f6df683f |
1 | /* |
2 | * Copyright (c) 2016 Oracle. All rights reserved. |
3 | * |
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: |
9 | * |
10 | * Redistribution and use in source and binary forms, with or |
11 | * without modification, are permitted provided that the following |
12 | * conditions are met: |
13 | * |
14 | * - Redistributions of source code must retain the above |
15 | * copyright notice, this list of conditions and the following |
16 | * disclaimer. |
17 | * |
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. |
22 | * |
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 |
30 | * SOFTWARE. |
31 | */ |
32 | |
33 | #include "ib_mr.h" |
34 | |
35 | struct rds_ib_mr *rds_ib_alloc_fmr(struct rds_ib_device *rds_ibdev, int npages) |
36 | { |
37 | struct rds_ib_mr_pool *pool; |
38 | struct rds_ib_mr *ibmr = NULL; |
a69365a3 |
39 | struct rds_ib_fmr *fmr; |
490ea596 |
40 | int err = 0; |
f6df683f |
41 | |
42 | if (npages <= RDS_MR_8K_MSG_SIZE) |
43 | pool = rds_ibdev->mr_8k_pool; |
44 | else |
45 | pool = rds_ibdev->mr_1m_pool; |
46 | |
490ea596 |
47 | ibmr = rds_ib_try_reuse_ibmr(pool); |
48 | if (ibmr) |
49 | return ibmr; |
f6df683f |
50 | |
51 | ibmr = kzalloc_node(sizeof(*ibmr), GFP_KERNEL, |
52 | rdsibdev_to_node(rds_ibdev)); |
53 | if (!ibmr) { |
54 | err = -ENOMEM; |
55 | goto out_no_cigar; |
56 | } |
57 | |
a69365a3 |
58 | fmr = &ibmr->u.fmr; |
59 | fmr->fmr = ib_alloc_fmr(rds_ibdev->pd, |
f6df683f |
60 | (IB_ACCESS_LOCAL_WRITE | |
61 | IB_ACCESS_REMOTE_READ | |
62 | IB_ACCESS_REMOTE_WRITE | |
63 | IB_ACCESS_REMOTE_ATOMIC), |
64 | &pool->fmr_attr); |
a69365a3 |
65 | if (IS_ERR(fmr->fmr)) { |
66 | err = PTR_ERR(fmr->fmr); |
67 | fmr->fmr = NULL; |
f6df683f |
68 | pr_warn("RDS/IB: %s failed (err=%d)\n", __func__, err); |
69 | goto out_no_cigar; |
70 | } |
71 | |
72 | ibmr->pool = pool; |
73 | if (pool->pool_type == RDS_IB_MR_8K_POOL) |
74 | rds_ib_stats_inc(s_ib_rdma_mr_8k_alloc); |
75 | else |
76 | rds_ib_stats_inc(s_ib_rdma_mr_1m_alloc); |
77 | |
78 | return ibmr; |
79 | |
80 | out_no_cigar: |
81 | if (ibmr) { |
a69365a3 |
82 | if (fmr->fmr) |
83 | ib_dealloc_fmr(fmr->fmr); |
f6df683f |
84 | kfree(ibmr); |
85 | } |
86 | atomic_dec(&pool->item_count); |
87 | return ERR_PTR(err); |
88 | } |
89 | |
90 | int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev, struct rds_ib_mr *ibmr, |
91 | struct scatterlist *sg, unsigned int nents) |
92 | { |
93 | struct ib_device *dev = rds_ibdev->dev; |
a69365a3 |
94 | struct rds_ib_fmr *fmr = &ibmr->u.fmr; |
f6df683f |
95 | struct scatterlist *scat = sg; |
96 | u64 io_addr = 0; |
97 | u64 *dma_pages; |
98 | u32 len; |
99 | int page_cnt, sg_dma_len; |
100 | int i, j; |
101 | int ret; |
102 | |
103 | sg_dma_len = ib_dma_map_sg(dev, sg, nents, DMA_BIDIRECTIONAL); |
104 | if (unlikely(!sg_dma_len)) { |
105 | pr_warn("RDS/IB: %s failed!\n", __func__); |
106 | return -EBUSY; |
107 | } |
108 | |
109 | len = 0; |
110 | page_cnt = 0; |
111 | |
112 | for (i = 0; i < sg_dma_len; ++i) { |
113 | unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]); |
114 | u64 dma_addr = ib_sg_dma_address(dev, &scat[i]); |
115 | |
116 | if (dma_addr & ~PAGE_MASK) { |
117 | if (i > 0) |
118 | return -EINVAL; |
119 | else |
120 | ++page_cnt; |
121 | } |
122 | if ((dma_addr + dma_len) & ~PAGE_MASK) { |
123 | if (i < sg_dma_len - 1) |
124 | return -EINVAL; |
125 | else |
126 | ++page_cnt; |
127 | } |
128 | |
129 | len += dma_len; |
130 | } |
131 | |
132 | page_cnt += len >> PAGE_SHIFT; |
133 | if (page_cnt > ibmr->pool->fmr_attr.max_pages) |
134 | return -EINVAL; |
135 | |
136 | dma_pages = kmalloc_node(sizeof(u64) * page_cnt, GFP_ATOMIC, |
137 | rdsibdev_to_node(rds_ibdev)); |
138 | if (!dma_pages) |
139 | return -ENOMEM; |
140 | |
141 | page_cnt = 0; |
142 | for (i = 0; i < sg_dma_len; ++i) { |
143 | unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]); |
144 | u64 dma_addr = ib_sg_dma_address(dev, &scat[i]); |
145 | |
146 | for (j = 0; j < dma_len; j += PAGE_SIZE) |
147 | dma_pages[page_cnt++] = |
148 | (dma_addr & PAGE_MASK) + j; |
149 | } |
150 | |
a69365a3 |
151 | ret = ib_map_phys_fmr(fmr->fmr, dma_pages, page_cnt, io_addr); |
f6df683f |
152 | if (ret) |
153 | goto out; |
154 | |
155 | /* Success - we successfully remapped the MR, so we can |
156 | * safely tear down the old mapping. |
157 | */ |
158 | rds_ib_teardown_mr(ibmr); |
159 | |
160 | ibmr->sg = scat; |
161 | ibmr->sg_len = nents; |
162 | ibmr->sg_dma_len = sg_dma_len; |
163 | ibmr->remap_count++; |
164 | |
165 | if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL) |
166 | rds_ib_stats_inc(s_ib_rdma_mr_8k_used); |
167 | else |
168 | rds_ib_stats_inc(s_ib_rdma_mr_1m_used); |
169 | ret = 0; |
170 | |
171 | out: |
172 | kfree(dma_pages); |
173 | |
174 | return ret; |
175 | } |
490ea596 |
176 | |
177 | struct rds_ib_mr *rds_ib_reg_fmr(struct rds_ib_device *rds_ibdev, |
178 | struct scatterlist *sg, |
179 | unsigned long nents, |
180 | u32 *key) |
181 | { |
182 | struct rds_ib_mr *ibmr = NULL; |
183 | struct rds_ib_fmr *fmr; |
184 | int ret; |
185 | |
186 | ibmr = rds_ib_alloc_fmr(rds_ibdev, nents); |
187 | if (IS_ERR(ibmr)) |
188 | return ibmr; |
189 | |
190 | ibmr->device = rds_ibdev; |
191 | fmr = &ibmr->u.fmr; |
192 | ret = rds_ib_map_fmr(rds_ibdev, ibmr, sg, nents); |
193 | if (ret == 0) |
194 | *key = fmr->fmr->rkey; |
195 | else |
196 | rds_ib_free_mr(ibmr, 0); |
197 | |
198 | return ibmr; |
199 | } |
200 | |
201 | void rds_ib_unreg_fmr(struct list_head *list, unsigned int *nfreed, |
202 | unsigned long *unpinned, unsigned int goal) |
203 | { |
204 | struct rds_ib_mr *ibmr, *next; |
205 | struct rds_ib_fmr *fmr; |
206 | LIST_HEAD(fmr_list); |
207 | int ret = 0; |
208 | unsigned int freed = *nfreed; |
209 | |
210 | /* String all ib_mr's onto one list and hand them to ib_unmap_fmr */ |
211 | list_for_each_entry(ibmr, list, unmap_list) { |
212 | fmr = &ibmr->u.fmr; |
213 | list_add(&fmr->fmr->list, &fmr_list); |
214 | } |
215 | |
216 | ret = ib_unmap_fmr(&fmr_list); |
217 | if (ret) |
218 | pr_warn("RDS/IB: FMR invalidation failed (err=%d)\n", ret); |
219 | |
220 | /* Now we can destroy the DMA mapping and unpin any pages */ |
221 | list_for_each_entry_safe(ibmr, next, list, unmap_list) { |
222 | fmr = &ibmr->u.fmr; |
223 | *unpinned += ibmr->sg_len; |
224 | __rds_ib_teardown_mr(ibmr); |
225 | if (freed < goal || |
226 | ibmr->remap_count >= ibmr->pool->fmr_attr.max_maps) { |
227 | if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL) |
228 | rds_ib_stats_inc(s_ib_rdma_mr_8k_free); |
229 | else |
230 | rds_ib_stats_inc(s_ib_rdma_mr_1m_free); |
231 | list_del(&ibmr->unmap_list); |
232 | ib_dealloc_fmr(fmr->fmr); |
233 | kfree(ibmr); |
234 | freed++; |
235 | } |
236 | } |
237 | *nfreed = freed; |
238 | } |
239 | |
240 | void rds_ib_free_fmr_list(struct rds_ib_mr *ibmr) |
241 | { |
242 | struct rds_ib_mr_pool *pool = ibmr->pool; |
243 | |
244 | if (ibmr->remap_count >= pool->fmr_attr.max_maps) |
245 | llist_add(&ibmr->llnode, &pool->drop_list); |
246 | else |
247 | llist_add(&ibmr->llnode, &pool->free_list); |
248 | } |