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 struct rds_ib_fmr *fmr;
42 if (npages <= RDS_MR_8K_MSG_SIZE)
43 pool = rds_ibdev->mr_8k_pool;
45 pool = rds_ibdev->mr_1m_pool;
47 ibmr = rds_ib_try_reuse_ibmr(pool);
51 ibmr = kzalloc_node(sizeof(*ibmr), GFP_KERNEL,
52 rdsibdev_to_node(rds_ibdev));
59 fmr->fmr = ib_alloc_fmr(rds_ibdev->pd,
60 (IB_ACCESS_LOCAL_WRITE |
61 IB_ACCESS_REMOTE_READ |
62 IB_ACCESS_REMOTE_WRITE |
63 IB_ACCESS_REMOTE_ATOMIC),
65 if (IS_ERR(fmr->fmr)) {
66 err = PTR_ERR(fmr->fmr);
68 pr_warn("RDS/IB: %s failed (err=%d)\n", __func__, err);
73 if (pool->pool_type == RDS_IB_MR_8K_POOL)
74 rds_ib_stats_inc(s_ib_rdma_mr_8k_alloc);
76 rds_ib_stats_inc(s_ib_rdma_mr_1m_alloc);
82 atomic_dec(&pool->item_count);
87 static int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev,
88 struct rds_ib_mr *ibmr, struct scatterlist *sg,
91 struct ib_device *dev = rds_ibdev->dev;
92 struct rds_ib_fmr *fmr = &ibmr->u.fmr;
93 struct scatterlist *scat = sg;
97 int page_cnt, sg_dma_len;
101 sg_dma_len = ib_dma_map_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
102 if (unlikely(!sg_dma_len)) {
103 pr_warn("RDS/IB: %s failed!\n", __func__);
110 for (i = 0; i < sg_dma_len; ++i) {
111 unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
112 u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);
114 if (dma_addr & ~PAGE_MASK) {
116 ib_dma_unmap_sg(dev, sg, nents,
123 if ((dma_addr + dma_len) & ~PAGE_MASK) {
124 if (i < sg_dma_len - 1) {
125 ib_dma_unmap_sg(dev, sg, nents,
136 page_cnt += len >> PAGE_SHIFT;
137 if (page_cnt > ibmr->pool->fmr_attr.max_pages) {
138 ib_dma_unmap_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
142 dma_pages = kmalloc_node(sizeof(u64) * page_cnt, GFP_ATOMIC,
143 rdsibdev_to_node(rds_ibdev));
145 ib_dma_unmap_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
150 for (i = 0; i < sg_dma_len; ++i) {
151 unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
152 u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);
154 for (j = 0; j < dma_len; j += PAGE_SIZE)
155 dma_pages[page_cnt++] =
156 (dma_addr & PAGE_MASK) + j;
159 ret = ib_map_phys_fmr(fmr->fmr, dma_pages, page_cnt, io_addr);
161 ib_dma_unmap_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
165 /* Success - we successfully remapped the MR, so we can
166 * safely tear down the old mapping.
168 rds_ib_teardown_mr(ibmr);
171 ibmr->sg_len = nents;
172 ibmr->sg_dma_len = sg_dma_len;
175 if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
176 rds_ib_stats_inc(s_ib_rdma_mr_8k_used);
178 rds_ib_stats_inc(s_ib_rdma_mr_1m_used);
187 struct rds_ib_mr *rds_ib_reg_fmr(struct rds_ib_device *rds_ibdev,
188 struct scatterlist *sg,
192 struct rds_ib_mr *ibmr = NULL;
193 struct rds_ib_fmr *fmr;
196 ibmr = rds_ib_alloc_fmr(rds_ibdev, nents);
200 ibmr->device = rds_ibdev;
202 ret = rds_ib_map_fmr(rds_ibdev, ibmr, sg, nents);
204 *key = fmr->fmr->rkey;
206 rds_ib_free_mr(ibmr, 0);
211 void rds_ib_unreg_fmr(struct list_head *list, unsigned int *nfreed,
212 unsigned long *unpinned, unsigned int goal)
214 struct rds_ib_mr *ibmr, *next;
215 struct rds_ib_fmr *fmr;
218 unsigned int freed = *nfreed;
220 /* String all ib_mr's onto one list and hand them to ib_unmap_fmr */
221 list_for_each_entry(ibmr, list, unmap_list) {
223 list_add(&fmr->fmr->list, &fmr_list);
226 ret = ib_unmap_fmr(&fmr_list);
228 pr_warn("RDS/IB: FMR invalidation failed (err=%d)\n", ret);
230 /* Now we can destroy the DMA mapping and unpin any pages */
231 list_for_each_entry_safe(ibmr, next, list, unmap_list) {
233 *unpinned += ibmr->sg_len;
234 __rds_ib_teardown_mr(ibmr);
236 ibmr->remap_count >= ibmr->pool->fmr_attr.max_maps) {
237 if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
238 rds_ib_stats_inc(s_ib_rdma_mr_8k_free);
240 rds_ib_stats_inc(s_ib_rdma_mr_1m_free);
241 list_del(&ibmr->unmap_list);
242 ib_dealloc_fmr(fmr->fmr);
250 void rds_ib_free_fmr_list(struct rds_ib_mr *ibmr)
252 struct rds_ib_mr_pool *pool = ibmr->pool;
254 if (ibmr->remap_count >= pool->fmr_attr.max_maps)
255 llist_add(&ibmr->llnode, &pool->drop_list);
257 llist_add(&ibmr->llnode, &pool->free_list);