2 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
34 #include <linux/init.h>
35 #include <linux/interrupt.h>
36 #include <linux/slab.h>
37 #include <linux/export.h>
39 #include <linux/dma-mapping.h>
41 #include <linux/mlx4/cmd.h>
42 #include <linux/cpu_rmap.h>
48 MLX4_IRQNAME_SIZE = 32
52 MLX4_NUM_ASYNC_EQE = 0x100,
53 MLX4_NUM_SPARE_EQE = 0x80,
54 MLX4_EQ_ENTRY_SIZE = 0x20
57 #define MLX4_EQ_STATUS_OK ( 0 << 28)
58 #define MLX4_EQ_STATUS_WRITE_FAIL (10 << 28)
59 #define MLX4_EQ_OWNER_SW ( 0 << 24)
60 #define MLX4_EQ_OWNER_HW ( 1 << 24)
61 #define MLX4_EQ_FLAG_EC ( 1 << 18)
62 #define MLX4_EQ_FLAG_OI ( 1 << 17)
63 #define MLX4_EQ_STATE_ARMED ( 9 << 8)
64 #define MLX4_EQ_STATE_FIRED (10 << 8)
65 #define MLX4_EQ_STATE_ALWAYS_ARMED (11 << 8)
67 #define MLX4_ASYNC_EVENT_MASK ((1ull << MLX4_EVENT_TYPE_PATH_MIG) | \
68 (1ull << MLX4_EVENT_TYPE_COMM_EST) | \
69 (1ull << MLX4_EVENT_TYPE_SQ_DRAINED) | \
70 (1ull << MLX4_EVENT_TYPE_CQ_ERROR) | \
71 (1ull << MLX4_EVENT_TYPE_WQ_CATAS_ERROR) | \
72 (1ull << MLX4_EVENT_TYPE_EEC_CATAS_ERROR) | \
73 (1ull << MLX4_EVENT_TYPE_PATH_MIG_FAILED) | \
74 (1ull << MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR) | \
75 (1ull << MLX4_EVENT_TYPE_WQ_ACCESS_ERROR) | \
76 (1ull << MLX4_EVENT_TYPE_PORT_CHANGE) | \
77 (1ull << MLX4_EVENT_TYPE_ECC_DETECT) | \
78 (1ull << MLX4_EVENT_TYPE_SRQ_CATAS_ERROR) | \
79 (1ull << MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE) | \
80 (1ull << MLX4_EVENT_TYPE_SRQ_LIMIT) | \
81 (1ull << MLX4_EVENT_TYPE_CMD) | \
82 (1ull << MLX4_EVENT_TYPE_COMM_CHANNEL) | \
83 (1ull << MLX4_EVENT_TYPE_FLR_EVENT) | \
84 (1ull << MLX4_EVENT_TYPE_FATAL_WARNING))
86 static u64 get_async_ev_mask(struct mlx4_dev *dev)
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);
95 static void eq_set_ci(struct mlx4_eq *eq, int req_not)
97 __raw_writel((__force u32) cpu_to_be32((eq->cons_index & 0xffffff) |
100 /* We still want ordering, just not swabbing, so add a barrier */
104 static struct mlx4_eqe *get_eqe(struct mlx4_eq *eq, u32 entry)
106 unsigned long off = (entry & (eq->nent - 1)) * MLX4_EQ_ENTRY_SIZE;
107 return eq->page_list[off / PAGE_SIZE].buf + off % PAGE_SIZE;
110 static struct mlx4_eqe *next_eqe_sw(struct mlx4_eq *eq)
112 struct mlx4_eqe *eqe = get_eqe(eq, eq->cons_index);
113 return !!(eqe->owner & 0x80) ^ !!(eq->cons_index & eq->nent) ? NULL : eqe;
116 static struct mlx4_eqe *next_slave_event_eqe(struct mlx4_slave_event_eq *slave_eq)
118 struct mlx4_eqe *eqe =
119 &slave_eq->event_eqe[slave_eq->cons & (SLAVE_EVENT_EQ_SIZE - 1)];
120 return (!!(eqe->owner & 0x80) ^
121 !!(slave_eq->cons & SLAVE_EVENT_EQ_SIZE)) ?
125 void mlx4_gen_slave_eqe(struct work_struct *work)
127 struct mlx4_mfunc_master_ctx *master =
128 container_of(work, struct mlx4_mfunc_master_ctx,
130 struct mlx4_mfunc *mfunc =
131 container_of(master, struct mlx4_mfunc, master);
132 struct mlx4_priv *priv = container_of(mfunc, struct mlx4_priv, mfunc);
133 struct mlx4_dev *dev = &priv->dev;
134 struct mlx4_slave_event_eq *slave_eq = &mfunc->master.slave_eq;
135 struct mlx4_eqe *eqe;
139 for (eqe = next_slave_event_eqe(slave_eq); eqe;
140 eqe = next_slave_event_eqe(slave_eq)) {
141 slave = eqe->slave_id;
143 /* All active slaves need to receive the event */
144 if (slave == ALL_SLAVES) {
145 for (i = 0; i < dev->num_slaves; i++) {
146 if (i != dev->caps.function &&
147 master->slave_state[i].active)
148 if (mlx4_GEN_EQE(dev, i, eqe))
149 mlx4_warn(dev, "Failed to "
151 "for slave %d\n", i);
154 if (mlx4_GEN_EQE(dev, slave, eqe))
155 mlx4_warn(dev, "Failed to generate event "
156 "for slave %d\n", slave);
163 static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
165 struct mlx4_priv *priv = mlx4_priv(dev);
166 struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq;
167 struct mlx4_eqe *s_eqe;
170 spin_lock_irqsave(&slave_eq->event_lock, flags);
171 s_eqe = &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
172 if ((!!(s_eqe->owner & 0x80)) ^
173 (!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) {
174 mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. "
175 "No free EQE on slave events queue\n", slave);
176 spin_unlock_irqrestore(&slave_eq->event_lock, flags);
180 memcpy(s_eqe, eqe, sizeof(struct mlx4_eqe) - 1);
181 s_eqe->slave_id = slave;
182 /* ensure all information is written before setting the ownersip bit */
184 s_eqe->owner = !!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE) ? 0x0 : 0x80;
187 queue_work(priv->mfunc.master.comm_wq,
188 &priv->mfunc.master.slave_event_work);
189 spin_unlock_irqrestore(&slave_eq->event_lock, flags);
192 static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
193 struct mlx4_eqe *eqe)
195 struct mlx4_priv *priv = mlx4_priv(dev);
196 struct mlx4_slave_state *s_slave =
197 &priv->mfunc.master.slave_state[slave];
199 if (!s_slave->active) {
200 /*mlx4_warn(dev, "Trying to pass event to inactive slave\n");*/
204 slave_event(dev, slave, eqe);
207 int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port)
211 struct mlx4_priv *priv = mlx4_priv(dev);
212 struct mlx4_slave_state *s_slave = &priv->mfunc.master.slave_state[slave];
214 if (!s_slave->active)
217 memset(&eqe, 0, sizeof eqe);
219 eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
220 eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE;
221 eqe.event.port_mgmt_change.port = port;
223 return mlx4_GEN_EQE(dev, slave, &eqe);
225 EXPORT_SYMBOL(mlx4_gen_pkey_eqe);
227 int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port)
231 /*don't send if we don't have the that slave */
232 if (dev->num_vfs < slave)
234 memset(&eqe, 0, sizeof eqe);
236 eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
237 eqe.subtype = MLX4_DEV_PMC_SUBTYPE_GUID_INFO;
238 eqe.event.port_mgmt_change.port = port;
240 return mlx4_GEN_EQE(dev, slave, &eqe);
242 EXPORT_SYMBOL(mlx4_gen_guid_change_eqe);
244 int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port,
245 u8 port_subtype_change)
249 /*don't send if we don't have the that slave */
250 if (dev->num_vfs < slave)
252 memset(&eqe, 0, sizeof eqe);
254 eqe.type = MLX4_EVENT_TYPE_PORT_CHANGE;
255 eqe.subtype = port_subtype_change;
256 eqe.event.port_change.port = cpu_to_be32(port << 28);
258 mlx4_dbg(dev, "%s: sending: %d to slave: %d on port: %d\n", __func__,
259 port_subtype_change, slave, port);
260 return mlx4_GEN_EQE(dev, slave, &eqe);
262 EXPORT_SYMBOL(mlx4_gen_port_state_change_eqe);
264 enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port)
266 struct mlx4_priv *priv = mlx4_priv(dev);
267 struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
268 if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS) {
269 pr_err("%s: Error: asking for slave:%d, port:%d\n",
270 __func__, slave, port);
271 return SLAVE_PORT_DOWN;
273 return s_state[slave].port_state[port];
275 EXPORT_SYMBOL(mlx4_get_slave_port_state);
277 static int mlx4_set_slave_port_state(struct mlx4_dev *dev, int slave, u8 port,
278 enum slave_port_state state)
280 struct mlx4_priv *priv = mlx4_priv(dev);
281 struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state;
283 if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) {
284 pr_err("%s: Error: asking for slave:%d, port:%d\n",
285 __func__, slave, port);
288 s_state[slave].port_state[port] = state;
293 static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event)
296 enum slave_port_gen_event gen_event;
298 for (i = 0; i < dev->num_slaves; i++)
299 set_and_calc_slave_port_state(dev, i, port, event, &gen_event);
301 /**************************************************************************
302 The function get as input the new event to that port,
303 and according to the prev state change the slave's port state.
305 MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
306 MLX4_PORT_STATE_DEV_EVENT_PORT_UP
307 MLX4_PORT_STATE_IB_EVENT_GID_VALID
308 MLX4_PORT_STATE_IB_EVENT_GID_INVALID
309 ***************************************************************************/
310 int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave,
312 enum slave_port_gen_event *gen_event)
314 struct mlx4_priv *priv = mlx4_priv(dev);
315 struct mlx4_slave_state *ctx = NULL;
318 enum slave_port_state cur_state =
319 mlx4_get_slave_port_state(dev, slave, port);
321 *gen_event = SLAVE_PORT_GEN_EVENT_NONE;
323 if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) {
324 pr_err("%s: Error: asking for slave:%d, port:%d\n",
325 __func__, slave, port);
329 ctx = &priv->mfunc.master.slave_state[slave];
330 spin_lock_irqsave(&ctx->lock, flags);
333 case SLAVE_PORT_DOWN:
334 if (MLX4_PORT_STATE_DEV_EVENT_PORT_UP == event)
335 mlx4_set_slave_port_state(dev, slave, port,
338 case SLAVE_PENDING_UP:
339 if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event)
340 mlx4_set_slave_port_state(dev, slave, port,
342 else if (MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID == event) {
343 mlx4_set_slave_port_state(dev, slave, port,
345 *gen_event = SLAVE_PORT_GEN_EVENT_UP;
349 if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) {
350 mlx4_set_slave_port_state(dev, slave, port,
352 *gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
353 } else if (MLX4_PORT_STATE_IB_EVENT_GID_INVALID ==
355 mlx4_set_slave_port_state(dev, slave, port,
357 *gen_event = SLAVE_PORT_GEN_EVENT_DOWN;
361 pr_err("%s: BUG!!! UNKNOWN state: "
362 "slave:%d, port:%d\n", __func__, slave, port);
365 ret = mlx4_get_slave_port_state(dev, slave, port);
368 spin_unlock_irqrestore(&ctx->lock, flags);
372 EXPORT_SYMBOL(set_and_calc_slave_port_state);
374 int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr)
378 memset(&eqe, 0, sizeof eqe);
380 eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT;
381 eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PORT_INFO;
382 eqe.event.port_mgmt_change.port = port;
383 eqe.event.port_mgmt_change.params.port_info.changed_attr =
384 cpu_to_be32((u32) attr);
386 slave_event(dev, ALL_SLAVES, &eqe);
389 EXPORT_SYMBOL(mlx4_gen_slaves_port_mgt_ev);
391 void mlx4_master_handle_slave_flr(struct work_struct *work)
393 struct mlx4_mfunc_master_ctx *master =
394 container_of(work, struct mlx4_mfunc_master_ctx,
395 slave_flr_event_work);
396 struct mlx4_mfunc *mfunc =
397 container_of(master, struct mlx4_mfunc, master);
398 struct mlx4_priv *priv =
399 container_of(mfunc, struct mlx4_priv, mfunc);
400 struct mlx4_dev *dev = &priv->dev;
401 struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
405 mlx4_dbg(dev, "mlx4_handle_slave_flr\n");
407 for (i = 0 ; i < dev->num_slaves; i++) {
409 if (MLX4_COMM_CMD_FLR == slave_state[i].last_cmd) {
410 mlx4_dbg(dev, "mlx4_handle_slave_flr: "
411 "clean slave: %d\n", i);
413 mlx4_delete_all_resources_for_slave(dev, i);
414 /*return the slave to running mode*/
415 spin_lock(&priv->mfunc.master.slave_state_lock);
416 slave_state[i].last_cmd = MLX4_COMM_CMD_RESET;
417 slave_state[i].is_slave_going_down = 0;
418 spin_unlock(&priv->mfunc.master.slave_state_lock);
420 err = mlx4_cmd(dev, 0, i, 0, MLX4_CMD_INFORM_FLR_DONE,
421 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
423 mlx4_warn(dev, "Failed to notify FW on "
424 "FLR done (slave:%d)\n", i);
429 static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
431 struct mlx4_priv *priv = mlx4_priv(dev);
432 struct mlx4_eqe *eqe;
440 u8 update_slave_state;
442 enum slave_port_gen_event gen_event;
444 while ((eqe = next_eqe_sw(eq))) {
446 * Make sure we read EQ entry contents after we've
447 * checked the ownership bit.
452 case MLX4_EVENT_TYPE_COMP:
453 cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
454 mlx4_cq_completion(dev, cqn);
457 case MLX4_EVENT_TYPE_PATH_MIG:
458 case MLX4_EVENT_TYPE_COMM_EST:
459 case MLX4_EVENT_TYPE_SQ_DRAINED:
460 case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE:
461 case MLX4_EVENT_TYPE_WQ_CATAS_ERROR:
462 case MLX4_EVENT_TYPE_PATH_MIG_FAILED:
463 case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
464 case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR:
465 mlx4_dbg(dev, "event %d arrived\n", eqe->type);
466 if (mlx4_is_master(dev)) {
467 /* forward only to slave owning the QP */
468 ret = mlx4_get_slave_from_resource_id(dev,
470 be32_to_cpu(eqe->event.qp.qpn)
472 if (ret && ret != -ENOENT) {
473 mlx4_dbg(dev, "QP event %02x(%02x) on "
474 "EQ %d at index %u: could "
475 "not get slave id (%d)\n",
476 eqe->type, eqe->subtype,
477 eq->eqn, eq->cons_index, ret);
481 if (!ret && slave != dev->caps.function) {
482 mlx4_slave_event(dev, slave, eqe);
487 mlx4_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) &
488 0xffffff, eqe->type);
491 case MLX4_EVENT_TYPE_SRQ_LIMIT:
492 mlx4_warn(dev, "%s: MLX4_EVENT_TYPE_SRQ_LIMIT\n",
494 case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
495 if (mlx4_is_master(dev)) {
496 /* forward only to slave owning the SRQ */
497 ret = mlx4_get_slave_from_resource_id(dev,
499 be32_to_cpu(eqe->event.srq.srqn)
502 if (ret && ret != -ENOENT) {
503 mlx4_warn(dev, "SRQ event %02x(%02x) "
504 "on EQ %d at index %u: could"
505 " not get slave id (%d)\n",
506 eqe->type, eqe->subtype,
507 eq->eqn, eq->cons_index, ret);
510 mlx4_warn(dev, "%s: slave:%d, srq_no:0x%x,"
511 " event: %02x(%02x)\n", __func__,
513 be32_to_cpu(eqe->event.srq.srqn),
514 eqe->type, eqe->subtype);
516 if (!ret && slave != dev->caps.function) {
517 mlx4_warn(dev, "%s: sending event "
518 "%02x(%02x) to slave:%d\n",
520 eqe->subtype, slave);
521 mlx4_slave_event(dev, slave, eqe);
525 mlx4_srq_event(dev, be32_to_cpu(eqe->event.srq.srqn) &
526 0xffffff, eqe->type);
529 case MLX4_EVENT_TYPE_CMD:
531 be16_to_cpu(eqe->event.cmd.token),
532 eqe->event.cmd.status,
533 be64_to_cpu(eqe->event.cmd.out_param));
536 case MLX4_EVENT_TYPE_PORT_CHANGE:
537 port = be32_to_cpu(eqe->event.port_change.port) >> 28;
538 if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) {
539 mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_DOWN,
541 mlx4_priv(dev)->sense.do_sense_port[port] = 1;
542 if (!mlx4_is_master(dev))
544 for (i = 0; i < dev->num_slaves; i++) {
545 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) {
546 if (i == mlx4_master_func_num(dev))
548 mlx4_dbg(dev, "%s: Sending MLX4_PORT_CHANGE_SUBTYPE_DOWN"
549 " to slave: %d, port:%d\n",
551 mlx4_slave_event(dev, i, eqe);
552 } else { /* IB port */
553 set_and_calc_slave_port_state(dev, i, port,
554 MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN,
556 /*we can be in pending state, then do not send port_down event*/
557 if (SLAVE_PORT_GEN_EVENT_DOWN == gen_event) {
558 if (i == mlx4_master_func_num(dev))
560 mlx4_slave_event(dev, i, eqe);
565 mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_UP, port);
567 mlx4_priv(dev)->sense.do_sense_port[port] = 0;
569 if (!mlx4_is_master(dev))
571 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
572 for (i = 0; i < dev->num_slaves; i++) {
573 if (i == mlx4_master_func_num(dev))
575 mlx4_slave_event(dev, i, eqe);
578 /* port-up event will be sent to a slave when the
579 * slave's alias-guid is set. This is done in alias_GUID.c
581 set_all_slave_state(dev, port, MLX4_DEV_EVENT_PORT_UP);
585 case MLX4_EVENT_TYPE_CQ_ERROR:
586 mlx4_warn(dev, "CQ %s on CQN %06x\n",
587 eqe->event.cq_err.syndrome == 1 ?
588 "overrun" : "access violation",
589 be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff);
590 if (mlx4_is_master(dev)) {
591 ret = mlx4_get_slave_from_resource_id(dev,
593 be32_to_cpu(eqe->event.cq_err.cqn)
595 if (ret && ret != -ENOENT) {
596 mlx4_dbg(dev, "CQ event %02x(%02x) on "
597 "EQ %d at index %u: could "
598 "not get slave id (%d)\n",
599 eqe->type, eqe->subtype,
600 eq->eqn, eq->cons_index, ret);
604 if (!ret && slave != dev->caps.function) {
605 mlx4_slave_event(dev, slave, eqe);
610 be32_to_cpu(eqe->event.cq_err.cqn)
615 case MLX4_EVENT_TYPE_EQ_OVERFLOW:
616 mlx4_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
619 case MLX4_EVENT_TYPE_COMM_CHANNEL:
620 if (!mlx4_is_master(dev)) {
621 mlx4_warn(dev, "Received comm channel event "
622 "for non master device\n");
625 memcpy(&priv->mfunc.master.comm_arm_bit_vector,
626 eqe->event.comm_channel_arm.bit_vec,
627 sizeof eqe->event.comm_channel_arm.bit_vec);
628 queue_work(priv->mfunc.master.comm_wq,
629 &priv->mfunc.master.comm_work);
632 case MLX4_EVENT_TYPE_FLR_EVENT:
633 flr_slave = be32_to_cpu(eqe->event.flr_event.slave_id);
634 if (!mlx4_is_master(dev)) {
635 mlx4_warn(dev, "Non-master function received"
640 mlx4_dbg(dev, "FLR event for slave: %d\n", flr_slave);
642 if (flr_slave >= dev->num_slaves) {
644 "Got FLR for unknown function: %d\n",
646 update_slave_state = 0;
648 update_slave_state = 1;
650 spin_lock(&priv->mfunc.master.slave_state_lock);
651 if (update_slave_state) {
652 priv->mfunc.master.slave_state[flr_slave].active = false;
653 priv->mfunc.master.slave_state[flr_slave].last_cmd = MLX4_COMM_CMD_FLR;
654 priv->mfunc.master.slave_state[flr_slave].is_slave_going_down = 1;
656 spin_unlock(&priv->mfunc.master.slave_state_lock);
657 queue_work(priv->mfunc.master.comm_wq,
658 &priv->mfunc.master.slave_flr_event_work);
661 case MLX4_EVENT_TYPE_FATAL_WARNING:
662 if (eqe->subtype == MLX4_FATAL_WARNING_SUBTYPE_WARMING) {
663 if (mlx4_is_master(dev))
664 for (i = 0; i < dev->num_slaves; i++) {
665 mlx4_dbg(dev, "%s: Sending "
666 "MLX4_FATAL_WARNING_SUBTYPE_WARMING"
667 " to slave: %d\n", __func__, i);
668 if (i == dev->caps.function)
670 mlx4_slave_event(dev, i, eqe);
672 mlx4_err(dev, "Temperature Threshold was reached! "
673 "Threshold: %d celsius degrees; "
674 "Current Temperature: %d\n",
675 be16_to_cpu(eqe->event.warming.warning_threshold),
676 be16_to_cpu(eqe->event.warming.current_temperature));
678 mlx4_warn(dev, "Unhandled event FATAL WARNING (%02x), "
679 "subtype %02x on EQ %d at index %u. owner=%x, "
680 "nent=0x%x, slave=%x, ownership=%s\n",
681 eqe->type, eqe->subtype, eq->eqn,
682 eq->cons_index, eqe->owner, eq->nent,
684 !!(eqe->owner & 0x80) ^
685 !!(eq->cons_index & eq->nent) ? "HW" : "SW");
689 case MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT:
690 mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_MGMT_CHANGE,
691 (unsigned long) eqe);
694 case MLX4_EVENT_TYPE_EEC_CATAS_ERROR:
695 case MLX4_EVENT_TYPE_ECC_DETECT:
697 mlx4_warn(dev, "Unhandled event %02x(%02x) on EQ %d at "
698 "index %u. owner=%x, nent=0x%x, slave=%x, "
700 eqe->type, eqe->subtype, eq->eqn,
701 eq->cons_index, eqe->owner, eq->nent,
703 !!(eqe->owner & 0x80) ^
704 !!(eq->cons_index & eq->nent) ? "HW" : "SW");
713 * The HCA will think the queue has overflowed if we
714 * don't tell it we've been processing events. We
715 * create our EQs with MLX4_NUM_SPARE_EQE extra
716 * entries, so we must update our consumer index at
719 if (unlikely(set_ci >= MLX4_NUM_SPARE_EQE)) {
730 static irqreturn_t mlx4_interrupt(int irq, void *dev_ptr)
732 struct mlx4_dev *dev = dev_ptr;
733 struct mlx4_priv *priv = mlx4_priv(dev);
737 writel(priv->eq_table.clr_mask, priv->eq_table.clr_int);
739 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
740 work |= mlx4_eq_int(dev, &priv->eq_table.eq[i]);
742 return IRQ_RETVAL(work);
745 static irqreturn_t mlx4_msi_x_interrupt(int irq, void *eq_ptr)
747 struct mlx4_eq *eq = eq_ptr;
748 struct mlx4_dev *dev = eq->dev;
750 mlx4_eq_int(dev, eq);
752 /* MSI-X vectors always belong to us */
756 int mlx4_MAP_EQ_wrapper(struct mlx4_dev *dev, int slave,
757 struct mlx4_vhcr *vhcr,
758 struct mlx4_cmd_mailbox *inbox,
759 struct mlx4_cmd_mailbox *outbox,
760 struct mlx4_cmd_info *cmd)
762 struct mlx4_priv *priv = mlx4_priv(dev);
763 struct mlx4_slave_event_eq_info *event_eq =
764 priv->mfunc.master.slave_state[slave].event_eq;
765 u32 in_modifier = vhcr->in_modifier;
766 u32 eqn = in_modifier & 0x1FF;
767 u64 in_param = vhcr->in_param;
771 if (slave == dev->caps.function)
772 err = mlx4_cmd(dev, in_param, (in_modifier & 0x80000000) | eqn,
773 0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
776 for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i)
777 if (in_param & (1LL << i))
778 event_eq[i].eqn = in_modifier >> 31 ? -1 : eqn;
783 static int mlx4_MAP_EQ(struct mlx4_dev *dev, u64 event_mask, int unmap,
786 return mlx4_cmd(dev, event_mask, (unmap << 31) | eq_num,
787 0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
791 static int mlx4_SW2HW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
794 return mlx4_cmd(dev, mailbox->dma, eq_num, 0,
795 MLX4_CMD_SW2HW_EQ, MLX4_CMD_TIME_CLASS_A,
799 static int mlx4_HW2SW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
802 return mlx4_cmd_box(dev, 0, mailbox->dma, eq_num,
803 0, MLX4_CMD_HW2SW_EQ, MLX4_CMD_TIME_CLASS_A,
807 static int mlx4_num_eq_uar(struct mlx4_dev *dev)
810 * Each UAR holds 4 EQ doorbells. To figure out how many UARs
811 * we need to map, take the difference of highest index and
812 * the lowest index we'll use and add 1.
814 return (dev->caps.num_comp_vectors + 1 + dev->caps.reserved_eqs +
815 dev->caps.comp_pool)/4 - dev->caps.reserved_eqs/4 + 1;
818 static void __iomem *mlx4_get_eq_uar(struct mlx4_dev *dev, struct mlx4_eq *eq)
820 struct mlx4_priv *priv = mlx4_priv(dev);
823 index = eq->eqn / 4 - dev->caps.reserved_eqs / 4;
825 if (!priv->eq_table.uar_map[index]) {
826 priv->eq_table.uar_map[index] =
827 ioremap(pci_resource_start(dev->pdev, 2) +
828 ((eq->eqn / 4) << PAGE_SHIFT),
830 if (!priv->eq_table.uar_map[index]) {
831 mlx4_err(dev, "Couldn't map EQ doorbell for EQN 0x%06x\n",
837 return priv->eq_table.uar_map[index] + 0x800 + 8 * (eq->eqn % 4);
840 static int mlx4_create_eq(struct mlx4_dev *dev, int nent,
841 u8 intr, struct mlx4_eq *eq)
843 struct mlx4_priv *priv = mlx4_priv(dev);
844 struct mlx4_cmd_mailbox *mailbox;
845 struct mlx4_eq_context *eq_context;
847 u64 *dma_list = NULL;
854 eq->nent = roundup_pow_of_two(max(nent, 2));
855 npages = PAGE_ALIGN(eq->nent * MLX4_EQ_ENTRY_SIZE) / PAGE_SIZE;
857 eq->page_list = kmalloc(npages * sizeof *eq->page_list,
862 for (i = 0; i < npages; ++i)
863 eq->page_list[i].buf = NULL;
865 dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
869 mailbox = mlx4_alloc_cmd_mailbox(dev);
872 eq_context = mailbox->buf;
874 for (i = 0; i < npages; ++i) {
875 eq->page_list[i].buf = dma_alloc_coherent(&dev->pdev->dev,
876 PAGE_SIZE, &t, GFP_KERNEL);
877 if (!eq->page_list[i].buf)
878 goto err_out_free_pages;
881 eq->page_list[i].map = t;
883 memset(eq->page_list[i].buf, 0, PAGE_SIZE);
886 eq->eqn = mlx4_bitmap_alloc(&priv->eq_table.bitmap);
888 goto err_out_free_pages;
890 eq->doorbell = mlx4_get_eq_uar(dev, eq);
893 goto err_out_free_eq;
896 err = mlx4_mtt_init(dev, npages, PAGE_SHIFT, &eq->mtt);
898 goto err_out_free_eq;
900 err = mlx4_write_mtt(dev, &eq->mtt, 0, npages, dma_list);
902 goto err_out_free_mtt;
904 memset(eq_context, 0, sizeof *eq_context);
905 eq_context->flags = cpu_to_be32(MLX4_EQ_STATUS_OK |
906 MLX4_EQ_STATE_ARMED);
907 eq_context->log_eq_size = ilog2(eq->nent);
908 eq_context->intr = intr;
909 eq_context->log_page_size = PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT;
911 mtt_addr = mlx4_mtt_addr(dev, &eq->mtt);
912 eq_context->mtt_base_addr_h = mtt_addr >> 32;
913 eq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff);
915 err = mlx4_SW2HW_EQ(dev, mailbox, eq->eqn);
917 mlx4_warn(dev, "SW2HW_EQ failed (%d)\n", err);
918 goto err_out_free_mtt;
922 mlx4_free_cmd_mailbox(dev, mailbox);
929 mlx4_mtt_cleanup(dev, &eq->mtt);
932 mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn);
935 for (i = 0; i < npages; ++i)
936 if (eq->page_list[i].buf)
937 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
938 eq->page_list[i].buf,
939 eq->page_list[i].map);
941 mlx4_free_cmd_mailbox(dev, mailbox);
944 kfree(eq->page_list);
951 static void mlx4_free_eq(struct mlx4_dev *dev,
954 struct mlx4_priv *priv = mlx4_priv(dev);
955 struct mlx4_cmd_mailbox *mailbox;
957 int npages = PAGE_ALIGN(MLX4_EQ_ENTRY_SIZE * eq->nent) / PAGE_SIZE;
960 mailbox = mlx4_alloc_cmd_mailbox(dev);
964 err = mlx4_HW2SW_EQ(dev, mailbox, eq->eqn);
966 mlx4_warn(dev, "HW2SW_EQ failed (%d)\n", err);
969 mlx4_dbg(dev, "Dumping EQ context %02x:\n", eq->eqn);
970 for (i = 0; i < sizeof (struct mlx4_eq_context) / 4; ++i) {
972 pr_cont("[%02x] ", i * 4);
973 pr_cont(" %08x", be32_to_cpup(mailbox->buf + i * 4));
974 if ((i + 1) % 4 == 0)
979 mlx4_mtt_cleanup(dev, &eq->mtt);
980 for (i = 0; i < npages; ++i)
981 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
982 eq->page_list[i].buf,
983 eq->page_list[i].map);
985 kfree(eq->page_list);
986 mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn);
987 mlx4_free_cmd_mailbox(dev, mailbox);
990 static void mlx4_free_irqs(struct mlx4_dev *dev)
992 struct mlx4_eq_table *eq_table = &mlx4_priv(dev)->eq_table;
993 struct mlx4_priv *priv = mlx4_priv(dev);
996 if (eq_table->have_irq)
997 free_irq(dev->pdev->irq, dev);
999 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1000 if (eq_table->eq[i].have_irq) {
1001 free_irq(eq_table->eq[i].irq, eq_table->eq + i);
1002 eq_table->eq[i].have_irq = 0;
1005 for (i = 0; i < dev->caps.comp_pool; i++) {
1007 * Freeing the assigned irq's
1008 * all bits should be 0, but we need to validate
1010 if (priv->msix_ctl.pool_bm & 1ULL << i) {
1011 /* NO need protecting*/
1012 vec = dev->caps.num_comp_vectors + 1 + i;
1013 free_irq(priv->eq_table.eq[vec].irq,
1014 &priv->eq_table.eq[vec]);
1019 kfree(eq_table->irq_names);
1022 static int mlx4_map_clr_int(struct mlx4_dev *dev)
1024 struct mlx4_priv *priv = mlx4_priv(dev);
1026 priv->clr_base = ioremap(pci_resource_start(dev->pdev, priv->fw.clr_int_bar) +
1027 priv->fw.clr_int_base, MLX4_CLR_INT_SIZE);
1028 if (!priv->clr_base) {
1029 mlx4_err(dev, "Couldn't map interrupt clear register, aborting.\n");
1036 static void mlx4_unmap_clr_int(struct mlx4_dev *dev)
1038 struct mlx4_priv *priv = mlx4_priv(dev);
1040 iounmap(priv->clr_base);
1043 int mlx4_alloc_eq_table(struct mlx4_dev *dev)
1045 struct mlx4_priv *priv = mlx4_priv(dev);
1047 priv->eq_table.eq = kcalloc(dev->caps.num_eqs - dev->caps.reserved_eqs,
1048 sizeof *priv->eq_table.eq, GFP_KERNEL);
1049 if (!priv->eq_table.eq)
1055 void mlx4_free_eq_table(struct mlx4_dev *dev)
1057 kfree(mlx4_priv(dev)->eq_table.eq);
1060 int mlx4_init_eq_table(struct mlx4_dev *dev)
1062 struct mlx4_priv *priv = mlx4_priv(dev);
1066 priv->eq_table.uar_map = kcalloc(mlx4_num_eq_uar(dev),
1067 sizeof *priv->eq_table.uar_map,
1069 if (!priv->eq_table.uar_map) {
1074 err = mlx4_bitmap_init(&priv->eq_table.bitmap, dev->caps.num_eqs,
1075 dev->caps.num_eqs - 1, dev->caps.reserved_eqs, 0);
1079 for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
1080 priv->eq_table.uar_map[i] = NULL;
1082 if (!mlx4_is_slave(dev)) {
1083 err = mlx4_map_clr_int(dev);
1085 goto err_out_bitmap;
1087 priv->eq_table.clr_mask =
1088 swab32(1 << (priv->eq_table.inta_pin & 31));
1089 priv->eq_table.clr_int = priv->clr_base +
1090 (priv->eq_table.inta_pin < 32 ? 4 : 0);
1093 priv->eq_table.irq_names =
1094 kmalloc(MLX4_IRQNAME_SIZE * (dev->caps.num_comp_vectors + 1 +
1095 dev->caps.comp_pool),
1097 if (!priv->eq_table.irq_names) {
1099 goto err_out_bitmap;
1102 for (i = 0; i < dev->caps.num_comp_vectors; ++i) {
1103 err = mlx4_create_eq(dev, dev->caps.num_cqs -
1104 dev->caps.reserved_cqs +
1106 (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
1107 &priv->eq_table.eq[i]);
1114 err = mlx4_create_eq(dev, MLX4_NUM_ASYNC_EQE + MLX4_NUM_SPARE_EQE,
1115 (dev->flags & MLX4_FLAG_MSI_X) ? dev->caps.num_comp_vectors : 0,
1116 &priv->eq_table.eq[dev->caps.num_comp_vectors]);
1120 /*if additional completion vectors poolsize is 0 this loop will not run*/
1121 for (i = dev->caps.num_comp_vectors + 1;
1122 i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i) {
1124 err = mlx4_create_eq(dev, dev->caps.num_cqs -
1125 dev->caps.reserved_cqs +
1127 (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
1128 &priv->eq_table.eq[i]);
1136 if (dev->flags & MLX4_FLAG_MSI_X) {
1137 const char *eq_name;
1139 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i) {
1140 if (i < dev->caps.num_comp_vectors) {
1141 snprintf(priv->eq_table.irq_names +
1142 i * MLX4_IRQNAME_SIZE,
1144 "mlx4-comp-%d@pci:%s", i,
1145 pci_name(dev->pdev));
1147 snprintf(priv->eq_table.irq_names +
1148 i * MLX4_IRQNAME_SIZE,
1150 "mlx4-async@pci:%s",
1151 pci_name(dev->pdev));
1154 eq_name = priv->eq_table.irq_names +
1155 i * MLX4_IRQNAME_SIZE;
1156 err = request_irq(priv->eq_table.eq[i].irq,
1157 mlx4_msi_x_interrupt, 0, eq_name,
1158 priv->eq_table.eq + i);
1162 priv->eq_table.eq[i].have_irq = 1;
1165 snprintf(priv->eq_table.irq_names,
1168 pci_name(dev->pdev));
1169 err = request_irq(dev->pdev->irq, mlx4_interrupt,
1170 IRQF_SHARED, priv->eq_table.irq_names, dev);
1174 priv->eq_table.have_irq = 1;
1177 err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1178 priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1180 mlx4_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
1181 priv->eq_table.eq[dev->caps.num_comp_vectors].eqn, err);
1183 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
1184 eq_set_ci(&priv->eq_table.eq[i], 1);
1189 mlx4_free_eq(dev, &priv->eq_table.eq[dev->caps.num_comp_vectors]);
1192 i = dev->caps.num_comp_vectors - 1;
1196 mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1199 if (!mlx4_is_slave(dev))
1200 mlx4_unmap_clr_int(dev);
1201 mlx4_free_irqs(dev);
1204 mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1207 kfree(priv->eq_table.uar_map);
1212 void mlx4_cleanup_eq_table(struct mlx4_dev *dev)
1214 struct mlx4_priv *priv = mlx4_priv(dev);
1217 mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 1,
1218 priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1220 mlx4_free_irqs(dev);
1222 for (i = 0; i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i)
1223 mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1225 if (!mlx4_is_slave(dev))
1226 mlx4_unmap_clr_int(dev);
1228 for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
1229 if (priv->eq_table.uar_map[i])
1230 iounmap(priv->eq_table.uar_map[i]);
1232 mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1234 kfree(priv->eq_table.uar_map);
1237 /* A test that verifies that we can accept interrupts on all
1238 * the irq vectors of the device.
1239 * Interrupts are checked using the NOP command.
1241 int mlx4_test_interrupts(struct mlx4_dev *dev)
1243 struct mlx4_priv *priv = mlx4_priv(dev);
1247 err = mlx4_NOP(dev);
1248 /* When not in MSI_X, there is only one irq to check */
1249 if (!(dev->flags & MLX4_FLAG_MSI_X) || mlx4_is_slave(dev))
1252 /* A loop over all completion vectors, for each vector we will check
1253 * whether it works by mapping command completions to that vector
1254 * and performing a NOP command
1256 for(i = 0; !err && (i < dev->caps.num_comp_vectors); ++i) {
1257 /* Temporary use polling for command completions */
1258 mlx4_cmd_use_polling(dev);
1260 /* Map the new eq to handle all asyncronous events */
1261 err = mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1262 priv->eq_table.eq[i].eqn);
1264 mlx4_warn(dev, "Failed mapping eq for interrupt test\n");
1265 mlx4_cmd_use_events(dev);
1269 /* Go back to using events */
1270 mlx4_cmd_use_events(dev);
1271 err = mlx4_NOP(dev);
1274 /* Return to default */
1275 mlx4_MAP_EQ(dev, get_async_ev_mask(dev), 0,
1276 priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1279 EXPORT_SYMBOL(mlx4_test_interrupts);
1281 int mlx4_assign_eq(struct mlx4_dev *dev, char *name, struct cpu_rmap *rmap,
1285 struct mlx4_priv *priv = mlx4_priv(dev);
1286 int vec = 0, err = 0, i;
1288 mutex_lock(&priv->msix_ctl.pool_lock);
1289 for (i = 0; !vec && i < dev->caps.comp_pool; i++) {
1290 if (~priv->msix_ctl.pool_bm & 1ULL << i) {
1291 priv->msix_ctl.pool_bm |= 1ULL << i;
1292 vec = dev->caps.num_comp_vectors + 1 + i;
1293 snprintf(priv->eq_table.irq_names +
1294 vec * MLX4_IRQNAME_SIZE,
1295 MLX4_IRQNAME_SIZE, "%s", name);
1296 #ifdef CONFIG_RFS_ACCEL
1298 err = irq_cpu_rmap_add(rmap,
1299 priv->eq_table.eq[vec].irq);
1301 mlx4_warn(dev, "Failed adding irq rmap\n");
1304 err = request_irq(priv->eq_table.eq[vec].irq,
1305 mlx4_msi_x_interrupt, 0,
1306 &priv->eq_table.irq_names[vec<<5],
1307 priv->eq_table.eq + vec);
1309 /*zero out bit by fliping it*/
1310 priv->msix_ctl.pool_bm ^= 1 << i;
1313 /*we dont want to break here*/
1315 eq_set_ci(&priv->eq_table.eq[vec], 1);
1318 mutex_unlock(&priv->msix_ctl.pool_lock);
1324 err = (i == dev->caps.comp_pool) ? -ENOSPC : err;
1328 EXPORT_SYMBOL(mlx4_assign_eq);
1330 void mlx4_release_eq(struct mlx4_dev *dev, int vec)
1332 struct mlx4_priv *priv = mlx4_priv(dev);
1334 int i = vec - dev->caps.num_comp_vectors - 1;
1336 if (likely(i >= 0)) {
1337 /*sanity check , making sure were not trying to free irq's
1338 Belonging to a legacy EQ*/
1339 mutex_lock(&priv->msix_ctl.pool_lock);
1340 if (priv->msix_ctl.pool_bm & 1ULL << i) {
1341 free_irq(priv->eq_table.eq[vec].irq,
1342 &priv->eq_table.eq[vec]);
1343 priv->msix_ctl.pool_bm &= ~(1ULL << i);
1345 mutex_unlock(&priv->msix_ctl.pool_lock);
1349 EXPORT_SYMBOL(mlx4_release_eq);