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>
47 MLX4_IRQNAME_SIZE = 32
51 MLX4_NUM_ASYNC_EQE = 0x100,
52 MLX4_NUM_SPARE_EQE = 0x80,
53 MLX4_EQ_ENTRY_SIZE = 0x20
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)
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_COMM_CHANNEL) | \
82 (1ull << MLX4_EVENT_TYPE_FLR_EVENT) | \
83 (1ull << MLX4_EVENT_TYPE_FATAL_WARNING))
85 static void eq_set_ci(struct mlx4_eq *eq, int req_not)
87 __raw_writel((__force u32) cpu_to_be32((eq->cons_index & 0xffffff) |
90 /* We still want ordering, just not swabbing, so add a barrier */
94 static struct mlx4_eqe *get_eqe(struct mlx4_eq *eq, u32 entry)
96 unsigned long off = (entry & (eq->nent - 1)) * MLX4_EQ_ENTRY_SIZE;
97 return eq->page_list[off / PAGE_SIZE].buf + off % PAGE_SIZE;
100 static struct mlx4_eqe *next_eqe_sw(struct mlx4_eq *eq)
102 struct mlx4_eqe *eqe = get_eqe(eq, eq->cons_index);
103 return !!(eqe->owner & 0x80) ^ !!(eq->cons_index & eq->nent) ? NULL : eqe;
106 static struct mlx4_eqe *next_slave_event_eqe(struct mlx4_slave_event_eq *slave_eq)
108 struct mlx4_eqe *eqe =
109 &slave_eq->event_eqe[slave_eq->cons & (SLAVE_EVENT_EQ_SIZE - 1)];
110 return (!!(eqe->owner & 0x80) ^
111 !!(slave_eq->cons & SLAVE_EVENT_EQ_SIZE)) ?
115 void mlx4_gen_slave_eqe(struct work_struct *work)
117 struct mlx4_mfunc_master_ctx *master =
118 container_of(work, struct mlx4_mfunc_master_ctx,
120 struct mlx4_mfunc *mfunc =
121 container_of(master, struct mlx4_mfunc, master);
122 struct mlx4_priv *priv = container_of(mfunc, struct mlx4_priv, mfunc);
123 struct mlx4_dev *dev = &priv->dev;
124 struct mlx4_slave_event_eq *slave_eq = &mfunc->master.slave_eq;
125 struct mlx4_eqe *eqe;
129 for (eqe = next_slave_event_eqe(slave_eq); eqe;
130 eqe = next_slave_event_eqe(slave_eq)) {
131 slave = eqe->slave_id;
133 /* All active slaves need to receive the event */
134 if (slave == ALL_SLAVES) {
135 for (i = 0; i < dev->num_slaves; i++) {
136 if (i != dev->caps.function &&
137 master->slave_state[i].active)
138 if (mlx4_GEN_EQE(dev, i, eqe))
139 mlx4_warn(dev, "Failed to "
141 "for slave %d\n", i);
144 if (mlx4_GEN_EQE(dev, slave, eqe))
145 mlx4_warn(dev, "Failed to generate event "
146 "for slave %d\n", slave);
153 static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe)
155 struct mlx4_priv *priv = mlx4_priv(dev);
156 struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq;
157 struct mlx4_eqe *s_eqe =
158 &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)];
160 if ((!!(s_eqe->owner & 0x80)) ^
161 (!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) {
162 mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. "
163 "No free EQE on slave events queue\n", slave);
167 memcpy(s_eqe, eqe, sizeof(struct mlx4_eqe) - 1);
168 s_eqe->slave_id = slave;
169 /* ensure all information is written before setting the ownersip bit */
171 s_eqe->owner = !!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE) ? 0x0 : 0x80;
174 queue_work(priv->mfunc.master.comm_wq,
175 &priv->mfunc.master.slave_event_work);
178 static void mlx4_slave_event(struct mlx4_dev *dev, int slave,
179 struct mlx4_eqe *eqe)
181 struct mlx4_priv *priv = mlx4_priv(dev);
182 struct mlx4_slave_state *s_slave =
183 &priv->mfunc.master.slave_state[slave];
185 if (!s_slave->active) {
186 /*mlx4_warn(dev, "Trying to pass event to inactive slave\n");*/
190 slave_event(dev, slave, eqe);
193 void mlx4_master_handle_slave_flr(struct work_struct *work)
195 struct mlx4_mfunc_master_ctx *master =
196 container_of(work, struct mlx4_mfunc_master_ctx,
197 slave_flr_event_work);
198 struct mlx4_mfunc *mfunc =
199 container_of(master, struct mlx4_mfunc, master);
200 struct mlx4_priv *priv =
201 container_of(mfunc, struct mlx4_priv, mfunc);
202 struct mlx4_dev *dev = &priv->dev;
203 struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
207 mlx4_dbg(dev, "mlx4_handle_slave_flr\n");
209 for (i = 0 ; i < dev->num_slaves; i++) {
211 if (MLX4_COMM_CMD_FLR == slave_state[i].last_cmd) {
212 mlx4_dbg(dev, "mlx4_handle_slave_flr: "
213 "clean slave: %d\n", i);
215 mlx4_delete_all_resources_for_slave(dev, i);
216 /*return the slave to running mode*/
217 spin_lock(&priv->mfunc.master.slave_state_lock);
218 slave_state[i].last_cmd = MLX4_COMM_CMD_RESET;
219 slave_state[i].is_slave_going_down = 0;
220 spin_unlock(&priv->mfunc.master.slave_state_lock);
222 err = mlx4_cmd(dev, 0, i, 0, MLX4_CMD_INFORM_FLR_DONE,
223 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
225 mlx4_warn(dev, "Failed to notify FW on "
226 "FLR done (slave:%d)\n", i);
231 static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq)
233 struct mlx4_priv *priv = mlx4_priv(dev);
234 struct mlx4_eqe *eqe;
242 u8 update_slave_state;
245 while ((eqe = next_eqe_sw(eq))) {
247 * Make sure we read EQ entry contents after we've
248 * checked the ownership bit.
253 case MLX4_EVENT_TYPE_COMP:
254 cqn = be32_to_cpu(eqe->event.comp.cqn) & 0xffffff;
255 mlx4_cq_completion(dev, cqn);
258 case MLX4_EVENT_TYPE_PATH_MIG:
259 case MLX4_EVENT_TYPE_COMM_EST:
260 case MLX4_EVENT_TYPE_SQ_DRAINED:
261 case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE:
262 case MLX4_EVENT_TYPE_WQ_CATAS_ERROR:
263 case MLX4_EVENT_TYPE_PATH_MIG_FAILED:
264 case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
265 case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR:
266 mlx4_dbg(dev, "event %d arrived\n", eqe->type);
267 if (mlx4_is_master(dev)) {
268 /* forward only to slave owning the QP */
269 ret = mlx4_get_slave_from_resource_id(dev,
271 be32_to_cpu(eqe->event.qp.qpn)
273 if (ret && ret != -ENOENT) {
274 mlx4_dbg(dev, "QP event %02x(%02x) on "
275 "EQ %d at index %u: could "
276 "not get slave id (%d)\n",
277 eqe->type, eqe->subtype,
278 eq->eqn, eq->cons_index, ret);
282 if (!ret && slave != dev->caps.function) {
283 mlx4_slave_event(dev, slave, eqe);
288 mlx4_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) &
289 0xffffff, eqe->type);
292 case MLX4_EVENT_TYPE_SRQ_LIMIT:
293 mlx4_warn(dev, "%s: MLX4_EVENT_TYPE_SRQ_LIMIT\n",
295 case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
296 if (mlx4_is_master(dev)) {
297 /* forward only to slave owning the SRQ */
298 ret = mlx4_get_slave_from_resource_id(dev,
300 be32_to_cpu(eqe->event.srq.srqn)
303 if (ret && ret != -ENOENT) {
304 mlx4_warn(dev, "SRQ event %02x(%02x) "
305 "on EQ %d at index %u: could"
306 " not get slave id (%d)\n",
307 eqe->type, eqe->subtype,
308 eq->eqn, eq->cons_index, ret);
311 mlx4_warn(dev, "%s: slave:%d, srq_no:0x%x,"
312 " event: %02x(%02x)\n", __func__,
314 be32_to_cpu(eqe->event.srq.srqn),
315 eqe->type, eqe->subtype);
317 if (!ret && slave != dev->caps.function) {
318 mlx4_warn(dev, "%s: sending event "
319 "%02x(%02x) to slave:%d\n",
321 eqe->subtype, slave);
322 mlx4_slave_event(dev, slave, eqe);
326 mlx4_srq_event(dev, be32_to_cpu(eqe->event.srq.srqn) &
327 0xffffff, eqe->type);
330 case MLX4_EVENT_TYPE_CMD:
332 be16_to_cpu(eqe->event.cmd.token),
333 eqe->event.cmd.status,
334 be64_to_cpu(eqe->event.cmd.out_param));
337 case MLX4_EVENT_TYPE_PORT_CHANGE:
338 port = be32_to_cpu(eqe->event.port_change.port) >> 28;
339 if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) {
340 mlx4_dispatch_event(dev,
341 MLX4_DEV_EVENT_PORT_DOWN,
343 mlx4_priv(dev)->sense.do_sense_port[port] = 1;
344 if (mlx4_is_master(dev))
345 /*change the state of all slave's port
347 for (i = 0; i < dev->num_slaves; i++) {
348 mlx4_dbg(dev, "%s: Sending "
349 "MLX4_PORT_CHANGE_SUBTYPE_DOWN"
350 " to slave: %d, port:%d\n",
352 if (i == dev->caps.function)
354 mlx4_slave_event(dev, i, eqe);
357 mlx4_dispatch_event(dev,
358 MLX4_DEV_EVENT_PORT_UP,
360 mlx4_priv(dev)->sense.do_sense_port[port] = 0;
362 if (mlx4_is_master(dev)) {
363 for (i = 0; i < dev->num_slaves; i++) {
364 if (i == dev->caps.function)
366 mlx4_slave_event(dev, i, eqe);
372 case MLX4_EVENT_TYPE_CQ_ERROR:
373 mlx4_warn(dev, "CQ %s on CQN %06x\n",
374 eqe->event.cq_err.syndrome == 1 ?
375 "overrun" : "access violation",
376 be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff);
377 if (mlx4_is_master(dev)) {
378 ret = mlx4_get_slave_from_resource_id(dev,
380 be32_to_cpu(eqe->event.cq_err.cqn)
382 if (ret && ret != -ENOENT) {
383 mlx4_dbg(dev, "CQ event %02x(%02x) on "
384 "EQ %d at index %u: could "
385 "not get slave id (%d)\n",
386 eqe->type, eqe->subtype,
387 eq->eqn, eq->cons_index, ret);
391 if (!ret && slave != dev->caps.function) {
392 mlx4_slave_event(dev, slave, eqe);
397 be32_to_cpu(eqe->event.cq_err.cqn)
402 case MLX4_EVENT_TYPE_EQ_OVERFLOW:
403 mlx4_warn(dev, "EQ overrun on EQN %d\n", eq->eqn);
406 case MLX4_EVENT_TYPE_COMM_CHANNEL:
407 if (!mlx4_is_master(dev)) {
408 mlx4_warn(dev, "Received comm channel event "
409 "for non master device\n");
412 memcpy(&priv->mfunc.master.comm_arm_bit_vector,
413 eqe->event.comm_channel_arm.bit_vec,
414 sizeof eqe->event.comm_channel_arm.bit_vec);
415 queue_work(priv->mfunc.master.comm_wq,
416 &priv->mfunc.master.comm_work);
419 case MLX4_EVENT_TYPE_FLR_EVENT:
420 flr_slave = be32_to_cpu(eqe->event.flr_event.slave_id);
421 if (!mlx4_is_master(dev)) {
422 mlx4_warn(dev, "Non-master function received"
427 mlx4_dbg(dev, "FLR event for slave: %d\n", flr_slave);
429 if (flr_slave > dev->num_slaves) {
431 "Got FLR for unknown function: %d\n",
433 update_slave_state = 0;
435 update_slave_state = 1;
437 spin_lock(&priv->mfunc.master.slave_state_lock);
438 if (update_slave_state) {
439 priv->mfunc.master.slave_state[flr_slave].active = false;
440 priv->mfunc.master.slave_state[flr_slave].last_cmd = MLX4_COMM_CMD_FLR;
441 priv->mfunc.master.slave_state[flr_slave].is_slave_going_down = 1;
443 spin_unlock(&priv->mfunc.master.slave_state_lock);
444 queue_work(priv->mfunc.master.comm_wq,
445 &priv->mfunc.master.slave_flr_event_work);
448 case MLX4_EVENT_TYPE_FATAL_WARNING:
449 if (eqe->subtype == MLX4_FATAL_WARNING_SUBTYPE_WARMING) {
450 if (mlx4_is_master(dev))
451 for (i = 0; i < dev->num_slaves; i++) {
452 mlx4_dbg(dev, "%s: Sending "
453 "MLX4_FATAL_WARNING_SUBTYPE_WARMING"
454 " to slave: %d\n", __func__, i);
455 if (i == dev->caps.function)
457 mlx4_slave_event(dev, i, eqe);
459 mlx4_err(dev, "Temperature Threshold was reached! "
460 "Threshold: %d celsius degrees; "
461 "Current Temperature: %d\n",
462 be16_to_cpu(eqe->event.warming.warning_threshold),
463 be16_to_cpu(eqe->event.warming.current_temperature));
465 mlx4_warn(dev, "Unhandled event FATAL WARNING (%02x), "
466 "subtype %02x on EQ %d at index %u. owner=%x, "
467 "nent=0x%x, slave=%x, ownership=%s\n",
468 eqe->type, eqe->subtype, eq->eqn,
469 eq->cons_index, eqe->owner, eq->nent,
471 !!(eqe->owner & 0x80) ^
472 !!(eq->cons_index & eq->nent) ? "HW" : "SW");
476 case MLX4_EVENT_TYPE_EEC_CATAS_ERROR:
477 case MLX4_EVENT_TYPE_ECC_DETECT:
479 mlx4_warn(dev, "Unhandled event %02x(%02x) on EQ %d at "
480 "index %u. owner=%x, nent=0x%x, slave=%x, "
482 eqe->type, eqe->subtype, eq->eqn,
483 eq->cons_index, eqe->owner, eq->nent,
485 !!(eqe->owner & 0x80) ^
486 !!(eq->cons_index & eq->nent) ? "HW" : "SW");
495 * The HCA will think the queue has overflowed if we
496 * don't tell it we've been processing events. We
497 * create our EQs with MLX4_NUM_SPARE_EQE extra
498 * entries, so we must update our consumer index at
501 if (unlikely(set_ci >= MLX4_NUM_SPARE_EQE)) {
512 static irqreturn_t mlx4_interrupt(int irq, void *dev_ptr)
514 struct mlx4_dev *dev = dev_ptr;
515 struct mlx4_priv *priv = mlx4_priv(dev);
519 writel(priv->eq_table.clr_mask, priv->eq_table.clr_int);
521 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
522 work |= mlx4_eq_int(dev, &priv->eq_table.eq[i]);
524 return IRQ_RETVAL(work);
527 static irqreturn_t mlx4_msi_x_interrupt(int irq, void *eq_ptr)
529 struct mlx4_eq *eq = eq_ptr;
530 struct mlx4_dev *dev = eq->dev;
532 mlx4_eq_int(dev, eq);
534 /* MSI-X vectors always belong to us */
538 int mlx4_MAP_EQ_wrapper(struct mlx4_dev *dev, int slave,
539 struct mlx4_vhcr *vhcr,
540 struct mlx4_cmd_mailbox *inbox,
541 struct mlx4_cmd_mailbox *outbox,
542 struct mlx4_cmd_info *cmd)
544 struct mlx4_priv *priv = mlx4_priv(dev);
545 struct mlx4_slave_event_eq_info *event_eq =
546 priv->mfunc.master.slave_state[slave].event_eq;
547 u32 in_modifier = vhcr->in_modifier;
548 u32 eqn = in_modifier & 0x1FF;
549 u64 in_param = vhcr->in_param;
553 if (slave == dev->caps.function)
554 err = mlx4_cmd(dev, in_param, (in_modifier & 0x80000000) | eqn,
555 0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
558 for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i)
559 if (in_param & (1LL << i))
560 event_eq[i].eqn = in_modifier >> 31 ? -1 : eqn;
565 static int mlx4_MAP_EQ(struct mlx4_dev *dev, u64 event_mask, int unmap,
568 return mlx4_cmd(dev, event_mask, (unmap << 31) | eq_num,
569 0, MLX4_CMD_MAP_EQ, MLX4_CMD_TIME_CLASS_B,
573 static int mlx4_SW2HW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
576 return mlx4_cmd(dev, mailbox->dma, eq_num, 0,
577 MLX4_CMD_SW2HW_EQ, MLX4_CMD_TIME_CLASS_A,
581 static int mlx4_HW2SW_EQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
584 return mlx4_cmd_box(dev, 0, mailbox->dma, eq_num,
585 0, MLX4_CMD_HW2SW_EQ, MLX4_CMD_TIME_CLASS_A,
589 static int mlx4_num_eq_uar(struct mlx4_dev *dev)
592 * Each UAR holds 4 EQ doorbells. To figure out how many UARs
593 * we need to map, take the difference of highest index and
594 * the lowest index we'll use and add 1.
596 return (dev->caps.num_comp_vectors + 1 + dev->caps.reserved_eqs +
597 dev->caps.comp_pool)/4 - dev->caps.reserved_eqs/4 + 1;
600 static void __iomem *mlx4_get_eq_uar(struct mlx4_dev *dev, struct mlx4_eq *eq)
602 struct mlx4_priv *priv = mlx4_priv(dev);
605 index = eq->eqn / 4 - dev->caps.reserved_eqs / 4;
607 if (!priv->eq_table.uar_map[index]) {
608 priv->eq_table.uar_map[index] =
609 ioremap(pci_resource_start(dev->pdev, 2) +
610 ((eq->eqn / 4) << PAGE_SHIFT),
612 if (!priv->eq_table.uar_map[index]) {
613 mlx4_err(dev, "Couldn't map EQ doorbell for EQN 0x%06x\n",
619 return priv->eq_table.uar_map[index] + 0x800 + 8 * (eq->eqn % 4);
622 static int mlx4_create_eq(struct mlx4_dev *dev, int nent,
623 u8 intr, struct mlx4_eq *eq)
625 struct mlx4_priv *priv = mlx4_priv(dev);
626 struct mlx4_cmd_mailbox *mailbox;
627 struct mlx4_eq_context *eq_context;
629 u64 *dma_list = NULL;
636 eq->nent = roundup_pow_of_two(max(nent, 2));
637 npages = PAGE_ALIGN(eq->nent * MLX4_EQ_ENTRY_SIZE) / PAGE_SIZE;
639 eq->page_list = kmalloc(npages * sizeof *eq->page_list,
644 for (i = 0; i < npages; ++i)
645 eq->page_list[i].buf = NULL;
647 dma_list = kmalloc(npages * sizeof *dma_list, GFP_KERNEL);
651 mailbox = mlx4_alloc_cmd_mailbox(dev);
654 eq_context = mailbox->buf;
656 for (i = 0; i < npages; ++i) {
657 eq->page_list[i].buf = dma_alloc_coherent(&dev->pdev->dev,
658 PAGE_SIZE, &t, GFP_KERNEL);
659 if (!eq->page_list[i].buf)
660 goto err_out_free_pages;
663 eq->page_list[i].map = t;
665 memset(eq->page_list[i].buf, 0, PAGE_SIZE);
668 eq->eqn = mlx4_bitmap_alloc(&priv->eq_table.bitmap);
670 goto err_out_free_pages;
672 eq->doorbell = mlx4_get_eq_uar(dev, eq);
675 goto err_out_free_eq;
678 err = mlx4_mtt_init(dev, npages, PAGE_SHIFT, &eq->mtt);
680 goto err_out_free_eq;
682 err = mlx4_write_mtt(dev, &eq->mtt, 0, npages, dma_list);
684 goto err_out_free_mtt;
686 memset(eq_context, 0, sizeof *eq_context);
687 eq_context->flags = cpu_to_be32(MLX4_EQ_STATUS_OK |
688 MLX4_EQ_STATE_ARMED);
689 eq_context->log_eq_size = ilog2(eq->nent);
690 eq_context->intr = intr;
691 eq_context->log_page_size = PAGE_SHIFT - MLX4_ICM_PAGE_SHIFT;
693 mtt_addr = mlx4_mtt_addr(dev, &eq->mtt);
694 eq_context->mtt_base_addr_h = mtt_addr >> 32;
695 eq_context->mtt_base_addr_l = cpu_to_be32(mtt_addr & 0xffffffff);
697 err = mlx4_SW2HW_EQ(dev, mailbox, eq->eqn);
699 mlx4_warn(dev, "SW2HW_EQ failed (%d)\n", err);
700 goto err_out_free_mtt;
704 mlx4_free_cmd_mailbox(dev, mailbox);
711 mlx4_mtt_cleanup(dev, &eq->mtt);
714 mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn);
717 for (i = 0; i < npages; ++i)
718 if (eq->page_list[i].buf)
719 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
720 eq->page_list[i].buf,
721 eq->page_list[i].map);
723 mlx4_free_cmd_mailbox(dev, mailbox);
726 kfree(eq->page_list);
733 static void mlx4_free_eq(struct mlx4_dev *dev,
736 struct mlx4_priv *priv = mlx4_priv(dev);
737 struct mlx4_cmd_mailbox *mailbox;
739 int npages = PAGE_ALIGN(MLX4_EQ_ENTRY_SIZE * eq->nent) / PAGE_SIZE;
742 mailbox = mlx4_alloc_cmd_mailbox(dev);
746 err = mlx4_HW2SW_EQ(dev, mailbox, eq->eqn);
748 mlx4_warn(dev, "HW2SW_EQ failed (%d)\n", err);
751 mlx4_dbg(dev, "Dumping EQ context %02x:\n", eq->eqn);
752 for (i = 0; i < sizeof (struct mlx4_eq_context) / 4; ++i) {
754 pr_cont("[%02x] ", i * 4);
755 pr_cont(" %08x", be32_to_cpup(mailbox->buf + i * 4));
756 if ((i + 1) % 4 == 0)
761 mlx4_mtt_cleanup(dev, &eq->mtt);
762 for (i = 0; i < npages; ++i)
763 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
764 eq->page_list[i].buf,
765 eq->page_list[i].map);
767 kfree(eq->page_list);
768 mlx4_bitmap_free(&priv->eq_table.bitmap, eq->eqn);
769 mlx4_free_cmd_mailbox(dev, mailbox);
772 static void mlx4_free_irqs(struct mlx4_dev *dev)
774 struct mlx4_eq_table *eq_table = &mlx4_priv(dev)->eq_table;
775 struct mlx4_priv *priv = mlx4_priv(dev);
778 if (eq_table->have_irq)
779 free_irq(dev->pdev->irq, dev);
781 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
782 if (eq_table->eq[i].have_irq) {
783 free_irq(eq_table->eq[i].irq, eq_table->eq + i);
784 eq_table->eq[i].have_irq = 0;
787 for (i = 0; i < dev->caps.comp_pool; i++) {
789 * Freeing the assigned irq's
790 * all bits should be 0, but we need to validate
792 if (priv->msix_ctl.pool_bm & 1ULL << i) {
793 /* NO need protecting*/
794 vec = dev->caps.num_comp_vectors + 1 + i;
795 free_irq(priv->eq_table.eq[vec].irq,
796 &priv->eq_table.eq[vec]);
801 kfree(eq_table->irq_names);
804 static int mlx4_map_clr_int(struct mlx4_dev *dev)
806 struct mlx4_priv *priv = mlx4_priv(dev);
808 priv->clr_base = ioremap(pci_resource_start(dev->pdev, priv->fw.clr_int_bar) +
809 priv->fw.clr_int_base, MLX4_CLR_INT_SIZE);
810 if (!priv->clr_base) {
811 mlx4_err(dev, "Couldn't map interrupt clear register, aborting.\n");
818 static void mlx4_unmap_clr_int(struct mlx4_dev *dev)
820 struct mlx4_priv *priv = mlx4_priv(dev);
822 iounmap(priv->clr_base);
825 int mlx4_alloc_eq_table(struct mlx4_dev *dev)
827 struct mlx4_priv *priv = mlx4_priv(dev);
829 priv->eq_table.eq = kcalloc(dev->caps.num_eqs - dev->caps.reserved_eqs,
830 sizeof *priv->eq_table.eq, GFP_KERNEL);
831 if (!priv->eq_table.eq)
837 void mlx4_free_eq_table(struct mlx4_dev *dev)
839 kfree(mlx4_priv(dev)->eq_table.eq);
842 int mlx4_init_eq_table(struct mlx4_dev *dev)
844 struct mlx4_priv *priv = mlx4_priv(dev);
848 priv->eq_table.uar_map = kcalloc(mlx4_num_eq_uar(dev),
849 sizeof *priv->eq_table.uar_map,
851 if (!priv->eq_table.uar_map) {
856 err = mlx4_bitmap_init(&priv->eq_table.bitmap, dev->caps.num_eqs,
857 dev->caps.num_eqs - 1, dev->caps.reserved_eqs, 0);
861 for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
862 priv->eq_table.uar_map[i] = NULL;
864 if (!mlx4_is_slave(dev)) {
865 err = mlx4_map_clr_int(dev);
869 priv->eq_table.clr_mask =
870 swab32(1 << (priv->eq_table.inta_pin & 31));
871 priv->eq_table.clr_int = priv->clr_base +
872 (priv->eq_table.inta_pin < 32 ? 4 : 0);
875 priv->eq_table.irq_names =
876 kmalloc(MLX4_IRQNAME_SIZE * (dev->caps.num_comp_vectors + 1 +
877 dev->caps.comp_pool),
879 if (!priv->eq_table.irq_names) {
884 for (i = 0; i < dev->caps.num_comp_vectors; ++i) {
885 err = mlx4_create_eq(dev, dev->caps.num_cqs -
886 dev->caps.reserved_cqs +
888 (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
889 &priv->eq_table.eq[i]);
896 err = mlx4_create_eq(dev, MLX4_NUM_ASYNC_EQE + MLX4_NUM_SPARE_EQE,
897 (dev->flags & MLX4_FLAG_MSI_X) ? dev->caps.num_comp_vectors : 0,
898 &priv->eq_table.eq[dev->caps.num_comp_vectors]);
902 /*if additional completion vectors poolsize is 0 this loop will not run*/
903 for (i = dev->caps.num_comp_vectors + 1;
904 i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i) {
906 err = mlx4_create_eq(dev, dev->caps.num_cqs -
907 dev->caps.reserved_cqs +
909 (dev->flags & MLX4_FLAG_MSI_X) ? i : 0,
910 &priv->eq_table.eq[i]);
918 if (dev->flags & MLX4_FLAG_MSI_X) {
921 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i) {
922 if (i < dev->caps.num_comp_vectors) {
923 snprintf(priv->eq_table.irq_names +
924 i * MLX4_IRQNAME_SIZE,
926 "mlx4-comp-%d@pci:%s", i,
927 pci_name(dev->pdev));
929 snprintf(priv->eq_table.irq_names +
930 i * MLX4_IRQNAME_SIZE,
933 pci_name(dev->pdev));
936 eq_name = priv->eq_table.irq_names +
937 i * MLX4_IRQNAME_SIZE;
938 err = request_irq(priv->eq_table.eq[i].irq,
939 mlx4_msi_x_interrupt, 0, eq_name,
940 priv->eq_table.eq + i);
944 priv->eq_table.eq[i].have_irq = 1;
947 snprintf(priv->eq_table.irq_names,
950 pci_name(dev->pdev));
951 err = request_irq(dev->pdev->irq, mlx4_interrupt,
952 IRQF_SHARED, priv->eq_table.irq_names, dev);
956 priv->eq_table.have_irq = 1;
959 err = mlx4_MAP_EQ(dev, MLX4_ASYNC_EVENT_MASK, 0,
960 priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
962 mlx4_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
963 priv->eq_table.eq[dev->caps.num_comp_vectors].eqn, err);
965 for (i = 0; i < dev->caps.num_comp_vectors + 1; ++i)
966 eq_set_ci(&priv->eq_table.eq[i], 1);
971 mlx4_free_eq(dev, &priv->eq_table.eq[dev->caps.num_comp_vectors]);
974 i = dev->caps.num_comp_vectors - 1;
978 mlx4_free_eq(dev, &priv->eq_table.eq[i]);
981 if (!mlx4_is_slave(dev))
982 mlx4_unmap_clr_int(dev);
986 mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
989 kfree(priv->eq_table.uar_map);
994 void mlx4_cleanup_eq_table(struct mlx4_dev *dev)
996 struct mlx4_priv *priv = mlx4_priv(dev);
999 mlx4_MAP_EQ(dev, MLX4_ASYNC_EVENT_MASK, 1,
1000 priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1002 mlx4_free_irqs(dev);
1004 for (i = 0; i < dev->caps.num_comp_vectors + dev->caps.comp_pool + 1; ++i)
1005 mlx4_free_eq(dev, &priv->eq_table.eq[i]);
1007 if (!mlx4_is_slave(dev))
1008 mlx4_unmap_clr_int(dev);
1010 for (i = 0; i < mlx4_num_eq_uar(dev); ++i)
1011 if (priv->eq_table.uar_map[i])
1012 iounmap(priv->eq_table.uar_map[i]);
1014 mlx4_bitmap_cleanup(&priv->eq_table.bitmap);
1016 kfree(priv->eq_table.uar_map);
1019 /* A test that verifies that we can accept interrupts on all
1020 * the irq vectors of the device.
1021 * Interrupts are checked using the NOP command.
1023 int mlx4_test_interrupts(struct mlx4_dev *dev)
1025 struct mlx4_priv *priv = mlx4_priv(dev);
1029 err = mlx4_NOP(dev);
1030 /* When not in MSI_X, there is only one irq to check */
1031 if (!(dev->flags & MLX4_FLAG_MSI_X) || mlx4_is_slave(dev))
1034 /* A loop over all completion vectors, for each vector we will check
1035 * whether it works by mapping command completions to that vector
1036 * and performing a NOP command
1038 for(i = 0; !err && (i < dev->caps.num_comp_vectors); ++i) {
1039 /* Temporary use polling for command completions */
1040 mlx4_cmd_use_polling(dev);
1042 /* Map the new eq to handle all asyncronous events */
1043 err = mlx4_MAP_EQ(dev, MLX4_ASYNC_EVENT_MASK, 0,
1044 priv->eq_table.eq[i].eqn);
1046 mlx4_warn(dev, "Failed mapping eq for interrupt test\n");
1047 mlx4_cmd_use_events(dev);
1051 /* Go back to using events */
1052 mlx4_cmd_use_events(dev);
1053 err = mlx4_NOP(dev);
1056 /* Return to default */
1057 mlx4_MAP_EQ(dev, MLX4_ASYNC_EVENT_MASK, 0,
1058 priv->eq_table.eq[dev->caps.num_comp_vectors].eqn);
1061 EXPORT_SYMBOL(mlx4_test_interrupts);
1063 int mlx4_assign_eq(struct mlx4_dev *dev, char* name, int * vector)
1066 struct mlx4_priv *priv = mlx4_priv(dev);
1067 int vec = 0, err = 0, i;
1069 spin_lock(&priv->msix_ctl.pool_lock);
1070 for (i = 0; !vec && i < dev->caps.comp_pool; i++) {
1071 if (~priv->msix_ctl.pool_bm & 1ULL << i) {
1072 priv->msix_ctl.pool_bm |= 1ULL << i;
1073 vec = dev->caps.num_comp_vectors + 1 + i;
1074 snprintf(priv->eq_table.irq_names +
1075 vec * MLX4_IRQNAME_SIZE,
1076 MLX4_IRQNAME_SIZE, "%s", name);
1077 err = request_irq(priv->eq_table.eq[vec].irq,
1078 mlx4_msi_x_interrupt, 0,
1079 &priv->eq_table.irq_names[vec<<5],
1080 priv->eq_table.eq + vec);
1082 /*zero out bit by fliping it*/
1083 priv->msix_ctl.pool_bm ^= 1 << i;
1086 /*we dont want to break here*/
1088 eq_set_ci(&priv->eq_table.eq[vec], 1);
1091 spin_unlock(&priv->msix_ctl.pool_lock);
1097 err = (i == dev->caps.comp_pool) ? -ENOSPC : err;
1101 EXPORT_SYMBOL(mlx4_assign_eq);
1103 void mlx4_release_eq(struct mlx4_dev *dev, int vec)
1105 struct mlx4_priv *priv = mlx4_priv(dev);
1107 int i = vec - dev->caps.num_comp_vectors - 1;
1109 if (likely(i >= 0)) {
1110 /*sanity check , making sure were not trying to free irq's
1111 Belonging to a legacy EQ*/
1112 spin_lock(&priv->msix_ctl.pool_lock);
1113 if (priv->msix_ctl.pool_bm & 1ULL << i) {
1114 free_irq(priv->eq_table.eq[vec].irq,
1115 &priv->eq_table.eq[vec]);
1116 priv->msix_ctl.pool_bm &= ~(1ULL << i);
1118 spin_unlock(&priv->msix_ctl.pool_lock);
1122 EXPORT_SYMBOL(mlx4_release_eq);