]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/net/ethernet/mellanox/mlx4/icm.c
2 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
3 * Copyright (c) 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/errno.h>
36 #include <linux/scatterlist.h>
37 #include <linux/slab.h>
39 #include <linux/mlx4/cmd.h>
46 * We allocate in as big chunks as we can, up to a maximum of 256 KB
50 MLX4_ICM_ALLOC_SIZE
= 1 << 18,
51 MLX4_TABLE_CHUNK_SIZE
= 1 << 18
54 static void mlx4_free_icm_pages(struct mlx4_dev
*dev
, struct mlx4_icm_chunk
*chunk
)
59 pci_unmap_sg(dev
->persist
->pdev
, chunk
->mem
, chunk
->npages
,
60 PCI_DMA_BIDIRECTIONAL
);
62 for (i
= 0; i
< chunk
->npages
; ++i
)
63 __free_pages(sg_page(&chunk
->mem
[i
]),
64 get_order(chunk
->mem
[i
].length
));
67 static void mlx4_free_icm_coherent(struct mlx4_dev
*dev
, struct mlx4_icm_chunk
*chunk
)
71 for (i
= 0; i
< chunk
->npages
; ++i
)
72 dma_free_coherent(&dev
->persist
->pdev
->dev
,
74 lowmem_page_address(sg_page(&chunk
->mem
[i
])),
75 sg_dma_address(&chunk
->mem
[i
]));
78 void mlx4_free_icm(struct mlx4_dev
*dev
, struct mlx4_icm
*icm
, int coherent
)
80 struct mlx4_icm_chunk
*chunk
, *tmp
;
85 list_for_each_entry_safe(chunk
, tmp
, &icm
->chunk_list
, list
) {
87 mlx4_free_icm_coherent(dev
, chunk
);
89 mlx4_free_icm_pages(dev
, chunk
);
97 static int mlx4_alloc_icm_pages(struct scatterlist
*mem
, int order
,
98 gfp_t gfp_mask
, int node
)
102 page
= alloc_pages_node(node
, gfp_mask
, order
);
104 page
= alloc_pages(gfp_mask
, order
);
109 sg_set_page(mem
, page
, PAGE_SIZE
<< order
, 0);
113 static int mlx4_alloc_icm_coherent(struct device
*dev
, struct scatterlist
*mem
,
114 int order
, gfp_t gfp_mask
)
116 void *buf
= dma_alloc_coherent(dev
, PAGE_SIZE
<< order
,
117 &sg_dma_address(mem
), gfp_mask
);
121 if (offset_in_page(buf
)) {
122 dma_free_coherent(dev
, PAGE_SIZE
<< order
,
123 buf
, sg_dma_address(mem
));
127 sg_set_buf(mem
, buf
, PAGE_SIZE
<< order
);
128 sg_dma_len(mem
) = PAGE_SIZE
<< order
;
132 struct mlx4_icm
*mlx4_alloc_icm(struct mlx4_dev
*dev
, int npages
,
133 gfp_t gfp_mask
, int coherent
)
135 struct mlx4_icm
*icm
;
136 struct mlx4_icm_chunk
*chunk
= NULL
;
140 /* We use sg_set_buf for coherent allocs, which assumes low memory */
141 BUG_ON(coherent
&& (gfp_mask
& __GFP_HIGHMEM
));
143 icm
= kmalloc_node(sizeof(*icm
),
144 gfp_mask
& ~(__GFP_HIGHMEM
| __GFP_NOWARN
),
147 icm
= kmalloc(sizeof(*icm
),
148 gfp_mask
& ~(__GFP_HIGHMEM
| __GFP_NOWARN
));
154 INIT_LIST_HEAD(&icm
->chunk_list
);
156 cur_order
= get_order(MLX4_ICM_ALLOC_SIZE
);
160 chunk
= kmalloc_node(sizeof(*chunk
),
161 gfp_mask
& ~(__GFP_HIGHMEM
|
165 chunk
= kmalloc(sizeof(*chunk
),
166 gfp_mask
& ~(__GFP_HIGHMEM
|
172 sg_init_table(chunk
->mem
, MLX4_ICM_CHUNK_LEN
);
175 list_add_tail(&chunk
->list
, &icm
->chunk_list
);
178 while (1 << cur_order
> npages
)
182 ret
= mlx4_alloc_icm_coherent(&dev
->persist
->pdev
->dev
,
183 &chunk
->mem
[chunk
->npages
],
184 cur_order
, gfp_mask
);
186 ret
= mlx4_alloc_icm_pages(&chunk
->mem
[chunk
->npages
],
201 else if (chunk
->npages
== MLX4_ICM_CHUNK_LEN
) {
202 chunk
->nsg
= pci_map_sg(dev
->persist
->pdev
, chunk
->mem
,
204 PCI_DMA_BIDIRECTIONAL
);
210 if (chunk
->npages
== MLX4_ICM_CHUNK_LEN
)
213 npages
-= 1 << cur_order
;
216 if (!coherent
&& chunk
) {
217 chunk
->nsg
= pci_map_sg(dev
->persist
->pdev
, chunk
->mem
,
219 PCI_DMA_BIDIRECTIONAL
);
228 mlx4_free_icm(dev
, icm
, coherent
);
232 static int mlx4_MAP_ICM(struct mlx4_dev
*dev
, struct mlx4_icm
*icm
, u64 virt
)
234 return mlx4_map_cmd(dev
, MLX4_CMD_MAP_ICM
, icm
, virt
);
237 static int mlx4_UNMAP_ICM(struct mlx4_dev
*dev
, u64 virt
, u32 page_count
)
239 return mlx4_cmd(dev
, virt
, page_count
, 0, MLX4_CMD_UNMAP_ICM
,
240 MLX4_CMD_TIME_CLASS_B
, MLX4_CMD_NATIVE
);
243 int mlx4_MAP_ICM_AUX(struct mlx4_dev
*dev
, struct mlx4_icm
*icm
)
245 return mlx4_map_cmd(dev
, MLX4_CMD_MAP_ICM_AUX
, icm
, -1);
248 int mlx4_UNMAP_ICM_AUX(struct mlx4_dev
*dev
)
250 return mlx4_cmd(dev
, 0, 0, 0, MLX4_CMD_UNMAP_ICM_AUX
,
251 MLX4_CMD_TIME_CLASS_B
, MLX4_CMD_NATIVE
);
254 int mlx4_table_get(struct mlx4_dev
*dev
, struct mlx4_icm_table
*table
, u32 obj
)
256 u32 i
= (obj
& (table
->num_obj
- 1)) /
257 (MLX4_TABLE_CHUNK_SIZE
/ table
->obj_size
);
260 mutex_lock(&table
->mutex
);
263 ++table
->icm
[i
]->refcount
;
267 table
->icm
[i
] = mlx4_alloc_icm(dev
, MLX4_TABLE_CHUNK_SIZE
>> PAGE_SHIFT
,
268 (table
->lowmem
? GFP_KERNEL
: GFP_HIGHUSER
) |
269 __GFP_NOWARN
, table
->coherent
);
270 if (!table
->icm
[i
]) {
275 if (mlx4_MAP_ICM(dev
, table
->icm
[i
], table
->virt
+
276 (u64
) i
* MLX4_TABLE_CHUNK_SIZE
)) {
277 mlx4_free_icm(dev
, table
->icm
[i
], table
->coherent
);
278 table
->icm
[i
] = NULL
;
283 ++table
->icm
[i
]->refcount
;
286 mutex_unlock(&table
->mutex
);
290 void mlx4_table_put(struct mlx4_dev
*dev
, struct mlx4_icm_table
*table
, u32 obj
)
295 i
= (obj
& (table
->num_obj
- 1)) / (MLX4_TABLE_CHUNK_SIZE
/ table
->obj_size
);
297 mutex_lock(&table
->mutex
);
299 if (--table
->icm
[i
]->refcount
== 0) {
300 offset
= (u64
) i
* MLX4_TABLE_CHUNK_SIZE
;
301 mlx4_UNMAP_ICM(dev
, table
->virt
+ offset
,
302 MLX4_TABLE_CHUNK_SIZE
/ MLX4_ICM_PAGE_SIZE
);
303 mlx4_free_icm(dev
, table
->icm
[i
], table
->coherent
);
304 table
->icm
[i
] = NULL
;
307 mutex_unlock(&table
->mutex
);
310 void *mlx4_table_find(struct mlx4_icm_table
*table
, u32 obj
,
311 dma_addr_t
*dma_handle
)
313 int offset
, dma_offset
, i
;
315 struct mlx4_icm_chunk
*chunk
;
316 struct mlx4_icm
*icm
;
317 struct page
*page
= NULL
;
322 mutex_lock(&table
->mutex
);
324 idx
= (u64
) (obj
& (table
->num_obj
- 1)) * table
->obj_size
;
325 icm
= table
->icm
[idx
/ MLX4_TABLE_CHUNK_SIZE
];
326 dma_offset
= offset
= idx
% MLX4_TABLE_CHUNK_SIZE
;
331 list_for_each_entry(chunk
, &icm
->chunk_list
, list
) {
332 for (i
= 0; i
< chunk
->npages
; ++i
) {
333 if (dma_handle
&& dma_offset
>= 0) {
334 if (sg_dma_len(&chunk
->mem
[i
]) > dma_offset
)
335 *dma_handle
= sg_dma_address(&chunk
->mem
[i
]) +
337 dma_offset
-= sg_dma_len(&chunk
->mem
[i
]);
340 * DMA mapping can merge pages but not split them,
341 * so if we found the page, dma_handle has already
344 if (chunk
->mem
[i
].length
> offset
) {
345 page
= sg_page(&chunk
->mem
[i
]);
348 offset
-= chunk
->mem
[i
].length
;
353 mutex_unlock(&table
->mutex
);
354 return page
? lowmem_page_address(page
) + offset
: NULL
;
357 int mlx4_table_get_range(struct mlx4_dev
*dev
, struct mlx4_icm_table
*table
,
360 int inc
= MLX4_TABLE_CHUNK_SIZE
/ table
->obj_size
;
364 for (i
= start
; i
<= end
; i
+= inc
) {
365 err
= mlx4_table_get(dev
, table
, i
);
375 mlx4_table_put(dev
, table
, i
);
381 void mlx4_table_put_range(struct mlx4_dev
*dev
, struct mlx4_icm_table
*table
,
386 for (i
= start
; i
<= end
; i
+= MLX4_TABLE_CHUNK_SIZE
/ table
->obj_size
)
387 mlx4_table_put(dev
, table
, i
);
390 int mlx4_init_icm_table(struct mlx4_dev
*dev
, struct mlx4_icm_table
*table
,
391 u64 virt
, int obj_size
, u32 nobj
, int reserved
,
392 int use_lowmem
, int use_coherent
)
400 obj_per_chunk
= MLX4_TABLE_CHUNK_SIZE
/ obj_size
;
401 num_icm
= (nobj
+ obj_per_chunk
- 1) / obj_per_chunk
;
403 table
->icm
= kcalloc(num_icm
, sizeof *table
->icm
, GFP_KERNEL
);
407 table
->num_icm
= num_icm
;
408 table
->num_obj
= nobj
;
409 table
->obj_size
= obj_size
;
410 table
->lowmem
= use_lowmem
;
411 table
->coherent
= use_coherent
;
412 mutex_init(&table
->mutex
);
414 size
= (u64
) nobj
* obj_size
;
415 for (i
= 0; i
* MLX4_TABLE_CHUNK_SIZE
< reserved
* obj_size
; ++i
) {
416 chunk_size
= MLX4_TABLE_CHUNK_SIZE
;
417 if ((i
+ 1) * MLX4_TABLE_CHUNK_SIZE
> size
)
418 chunk_size
= PAGE_ALIGN(size
-
419 i
* MLX4_TABLE_CHUNK_SIZE
);
421 table
->icm
[i
] = mlx4_alloc_icm(dev
, chunk_size
>> PAGE_SHIFT
,
422 (use_lowmem
? GFP_KERNEL
: GFP_HIGHUSER
) |
423 __GFP_NOWARN
, use_coherent
);
426 if (mlx4_MAP_ICM(dev
, table
->icm
[i
], virt
+ i
* MLX4_TABLE_CHUNK_SIZE
)) {
427 mlx4_free_icm(dev
, table
->icm
[i
], use_coherent
);
428 table
->icm
[i
] = NULL
;
433 * Add a reference to this ICM chunk so that it never
434 * gets freed (since it contains reserved firmware objects).
436 ++table
->icm
[i
]->refcount
;
442 for (i
= 0; i
< num_icm
; ++i
)
444 mlx4_UNMAP_ICM(dev
, virt
+ i
* MLX4_TABLE_CHUNK_SIZE
,
445 MLX4_TABLE_CHUNK_SIZE
/ MLX4_ICM_PAGE_SIZE
);
446 mlx4_free_icm(dev
, table
->icm
[i
], use_coherent
);
454 void mlx4_cleanup_icm_table(struct mlx4_dev
*dev
, struct mlx4_icm_table
*table
)
458 for (i
= 0; i
< table
->num_icm
; ++i
)
460 mlx4_UNMAP_ICM(dev
, table
->virt
+ i
* MLX4_TABLE_CHUNK_SIZE
,
461 MLX4_TABLE_CHUNK_SIZE
/ MLX4_ICM_PAGE_SIZE
);
462 mlx4_free_icm(dev
, table
->icm
[i
], table
->coherent
);