2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
4 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #include <linux/errno.h>
36 #include <linux/export.h>
37 #include <linux/slab.h>
38 #include <linux/kernel.h>
39 #include <linux/vmalloc.h>
41 #include <linux/mlx4/cmd.h>
46 static u32
mlx4_buddy_alloc(struct mlx4_buddy
*buddy
, int order
)
52 spin_lock(&buddy
->lock
);
54 for (o
= order
; o
<= buddy
->max_order
; ++o
)
55 if (buddy
->num_free
[o
]) {
56 m
= 1 << (buddy
->max_order
- o
);
57 seg
= find_first_bit(buddy
->bits
[o
], m
);
62 spin_unlock(&buddy
->lock
);
66 clear_bit(seg
, buddy
->bits
[o
]);
72 set_bit(seg
^ 1, buddy
->bits
[o
]);
76 spin_unlock(&buddy
->lock
);
83 static void mlx4_buddy_free(struct mlx4_buddy
*buddy
, u32 seg
, int order
)
87 spin_lock(&buddy
->lock
);
89 while (test_bit(seg
^ 1, buddy
->bits
[order
])) {
90 clear_bit(seg
^ 1, buddy
->bits
[order
]);
91 --buddy
->num_free
[order
];
96 set_bit(seg
, buddy
->bits
[order
]);
97 ++buddy
->num_free
[order
];
99 spin_unlock(&buddy
->lock
);
102 static int mlx4_buddy_init(struct mlx4_buddy
*buddy
, int max_order
)
106 buddy
->max_order
= max_order
;
107 spin_lock_init(&buddy
->lock
);
109 buddy
->bits
= kcalloc(buddy
->max_order
+ 1, sizeof (long *),
111 buddy
->num_free
= kcalloc((buddy
->max_order
+ 1), sizeof *buddy
->num_free
,
113 if (!buddy
->bits
|| !buddy
->num_free
)
116 for (i
= 0; i
<= buddy
->max_order
; ++i
) {
117 s
= BITS_TO_LONGS(1 << (buddy
->max_order
- i
));
118 buddy
->bits
[i
] = kcalloc(s
, sizeof (long), GFP_KERNEL
| __GFP_NOWARN
);
119 if (!buddy
->bits
[i
]) {
120 buddy
->bits
[i
] = vzalloc(s
* sizeof(long));
126 set_bit(0, buddy
->bits
[buddy
->max_order
]);
127 buddy
->num_free
[buddy
->max_order
] = 1;
132 for (i
= 0; i
<= buddy
->max_order
; ++i
)
133 if (buddy
->bits
[i
] && is_vmalloc_addr(buddy
->bits
[i
]))
134 vfree(buddy
->bits
[i
]);
136 kfree(buddy
->bits
[i
]);
140 kfree(buddy
->num_free
);
145 static void mlx4_buddy_cleanup(struct mlx4_buddy
*buddy
)
149 for (i
= 0; i
<= buddy
->max_order
; ++i
)
150 if (is_vmalloc_addr(buddy
->bits
[i
]))
151 vfree(buddy
->bits
[i
]);
153 kfree(buddy
->bits
[i
]);
156 kfree(buddy
->num_free
);
159 u32
__mlx4_alloc_mtt_range(struct mlx4_dev
*dev
, int order
)
161 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
166 seg_order
= max_t(int, order
- log_mtts_per_seg
, 0);
168 seg
= mlx4_buddy_alloc(&mr_table
->mtt_buddy
, seg_order
);
172 offset
= seg
* (1 << log_mtts_per_seg
);
174 if (mlx4_table_get_range(dev
, &mr_table
->mtt_table
, offset
,
175 offset
+ (1 << order
) - 1)) {
176 mlx4_buddy_free(&mr_table
->mtt_buddy
, seg
, seg_order
);
183 static u32
mlx4_alloc_mtt_range(struct mlx4_dev
*dev
, int order
)
189 if (mlx4_is_mfunc(dev
)) {
190 set_param_l(&in_param
, order
);
191 err
= mlx4_cmd_imm(dev
, in_param
, &out_param
, RES_MTT
,
192 RES_OP_RESERVE_AND_MAP
,
194 MLX4_CMD_TIME_CLASS_A
,
198 return get_param_l(&out_param
);
200 return __mlx4_alloc_mtt_range(dev
, order
);
203 int mlx4_mtt_init(struct mlx4_dev
*dev
, int npages
, int page_shift
,
204 struct mlx4_mtt
*mtt
)
210 mtt
->page_shift
= MLX4_ICM_PAGE_SHIFT
;
213 mtt
->page_shift
= page_shift
;
215 for (mtt
->order
= 0, i
= 1; i
< npages
; i
<<= 1)
218 mtt
->offset
= mlx4_alloc_mtt_range(dev
, mtt
->order
);
219 if (mtt
->offset
== -1)
224 EXPORT_SYMBOL_GPL(mlx4_mtt_init
);
226 void __mlx4_free_mtt_range(struct mlx4_dev
*dev
, u32 offset
, int order
)
230 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
232 seg_order
= max_t(int, order
- log_mtts_per_seg
, 0);
233 first_seg
= offset
/ (1 << log_mtts_per_seg
);
235 mlx4_buddy_free(&mr_table
->mtt_buddy
, first_seg
, seg_order
);
236 mlx4_table_put_range(dev
, &mr_table
->mtt_table
, offset
,
237 offset
+ (1 << order
) - 1);
240 static void mlx4_free_mtt_range(struct mlx4_dev
*dev
, u32 offset
, int order
)
245 if (mlx4_is_mfunc(dev
)) {
246 set_param_l(&in_param
, offset
);
247 set_param_h(&in_param
, order
);
248 err
= mlx4_cmd(dev
, in_param
, RES_MTT
, RES_OP_RESERVE_AND_MAP
,
250 MLX4_CMD_TIME_CLASS_A
,
253 mlx4_warn(dev
, "Failed to free mtt range at:%d order:%d\n",
257 __mlx4_free_mtt_range(dev
, offset
, order
);
260 void mlx4_mtt_cleanup(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
)
265 mlx4_free_mtt_range(dev
, mtt
->offset
, mtt
->order
);
267 EXPORT_SYMBOL_GPL(mlx4_mtt_cleanup
);
269 u64
mlx4_mtt_addr(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
)
271 return (u64
) mtt
->offset
* dev
->caps
.mtt_entry_sz
;
273 EXPORT_SYMBOL_GPL(mlx4_mtt_addr
);
275 static u32
hw_index_to_key(u32 ind
)
277 return (ind
>> 24) | (ind
<< 8);
280 static u32
key_to_hw_index(u32 key
)
282 return (key
<< 24) | (key
>> 8);
285 static int mlx4_SW2HW_MPT(struct mlx4_dev
*dev
, struct mlx4_cmd_mailbox
*mailbox
,
288 return mlx4_cmd(dev
, mailbox
->dma
, mpt_index
,
289 0, MLX4_CMD_SW2HW_MPT
, MLX4_CMD_TIME_CLASS_B
,
293 static int mlx4_HW2SW_MPT(struct mlx4_dev
*dev
, struct mlx4_cmd_mailbox
*mailbox
,
296 return mlx4_cmd_box(dev
, 0, mailbox
? mailbox
->dma
: 0, mpt_index
,
297 !mailbox
, MLX4_CMD_HW2SW_MPT
,
298 MLX4_CMD_TIME_CLASS_B
, MLX4_CMD_WRAPPED
);
301 int mlx4_mr_hw_get_mpt(struct mlx4_dev
*dev
, struct mlx4_mr
*mmr
,
302 struct mlx4_mpt_entry
***mpt_entry
)
305 int key
= key_to_hw_index(mmr
->key
) & (dev
->caps
.num_mpts
- 1);
306 struct mlx4_cmd_mailbox
*mailbox
= NULL
;
308 /* Make sure that at this point we have single-threaded access only */
310 if (mmr
->enabled
!= MLX4_MPT_EN_HW
)
313 err
= mlx4_HW2SW_MPT(dev
, NULL
, key
);
316 mlx4_warn(dev
, "HW2SW_MPT failed (%d).", err
);
317 mlx4_warn(dev
, "Most likely the MR has MWs bound to it.\n");
321 mmr
->enabled
= MLX4_MPT_EN_SW
;
323 if (!mlx4_is_mfunc(dev
)) {
324 **mpt_entry
= mlx4_table_find(
325 &mlx4_priv(dev
)->mr_table
.dmpt_table
,
328 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
329 if (IS_ERR_OR_NULL(mailbox
))
330 return PTR_ERR(mailbox
);
332 err
= mlx4_cmd_box(dev
, 0, mailbox
->dma
, key
,
333 0, MLX4_CMD_QUERY_MPT
,
334 MLX4_CMD_TIME_CLASS_B
,
340 *mpt_entry
= (struct mlx4_mpt_entry
**)&mailbox
->buf
;
343 if (!(*mpt_entry
) || !(**mpt_entry
)) {
351 mlx4_free_cmd_mailbox(dev
, mailbox
);
354 EXPORT_SYMBOL_GPL(mlx4_mr_hw_get_mpt
);
356 int mlx4_mr_hw_write_mpt(struct mlx4_dev
*dev
, struct mlx4_mr
*mmr
,
357 struct mlx4_mpt_entry
**mpt_entry
)
361 if (!mlx4_is_mfunc(dev
)) {
362 /* Make sure any changes to this entry are flushed */
365 *(u8
*)(*mpt_entry
) = MLX4_MPT_STATUS_HW
;
367 /* Make sure the new status is written */
370 err
= mlx4_SYNC_TPT(dev
);
372 int key
= key_to_hw_index(mmr
->key
) & (dev
->caps
.num_mpts
- 1);
374 struct mlx4_cmd_mailbox
*mailbox
=
375 container_of((void *)mpt_entry
, struct mlx4_cmd_mailbox
,
378 err
= mlx4_SW2HW_MPT(dev
, mailbox
, key
);
381 mmr
->pd
= be32_to_cpu((*mpt_entry
)->pd_flags
) & MLX4_MPT_PD_MASK
;
383 mmr
->enabled
= MLX4_MPT_EN_HW
;
386 EXPORT_SYMBOL_GPL(mlx4_mr_hw_write_mpt
);
388 void mlx4_mr_hw_put_mpt(struct mlx4_dev
*dev
,
389 struct mlx4_mpt_entry
**mpt_entry
)
391 if (mlx4_is_mfunc(dev
)) {
392 struct mlx4_cmd_mailbox
*mailbox
=
393 container_of((void *)mpt_entry
, struct mlx4_cmd_mailbox
,
395 mlx4_free_cmd_mailbox(dev
, mailbox
);
398 EXPORT_SYMBOL_GPL(mlx4_mr_hw_put_mpt
);
400 int mlx4_mr_hw_change_pd(struct mlx4_dev
*dev
, struct mlx4_mpt_entry
*mpt_entry
,
403 u32 pd_flags
= be32_to_cpu(mpt_entry
->pd_flags
);
404 /* The wrapper function will put the slave's id here */
405 if (mlx4_is_mfunc(dev
))
406 pd_flags
&= ~MLX4_MPT_PD_VF_MASK
;
407 mpt_entry
->pd_flags
= cpu_to_be32((pd_flags
& ~MLX4_MPT_PD_MASK
) |
408 (pdn
& MLX4_MPT_PD_MASK
)
409 | MLX4_MPT_PD_FLAG_EN_INV
);
412 EXPORT_SYMBOL_GPL(mlx4_mr_hw_change_pd
);
414 int mlx4_mr_hw_change_access(struct mlx4_dev
*dev
,
415 struct mlx4_mpt_entry
*mpt_entry
,
418 u32 flags
= (be32_to_cpu(mpt_entry
->flags
) & ~MLX4_PERM_MASK
) |
419 (access
& MLX4_PERM_MASK
);
421 mpt_entry
->flags
= cpu_to_be32(flags
);
424 EXPORT_SYMBOL_GPL(mlx4_mr_hw_change_access
);
426 static int mlx4_mr_alloc_reserved(struct mlx4_dev
*dev
, u32 mridx
, u32 pd
,
427 u64 iova
, u64 size
, u32 access
, int npages
,
428 int page_shift
, struct mlx4_mr
*mr
)
434 mr
->enabled
= MLX4_MPT_DISABLED
;
435 mr
->key
= hw_index_to_key(mridx
);
437 return mlx4_mtt_init(dev
, npages
, page_shift
, &mr
->mtt
);
440 static int mlx4_WRITE_MTT(struct mlx4_dev
*dev
,
441 struct mlx4_cmd_mailbox
*mailbox
,
444 return mlx4_cmd(dev
, mailbox
->dma
, num_entries
, 0, MLX4_CMD_WRITE_MTT
,
445 MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_WRAPPED
);
448 int __mlx4_mpt_reserve(struct mlx4_dev
*dev
)
450 struct mlx4_priv
*priv
= mlx4_priv(dev
);
452 return mlx4_bitmap_alloc(&priv
->mr_table
.mpt_bitmap
);
455 static int mlx4_mpt_reserve(struct mlx4_dev
*dev
)
459 if (mlx4_is_mfunc(dev
)) {
460 if (mlx4_cmd_imm(dev
, 0, &out_param
, RES_MPT
, RES_OP_RESERVE
,
462 MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_WRAPPED
))
464 return get_param_l(&out_param
);
466 return __mlx4_mpt_reserve(dev
);
469 void __mlx4_mpt_release(struct mlx4_dev
*dev
, u32 index
)
471 struct mlx4_priv
*priv
= mlx4_priv(dev
);
473 mlx4_bitmap_free(&priv
->mr_table
.mpt_bitmap
, index
, MLX4_NO_RR
);
476 static void mlx4_mpt_release(struct mlx4_dev
*dev
, u32 index
)
480 if (mlx4_is_mfunc(dev
)) {
481 set_param_l(&in_param
, index
);
482 if (mlx4_cmd(dev
, in_param
, RES_MPT
, RES_OP_RESERVE
,
484 MLX4_CMD_TIME_CLASS_A
, MLX4_CMD_WRAPPED
))
485 mlx4_warn(dev
, "Failed to release mr index:%d\n",
489 __mlx4_mpt_release(dev
, index
);
492 int __mlx4_mpt_alloc_icm(struct mlx4_dev
*dev
, u32 index
, gfp_t gfp
)
494 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
496 return mlx4_table_get(dev
, &mr_table
->dmpt_table
, index
, gfp
);
499 static int mlx4_mpt_alloc_icm(struct mlx4_dev
*dev
, u32 index
, gfp_t gfp
)
503 if (mlx4_is_mfunc(dev
)) {
504 set_param_l(¶m
, index
);
505 return mlx4_cmd_imm(dev
, param
, ¶m
, RES_MPT
, RES_OP_MAP_ICM
,
507 MLX4_CMD_TIME_CLASS_A
,
510 return __mlx4_mpt_alloc_icm(dev
, index
, gfp
);
513 void __mlx4_mpt_free_icm(struct mlx4_dev
*dev
, u32 index
)
515 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
517 mlx4_table_put(dev
, &mr_table
->dmpt_table
, index
);
520 static void mlx4_mpt_free_icm(struct mlx4_dev
*dev
, u32 index
)
524 if (mlx4_is_mfunc(dev
)) {
525 set_param_l(&in_param
, index
);
526 if (mlx4_cmd(dev
, in_param
, RES_MPT
, RES_OP_MAP_ICM
,
527 MLX4_CMD_FREE_RES
, MLX4_CMD_TIME_CLASS_A
,
529 mlx4_warn(dev
, "Failed to free icm of mr index:%d\n",
533 return __mlx4_mpt_free_icm(dev
, index
);
536 int mlx4_mr_alloc(struct mlx4_dev
*dev
, u32 pd
, u64 iova
, u64 size
, u32 access
,
537 int npages
, int page_shift
, struct mlx4_mr
*mr
)
542 index
= mlx4_mpt_reserve(dev
);
546 err
= mlx4_mr_alloc_reserved(dev
, index
, pd
, iova
, size
,
547 access
, npages
, page_shift
, mr
);
549 mlx4_mpt_release(dev
, index
);
553 EXPORT_SYMBOL_GPL(mlx4_mr_alloc
);
555 static int mlx4_mr_free_reserved(struct mlx4_dev
*dev
, struct mlx4_mr
*mr
)
559 if (mr
->enabled
== MLX4_MPT_EN_HW
) {
560 err
= mlx4_HW2SW_MPT(dev
, NULL
,
561 key_to_hw_index(mr
->key
) &
562 (dev
->caps
.num_mpts
- 1));
564 mlx4_warn(dev
, "HW2SW_MPT failed (%d), MR has MWs bound to it\n",
569 mr
->enabled
= MLX4_MPT_EN_SW
;
571 mlx4_mtt_cleanup(dev
, &mr
->mtt
);
576 int mlx4_mr_free(struct mlx4_dev
*dev
, struct mlx4_mr
*mr
)
580 ret
= mlx4_mr_free_reserved(dev
, mr
);
584 mlx4_mpt_free_icm(dev
, key_to_hw_index(mr
->key
));
585 mlx4_mpt_release(dev
, key_to_hw_index(mr
->key
));
589 EXPORT_SYMBOL_GPL(mlx4_mr_free
);
591 void mlx4_mr_rereg_mem_cleanup(struct mlx4_dev
*dev
, struct mlx4_mr
*mr
)
593 mlx4_mtt_cleanup(dev
, &mr
->mtt
);
595 EXPORT_SYMBOL_GPL(mlx4_mr_rereg_mem_cleanup
);
597 int mlx4_mr_rereg_mem_write(struct mlx4_dev
*dev
, struct mlx4_mr
*mr
,
598 u64 iova
, u64 size
, int npages
,
599 int page_shift
, struct mlx4_mpt_entry
*mpt_entry
)
603 mpt_entry
->start
= cpu_to_be64(mr
->iova
);
604 mpt_entry
->length
= cpu_to_be64(mr
->size
);
605 mpt_entry
->entity_size
= cpu_to_be32(mr
->mtt
.page_shift
);
607 err
= mlx4_mtt_init(dev
, npages
, page_shift
, &mr
->mtt
);
611 if (mr
->mtt
.order
< 0) {
612 mpt_entry
->flags
|= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL
);
613 mpt_entry
->mtt_addr
= 0;
615 mpt_entry
->mtt_addr
= cpu_to_be64(mlx4_mtt_addr(dev
,
617 if (mr
->mtt
.page_shift
== 0)
618 mpt_entry
->mtt_sz
= cpu_to_be32(1 << mr
->mtt
.order
);
620 mr
->enabled
= MLX4_MPT_EN_SW
;
624 EXPORT_SYMBOL_GPL(mlx4_mr_rereg_mem_write
);
626 int mlx4_mr_enable(struct mlx4_dev
*dev
, struct mlx4_mr
*mr
)
628 struct mlx4_cmd_mailbox
*mailbox
;
629 struct mlx4_mpt_entry
*mpt_entry
;
632 err
= mlx4_mpt_alloc_icm(dev
, key_to_hw_index(mr
->key
), GFP_KERNEL
);
636 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
637 if (IS_ERR(mailbox
)) {
638 err
= PTR_ERR(mailbox
);
641 mpt_entry
= mailbox
->buf
;
642 mpt_entry
->flags
= cpu_to_be32(MLX4_MPT_FLAG_MIO
|
643 MLX4_MPT_FLAG_REGION
|
646 mpt_entry
->key
= cpu_to_be32(key_to_hw_index(mr
->key
));
647 mpt_entry
->pd_flags
= cpu_to_be32(mr
->pd
| MLX4_MPT_PD_FLAG_EN_INV
);
648 mpt_entry
->start
= cpu_to_be64(mr
->iova
);
649 mpt_entry
->length
= cpu_to_be64(mr
->size
);
650 mpt_entry
->entity_size
= cpu_to_be32(mr
->mtt
.page_shift
);
652 if (mr
->mtt
.order
< 0) {
653 mpt_entry
->flags
|= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL
);
654 mpt_entry
->mtt_addr
= 0;
656 mpt_entry
->mtt_addr
= cpu_to_be64(mlx4_mtt_addr(dev
,
660 if (mr
->mtt
.order
>= 0 && mr
->mtt
.page_shift
== 0) {
661 /* fast register MR in free state */
662 mpt_entry
->flags
|= cpu_to_be32(MLX4_MPT_FLAG_FREE
);
663 mpt_entry
->pd_flags
|= cpu_to_be32(MLX4_MPT_PD_FLAG_FAST_REG
|
664 MLX4_MPT_PD_FLAG_RAE
);
665 mpt_entry
->mtt_sz
= cpu_to_be32(1 << mr
->mtt
.order
);
667 mpt_entry
->flags
|= cpu_to_be32(MLX4_MPT_FLAG_SW_OWNS
);
670 err
= mlx4_SW2HW_MPT(dev
, mailbox
,
671 key_to_hw_index(mr
->key
) & (dev
->caps
.num_mpts
- 1));
673 mlx4_warn(dev
, "SW2HW_MPT failed (%d)\n", err
);
676 mr
->enabled
= MLX4_MPT_EN_HW
;
678 mlx4_free_cmd_mailbox(dev
, mailbox
);
683 mlx4_free_cmd_mailbox(dev
, mailbox
);
686 mlx4_mpt_free_icm(dev
, key_to_hw_index(mr
->key
));
689 EXPORT_SYMBOL_GPL(mlx4_mr_enable
);
691 static int mlx4_write_mtt_chunk(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
,
692 int start_index
, int npages
, u64
*page_list
)
694 struct mlx4_priv
*priv
= mlx4_priv(dev
);
696 dma_addr_t dma_handle
;
699 mtts
= mlx4_table_find(&priv
->mr_table
.mtt_table
, mtt
->offset
+
700 start_index
, &dma_handle
);
705 dma_sync_single_for_cpu(&dev
->pdev
->dev
, dma_handle
,
706 npages
* sizeof (u64
), DMA_TO_DEVICE
);
708 for (i
= 0; i
< npages
; ++i
)
709 mtts
[i
] = cpu_to_be64(page_list
[i
] | MLX4_MTT_FLAG_PRESENT
);
711 dma_sync_single_for_device(&dev
->pdev
->dev
, dma_handle
,
712 npages
* sizeof (u64
), DMA_TO_DEVICE
);
717 int __mlx4_write_mtt(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
,
718 int start_index
, int npages
, u64
*page_list
)
723 int max_mtts_first_page
;
725 /* compute how may mtts fit in the first page */
726 mtts_per_page
= PAGE_SIZE
/ sizeof(u64
);
727 max_mtts_first_page
= mtts_per_page
- (mtt
->offset
+ start_index
)
730 chunk
= min_t(int, max_mtts_first_page
, npages
);
733 err
= mlx4_write_mtt_chunk(dev
, mtt
, start_index
, chunk
, page_list
);
737 start_index
+= chunk
;
740 chunk
= min_t(int, mtts_per_page
, npages
);
745 int mlx4_write_mtt(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
,
746 int start_index
, int npages
, u64
*page_list
)
748 struct mlx4_cmd_mailbox
*mailbox
= NULL
;
749 __be64
*inbox
= NULL
;
757 if (mlx4_is_mfunc(dev
)) {
758 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
760 return PTR_ERR(mailbox
);
761 inbox
= mailbox
->buf
;
764 chunk
= min_t(int, MLX4_MAILBOX_SIZE
/ sizeof(u64
) - 2,
766 inbox
[0] = cpu_to_be64(mtt
->offset
+ start_index
);
768 for (i
= 0; i
< chunk
; ++i
)
769 inbox
[i
+ 2] = cpu_to_be64(page_list
[i
] |
770 MLX4_MTT_FLAG_PRESENT
);
771 err
= mlx4_WRITE_MTT(dev
, mailbox
, chunk
);
773 mlx4_free_cmd_mailbox(dev
, mailbox
);
778 start_index
+= chunk
;
781 mlx4_free_cmd_mailbox(dev
, mailbox
);
785 return __mlx4_write_mtt(dev
, mtt
, start_index
, npages
, page_list
);
787 EXPORT_SYMBOL_GPL(mlx4_write_mtt
);
789 int mlx4_buf_write_mtt(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
,
790 struct mlx4_buf
*buf
, gfp_t gfp
)
796 page_list
= kmalloc(buf
->npages
* sizeof *page_list
,
801 for (i
= 0; i
< buf
->npages
; ++i
)
803 page_list
[i
] = buf
->direct
.map
+ (i
<< buf
->page_shift
);
805 page_list
[i
] = buf
->page_list
[i
].map
;
807 err
= mlx4_write_mtt(dev
, mtt
, 0, buf
->npages
, page_list
);
812 EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt
);
814 int mlx4_mw_alloc(struct mlx4_dev
*dev
, u32 pd
, enum mlx4_mw_type type
,
819 if ((type
== MLX4_MW_TYPE_1
&&
820 !(dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_MEM_WINDOW
)) ||
821 (type
== MLX4_MW_TYPE_2
&&
822 !(dev
->caps
.bmme_flags
& MLX4_BMME_FLAG_TYPE_2_WIN
)))
825 index
= mlx4_mpt_reserve(dev
);
829 mw
->key
= hw_index_to_key(index
);
832 mw
->enabled
= MLX4_MPT_DISABLED
;
836 EXPORT_SYMBOL_GPL(mlx4_mw_alloc
);
838 int mlx4_mw_enable(struct mlx4_dev
*dev
, struct mlx4_mw
*mw
)
840 struct mlx4_cmd_mailbox
*mailbox
;
841 struct mlx4_mpt_entry
*mpt_entry
;
844 err
= mlx4_mpt_alloc_icm(dev
, key_to_hw_index(mw
->key
), GFP_KERNEL
);
848 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
849 if (IS_ERR(mailbox
)) {
850 err
= PTR_ERR(mailbox
);
853 mpt_entry
= mailbox
->buf
;
855 /* Note that the MLX4_MPT_FLAG_REGION bit in mpt_entry->flags is turned
856 * off, thus creating a memory window and not a memory region.
858 mpt_entry
->key
= cpu_to_be32(key_to_hw_index(mw
->key
));
859 mpt_entry
->pd_flags
= cpu_to_be32(mw
->pd
);
860 if (mw
->type
== MLX4_MW_TYPE_2
) {
861 mpt_entry
->flags
|= cpu_to_be32(MLX4_MPT_FLAG_FREE
);
862 mpt_entry
->qpn
= cpu_to_be32(MLX4_MPT_QP_FLAG_BOUND_QP
);
863 mpt_entry
->pd_flags
|= cpu_to_be32(MLX4_MPT_PD_FLAG_EN_INV
);
866 err
= mlx4_SW2HW_MPT(dev
, mailbox
,
867 key_to_hw_index(mw
->key
) &
868 (dev
->caps
.num_mpts
- 1));
870 mlx4_warn(dev
, "SW2HW_MPT failed (%d)\n", err
);
873 mw
->enabled
= MLX4_MPT_EN_HW
;
875 mlx4_free_cmd_mailbox(dev
, mailbox
);
880 mlx4_free_cmd_mailbox(dev
, mailbox
);
883 mlx4_mpt_free_icm(dev
, key_to_hw_index(mw
->key
));
886 EXPORT_SYMBOL_GPL(mlx4_mw_enable
);
888 void mlx4_mw_free(struct mlx4_dev
*dev
, struct mlx4_mw
*mw
)
892 if (mw
->enabled
== MLX4_MPT_EN_HW
) {
893 err
= mlx4_HW2SW_MPT(dev
, NULL
,
894 key_to_hw_index(mw
->key
) &
895 (dev
->caps
.num_mpts
- 1));
897 mlx4_warn(dev
, "xxx HW2SW_MPT failed (%d)\n", err
);
899 mw
->enabled
= MLX4_MPT_EN_SW
;
902 mlx4_mpt_free_icm(dev
, key_to_hw_index(mw
->key
));
903 mlx4_mpt_release(dev
, key_to_hw_index(mw
->key
));
905 EXPORT_SYMBOL_GPL(mlx4_mw_free
);
907 int mlx4_init_mr_table(struct mlx4_dev
*dev
)
909 struct mlx4_priv
*priv
= mlx4_priv(dev
);
910 struct mlx4_mr_table
*mr_table
= &priv
->mr_table
;
913 /* Nothing to do for slaves - all MR handling is forwarded
915 if (mlx4_is_slave(dev
))
918 if (!is_power_of_2(dev
->caps
.num_mpts
))
921 err
= mlx4_bitmap_init(&mr_table
->mpt_bitmap
, dev
->caps
.num_mpts
,
922 ~0, dev
->caps
.reserved_mrws
, 0);
926 err
= mlx4_buddy_init(&mr_table
->mtt_buddy
,
927 ilog2((u32
)dev
->caps
.num_mtts
/
928 (1 << log_mtts_per_seg
)));
932 if (dev
->caps
.reserved_mtts
) {
933 priv
->reserved_mtts
=
934 mlx4_alloc_mtt_range(dev
,
935 fls(dev
->caps
.reserved_mtts
- 1));
936 if (priv
->reserved_mtts
< 0) {
937 mlx4_warn(dev
, "MTT table of order %u is too small\n",
938 mr_table
->mtt_buddy
.max_order
);
940 goto err_reserve_mtts
;
947 mlx4_buddy_cleanup(&mr_table
->mtt_buddy
);
950 mlx4_bitmap_cleanup(&mr_table
->mpt_bitmap
);
955 void mlx4_cleanup_mr_table(struct mlx4_dev
*dev
)
957 struct mlx4_priv
*priv
= mlx4_priv(dev
);
958 struct mlx4_mr_table
*mr_table
= &priv
->mr_table
;
960 if (mlx4_is_slave(dev
))
962 if (priv
->reserved_mtts
>= 0)
963 mlx4_free_mtt_range(dev
, priv
->reserved_mtts
,
964 fls(dev
->caps
.reserved_mtts
- 1));
965 mlx4_buddy_cleanup(&mr_table
->mtt_buddy
);
966 mlx4_bitmap_cleanup(&mr_table
->mpt_bitmap
);
969 static inline int mlx4_check_fmr(struct mlx4_fmr
*fmr
, u64
*page_list
,
970 int npages
, u64 iova
)
974 if (npages
> fmr
->max_pages
)
977 page_mask
= (1 << fmr
->page_shift
) - 1;
979 /* We are getting page lists, so va must be page aligned. */
980 if (iova
& page_mask
)
983 /* Trust the user not to pass misaligned data in page_list */
985 for (i
= 0; i
< npages
; ++i
) {
986 if (page_list
[i
] & ~page_mask
)
990 if (fmr
->maps
>= fmr
->max_maps
)
996 int mlx4_map_phys_fmr(struct mlx4_dev
*dev
, struct mlx4_fmr
*fmr
, u64
*page_list
,
997 int npages
, u64 iova
, u32
*lkey
, u32
*rkey
)
1002 err
= mlx4_check_fmr(fmr
, page_list
, npages
, iova
);
1008 key
= key_to_hw_index(fmr
->mr
.key
);
1009 key
+= dev
->caps
.num_mpts
;
1010 *lkey
= *rkey
= fmr
->mr
.key
= hw_index_to_key(key
);
1012 *(u8
*) fmr
->mpt
= MLX4_MPT_STATUS_SW
;
1014 /* Make sure MPT status is visible before writing MTT entries */
1017 dma_sync_single_for_cpu(&dev
->pdev
->dev
, fmr
->dma_handle
,
1018 npages
* sizeof(u64
), DMA_TO_DEVICE
);
1020 for (i
= 0; i
< npages
; ++i
)
1021 fmr
->mtts
[i
] = cpu_to_be64(page_list
[i
] | MLX4_MTT_FLAG_PRESENT
);
1023 dma_sync_single_for_device(&dev
->pdev
->dev
, fmr
->dma_handle
,
1024 npages
* sizeof(u64
), DMA_TO_DEVICE
);
1026 fmr
->mpt
->key
= cpu_to_be32(key
);
1027 fmr
->mpt
->lkey
= cpu_to_be32(key
);
1028 fmr
->mpt
->length
= cpu_to_be64(npages
* (1ull << fmr
->page_shift
));
1029 fmr
->mpt
->start
= cpu_to_be64(iova
);
1031 /* Make MTT entries are visible before setting MPT status */
1034 *(u8
*) fmr
->mpt
= MLX4_MPT_STATUS_HW
;
1036 /* Make sure MPT status is visible before consumer can use FMR */
1041 EXPORT_SYMBOL_GPL(mlx4_map_phys_fmr
);
1043 int mlx4_fmr_alloc(struct mlx4_dev
*dev
, u32 pd
, u32 access
, int max_pages
,
1044 int max_maps
, u8 page_shift
, struct mlx4_fmr
*fmr
)
1046 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1049 if (max_maps
> dev
->caps
.max_fmr_maps
)
1052 if (page_shift
< (ffs(dev
->caps
.page_size_cap
) - 1) || page_shift
>= 32)
1055 /* All MTTs must fit in the same page */
1056 if (max_pages
* sizeof *fmr
->mtts
> PAGE_SIZE
)
1059 fmr
->page_shift
= page_shift
;
1060 fmr
->max_pages
= max_pages
;
1061 fmr
->max_maps
= max_maps
;
1064 err
= mlx4_mr_alloc(dev
, pd
, 0, 0, access
, max_pages
,
1065 page_shift
, &fmr
->mr
);
1069 fmr
->mtts
= mlx4_table_find(&priv
->mr_table
.mtt_table
,
1081 (void) mlx4_mr_free(dev
, &fmr
->mr
);
1084 EXPORT_SYMBOL_GPL(mlx4_fmr_alloc
);
1086 int mlx4_fmr_enable(struct mlx4_dev
*dev
, struct mlx4_fmr
*fmr
)
1088 struct mlx4_priv
*priv
= mlx4_priv(dev
);
1091 err
= mlx4_mr_enable(dev
, &fmr
->mr
);
1095 fmr
->mpt
= mlx4_table_find(&priv
->mr_table
.dmpt_table
,
1096 key_to_hw_index(fmr
->mr
.key
), NULL
);
1102 EXPORT_SYMBOL_GPL(mlx4_fmr_enable
);
1104 void mlx4_fmr_unmap(struct mlx4_dev
*dev
, struct mlx4_fmr
*fmr
,
1105 u32
*lkey
, u32
*rkey
)
1107 struct mlx4_cmd_mailbox
*mailbox
;
1115 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
1116 if (IS_ERR(mailbox
)) {
1117 err
= PTR_ERR(mailbox
);
1118 pr_warn("mlx4_ib: mlx4_alloc_cmd_mailbox failed (%d)\n", err
);
1122 err
= mlx4_HW2SW_MPT(dev
, NULL
,
1123 key_to_hw_index(fmr
->mr
.key
) &
1124 (dev
->caps
.num_mpts
- 1));
1125 mlx4_free_cmd_mailbox(dev
, mailbox
);
1127 pr_warn("mlx4_ib: mlx4_HW2SW_MPT failed (%d)\n", err
);
1130 fmr
->mr
.enabled
= MLX4_MPT_EN_SW
;
1132 EXPORT_SYMBOL_GPL(mlx4_fmr_unmap
);
1134 int mlx4_fmr_free(struct mlx4_dev
*dev
, struct mlx4_fmr
*fmr
)
1141 ret
= mlx4_mr_free(dev
, &fmr
->mr
);
1144 fmr
->mr
.enabled
= MLX4_MPT_DISABLED
;
1148 EXPORT_SYMBOL_GPL(mlx4_fmr_free
);
1150 int mlx4_SYNC_TPT(struct mlx4_dev
*dev
)
1152 return mlx4_cmd(dev
, 0, 0, 0, MLX4_CMD_SYNC_TPT
, 1000,
1155 EXPORT_SYMBOL_GPL(mlx4_SYNC_TPT
);