2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 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/init.h>
36 #include <linux/errno.h>
38 #include <linux/mlx4/cmd.h>
44 * Must be packed because mtt_seg is 64 bits but only aligned to 32 bits.
46 struct mlx4_mpt_entry
{
60 __be32 first_byte_offset
;
61 } __attribute__((packed
));
63 #define MLX4_MPT_FLAG_SW_OWNS (0xfUL << 28)
64 #define MLX4_MPT_FLAG_MIO (1 << 17)
65 #define MLX4_MPT_FLAG_BIND_ENABLE (1 << 15)
66 #define MLX4_MPT_FLAG_PHYSICAL (1 << 9)
67 #define MLX4_MPT_FLAG_REGION (1 << 8)
69 #define MLX4_MTT_FLAG_PRESENT 1
71 #define MLX4_MPT_STATUS_SW 0xF0
72 #define MLX4_MPT_STATUS_HW 0x00
74 static u32
mlx4_buddy_alloc(struct mlx4_buddy
*buddy
, int order
)
80 spin_lock(&buddy
->lock
);
82 for (o
= order
; o
<= buddy
->max_order
; ++o
) {
83 m
= 1 << (buddy
->max_order
- o
);
84 seg
= find_first_bit(buddy
->bits
[o
], m
);
89 spin_unlock(&buddy
->lock
);
93 clear_bit(seg
, buddy
->bits
[o
]);
98 set_bit(seg
^ 1, buddy
->bits
[o
]);
101 spin_unlock(&buddy
->lock
);
108 static void mlx4_buddy_free(struct mlx4_buddy
*buddy
, u32 seg
, int order
)
112 spin_lock(&buddy
->lock
);
114 while (test_bit(seg
^ 1, buddy
->bits
[order
])) {
115 clear_bit(seg
^ 1, buddy
->bits
[order
]);
120 set_bit(seg
, buddy
->bits
[order
]);
122 spin_unlock(&buddy
->lock
);
125 static int mlx4_buddy_init(struct mlx4_buddy
*buddy
, int max_order
)
129 buddy
->max_order
= max_order
;
130 spin_lock_init(&buddy
->lock
);
132 buddy
->bits
= kzalloc((buddy
->max_order
+ 1) * sizeof (long *),
137 for (i
= 0; i
<= buddy
->max_order
; ++i
) {
138 s
= BITS_TO_LONGS(1 << (buddy
->max_order
- i
));
139 buddy
->bits
[i
] = kmalloc(s
* sizeof (long), GFP_KERNEL
);
142 bitmap_zero(buddy
->bits
[i
], 1 << (buddy
->max_order
- i
));
145 set_bit(0, buddy
->bits
[buddy
->max_order
]);
150 for (i
= 0; i
<= buddy
->max_order
; ++i
)
151 kfree(buddy
->bits
[i
]);
159 static void mlx4_buddy_cleanup(struct mlx4_buddy
*buddy
)
163 for (i
= 0; i
<= buddy
->max_order
; ++i
)
164 kfree(buddy
->bits
[i
]);
169 static u32
mlx4_alloc_mtt_range(struct mlx4_dev
*dev
, int order
)
171 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
174 seg
= mlx4_buddy_alloc(&mr_table
->mtt_buddy
, order
);
178 if (mlx4_table_get_range(dev
, &mr_table
->mtt_table
, seg
,
179 seg
+ (1 << order
) - 1)) {
180 mlx4_buddy_free(&mr_table
->mtt_buddy
, seg
, order
);
187 int mlx4_mtt_init(struct mlx4_dev
*dev
, int npages
, int page_shift
,
188 struct mlx4_mtt
*mtt
)
194 mtt
->page_shift
= MLX4_ICM_PAGE_SHIFT
;
197 mtt
->page_shift
= page_shift
;
199 for (mtt
->order
= 0, i
= MLX4_MTT_ENTRY_PER_SEG
; i
< npages
; i
<<= 1)
202 mtt
->first_seg
= mlx4_alloc_mtt_range(dev
, mtt
->order
);
203 if (mtt
->first_seg
== -1)
208 EXPORT_SYMBOL_GPL(mlx4_mtt_init
);
210 void mlx4_mtt_cleanup(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
)
212 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
217 mlx4_buddy_free(&mr_table
->mtt_buddy
, mtt
->first_seg
, mtt
->order
);
218 mlx4_table_put_range(dev
, &mr_table
->mtt_table
, mtt
->first_seg
,
219 mtt
->first_seg
+ (1 << mtt
->order
) - 1);
221 EXPORT_SYMBOL_GPL(mlx4_mtt_cleanup
);
223 u64
mlx4_mtt_addr(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
)
225 return (u64
) mtt
->first_seg
* dev
->caps
.mtt_entry_sz
;
227 EXPORT_SYMBOL_GPL(mlx4_mtt_addr
);
229 static u32
hw_index_to_key(u32 ind
)
231 return (ind
>> 24) | (ind
<< 8);
234 static u32
key_to_hw_index(u32 key
)
236 return (key
<< 24) | (key
>> 8);
239 static int mlx4_SW2HW_MPT(struct mlx4_dev
*dev
, struct mlx4_cmd_mailbox
*mailbox
,
242 return mlx4_cmd(dev
, mailbox
->dma
, mpt_index
, 0, MLX4_CMD_SW2HW_MPT
,
243 MLX4_CMD_TIME_CLASS_B
);
246 static int mlx4_HW2SW_MPT(struct mlx4_dev
*dev
, struct mlx4_cmd_mailbox
*mailbox
,
249 return mlx4_cmd_box(dev
, 0, mailbox
? mailbox
->dma
: 0, mpt_index
,
250 !mailbox
, MLX4_CMD_HW2SW_MPT
, MLX4_CMD_TIME_CLASS_B
);
253 int mlx4_mr_alloc(struct mlx4_dev
*dev
, u32 pd
, u64 iova
, u64 size
, u32 access
,
254 int npages
, int page_shift
, struct mlx4_mr
*mr
)
256 struct mlx4_priv
*priv
= mlx4_priv(dev
);
260 index
= mlx4_bitmap_alloc(&priv
->mr_table
.mpt_bitmap
);
269 mr
->key
= hw_index_to_key(index
);
271 err
= mlx4_mtt_init(dev
, npages
, page_shift
, &mr
->mtt
);
273 mlx4_bitmap_free(&priv
->mr_table
.mpt_bitmap
, index
);
277 EXPORT_SYMBOL_GPL(mlx4_mr_alloc
);
279 void mlx4_mr_free(struct mlx4_dev
*dev
, struct mlx4_mr
*mr
)
281 struct mlx4_priv
*priv
= mlx4_priv(dev
);
285 err
= mlx4_HW2SW_MPT(dev
, NULL
,
286 key_to_hw_index(mr
->key
) &
287 (dev
->caps
.num_mpts
- 1));
289 mlx4_warn(dev
, "HW2SW_MPT failed (%d)\n", err
);
292 mlx4_mtt_cleanup(dev
, &mr
->mtt
);
293 mlx4_bitmap_free(&priv
->mr_table
.mpt_bitmap
, key_to_hw_index(mr
->key
));
295 EXPORT_SYMBOL_GPL(mlx4_mr_free
);
297 int mlx4_mr_enable(struct mlx4_dev
*dev
, struct mlx4_mr
*mr
)
299 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
300 struct mlx4_cmd_mailbox
*mailbox
;
301 struct mlx4_mpt_entry
*mpt_entry
;
304 err
= mlx4_table_get(dev
, &mr_table
->dmpt_table
, key_to_hw_index(mr
->key
));
308 mailbox
= mlx4_alloc_cmd_mailbox(dev
);
309 if (IS_ERR(mailbox
)) {
310 err
= PTR_ERR(mailbox
);
313 mpt_entry
= mailbox
->buf
;
315 memset(mpt_entry
, 0, sizeof *mpt_entry
);
317 mpt_entry
->flags
= cpu_to_be32(MLX4_MPT_FLAG_SW_OWNS
|
319 MLX4_MPT_FLAG_REGION
|
322 mpt_entry
->key
= cpu_to_be32(key_to_hw_index(mr
->key
));
323 mpt_entry
->pd
= cpu_to_be32(mr
->pd
);
324 mpt_entry
->start
= cpu_to_be64(mr
->iova
);
325 mpt_entry
->length
= cpu_to_be64(mr
->size
);
326 mpt_entry
->entity_size
= cpu_to_be32(mr
->mtt
.page_shift
);
327 if (mr
->mtt
.order
< 0) {
328 mpt_entry
->flags
|= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL
);
329 mpt_entry
->mtt_seg
= 0;
331 mpt_entry
->mtt_seg
= cpu_to_be64(mlx4_mtt_addr(dev
, &mr
->mtt
));
333 err
= mlx4_SW2HW_MPT(dev
, mailbox
,
334 key_to_hw_index(mr
->key
) & (dev
->caps
.num_mpts
- 1));
336 mlx4_warn(dev
, "SW2HW_MPT failed (%d)\n", err
);
342 mlx4_free_cmd_mailbox(dev
, mailbox
);
347 mlx4_free_cmd_mailbox(dev
, mailbox
);
350 mlx4_table_put(dev
, &mr_table
->dmpt_table
, key_to_hw_index(mr
->key
));
353 EXPORT_SYMBOL_GPL(mlx4_mr_enable
);
355 static int mlx4_write_mtt_chunk(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
,
356 int start_index
, int npages
, u64
*page_list
)
358 struct mlx4_priv
*priv
= mlx4_priv(dev
);
360 dma_addr_t dma_handle
;
362 int s
= start_index
* sizeof (u64
);
364 /* All MTTs must fit in the same page */
365 if (start_index
/ (PAGE_SIZE
/ sizeof (u64
)) !=
366 (start_index
+ npages
- 1) / (PAGE_SIZE
/ sizeof (u64
)))
369 if (start_index
& (MLX4_MTT_ENTRY_PER_SEG
- 1))
372 mtts
= mlx4_table_find(&priv
->mr_table
.mtt_table
, mtt
->first_seg
+
373 s
/ dev
->caps
.mtt_entry_sz
, &dma_handle
);
377 for (i
= 0; i
< npages
; ++i
)
378 mtts
[i
] = cpu_to_be64(page_list
[i
] | MLX4_MTT_FLAG_PRESENT
);
380 dma_sync_single(&dev
->pdev
->dev
, dma_handle
, npages
* sizeof (u64
), DMA_TO_DEVICE
);
385 int mlx4_write_mtt(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
,
386 int start_index
, int npages
, u64
*page_list
)
395 chunk
= min_t(int, PAGE_SIZE
/ sizeof(u64
), npages
);
396 err
= mlx4_write_mtt_chunk(dev
, mtt
, start_index
, chunk
, page_list
);
401 start_index
+= chunk
;
407 EXPORT_SYMBOL_GPL(mlx4_write_mtt
);
409 int mlx4_buf_write_mtt(struct mlx4_dev
*dev
, struct mlx4_mtt
*mtt
,
410 struct mlx4_buf
*buf
)
416 page_list
= kmalloc(buf
->npages
* sizeof *page_list
, GFP_KERNEL
);
420 for (i
= 0; i
< buf
->npages
; ++i
)
422 page_list
[i
] = buf
->direct
.map
+ (i
<< buf
->page_shift
);
424 page_list
[i
] = buf
->page_list
[i
].map
;
426 err
= mlx4_write_mtt(dev
, mtt
, 0, buf
->npages
, page_list
);
431 EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt
);
433 int mlx4_init_mr_table(struct mlx4_dev
*dev
)
435 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
438 err
= mlx4_bitmap_init(&mr_table
->mpt_bitmap
, dev
->caps
.num_mpts
,
439 ~0, dev
->caps
.reserved_mrws
);
443 err
= mlx4_buddy_init(&mr_table
->mtt_buddy
,
444 ilog2(dev
->caps
.num_mtt_segs
));
448 if (dev
->caps
.reserved_mtts
) {
449 if (mlx4_alloc_mtt_range(dev
, fls(dev
->caps
.reserved_mtts
- 1)) == -1) {
450 mlx4_warn(dev
, "MTT table of order %d is too small.\n",
451 mr_table
->mtt_buddy
.max_order
);
453 goto err_reserve_mtts
;
460 mlx4_buddy_cleanup(&mr_table
->mtt_buddy
);
463 mlx4_bitmap_cleanup(&mr_table
->mpt_bitmap
);
468 void mlx4_cleanup_mr_table(struct mlx4_dev
*dev
)
470 struct mlx4_mr_table
*mr_table
= &mlx4_priv(dev
)->mr_table
;
472 mlx4_buddy_cleanup(&mr_table
->mtt_buddy
);
473 mlx4_bitmap_cleanup(&mr_table
->mpt_bitmap
);
476 static inline int mlx4_check_fmr(struct mlx4_fmr
*fmr
, u64
*page_list
,
477 int npages
, u64 iova
)
481 if (npages
> fmr
->max_pages
)
484 page_mask
= (1 << fmr
->page_shift
) - 1;
486 /* We are getting page lists, so va must be page aligned. */
487 if (iova
& page_mask
)
490 /* Trust the user not to pass misaligned data in page_list */
492 for (i
= 0; i
< npages
; ++i
) {
493 if (page_list
[i
] & ~page_mask
)
497 if (fmr
->maps
>= fmr
->max_maps
)
503 int mlx4_map_phys_fmr(struct mlx4_dev
*dev
, struct mlx4_fmr
*fmr
, u64
*page_list
,
504 int npages
, u64 iova
, u32
*lkey
, u32
*rkey
)
509 err
= mlx4_check_fmr(fmr
, page_list
, npages
, iova
);
515 key
= key_to_hw_index(fmr
->mr
.key
);
516 key
+= dev
->caps
.num_mpts
;
517 *lkey
= *rkey
= fmr
->mr
.key
= hw_index_to_key(key
);
519 *(u8
*) fmr
->mpt
= MLX4_MPT_STATUS_SW
;
521 /* Make sure MPT status is visible before writing MTT entries */
524 for (i
= 0; i
< npages
; ++i
)
525 fmr
->mtts
[i
] = cpu_to_be64(page_list
[i
] | MLX4_MTT_FLAG_PRESENT
);
527 dma_sync_single(&dev
->pdev
->dev
, fmr
->dma_handle
,
528 npages
* sizeof(u64
), DMA_TO_DEVICE
);
530 fmr
->mpt
->key
= cpu_to_be32(key
);
531 fmr
->mpt
->lkey
= cpu_to_be32(key
);
532 fmr
->mpt
->length
= cpu_to_be64(npages
* (1ull << fmr
->page_shift
));
533 fmr
->mpt
->start
= cpu_to_be64(iova
);
535 /* Make MTT entries are visible before setting MPT status */
538 *(u8
*) fmr
->mpt
= MLX4_MPT_STATUS_HW
;
540 /* Make sure MPT status is visible before consumer can use FMR */
545 EXPORT_SYMBOL_GPL(mlx4_map_phys_fmr
);
547 int mlx4_fmr_alloc(struct mlx4_dev
*dev
, u32 pd
, u32 access
, int max_pages
,
548 int max_maps
, u8 page_shift
, struct mlx4_fmr
*fmr
)
550 struct mlx4_priv
*priv
= mlx4_priv(dev
);
554 if (page_shift
< (ffs(dev
->caps
.page_size_cap
) - 1) || page_shift
>= 32)
557 /* All MTTs must fit in the same page */
558 if (max_pages
* sizeof *fmr
->mtts
> PAGE_SIZE
)
561 fmr
->page_shift
= page_shift
;
562 fmr
->max_pages
= max_pages
;
563 fmr
->max_maps
= max_maps
;
566 err
= mlx4_mr_alloc(dev
, pd
, 0, 0, access
, max_pages
,
567 page_shift
, &fmr
->mr
);
571 mtt_seg
= fmr
->mr
.mtt
.first_seg
* dev
->caps
.mtt_entry_sz
;
573 fmr
->mtts
= mlx4_table_find(&priv
->mr_table
.mtt_table
,
574 fmr
->mr
.mtt
.first_seg
,
584 mlx4_mr_free(dev
, &fmr
->mr
);
587 EXPORT_SYMBOL_GPL(mlx4_fmr_alloc
);
589 int mlx4_fmr_enable(struct mlx4_dev
*dev
, struct mlx4_fmr
*fmr
)
591 struct mlx4_priv
*priv
= mlx4_priv(dev
);
594 err
= mlx4_mr_enable(dev
, &fmr
->mr
);
598 fmr
->mpt
= mlx4_table_find(&priv
->mr_table
.dmpt_table
,
599 key_to_hw_index(fmr
->mr
.key
), NULL
);
605 EXPORT_SYMBOL_GPL(mlx4_fmr_enable
);
607 void mlx4_fmr_unmap(struct mlx4_dev
*dev
, struct mlx4_fmr
*fmr
,
608 u32
*lkey
, u32
*rkey
)
615 *(u8
*) fmr
->mpt
= MLX4_MPT_STATUS_SW
;
617 EXPORT_SYMBOL_GPL(mlx4_fmr_unmap
);
619 int mlx4_fmr_free(struct mlx4_dev
*dev
, struct mlx4_fmr
*fmr
)
625 mlx4_mr_free(dev
, &fmr
->mr
);
629 EXPORT_SYMBOL_GPL(mlx4_fmr_free
);
631 int mlx4_SYNC_TPT(struct mlx4_dev
*dev
)
633 return mlx4_cmd(dev
, 0, 0, 0, MLX4_CMD_SYNC_TPT
, 1000);
635 EXPORT_SYMBOL_GPL(mlx4_SYNC_TPT
);