net/rds/ib_fmr.c

   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;
  39         struct rds_ib_fmr *fmr;
  40         int err = 0;
  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
  47         ibmr = rds_ib_try_reuse_ibmr(pool);
  48         if (ibmr)
  49                 return ibmr;
  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
  58         fmr = &ibmr->u.fmr;
  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),
  64                         &pool->fmr_attr);
  65         if (IS_ERR(fmr->fmr)) {
  66                 err = PTR_ERR(fmr->fmr);
  67                 fmr->fmr = NULL;
  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         kfree(ibmr);
  82         atomic_dec(&pool->item_count);
  83
  84         return ERR_PTR(err);
  85 }
  86
  87 static int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev,
  88                           struct rds_ib_mr *ibmr, struct scatterlist *sg,
  89                           unsigned int nents)
  90 {
  91         struct ib_device *dev = rds_ibdev->dev;
  92         struct rds_ib_fmr *fmr = &ibmr->u.fmr;
  93         struct scatterlist *scat = sg;
  94         u64 io_addr = 0;
  95         u64 *dma_pages;
  96         u32 len;
  97         int page_cnt, sg_dma_len;
  98         int i, j;
  99         int ret;
 100
 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__);
 104                 return -EBUSY;
 105         }
 106
 107         len = 0;
 108         page_cnt = 0;
 109
 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]);
 113
 114                 if (dma_addr & ~PAGE_MASK) {
 115                         if (i > 0) {
 116                                 ib_dma_unmap_sg(dev, sg, nents,
 117                                                 DMA_BIDIRECTIONAL);
 118                                 return -EINVAL;
 119                         } else {
 120                                 ++page_cnt;
 121                         }
 122                 }
 123                 if ((dma_addr + dma_len) & ~PAGE_MASK) {
 124                         if (i < sg_dma_len - 1) {
 125                                 ib_dma_unmap_sg(dev, sg, nents,
 126                                                 DMA_BIDIRECTIONAL);
 127                                 return -EINVAL;
 128                         } else {
 129                                 ++page_cnt;
 130                         }
 131                 }
 132
 133                 len += dma_len;
 134         }
 135
 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);
 139                 return -EINVAL;
 140         }
 141
 142         dma_pages = kmalloc_array_node(sizeof(u64), page_cnt, GFP_ATOMIC,
 143                                        rdsibdev_to_node(rds_ibdev));
 144         if (!dma_pages) {
 145                 ib_dma_unmap_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
 146                 return -ENOMEM;
 147         }
 148
 149         page_cnt = 0;
 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]);
 153
 154                 for (j = 0; j < dma_len; j += PAGE_SIZE)
 155                         dma_pages[page_cnt++] =
 156                                 (dma_addr & PAGE_MASK) + j;
 157         }
 158
 159         ret = ib_map_phys_fmr(fmr->fmr, dma_pages, page_cnt, io_addr);
 160         if (ret) {
 161                 ib_dma_unmap_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
 162                 goto out;
 163         }
 164
 165         /* Success - we successfully remapped the MR, so we can
 166          * safely tear down the old mapping.
 167          */
 168         rds_ib_teardown_mr(ibmr);
 169
 170         ibmr->sg = scat;
 171         ibmr->sg_len = nents;
 172         ibmr->sg_dma_len = sg_dma_len;
 173         ibmr->remap_count++;
 174
 175         if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
 176                 rds_ib_stats_inc(s_ib_rdma_mr_8k_used);
 177         else
 178                 rds_ib_stats_inc(s_ib_rdma_mr_1m_used);
 179         ret = 0;
 180
 181 out:
 182         kfree(dma_pages);
 183
 184         return ret;
 185 }
 186
 187 struct rds_ib_mr *rds_ib_reg_fmr(struct rds_ib_device *rds_ibdev,
 188                                  struct scatterlist *sg,
 189                                  unsigned long nents,
 190                                  u32 *key)
 191 {
 192         struct rds_ib_mr *ibmr = NULL;
 193         struct rds_ib_fmr *fmr;
 194         int ret;
 195
 196         ibmr = rds_ib_alloc_fmr(rds_ibdev, nents);
 197         if (IS_ERR(ibmr))
 198                 return ibmr;
 199
 200         ibmr->device = rds_ibdev;
 201         fmr = &ibmr->u.fmr;
 202         ret = rds_ib_map_fmr(rds_ibdev, ibmr, sg, nents);
 203         if (ret == 0)
 204                 *key = fmr->fmr->rkey;
 205         else
 206                 rds_ib_free_mr(ibmr, 0);
 207
 208         return ibmr;
 209 }
 210
 211 void rds_ib_unreg_fmr(struct list_head *list, unsigned int *nfreed,
 212                       unsigned long *unpinned, unsigned int goal)
 213 {
 214         struct rds_ib_mr *ibmr, *next;
 215         struct rds_ib_fmr *fmr;
 216         LIST_HEAD(fmr_list);
 217         int ret = 0;
 218         unsigned int freed = *nfreed;
 219
 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) {
 222                 fmr = &ibmr->u.fmr;
 223                 list_add(&fmr->fmr->list, &fmr_list);
 224         }
 225
 226         ret = ib_unmap_fmr(&fmr_list);
 227         if (ret)
 228                 pr_warn("RDS/IB: FMR invalidation failed (err=%d)\n", ret);
 229
 230         /* Now we can destroy the DMA mapping and unpin any pages */
 231         list_for_each_entry_safe(ibmr, next, list, unmap_list) {
 232                 fmr = &ibmr->u.fmr;
 233                 *unpinned += ibmr->sg_len;
 234                 __rds_ib_teardown_mr(ibmr);
 235                 if (freed < goal ||
 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);
 239                         else
 240                                 rds_ib_stats_inc(s_ib_rdma_mr_1m_free);
 241                         list_del(&ibmr->unmap_list);
 242                         ib_dealloc_fmr(fmr->fmr);
 243                         kfree(ibmr);
 244                         freed++;
 245                 }
 246         }
 247         *nfreed = freed;
 248 }
 249
 250 void rds_ib_free_fmr_list(struct rds_ib_mr *ibmr)
 251 {
 252         struct rds_ib_mr_pool *pool = ibmr->pool;
 253
 254         if (ibmr->remap_count >= pool->fmr_attr.max_maps)
 255                 llist_add(&ibmr->llnode, &pool->drop_list);
 256         else
 257                 llist_add(&ibmr->llnode, &pool->free_list);
 258 }