2 * Copyright (c) 2016 Oracle. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 struct rds_ib_mr
*rds_ib_alloc_fmr(struct rds_ib_device
*rds_ibdev
, int npages
)
37 struct rds_ib_mr_pool
*pool
;
38 struct rds_ib_mr
*ibmr
= NULL
;
39 int err
= 0, iter
= 0;
41 if (npages
<= RDS_MR_8K_MSG_SIZE
)
42 pool
= rds_ibdev
->mr_8k_pool
;
44 pool
= rds_ibdev
->mr_1m_pool
;
46 if (atomic_read(&pool
->dirty_count
) >= pool
->max_items
/ 10)
47 queue_delayed_work(rds_ib_mr_wq
, &pool
->flush_worker
, 10);
49 /* Switch pools if one of the pool is reaching upper limit */
50 if (atomic_read(&pool
->dirty_count
) >= pool
->max_items
* 9 / 10) {
51 if (pool
->pool_type
== RDS_IB_MR_8K_POOL
)
52 pool
= rds_ibdev
->mr_1m_pool
;
54 pool
= rds_ibdev
->mr_8k_pool
;
58 ibmr
= rds_ib_reuse_mr(pool
);
62 /* No clean MRs - now we have the choice of either
63 * allocating a fresh MR up to the limit imposed by the
64 * driver, or flush any dirty unused MRs.
65 * We try to avoid stalling in the send path if possible,
66 * so we allocate as long as we're allowed to.
68 * We're fussy with enforcing the FMR limit, though. If the
69 * driver tells us we can't use more than N fmrs, we shouldn't
70 * start arguing with it
72 if (atomic_inc_return(&pool
->item_count
) <= pool
->max_items
)
75 atomic_dec(&pool
->item_count
);
78 if (pool
->pool_type
== RDS_IB_MR_8K_POOL
)
79 rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_depleted
);
81 rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_depleted
);
82 return ERR_PTR(-EAGAIN
);
85 /* We do have some empty MRs. Flush them out. */
86 if (pool
->pool_type
== RDS_IB_MR_8K_POOL
)
87 rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_wait
);
89 rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_wait
);
90 rds_ib_flush_mr_pool(pool
, 0, &ibmr
);
95 ibmr
= kzalloc_node(sizeof(*ibmr
), GFP_KERNEL
,
96 rdsibdev_to_node(rds_ibdev
));
102 ibmr
->fmr
= ib_alloc_fmr(rds_ibdev
->pd
,
103 (IB_ACCESS_LOCAL_WRITE
|
104 IB_ACCESS_REMOTE_READ
|
105 IB_ACCESS_REMOTE_WRITE
|
106 IB_ACCESS_REMOTE_ATOMIC
),
108 if (IS_ERR(ibmr
->fmr
)) {
109 err
= PTR_ERR(ibmr
->fmr
);
111 pr_warn("RDS/IB: %s failed (err=%d)\n", __func__
, err
);
116 if (pool
->pool_type
== RDS_IB_MR_8K_POOL
)
117 rds_ib_stats_inc(s_ib_rdma_mr_8k_alloc
);
119 rds_ib_stats_inc(s_ib_rdma_mr_1m_alloc
);
126 ib_dealloc_fmr(ibmr
->fmr
);
129 atomic_dec(&pool
->item_count
);
133 int rds_ib_map_fmr(struct rds_ib_device
*rds_ibdev
, struct rds_ib_mr
*ibmr
,
134 struct scatterlist
*sg
, unsigned int nents
)
136 struct ib_device
*dev
= rds_ibdev
->dev
;
137 struct scatterlist
*scat
= sg
;
141 int page_cnt
, sg_dma_len
;
145 sg_dma_len
= ib_dma_map_sg(dev
, sg
, nents
, DMA_BIDIRECTIONAL
);
146 if (unlikely(!sg_dma_len
)) {
147 pr_warn("RDS/IB: %s failed!\n", __func__
);
154 for (i
= 0; i
< sg_dma_len
; ++i
) {
155 unsigned int dma_len
= ib_sg_dma_len(dev
, &scat
[i
]);
156 u64 dma_addr
= ib_sg_dma_address(dev
, &scat
[i
]);
158 if (dma_addr
& ~PAGE_MASK
) {
164 if ((dma_addr
+ dma_len
) & ~PAGE_MASK
) {
165 if (i
< sg_dma_len
- 1)
174 page_cnt
+= len
>> PAGE_SHIFT
;
175 if (page_cnt
> ibmr
->pool
->fmr_attr
.max_pages
)
178 dma_pages
= kmalloc_node(sizeof(u64
) * page_cnt
, GFP_ATOMIC
,
179 rdsibdev_to_node(rds_ibdev
));
184 for (i
= 0; i
< sg_dma_len
; ++i
) {
185 unsigned int dma_len
= ib_sg_dma_len(dev
, &scat
[i
]);
186 u64 dma_addr
= ib_sg_dma_address(dev
, &scat
[i
]);
188 for (j
= 0; j
< dma_len
; j
+= PAGE_SIZE
)
189 dma_pages
[page_cnt
++] =
190 (dma_addr
& PAGE_MASK
) + j
;
193 ret
= ib_map_phys_fmr(ibmr
->fmr
, dma_pages
, page_cnt
, io_addr
);
197 /* Success - we successfully remapped the MR, so we can
198 * safely tear down the old mapping.
200 rds_ib_teardown_mr(ibmr
);
203 ibmr
->sg_len
= nents
;
204 ibmr
->sg_dma_len
= sg_dma_len
;
207 if (ibmr
->pool
->pool_type
== RDS_IB_MR_8K_POOL
)
208 rds_ib_stats_inc(s_ib_rdma_mr_8k_used
);
210 rds_ib_stats_inc(s_ib_rdma_mr_1m_used
);