notify FW on FLR done (slave:%d)\n",
484 i);
485 }
486 }
487 }
488
mlx4_eq_int(struct mlx4_dev * dev,struct mlx4_eq * eq)489 static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
490 {
491 struct mlx4_priv *priv = mlx4_priv(dev);
492 struct mlx4_eqe *eqe;
493 int cqn = -1;
494 int eqes_found = 0;
495 int set_ci = 0;
496 int port;
497 int slave = 0;
498 int ret;
499 u32 flr_slave;
500 u8 update_slave_state;
501 int i;
502 enum slave_port_gen_event gen_event;
503 unsigned long flags;
504 struct mlx4_vport_state *s_info;
505 int eqe_size = dev->caps.eqe_size;
506
507 while ((eqe = next_eqe_sw(eq, dev->caps.eqe_factor, eqe_size))) {
508 /*
509 * Make sure we read EQ entry contents after we've
510 * checked the ownership bit.
511 */
512 rmb();
513
514 switch (eqe->type) {
515 case MLX4_EVENT_TYPE_COMP:
516 cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
517 mlx4_cq_completion(dev, cqn);
518 break;
519
520 case MLX4_EVENT_TYPE_PATH_MIG:
521 case MLX4_EVENT_TYPE_COMM_EST:
522 case MLX4_EVENT_TYPE_SQ_DRAINED:
523 case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE:
524 case MLX4_EVENT_TYPE_WQ_CATAS_ERROR:
525 case MLX4_EVENT_TYPE_PATH_MIG_FAILED:
526 case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
527 case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR:
528 mlx4_dbg(dev, "event %d arrived\n", eqe->type);
529 if (mlx4_is_master(dev)) {
530 /* forward only to slave owning the QP */
531 ret = mlx4_get_slave_from_resource_id(dev,
532 RES_QP,
533 be32_to_cpu(eqe->event.qp.qpn)
534 & 0xffffff, &slave);
535 if (ret && ret != -ENOENT) {
536 mlx4_dbg(dev, "QP event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
537 eqe->type, eqe->subtype,
538 eq->eqn, eq->cons_index, ret);
539 break;
540 }
541
542 if (!ret && slave != dev->caps.function) {
543 mlx4_slave_event(dev, slave, eqe);
544 break;
545 }
546
547 }
548 mlx4_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) &
549 0xffffff, eqe->type);
550 break;
551
552 case MLX4_EVENT_TYPE_SRQ_LIMIT:
553 mlx4_dbg(dev, "%s: MLX4_EVENT_TYPE_SRQ_LIMIT\n",
554 __func__);
555 case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
556 if (mlx4_is_master(dev)) {
557 /* forward only to slave owning the SRQ */
558 ret = mlx4_get_slave_from_resource_id(dev,
559 RES_SRQ,
560 be32_to_cpu(eqe->event.srq.srqn)
561 & 0xffffff,
562 &slave);
563 if (ret && ret != -ENOENT) {
564 mlx4_warn(dev, "SRQ event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
565 eqe->type, eqe->subtype,
566 eq->eqn, eq->cons_index, ret);
567 break;
568 }
569 mlx4_warn(dev, "%s: slave:%d, srq_no:0x%x, event: %02x(%02x)\n",
570 __func__, slave,
571 be32_to_cpu(eqe->event.srq.srqn),
572 eqe->type, eqe->subtype);
573
574 if (!ret && slave != dev->caps.function) {
575 mlx4_warn(dev, "%s: sending event %02x(%02x) to slave:%d\n",
576 __func__, eqe->type,
577 eqe->subtype, slave);
578 mlx4_slave_event(dev, slave, eqe);
579 break;
580 }
581 }
582 mlx4_srq_event(dev, be32_to_cpu(eqe->event.srq.srqn) &
583 0xffffff, eqe->type);
584 break;
585
586 case MLX4_EVENT_TYPE_CMD:
587 mlx4_cmd_event(dev,
588 be16_to_cpu(eqe->event.cmd.token),
589 eqe->event.cmd.status,
590 be64_to_cpu(eqe->event.cmd.out_param));
591 break;
592
593 case MLX4_EVENT_TYPE_PORT_CHANGE: {
594 struct mlx4_slaves_pport slaves_port;
595 port = be32_to_cpu(eqe->event.port_change.port) >> 28;
596 slaves_port = mlx4_phys_to_slaves_pport(dev, port);
597 if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) {
598 mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_DOWN,
599 port);
600 mlx4_priv(dev)->sense.do_sense_port[port] = 1;
601 if (!mlx4_is_master(dev))
602 break;
603 for (i = 0; i < dev->persist->num_vfs + 1;
604 i++) {
605 int reported_port = mlx4_is_bonded(dev) ? 1 : mlx4_phys_to_slave_port(dev, i, port);
606
607 if (!test_bit(i, slaves_port.slaves) && !mlx4_is_bonded(dev))
608 continue;
609 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) {
610 if (i == mlx4_master_func_num(dev))
611 continue;
612 mlx4_dbg(dev, "%s: Sending MLX4_PORT_CHANGE_SUBTYPE_DOWN to slave: %d, port:%d\n",
613 __func__, i, port);
614 s_info = &priv->mfunc.master.vf_oper[i].vport[port].state;
615 if (0 /*IFLA_VF_LINK_STATE_AUTO == s_info->link_state*/) {
616 eqe->event.port_change.port =
617 cpu_to_be32(
618 (be32_to_cpu(eqe->event.port_change.port) & 0xFFFFFFF)
619 | (reported_port << 28));
620 mlx4_slave_event(dev, i, eqe);
621 }
622 } else { /* IB port */
623 set_and_calc_slave_port_state(dev, i, port,
624 MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
625 &gen_event);
626 /*we can be in pending state, then do not send port_down event*/
627 if (SLAVE_PORT_GEN_EVENT_DOWN == gen_event) {
628 if (i == mlx4_master_func_num(dev))
629 continue;
630 eqe->event.port_change.port =
631 cpu_to_be32(
632 (be32_to_cpu(eqe->event.port_change.port) & 0xFFFFFFF)
633 | (mlx4_phys_to_slave_port(dev, i, port) << 28));
634 mlx4_slave_event(dev, i, eqe);
635 }
636 }
637 }
638 } else {
639 mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_UP, port);
640
641 mlx4_priv(dev)->sense.do_sense_port[port] = 0;
642
643 if (!mlx4_is_master(dev))
644 break;
645 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
646 for (i = 0;
647 i < dev->persist->num_vfs + 1;
648 i++) {
649 int reported_port = mlx4_is_bonded(dev) ? 1 : mlx4_phys_to_slave_port(dev, i, port);
650
651 if (!test_bit(i, slaves_port.slaves) && !mlx4_is_bonded(dev))
652 continue;
653 if (i == mlx4_master_func_num(dev))
654 continue;
655 s_info = &priv->mfunc.master.vf_oper[i].vport[port].state;
656 if (0 /*IFLA_VF_LINK_STATE_AUTO == s_info->link_state*/) {
657 eqe->event.port_change.port =
658 cpu_to_be32(
659 (be32_to_cpu(eqe->event.port_change.port) & 0xFFFFFFF)
660 | (reported_port << 28));
661 mlx4_slave_event(dev, i, eqe);
662 }
663 }
664 else /* IB port */
665 /* port-up event will be sent to a slave when the
666 * slave's alias-guid is set. This is done in alias_GUID.c
667 */
668 set_all_slave_state(dev, port, MLX4_DEV_EVENT_PORT_UP);
669 }
670 break;
671 }
672
673 case MLX4_EVENT_TYPE_CQ_ERROR:
674 mlx4_warn(dev, "CQ %s on CQN %06x\n",
675 eqe->event.cq_err.syndrome == 1 ?
676 "overrun" : "access violation",
677 be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff);
678 if (mlx4_is_master(dev)) {
679 ret = mlx4_get_slave_from_resource_id(dev,
680 RES_CQ,
681 be32_to_cpu(eqe->event.cq_err.cqn)
682 & 0xffffff, &slave);
683 if (ret && ret != -ENOENT) {
684 mlx4_dbg(dev, "CQ event %02x(%02x) on EQ %d at index %u: could not get slave id (%d)\n",
685 eqe->type, eqe->subtype,
686 eq->eqn, eq->cons_index, ret);
687 break;
688 }
689
690 if (!ret && slave != dev->caps.function) {
691 mlx4_slave_event(dev, slave, eqe);
692 break;
693 }
694 }
695 mlx4_cq_event(dev,
696 be32_to_cpu(eqe->event.cq_err.cqn)
697 & 0xffffff,
698 eqe->type);
699 break;
700
701 case MLX4_EVENT_TYPE_EQ_OVERFLOW:
702 mlx4_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
703 break;
704
705 case MLX4_EVENT_TYPE_OP_REQUIRED:
706 atomic_inc(&priv->opreq_count);
707 /* FW commands can't be executed from interrupt context
708 * working in deferred task
709 */
710 queue_work(mlx4_wq, &priv->opreq_task);
711 break;
712
713 case MLX4_EVENT_TYPE_COMM_CHANNEL:
714 if (!mlx4_is_master(dev)) {
715 mlx4_warn(dev, "Received comm channel event for non master device\n");
716 break;
717 }
718 memcpy(&priv->mfunc.master.comm_arm_bit_vector,
719 eqe->event.comm_channel_arm.bit_vec,
720 sizeof eqe->event.comm_channel_arm.bit_vec);
721 queue_work(priv->mfunc.master.comm_wq,
722 &priv->mfunc.master.comm_work);
723 break;
724
725 case MLX4_EVENT_TYPE_FLR_EVENT:
726 flr_slave = be32_to_cpu(eqe->event.flr_event.slave_id);
727 if (!mlx4_is_master(dev)) {
728 mlx4_warn(dev, "Non-master function received FLR event\n");
729 break;
730 }
731
732 mlx4_dbg(dev, "FLR event for slave: %d\n", flr_slave);
733
734 if (flr_slave >= dev->num_slaves) {
735 mlx4_warn(dev,
736 "Got FLR for unknown function: %d\n",
737 flr_slave);
738 update_slave_state = 0;
739 } else
740 update_slave_state = 1;
741
742 spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
743 if (update_slave_state) {
744 priv->mfunc.master.slave_state[flr_slave].active = false;
745 priv->mfunc.master.slave_state[flr_slave].last_cmd = MLX4_COMM_CMD_FLR;
746 priv->mfunc.master.slave_state[flr_slave].is_slave_going_down = 1;
747 }
748 spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
749 mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_SHUTDOWN,
750 flr_slave);
751 queue_work(priv->mfunc.master.comm_wq,
752 &priv->mfunc.master.slave_flr_event_work);
753 break;
754
755 case MLX4_EVENT_TYPE_FATAL_WARNING:
756 if (eqe->subtype == MLX4_FATAL_WARNING_SUBTYPE_WARMING) {
757 if (mlx4_is_master(dev))
758 for (i = 0; i < dev->num_slaves; i++) {
759 mlx4_dbg(dev, "%s: Sending MLX4_FATAL_WARNING_SUBTYPE_WARMING to slave: %d\n",
760 __func__, i);
761 if (i == dev->caps.function)
762 continue;
763 mlx4_slave_event(dev, i, eqe);
764 }
765 mlx4_err(dev, "Temperature Threshold was reached! Threshold: %d celsius degrees; Current Temperature: %d\n",
766 be16_to_cpu(eqe->event.warming.warning_threshold),
767 be16_to_cpu(eqe->event.warming.current_temperature));
768 } else
769 mlx4_warn(dev, "Unhandled event FATAL WARNING (%02x), subtype %02x on EQ %d at index %u. owner=%x, nent=0x%x, slave=%x, ownership=%s\n",
770 eqe->type, eqe->subtype, eq->eqn,
771 eq->cons_index, eqe->owner, eq->nent,
772 eqe->slave_id,
773 !!(eqe->owner & 0x80) ^
774 !!(eq->cons_index & eq->nent) ? "HW" : "SW");
775
776 break;
777
778 case MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT:
779 mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_MGMT_CHANGE,
780 (unsigned long) eqe);
781 break;
782
783 case MLX4_EVENT_TYPE_RECOVERABLE_ERROR_EVENT:
784 switch (eqe->subtype) {
785 case MLX4_RECOVERABLE_ERROR_EVENT_SUBTYPE_BAD_CABLE:
786 mlx4_warn(dev, "Bad cable detected on port %u\n",
787 eqe->event.bad_cable.port);
788 break;
789 case MLX4_RECOVERABLE_ERROR_EVENT_SUBTYPE_UNSUPPORTED_CABLE:
790 mlx4_warn(dev, "Unsupported cable detected\n");
791 break;
792 default:
793 mlx4_dbg(dev,
794 "Unhandled recoverable error event detected: %02x(%02x) on EQ %d at index %u. owner=%x, nent=0x%x, ownership=%s\n",
795 eqe->type, eqe->subtype, eq->eqn,
796 eq->cons_index, eqe->owner, eq->nent,
797 !!(eqe->owner & 0x80) ^
798 !!(eq->cons_index & eq->nent) ? "HW" : "SW");
799 break;
800 }
801 break;
802
803 case MLX4_EVENT_TYPE_EEC_CATAS_ERROR:
804 case MLX4_EVENT_TYPE_ECC_DETECT:
805 default:
806 mlx4_warn(dev, "Unhandled event %02x(%02x) on EQ %d at index %u. owner=%x, nent=0x%x, slave=%x, ownership=%s\n",
807 eqe->type, eqe->subtype, eq->eqn,
808 eq->cons_index, eqe->owner, eq->nent,
809 eqe->slave_id,
810 !!(eqe->owner & 0x80) ^
811 !!(eq->cons_index & eq->nent) ? "HW" : "SW");
812 break;
813 }
814
815 ++eq->cons_index;
816 eqes_found = 1;
817 ++set_ci;
818
819 /*
820 * The HCA will think the queue has overflowed if we
821 * don't tell it we've been processing events. We
822 * create our EQs with MLX4_NUM_SPARE_EQE extra
823 * entries, so we must update our consumer index at
824 * least that often.
825 */
826 if (unlikely(set_ci >= MLX4_NUM_SPARE_EQE)) {
827 eq_set_ci(eq, 0);
828 set_ci = 0;
829 }
830 }
831
832 eq_set_ci(eq, 1);
833
834 return eqes_found;
835 }
836
mlx4_interrupt(int irq,void * dev_ptr)837 static irqreturn_t mlx4_interrupt(int irq, void *dev_ptr)
838 {
839 struct mlx4_dev *dev = dev_ptr;
840 struct mlx4_priv *priv = mlx4_priv(dev);
841 int work = 0;
842 int i;
843
844 writel(priv->eq_table.clr_mask, priv->eq_table.clr_int);
845
846 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
847 work |= mlx4_eq_int(dev, &priv->eq_table.eq[i]);
848
849 return IRQ_RETVAL(work);
850 }
851
mlx4_msi_x_interrupt(int irq,void * eq_ptr)852 static irqreturn_t mlx4_msi_x_interrupt(int irq, void *eq_ptr)
853 {
854 struct mlx4_eq *eq = eq_ptr;
855 struct mlx4_dev *dev = eq->dev;
856
857 mlx4_eq_int(dev, eq);
858
859 /* MSI-X vectors always belong to us */
860 return IRQ_HANDLED;
861 }
862
mlx4_MAP_EQ_wrapper(struct mlx4_dev * dev,int slave,struct mlx4_vhcr * vhcr,struct mlx4_cmd_mailbox * inbox,struct mlx4_cmd_mailbox * outbox,struct mlx4_cmd_info * cmd)863 int mlx4_MAP_EQ_wrapper(struct mlx4_dev *dev, int slave,
864 struct mlx4_vhcr *vhcr,
865 struct mlx4_cmd_mailbox *inbox,
866 struct mlx4_cmd_mailbox *outbox,
867 struct mlx4_cmd_info *cmd)
868 {
869 struct mlx4_priv *priv = mlx4_priv(dev);
870 struct mlx4_slave_event_eq_info *event_eq =
871 priv->mfunc.master.slave_state[slave].event_eq;
872 u32 in_modifier = vhcr->in_modifier;
873 u32 eqn = in_modifier & 0x3FF;
874 u64 in_param = vhcr->in_param;
875 int err = 0;
876 int i;
877
878 if (slave == dev->caps.function)
879 err = mlx4_cmd(dev, in_param, (in_modifier & 0x80000000) | eqn,
880 0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
881 MLX4_CMD_NATIVE);
882 if (!err)
883 for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i)
884 if (in_param & (1LL << i))
885 event_eq[i].eqn = in_modifier >> 31 ? -1 : eqn;
886
887 return err;
888 }
889
mlx4_MAP_EQ(struct mlx4_dev * dev,u64 event_mask,int unmap,int eq_num)890 static int mlx4_MAP_EQ(struct mlx4_dev *dev, u64 event_mask, int unmap,
891 int eq_num)
892 {
893 return mlx4_cmd(dev, event_mask, (unmap << 31) | eq_num,
894 0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
895 MLX4_CMD_WRAPPED);
896 }
897
mlx4_SW2HW_EQ(struct mlx4_dev * dev,struct mlx4_cmd_mailbox * mailbox,int eq_num)898 static int mlx4_SW2HW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
899 int eq_num)
900 {
901 return mlx4_cmd(dev, mailbox->dma, eq_num, 0,
902 MLX4_CMD_SW2HW_EQ, MLX4_CMD_TIME_CLASS_A,
903 MLX4_CMD_WRAPPED);
904 }
905
mlx4_HW2SW_EQ(struct mlx4_dev * dev,int eq_num)906 static int mlx4_HW2SW_EQ(struct mlx4_dev *dev, int eq_num)
907 {
908 return mlx4_cmd(dev, 0, eq_num, 1, MLX4_CMD_HW2SW_EQ,
909 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
910 }
911
mlx4_num_eq_uar(struct mlx4_dev * dev)912 static int mlx4_num_eq_uar(struct mlx4_dev *dev)
913 {
914 /*
915 * Each UAR holds 4 EQ doorbells. To figure out how many UARs
916 * we need to map, take the difference of highest index and
917 * the lowest index we'll use and add 1.
918 */
919 return (dev->caps.num_comp_vectors + 1 + dev->caps.reserved_eqs) / 4 -
920 dev->caps.reserved_eqs / 4 + 1;
921 }
922
mlx4_get_eq_uar(struct mlx4_dev * dev,struct mlx4_eq * eq)923 static void __iomem *mlx4_get_eq_uar(struct mlx4_dev *dev, struct mlx4_eq *eq)
924 {
925 struct mlx4_priv *priv = mlx4_priv(dev);
926 int index;
927
928 index = eq->eqn / 4 - dev->caps.reserved_eqs / 4;
929
930 if (!priv->eq_table.uar_map[index]) {
931 priv->eq_table.uar_map[index] =
932 ioremap(pci_resource_start(dev->persist->pdev, 2) +
933 ((eq->eqn / 4) << (dev->uar_page_shift)),
934 (1 << (dev->uar_page_shift)));
935 if (!priv->eq_table.uar_map[index]) {
936 mlx4_err(dev, "Couldn't map EQ doorbell for EQN 0x%06x\n",
937 eq->eqn);
938 return NULL;
939 }
940 }
941
942 return priv->eq_table.uar_map[index] + 0x800 + 8 * (eq->eqn % 4);
943 }
944
mlx4_unmap_uar(struct mlx4_dev * dev)945 static void mlx4_unmap_uar(struct mlx4_dev *dev)
946 {
947 struct mlx4_priv *priv = mlx4_priv(dev);
948 int i;
949
950 for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
951 if (priv->eq_table.uar_map[i]) {
952 iounmap(priv->eq_table.uar_map[i]);
953 priv->eq_table.uar_map[i] = NULL;
954 }
955 }
956
mlx4_create_eq(struct mlx4_dev * dev,int nent,u8 intr,struct mlx4_eq * eq)957 static int mlx4_create_eq(struct mlx4_dev *dev, int nent,
958 u8 intr, struct mlx4_eq *eq)
959 {
960 struct mlx4_priv *priv = mlx4_priv(dev);
961 struct mlx4_cmd_mailbox *mailbox;
962 struct mlx4_eq_context *eq_context;
963 int npages;
964 u64 *dma_list = NULL;
965 dma_addr_t t;
966 u64 mtt_addr;
967 int err = -ENOMEM;
968 int i;
969
970 eq->dev = dev;
971 eq->nent = roundup_pow_of_two(max(nent, 2));
972 /* CX3 is capable of extending the CQE/EQE from 32 to 64 bytes, with
973 * strides of 64B,128B and 256B.
974 */
975 npages = PAGE_ALIGN(eq->nent * dev->caps.eqe_size) / PAGE_SIZE;
976
977 eq->page_list = kmalloc(npages * sizeof *eq->page_list,
978 GFP_KERNEL);
979 if (!eq->page_list)
980 goto err_out;
981
982 for (i = 0; i < npages; ++i)
983 eq->page_list[i].buf = NULL;
984
985 dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
986 if (!dma_list)
987 goto err_out_free;
988
989 mailbox = mlx4_alloc_cmd_mailbox(dev);
990 if (IS_ERR(mailbox))
991 goto err_out_free;
992 eq_context = mailbox->buf;
993
994 for (i = 0; i < npages; ++i) {
995 eq->page_list[i].buf = dma_alloc_coherent(&dev->persist->
996 pdev->dev,
997 PAGE_SIZE, &t,
998 GFP_KERNEL);
999 if (!eq->page_list[i].buf)
1000 goto err_out_free_pages;
1001
1002 dma_list[i] = t;
1003 eq->page_list[i].map = t;
1004
1005 memset(eq->page_list[i].buf, 0, PAGE_SIZE);
1006 }
1007
1008 eq->eqn = mlx4_bitmap_alloc(&priv->eq_table.bitmap);
1009 if (eq->eqn == -1)
1010 goto err_out_free_pages;
1011
1012 eq->doorbell = mlx4_get_eq_uar(dev, eq);
1013 if (!eq->doorbell) {
1014 err = -ENOMEM;
1015 goto err_out_free_eq;
1016 }
1017
1018 err = mlx4_mtt_init(dev, npages, PAGE_SHIFT, &eq->mtt);
1019 if (err)
1020 goto err_out_free_eq;
1021
1022 err = mlx4_write_mtt(dev, &eq->mtt, 0, npages, dma_list);
1023 if (err)
1024 goto err_out_free_mtt;
1025
1026 eq_context->flags = cpu_to_be32(MLX4_EQ_STATUS_OK |
1027 MLX4_EQ_STATE_ARMED);
1028 eq_context->log_eq_size = ilog2(eq->nent);
1029 eq_context->intr = intr;
1030 eq_context->log_page_size = PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT;
1031
1032 mtt_addr = mlx4_mtt_addr(dev, &eq->mtt);
1033 eq_context->mtt_base_addr_h = mtt_addr >> 32;
1034 eq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff);
1035
1036 err = mlx4_SW2HW_EQ(dev, mailbox, eq->eqn);
1037 if (err) {
1038 mlx4_warn(dev, "SW2HW_EQ failed (%d)\n", err);
1039 goto err_out_free_mtt;
1040 }
1041
1042 kfree(dma_list);
1043 mlx4_free_cmd_mailbox(dev, mailbox);
1044
1045 eq->cons_index = 0;
1046
1047 return err;
1048
1049 err_out_free_mtt:
1050 mlx4_mtt_cleanup(dev, &eq->mtt);
1051
1052 err_out_free_eq:
1053 mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn, MLX4_USE_RR);
1054
1055 err_out_free_pages:
1056 for (i = 0; i < npages; ++i)
1057 if (eq->page_list[i].buf)
1058 dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
1059 eq->page_list[i].buf,
1060 eq->page_list[i].map);
1061
1062 mlx4_free_cmd_mailbox(dev, mailbox);
1063
1064 err_out_free:
1065 kfree(eq->page_list);
1066 kfree(dma_list);
1067
1068 err_out:
1069 return err;
1070 }
1071
mlx4_free_eq(struct mlx4_dev * dev,struct mlx4_eq * eq)1072 static void mlx4_free_eq(struct mlx4_dev *dev,
1073 struct mlx4_eq *eq)
1074 {
1075 struct mlx4_priv *priv = mlx4_priv(dev);
1076 int err;
1077 int i;
1078 /* CX3 is capable of extending the CQE/EQE from 32 to 64 bytes, with
1079 * strides of 64B,128B and 256B
1080 */
1081 int npages = PAGE_ALIGN(dev->caps.eqe_size * eq->nent) / PAGE_SIZE;
1082
1083 err = mlx4_HW2SW_EQ(dev, eq->eqn);
1084 if (err)
1085 mlx4_warn(dev, "HW2SW_EQ failed (%d)\n", err);
1086
1087 synchronize_irq(eq->irq);
1088
1089 mlx4_mtt_cleanup(dev, &eq->mtt);
1090 for (i = 0; i < npages; ++i)
1091 dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
1092 eq->page_list[i].buf,
1093 eq->page_list[i].map);
1094
1095 kfree(eq->page_list);
1096 mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn, MLX4_USE_RR);
1097 }
1098
mlx4_free_irqs(struct mlx4_dev * dev)1099 static void mlx4_free_irqs(struct mlx4_dev *dev)
1100 {
1101 struct mlx4_eq_table *eq_table = &mlx4_priv(dev)->eq_table;
1102 int i;
1103
1104 if (eq_table->have_irq)
1105 free_irq(dev->persist->pdev->irq, dev);
1106
1107 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1108 if (eq_table->eq[i].have_irq) {
1109 eq_table->eq[i].affinity_cpu_id = NOCPU;
1110 free_irq(eq_table->eq[i].irq, eq_table->eq + i);
1111 eq_table->eq[i].have_irq = 0;
1112 }
1113
1114 kfree(eq_table->irq_names);
1115 }
1116
mlx4_map_clr_int(struct mlx4_dev * dev)1117 static int mlx4_map_clr_int(struct mlx4_dev *dev)
1118 {
1119 struct mlx4_priv *priv = mlx4_priv(dev);
1120
1121 priv->clr_base = ioremap(pci_resource_start(dev->persist->pdev,
1122 priv->fw.clr_int_bar) +
1123 priv->fw.clr_int_base, MLX4_CLR_INT_SIZE);
1124 if (!priv->clr_base) {
1125 mlx4_err(dev, "Couldn't map interrupt clear register, aborting\n");
1126 return -ENOMEM;
1127 }
1128
1129 return 0;
1130 }
1131
mlx4_unmap_clr_int(struct mlx4_dev * dev)1132 static void mlx4_unmap_clr_int(struct mlx4_dev *dev)
1133 {
1134 struct mlx4_priv *priv = mlx4_priv(dev);
1135
1136 iounmap(priv->clr_base);
1137 }
1138
mlx4_alloc_eq_table(struct mlx4_dev * dev)1139 int mlx4_alloc_eq_table(struct mlx4_dev *dev)
1140 {
1141 struct mlx4_priv *priv = mlx4_priv(dev);
1142
1143 priv->eq_table.eq = kcalloc(dev->caps.num_eqs - dev->caps.reserved_eqs,
1144 sizeof *priv->eq_table.eq, GFP_KERNEL);
1145 if (!priv->eq_table.eq)
1146 return -ENOMEM;
1147
1148 return 0;
1149 }
1150
mlx4_free_eq_table(struct mlx4_dev * dev)1151 void mlx4_free_eq_table(struct mlx4_dev *dev)
1152 {
1153 kfree(mlx4_priv(dev)->eq_table.eq);
1154 }
1155
mlx4_init_eq_table(struct mlx4_dev * dev)1156 int mlx4_init_eq_table(struct mlx4_dev *dev)
1157 {
1158 struct mlx4_priv *priv = mlx4_priv(dev);
1159 int err;
1160 int i;
1161
1162 priv->eq_table.uar_map = kcalloc(mlx4_num_eq_uar(dev),
1163 sizeof *priv->eq_table.uar_map,
1164 GFP_KERNEL);
1165 if (!priv->eq_table.uar_map) {
1166 err = -ENOMEM;
1167 goto err_out_free;
1168 }
1169
1170 err = mlx4_bitmap_init(&priv->eq_table.bitmap,
1171 roundup_pow_of_two(dev->caps.num_eqs),
1172 dev->caps.num_eqs - 1,
1173 dev->caps.reserved_eqs,
1174 roundup_pow_of_two(dev->caps.num_eqs) -
1175 dev->caps.num_eqs);
1176 if (err)
1177 goto err_out_free;
1178
1179 for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
1180 priv->eq_table.uar_map[i] = NULL;
1181
1182 if (!mlx4_is_slave(dev)) {
1183 err = mlx4_map_clr_int(dev);
1184 if (err)
1185 goto err_out_bitmap;
1186
1187 priv->eq_table.clr_mask =
1188 swab32(1 << (priv->eq_table.inta_pin & 31));
1189 priv->eq_table.clr_int = priv->clr_base +
1190 (priv->eq_table.inta_pin < 32 ? 4 : 0);
1191 }
1192
1193 priv->eq_table.irq_names =
1194 kmalloc(MLX4_IRQNAME_SIZE * (dev->caps.num_comp_vectors + 1),
1195 GFP_KERNEL);
1196 if (!priv->eq_table.irq_names) {
1197 err = -ENOMEM;
1198 goto err_out_clr_int;
1199 }
1200
1201 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i) {
1202 if (i == MLX4_EQ_ASYNC) {
1203 err = mlx4_create_eq(dev,
1204 MLX4_NUM_ASYNC_EQE + MLX4_NUM_SPARE_EQE,
1205 0, &priv->eq_table.eq[MLX4_EQ_ASYNC]);
1206 } else {
1207 struct mlx4_eq *eq = &priv->eq_table.eq[i];
1208 #ifdef CONFIG_RFS_ACCEL
1209 int port = find_first_bit(eq->actv_ports.ports,
1210 dev->caps.num_ports) + 1;
1211
1212 if (port <= dev->caps.num_ports) {
1213 struct mlx4_port_info *info =
1214 &mlx4_priv(dev)->port[port];
1215
1216 if (!info->rmap) {
1217 info->rmap = alloc_irq_cpu_rmap(
1218 mlx4_get_eqs_per_port(dev, port));
1219 if (!info->rmap) {
1220 mlx4_warn(dev, "Failed to allocate cpu rmap\n");
1221 err = -ENOMEM;
1222 goto err_out_unmap;
1223 }
1224 }
1225
1226 err = irq_cpu_rmap_add(
1227 info->rmap, eq->irq);
1228 if (err)
1229 mlx4_warn(dev, "Failed adding irq rmap\n");
1230 }
1231 #endif
1232 err = mlx4_create_eq(dev, dev->quotas.cq +
1233 MLX4_NUM_SPARE_EQE,
1234 (dev->flags & MLX4_FLAG_MSI_X) ?
1235 i + 1 - !!(i > MLX4_EQ_ASYNC) : 0,
1236 eq);
1237 }
1238 if (err)
1239 goto err_out_unmap;
1240 }
1241
1242 if (dev->flags & MLX4_FLAG_MSI_X) {
1243 const char *eq_name;
1244
1245 snprintf(priv->eq_table.irq_names +
1246 MLX4_EQ_ASYNC * MLX4_IRQNAME_SIZE,
1247 MLX4_IRQNAME_SIZE,
1248 "mlx4-async@pci:%s",
1249 pci_name(dev->persist->pdev));
1250 eq_name = priv->eq_table.irq_names +
1251 MLX4_EQ_ASYNC * MLX4_IRQNAME_SIZE;
1252
1253 err = request_irq(priv->eq_table.eq[MLX4_EQ_ASYNC].irq,
1254 mlx4_msi_x_interrupt, 0, eq_name,
1255 priv->eq_table.eq + MLX4_EQ_ASYNC);
1256 if (err)
1257 goto err_out_unmap;
1258
1259 priv->eq_table.eq[MLX4_EQ_ASYNC].have_irq = 1;
1260 } else {
1261 snprintf(priv->eq_table.irq_names,
1262 MLX4_IRQNAME_SIZE,
1263 DRV_NAME "@pci:%s",
1264 pci_name(dev->persist->pdev));
1265 err = request_irq(dev->persist->pdev->irq, mlx4_interrupt,
1266 IRQF_SHARED, priv->eq_table.irq_names, dev);
1267 if (err)
1268 goto err_out_unmap;
1269
1270 priv->eq_table.have_irq = 1;
1271 }
1272
1273 err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1274 priv->eq_table.eq[MLX4_EQ_ASYNC].eqn);
1275 if (err)
1276 mlx4_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
1277 priv->eq_table.eq[MLX4_EQ_ASYNC].eqn, err);
1278
1279 /* arm ASYNC eq */
1280 eq_set_ci(&priv->eq_table.eq[MLX4_EQ_ASYNC], 1);
1281
1282 return 0;
1283
1284 err_out_unmap:
1285 while (i > 0)
1286 mlx4_free_eq(dev, &priv->eq_table.eq[--i]);
1287 #ifdef CONFIG_RFS_ACCEL
1288 for (i = 1; i <= dev->caps.num_ports; i++) {
1289 if (mlx4_priv(dev)->port[i].rmap) {
1290 free_irq_cpu_rmap(mlx4_priv(dev)->port[i].rmap);
1291 mlx4_priv(dev)->port[i].rmap = NULL;
1292 }
1293 }
1294 #endif
1295 mlx4_free_irqs(dev);
1296
1297 err_out_clr_int:
1298 if (!mlx4_is_slave(dev))
1299 mlx4_unmap_clr_int(dev);
1300
1301 err_out_bitmap:
1302 mlx4_unmap_uar(dev);
1303 mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1304
1305 err_out_free:
1306 kfree(priv->eq_table.uar_map);
1307
1308 return err;
1309 }
1310
mlx4_cleanup_eq_table(struct mlx4_dev * dev)1311 void mlx4_cleanup_eq_table(struct mlx4_dev *dev)
1312 {
1313 struct mlx4_priv *priv = mlx4_priv(dev);
1314 int i;
1315
1316 mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 1,
1317 priv->eq_table.eq[MLX4_EQ_ASYNC].eqn);
1318
1319 #ifdef CONFIG_RFS_ACCEL
1320 for (i = 1; i <= dev->caps.num_ports; i++) {
1321 if (mlx4_priv(dev)->port[i].rmap) {
1322 free_irq_cpu_rmap(mlx4_priv(dev)->port[i].rmap);
1323 mlx4_priv(dev)->port[i].rmap = NULL;
1324 }
1325 }
1326 #endif
1327 mlx4_free_irqs(dev);
1328
1329 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1330 mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1331
1332 if (!mlx4_is_slave(dev))
1333 mlx4_unmap_clr_int(dev);
1334
1335 mlx4_unmap_uar(dev);
1336 mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1337
1338 kfree(priv->eq_table.uar_map);
1339 }
1340
1341 /* A test that verifies that we can accept interrupts
1342 * on the vector allocated for asynchronous events
1343 */
mlx4_test_async(struct mlx4_dev * dev)1344 int mlx4_test_async(struct mlx4_dev *dev)
1345 {
1346 return mlx4_NOP(dev);
1347 }
1348 EXPORT_SYMBOL(mlx4_test_async);
1349
1350 /* A test that verifies that we can accept interrupts
1351 * on the given irq vector of the tested port.
1352 * Interrupts are checked using the NOP command.
1353 */
mlx4_test_interrupt(struct mlx4_dev * dev,int vector)1354 int mlx4_test_interrupt(struct mlx4_dev *dev, int vector)
1355 {
1356 struct mlx4_priv *priv = mlx4_priv(dev);
1357 int err;
1358
1359 /* Temporary use polling for command completions */
1360 mlx4_cmd_use_polling(dev);
1361
1362 /* Map the new eq to handle all asynchronous events */
1363 err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1364 priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(vector)].eqn);
1365 if (err) {
1366 mlx4_warn(dev, "Failed mapping eq for interrupt test\n");
1367 goto out;
1368 }
1369
1370 /* Go back to using events */
1371 mlx4_cmd_use_events(dev);
1372 err = mlx4_NOP(dev);
1373
1374 /* Return to default */
1375 mlx4_cmd_use_polling(dev);
1376 out:
1377 mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1378 priv->eq_table.eq[MLX4_EQ_ASYNC].eqn);
1379 mlx4_cmd_use_events(dev);
1380
1381 return err;
1382 }
1383 EXPORT_SYMBOL(mlx4_test_interrupt);
1384
mlx4_is_eq_vector_valid(struct mlx4_dev * dev,u8 port,int vector)1385 bool mlx4_is_eq_vector_valid(struct mlx4_dev *dev, u8 port, int vector)
1386 {
1387 struct mlx4_priv *priv = mlx4_priv(dev);
1388
1389 vector = MLX4_CQ_TO_EQ_VECTOR(vector);
1390 if (vector < 0 || (vector >= dev->caps.num_comp_vectors + 1) ||
1391 (vector == MLX4_EQ_ASYNC))
1392 return false;
1393
1394 return test_bit(port - 1, priv->eq_table.eq[vector].actv_ports.ports);
1395 }
1396 EXPORT_SYMBOL(mlx4_is_eq_vector_valid);
1397
mlx4_get_eqs_per_port(struct mlx4_dev * dev,u8 port)1398 u32 mlx4_get_eqs_per_port(struct mlx4_dev *dev, u8 port)
1399 {
1400 struct mlx4_priv *priv = mlx4_priv(dev);
1401 unsigned int i;
1402 unsigned int sum = 0;
1403
1404 for (i = 0; i < dev->caps.num_comp_vectors + 1; i++)
1405 sum += !!test_bit(port - 1,
1406 priv->eq_table.eq[i].actv_ports.ports);
1407
1408 return sum;
1409 }
1410 EXPORT_SYMBOL(mlx4_get_eqs_per_port);
1411
mlx4_is_eq_shared(struct mlx4_dev * dev,int vector)1412 int mlx4_is_eq_shared(struct mlx4_dev *dev, int vector)
1413 {
1414 struct mlx4_priv *priv = mlx4_priv(dev);
1415
1416 vector = MLX4_CQ_TO_EQ_VECTOR(vector);
1417 if (vector <= 0 || (vector >= dev->caps.num_comp_vectors + 1))
1418 return -EINVAL;
1419
1420 return !!(bitmap_weight(priv->eq_table.eq[vector].actv_ports.ports,
1421 dev->caps.num_ports) > 1);
1422 }
1423 EXPORT_SYMBOL(mlx4_is_eq_shared);
1424
mlx4_assign_eq(struct mlx4_dev * dev,u8 port,int * vector)1425 int mlx4_assign_eq(struct mlx4_dev *dev, u8 port, int *vector)
1426 {
1427 struct mlx4_priv *priv = mlx4_priv(dev);
1428 int err = 0, i = 0;
1429 u32 min_ref_count_val = (u32)-1;
1430 int requested_vector = MLX4_CQ_TO_EQ_VECTOR(*vector);
1431 int *prequested_vector = NULL;
1432
1433
1434 mutex_lock(&priv->msix_ctl.pool_lock);
1435 if (requested_vector < (dev->caps.num_comp_vectors + 1) &&
1436 (requested_vector >= 0) &&
1437 (requested_vector != MLX4_EQ_ASYNC)) {
1438 if (test_bit(port - 1,
1439 priv->eq_table.eq[requested_vector].actv_ports.ports)) {
1440 prequested_vector = &requested_vector;
1441 } else {
1442 struct mlx4_eq *eq;
1443
1444 for (i = 1; i < port;
1445 requested_vector += mlx4_get_eqs_per_port(dev, i++))
1446 ;
1447
1448 eq = &priv->eq_table.eq[requested_vector];
1449 if (requested_vector < dev->caps.num_comp_vectors + 1 &&
1450 test_bit(port - 1, eq->actv_ports.ports)) {
1451 prequested_vector = &requested_vector;
1452 }
1453 }
1454 }
1455
1456 if (!prequested_vector) {
1457 requested_vector = -1;
1458 for (i = 0; min_ref_count_val && i < dev->caps.num_comp_vectors + 1;
1459 i++) {
1460 struct mlx4_eq *eq = &priv->eq_table.eq[i];
1461
1462 if (min_ref_count_val > eq->ref_count &&
1463 test_bit(port - 1, eq->actv_ports.ports)) {
1464 min_ref_count_val = eq->ref_count;
1465 requested_vector = i;
1466 }
1467 }
1468
1469 if (requested_vector < 0) {
1470 err = -ENOSPC;
1471 goto err_unlock;
1472 }
1473
1474 prequested_vector = &requested_vector;
1475 }
1476
1477 if (!test_bit(*prequested_vector, priv->msix_ctl.pool_bm) &&
1478 dev->flags & MLX4_FLAG_MSI_X) {
1479 set_bit(*prequested_vector, priv->msix_ctl.pool_bm);
1480 snprintf(priv->eq_table.irq_names +
1481 *prequested_vector * MLX4_IRQNAME_SIZE,
1482 MLX4_IRQNAME_SIZE, "mlx4-%d@%s",
1483 *prequested_vector, dev_name(&dev->persist->pdev->dev));
1484
1485 err = request_irq(priv->eq_table.eq[*prequested_vector].irq,
1486 mlx4_msi_x_interrupt, 0,
1487 &priv->eq_table.irq_names[*prequested_vector << 5],
1488 priv->eq_table.eq + *prequested_vector);
1489
1490 if (err) {
1491 clear_bit(*prequested_vector, priv->msix_ctl.pool_bm);
1492 *prequested_vector = -1;
1493 } else {
1494 mlx4_set_eq_affinity_hint(priv, *prequested_vector);
1495 eq_set_ci(&priv->eq_table.eq[*prequested_vector], 1);
1496 priv->eq_table.eq[*prequested_vector].have_irq = 1;
1497 }
1498 }
1499
1500 if (!err && *prequested_vector >= 0)
1501 priv->eq_table.eq[*prequested_vector].ref_count++;
1502
1503 err_unlock:
1504 mutex_unlock(&priv->msix_ctl.pool_lock);
1505
1506 if (!err && *prequested_vector >= 0)
1507 *vector = MLX4_EQ_TO_CQ_VECTOR(*prequested_vector);
1508 else
1509 *vector = 0;
1510
1511 return err;
1512 }
1513 EXPORT_SYMBOL(mlx4_assign_eq);
1514
mlx4_eq_get_irq(struct mlx4_dev * dev,int cq_vec)1515 int mlx4_eq_get_irq(struct mlx4_dev *dev, int cq_vec)
1516 {
1517 struct mlx4_priv *priv = mlx4_priv(dev);
1518
1519 return priv->eq_table.eq[MLX4_CQ_TO_EQ_VECTOR(cq_vec)].irq;
1520 }
1521 EXPORT_SYMBOL(mlx4_eq_get_irq);
1522
mlx4_release_eq(struct mlx4_dev * dev,int vec)1523 void mlx4_release_eq(struct mlx4_dev *dev, int vec)
1524 {
1525 struct mlx4_priv *priv = mlx4_priv(dev);
1526 int eq_vec = MLX4_CQ_TO_EQ_VECTOR(vec);
1527
1528 mutex_lock(&priv->msix_ctl.pool_lock);
1529 priv->eq_table.eq[eq_vec].ref_count--;
1530
1531 /* once we allocated EQ, we don't release it because it might be binded
1532 * to cpu_rmap.
1533 */
1534 mutex_unlock(&priv->msix_ctl.pool_lock);
1535 }
1536 EXPORT_SYMBOL(mlx4_release_eq);
1537
1538 void
mlx4_disable_interrupts(struct mlx4_dev * dev)1539 mlx4_disable_interrupts(struct mlx4_dev *dev)
1540 {
1541 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
1542 int i;
1543
1544 if (dev->flags & MLX4_FLAG_MSI_X) {
1545 for (i = 0; i < (dev->caps.num_comp_vectors + 1); ++i)
1546 disable_irq(priv->eq_table.eq[i].irq);
1547 } else {
1548 disable_irq(dev->persist->pdev->irq);
1549 }
1550 }
1551 EXPORT_SYMBOL(mlx4_disable_interrupts);
1552
1553 void
mlx4_poll_interrupts(struct mlx4_dev * dev)1554 mlx4_poll_interrupts(struct mlx4_dev *dev)
1555 {
1556 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
1557 int i;
1558
1559 if (dev->flags & MLX4_FLAG_MSI_X) {
1560 for (i = 0; i < (dev->caps.num_comp_vectors + 1); ++i) {
1561 mlx4_msi_x_interrupt(priv->eq_table.eq[i].irq,
1562 priv->eq_table.eq + i);
1563 }
1564 } else {
1565 mlx4_interrupt(dev->persist->pdev->irq, dev);
1566 }
1567 }
1568 EXPORT_SYMBOL(mlx4_poll_interrupts);
1569