2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc. All rights reserved.
5 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
6 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
37 #include <linux/hardirq.h>
38 #include <linux/export.h>
40 #include <linux/mlx4/cmd.h>
41 #include <linux/mlx4/cq.h>
46 #define MLX4_CQ_STATUS_OK ( 0 << 28)
47 #define MLX4_CQ_STATUS_OVERFLOW ( 9 << 28)
48 #define MLX4_CQ_STATUS_WRITE_FAIL (10 << 28)
49 #define MLX4_CQ_FLAG_CC ( 1 << 18)
50 #define MLX4_CQ_FLAG_OI ( 1 << 17)
51 #define MLX4_CQ_STATE_ARMED ( 9 << 8)
52 #define MLX4_CQ_STATE_ARMED_SOL ( 6 << 8)
53 #define MLX4_EQ_STATE_FIRED (10 << 8)
55 #define TASKLET_MAX_TIME 2
56 #define TASKLET_MAX_TIME_JIFFIES msecs_to_jiffies(TASKLET_MAX_TIME)
58 void mlx4_cq_tasklet_cb(struct tasklet_struct
*t
)
61 unsigned long end
= jiffies
+ TASKLET_MAX_TIME_JIFFIES
;
62 struct mlx4_eq_tasklet
*ctx
= from_tasklet(ctx
, t
, task
);
63 struct mlx4_cq
*mcq
, *temp
;
65 spin_lock_irqsave(&ctx
->lock
, flags
);
66 list_splice_tail_init(&ctx
->list
, &ctx
->process_list
);
67 spin_unlock_irqrestore(&ctx
->lock
, flags
);
69 list_for_each_entry_safe(mcq
, temp
, &ctx
->process_list
, tasklet_ctx
.list
) {
70 list_del_init(&mcq
->tasklet_ctx
.list
);
71 mcq
->tasklet_ctx
.comp(mcq
);
72 if (refcount_dec_and_test(&mcq
->refcount
))
74 if (time_after(jiffies
, end
))
78 if (!list_empty(&ctx
->process_list
))
79 tasklet_schedule(&ctx
->task
);
82 static void mlx4_add_cq_to_tasklet(struct mlx4_cq
*cq
)
84 struct mlx4_eq_tasklet
*tasklet_ctx
= cq
->tasklet_ctx
.priv
;
88 spin_lock_irqsave(&tasklet_ctx
->lock
, flags
);
89 /* When migrating CQs between EQs will be implemented, please note
90 * that you need to sync this point. It is possible that
91 * while migrating a CQ, completions on the old EQs could
94 if (list_empty_careful(&cq
->tasklet_ctx
.list
)) {
95 refcount_inc(&cq
->refcount
);
96 kick
= list_empty(&tasklet_ctx
->list
);
97 list_add_tail(&cq
->tasklet_ctx
.list
, &tasklet_ctx
->list
);
99 tasklet_schedule(&tasklet_ctx
->task
);
101 spin_unlock_irqrestore(&tasklet_ctx
->lock
, flags
);
104 void mlx4_cq_completion(struct mlx4_dev
*dev
, u32 cqn
)
109 cq
= radix_tree_lookup(&mlx4_priv(dev
)->cq_table
.tree
,
110 cqn
& (dev
->caps
.num_cqs
- 1));
114 mlx4_dbg(dev
, "Completion event for bogus CQ %08x\n", cqn
);
118 /* Acessing the CQ outside of rcu_read_lock is safe, because
119 * the CQ is freed only after interrupt handling is completed.
126 void mlx4_cq_event(struct mlx4_dev
*dev
, u32 cqn
, int event_type
)
128 struct mlx4_cq_table
*cq_table
= &mlx4_priv(dev
)->cq_table
;
132 cq
= radix_tree_lookup(&cq_table
->tree
, cqn
& (dev
->caps
.num_cqs
- 1));
136 mlx4_dbg(dev
, "Async event for bogus CQ %08x\n", cqn
);
140 /* Acessing the CQ outside of rcu_read_lock is safe, because
141 * the CQ is freed only after interrupt handling is completed.
143 cq
->event(cq
, event_type
);
146 static int mlx4_SW2HW_CQ(struct mlx4_dev
*dev
, struct mlx4_cmd_mailbox
*mailbox
,
147 int cq_num
, u8 opmod
)
149 return mlx4_cmd(dev
, mailbox
->dma
, cq_num
, opmod
,
150 MLX4_CMD_SW2HW_CQ
, MLX4_CMD_TIME_CLASS_A
,
154 static int mlx4_MODIFY_CQ(struct mlx4_dev
*dev
, struct mlx4_cmd_mailbox
*mailbox
,
155 int cq_num
, u32 opmod
)
157 return mlx4_cmd(dev
, mailbox
->dma
, cq_num
, opmod
, MLX4_CMD_MODIFY_CQ
,
158 MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_WRAPPED
);
161 static int mlx4_HW2SW_CQ(struct mlx4_dev
*dev
, struct mlx4_cmd_mailbox
*mailbox
,
164 return mlx4_cmd_box(dev
, 0, mailbox
? mailbox
->dma
: 0,
165 cq_num
, mailbox
? 0 : 1, MLX4_CMD_HW2SW_CQ
,
166 MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_WRAPPED
);
169 int mlx4_cq_modify(struct mlx4_dev
*dev
, struct mlx4_cq
*cq
,
170 u16 count
, u16 period
)
172 struct mlx4_cmd_mailbox
*mailbox
;
173 struct mlx4_cq_context
*cq_context
;
176 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
178 return PTR_ERR(mailbox
);
180 cq_context
= mailbox
->buf
;
181 cq_context
->cq_max_count
= cpu_to_be16(count
);
182 cq_context
->cq_period
= cpu_to_be16(period
);
184 err
= mlx4_MODIFY_CQ(dev
, mailbox
, cq
->cqn
, 1);
186 mlx4_free_cmd_mailbox(dev
, mailbox
);
189 EXPORT_SYMBOL_GPL(mlx4_cq_modify
);
191 int mlx4_cq_resize(struct mlx4_dev
*dev
, struct mlx4_cq
*cq
,
192 int entries
, struct mlx4_mtt
*mtt
)
194 struct mlx4_cmd_mailbox
*mailbox
;
195 struct mlx4_cq_context
*cq_context
;
199 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
201 return PTR_ERR(mailbox
);
203 cq_context
= mailbox
->buf
;
204 cq_context
->logsize_usrpage
= cpu_to_be32(ilog2(entries
) << 24);
205 cq_context
->log_page_size
= mtt
->page_shift
- 12;
206 mtt_addr
= mlx4_mtt_addr(dev
, mtt
);
207 cq_context
->mtt_base_addr_h
= mtt_addr
>> 32;
208 cq_context
->mtt_base_addr_l
= cpu_to_be32(mtt_addr
& 0xffffffff);
210 err
= mlx4_MODIFY_CQ(dev
, mailbox
, cq
->cqn
, 0);
212 mlx4_free_cmd_mailbox(dev
, mailbox
);
215 EXPORT_SYMBOL_GPL(mlx4_cq_resize
);
217 int __mlx4_cq_alloc_icm(struct mlx4_dev
*dev
, int *cqn
)
219 struct mlx4_priv
*priv
= mlx4_priv(dev
);
220 struct mlx4_cq_table
*cq_table
= &priv
->cq_table
;
223 *cqn
= mlx4_bitmap_alloc(&cq_table
->bitmap
);
227 err
= mlx4_table_get(dev
, &cq_table
->table
, *cqn
);
231 err
= mlx4_table_get(dev
, &cq_table
->cmpt_table
, *cqn
);
237 mlx4_table_put(dev
, &cq_table
->table
, *cqn
);
240 mlx4_bitmap_free(&cq_table
->bitmap
, *cqn
, MLX4_NO_RR
);
244 static int mlx4_cq_alloc_icm(struct mlx4_dev
*dev
, int *cqn
, u8 usage
)
246 u32 in_modifier
= RES_CQ
| (((u32
)usage
& 3) << 30);
250 if (mlx4_is_mfunc(dev
)) {
251 err
= mlx4_cmd_imm(dev
, 0, &out_param
, in_modifier
,
252 RES_OP_RESERVE_AND_MAP
, MLX4_CMD_ALLOC_RES
,
253 MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_WRAPPED
);
257 *cqn
= get_param_l(&out_param
);
261 return __mlx4_cq_alloc_icm(dev
, cqn
);
264 void __mlx4_cq_free_icm(struct mlx4_dev
*dev
, int cqn
)
266 struct mlx4_priv
*priv
= mlx4_priv(dev
);
267 struct mlx4_cq_table
*cq_table
= &priv
->cq_table
;
269 mlx4_table_put(dev
, &cq_table
->cmpt_table
, cqn
);
270 mlx4_table_put(dev
, &cq_table
->table
, cqn
);
271 mlx4_bitmap_free(&cq_table
->bitmap
, cqn
, MLX4_NO_RR
);
274 static void mlx4_cq_free_icm(struct mlx4_dev
*dev
, int cqn
)
279 if (mlx4_is_mfunc(dev
)) {
280 set_param_l(&in_param
, cqn
);
281 err
= mlx4_cmd(dev
, in_param
, RES_CQ
, RES_OP_RESERVE_AND_MAP
,
283 MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_WRAPPED
);
285 mlx4_warn(dev
, "Failed freeing cq:%d\n", cqn
);
287 __mlx4_cq_free_icm(dev
, cqn
);
290 static int mlx4_init_user_cqes(void *buf
, int entries
, int cqe_size
)
292 int entries_per_copy
= PAGE_SIZE
/ cqe_size
;
297 init_ents
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
301 /* Populate a list of CQ entries to reduce the number of
302 * copy_to_user calls. 0xcc is the initialization value
303 * required by the FW.
305 memset(init_ents
, 0xcc, PAGE_SIZE
);
307 if (entries_per_copy
< entries
) {
308 for (i
= 0; i
< entries
/ entries_per_copy
; i
++) {
309 err
= copy_to_user((void __user
*)buf
, init_ents
, PAGE_SIZE
) ?
317 err
= copy_to_user((void __user
*)buf
, init_ents
, entries
* cqe_size
) ?
327 static void mlx4_init_kernel_cqes(struct mlx4_buf
*buf
,
334 memset(buf
->direct
.buf
, 0xcc, entries
* cqe_size
);
336 for (i
= 0; i
< buf
->npages
; i
++)
337 memset(buf
->page_list
[i
].buf
, 0xcc,
338 1UL << buf
->page_shift
);
341 int mlx4_cq_alloc(struct mlx4_dev
*dev
, int nent
,
342 struct mlx4_mtt
*mtt
, struct mlx4_uar
*uar
, u64 db_rec
,
343 struct mlx4_cq
*cq
, unsigned vector
, int collapsed
,
344 int timestamp_en
, void *buf_addr
, bool user_cq
)
346 bool sw_cq_init
= dev
->caps
.flags2
& MLX4_DEV_CAP_FLAG2_SW_CQ_INIT
;
347 struct mlx4_priv
*priv
= mlx4_priv(dev
);
348 struct mlx4_cq_table
*cq_table
= &priv
->cq_table
;
349 struct mlx4_cmd_mailbox
*mailbox
;
350 struct mlx4_cq_context
*cq_context
;
354 if (vector
>= dev
->caps
.num_comp_vectors
)
359 err
= mlx4_cq_alloc_icm(dev
, &cq
->cqn
, cq
->usage
);
363 spin_lock(&cq_table
->lock
);
364 err
= radix_tree_insert(&cq_table
->tree
, cq
->cqn
, cq
);
365 spin_unlock(&cq_table
->lock
);
369 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
370 if (IS_ERR(mailbox
)) {
371 err
= PTR_ERR(mailbox
);
375 cq_context
= mailbox
->buf
;
376 cq_context
->flags
= cpu_to_be32(!!collapsed
<< 18);
378 cq_context
->flags
|= cpu_to_be32(1 << 19);
380 cq_context
->logsize_usrpage
=
381 cpu_to_be32((ilog2(nent
) << 24) |
382 mlx4_to_hw_uar_index(dev
, uar
->index
));
383 cq_context
->comp_eqn
= priv
->eq_table
.eq
[MLX4_CQ_TO_EQ_VECTOR(vector
)].eqn
;
384 cq_context
->log_page_size
= mtt
->page_shift
- MLX4_ICM_PAGE_SHIFT
;
386 mtt_addr
= mlx4_mtt_addr(dev
, mtt
);
387 cq_context
->mtt_base_addr_h
= mtt_addr
>> 32;
388 cq_context
->mtt_base_addr_l
= cpu_to_be32(mtt_addr
& 0xffffffff);
389 cq_context
->db_rec_addr
= cpu_to_be64(db_rec
);
393 err
= mlx4_init_user_cqes(buf_addr
, nent
,
398 mlx4_init_kernel_cqes(buf_addr
, nent
,
403 err
= mlx4_SW2HW_CQ(dev
, mailbox
, cq
->cqn
, sw_cq_init
);
405 mlx4_free_cmd_mailbox(dev
, mailbox
);
412 refcount_set(&cq
->refcount
, 1);
413 init_completion(&cq
->free
);
414 cq
->comp
= mlx4_add_cq_to_tasklet
;
415 cq
->tasklet_ctx
.priv
=
416 &priv
->eq_table
.eq
[MLX4_CQ_TO_EQ_VECTOR(vector
)].tasklet_ctx
;
417 INIT_LIST_HEAD(&cq
->tasklet_ctx
.list
);
420 cq
->irq
= priv
->eq_table
.eq
[MLX4_CQ_TO_EQ_VECTOR(vector
)].irq
;
424 spin_lock(&cq_table
->lock
);
425 radix_tree_delete(&cq_table
->tree
, cq
->cqn
);
426 spin_unlock(&cq_table
->lock
);
429 mlx4_cq_free_icm(dev
, cq
->cqn
);
433 EXPORT_SYMBOL_GPL(mlx4_cq_alloc
);
435 void mlx4_cq_free(struct mlx4_dev
*dev
, struct mlx4_cq
*cq
)
437 struct mlx4_priv
*priv
= mlx4_priv(dev
);
438 struct mlx4_cq_table
*cq_table
= &priv
->cq_table
;
441 err
= mlx4_HW2SW_CQ(dev
, NULL
, cq
->cqn
);
443 mlx4_warn(dev
, "HW2SW_CQ failed (%d) for CQN %06x\n", err
, cq
->cqn
);
445 spin_lock(&cq_table
->lock
);
446 radix_tree_delete(&cq_table
->tree
, cq
->cqn
);
447 spin_unlock(&cq_table
->lock
);
449 synchronize_irq(priv
->eq_table
.eq
[MLX4_CQ_TO_EQ_VECTOR(cq
->vector
)].irq
);
450 if (priv
->eq_table
.eq
[MLX4_CQ_TO_EQ_VECTOR(cq
->vector
)].irq
!=
451 priv
->eq_table
.eq
[MLX4_EQ_ASYNC
].irq
)
452 synchronize_irq(priv
->eq_table
.eq
[MLX4_EQ_ASYNC
].irq
);
454 if (refcount_dec_and_test(&cq
->refcount
))
456 wait_for_completion(&cq
->free
);
458 mlx4_cq_free_icm(dev
, cq
->cqn
);
460 EXPORT_SYMBOL_GPL(mlx4_cq_free
);
462 int mlx4_init_cq_table(struct mlx4_dev
*dev
)
464 struct mlx4_cq_table
*cq_table
= &mlx4_priv(dev
)->cq_table
;
466 spin_lock_init(&cq_table
->lock
);
467 INIT_RADIX_TREE(&cq_table
->tree
, GFP_ATOMIC
);
468 if (mlx4_is_slave(dev
))
471 return mlx4_bitmap_init(&cq_table
->bitmap
, dev
->caps
.num_cqs
,
472 dev
->caps
.num_cqs
- 1, dev
->caps
.reserved_cqs
, 0);
475 void mlx4_cleanup_cq_table(struct mlx4_dev
*dev
)
477 if (mlx4_is_slave(dev
))
479 /* Nothing to do to clean up radix_tree */
480 mlx4_bitmap_cleanup(&mlx4_priv(dev
)->cq_table
.bitmap
);