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         if (atomic_read(&pool->dirty_count) >= pool->max_items / 10)
  48                 queue_delayed_work(rds_ib_mr_wq, &pool->flush_worker, 10);
  49
  50         /* Switch pools if one of the pool is reaching upper limit */
  51         if (atomic_read(&pool->dirty_count) >=  pool->max_items * 9 / 10) {
  52                 if (pool->pool_type == RDS_IB_MR_8K_POOL)
  53                         pool = rds_ibdev->mr_1m_pool;
  54                 else
  55                         pool = rds_ibdev->mr_8k_pool;
  56         }
  57
  58         ibmr = rds_ib_try_reuse_ibmr(pool);
  59         if (ibmr)
  60                 return ibmr;
  61
  62         ibmr = kzalloc_node(sizeof(*ibmr), GFP_KERNEL,
  63                             rdsibdev_to_node(rds_ibdev));
  64         if (!ibmr) {
  65                 err = -ENOMEM;
  66                 goto out_no_cigar;
  67         }
  68
  69         fmr = &ibmr->u.fmr;
  70         fmr->fmr = ib_alloc_fmr(rds_ibdev->pd,
  71                         (IB_ACCESS_LOCAL_WRITE |
  72                          IB_ACCESS_REMOTE_READ |
  73                          IB_ACCESS_REMOTE_WRITE |
  74                          IB_ACCESS_REMOTE_ATOMIC),
  75                         &pool->fmr_attr);
  76         if (IS_ERR(fmr->fmr)) {
  77                 err = PTR_ERR(fmr->fmr);
  78                 fmr->fmr = NULL;
  79                 pr_warn("RDS/IB: %s failed (err=%d)\n", __func__, err);
  80                 goto out_no_cigar;
  81         }
  82
  83         ibmr->pool = pool;
  84         if (pool->pool_type == RDS_IB_MR_8K_POOL)
  85                 rds_ib_stats_inc(s_ib_rdma_mr_8k_alloc);
  86         else
  87                 rds_ib_stats_inc(s_ib_rdma_mr_1m_alloc);
  88
  89         return ibmr;
  90
  91 out_no_cigar:
  92         if (ibmr) {
  93                 if (fmr->fmr)
  94                         ib_dealloc_fmr(fmr->fmr);
  95                 kfree(ibmr);
  96         }
  97         atomic_dec(&pool->item_count);
  98         return ERR_PTR(err);
  99 }
 100
 101 int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev, struct rds_ib_mr *ibmr,
 102                    struct scatterlist *sg, unsigned int nents)
 103 {
 104         struct ib_device *dev = rds_ibdev->dev;
 105         struct rds_ib_fmr *fmr = &ibmr->u.fmr;
 106         struct scatterlist *scat = sg;
 107         u64 io_addr = 0;
 108         u64 *dma_pages;
 109         u32 len;
 110         int page_cnt, sg_dma_len;
 111         int i, j;
 112         int ret;
 113
 114         sg_dma_len = ib_dma_map_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
 115         if (unlikely(!sg_dma_len)) {
 116                 pr_warn("RDS/IB: %s failed!\n", __func__);
 117                 return -EBUSY;
 118         }
 119
 120         len = 0;
 121         page_cnt = 0;
 122
 123         for (i = 0; i < sg_dma_len; ++i) {
 124                 unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
 125                 u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);
 126
 127                 if (dma_addr & ~PAGE_MASK) {
 128                         if (i > 0)
 129                                 return -EINVAL;
 130                         else
 131                                 ++page_cnt;
 132                 }
 133                 if ((dma_addr + dma_len) & ~PAGE_MASK) {
 134                         if (i < sg_dma_len - 1)
 135                                 return -EINVAL;
 136                         else
 137                                 ++page_cnt;
 138                 }
 139
 140                 len += dma_len;
 141         }
 142
 143         page_cnt += len >> PAGE_SHIFT;
 144         if (page_cnt > ibmr->pool->fmr_attr.max_pages)
 145                 return -EINVAL;
 146
 147         dma_pages = kmalloc_node(sizeof(u64) * page_cnt, GFP_ATOMIC,
 148                                  rdsibdev_to_node(rds_ibdev));
 149         if (!dma_pages)
 150                 return -ENOMEM;
 151
 152         page_cnt = 0;
 153         for (i = 0; i < sg_dma_len; ++i) {
 154                 unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
 155                 u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);
 156
 157                 for (j = 0; j < dma_len; j += PAGE_SIZE)
 158                         dma_pages[page_cnt++] =
 159                                 (dma_addr & PAGE_MASK) + j;
 160         }
 161
 162         ret = ib_map_phys_fmr(fmr->fmr, dma_pages, page_cnt, io_addr);
 163         if (ret)
 164                 goto out;
 165
 166         /* Success - we successfully remapped the MR, so we can
 167          * safely tear down the old mapping.
 168          */
 169         rds_ib_teardown_mr(ibmr);
 170
 171         ibmr->sg = scat;
 172         ibmr->sg_len = nents;
 173         ibmr->sg_dma_len = sg_dma_len;
 174         ibmr->remap_count++;
 175
 176         if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
 177                 rds_ib_stats_inc(s_ib_rdma_mr_8k_used);
 178         else
 179                 rds_ib_stats_inc(s_ib_rdma_mr_1m_used);
 180         ret = 0;
 181
 182 out:
 183         kfree(dma_pages);
 184
 185         return ret;
 186 }
 187
 188 struct rds_ib_mr *rds_ib_reg_fmr(struct rds_ib_device *rds_ibdev,
 189                                  struct scatterlist *sg,
 190                                  unsigned long nents,
 191                                  u32 *key)
 192 {
 193         struct rds_ib_mr *ibmr = NULL;
 194         struct rds_ib_fmr *fmr;
 195         int ret;
 196
 197         ibmr = rds_ib_alloc_fmr(rds_ibdev, nents);
 198         if (IS_ERR(ibmr))
 199                 return ibmr;
 200
 201         ibmr->device = rds_ibdev;
 202         fmr = &ibmr->u.fmr;
 203         ret = rds_ib_map_fmr(rds_ibdev, ibmr, sg, nents);
 204         if (ret == 0)
 205                 *key = fmr->fmr->rkey;
 206         else
 207                 rds_ib_free_mr(ibmr, 0);
 208
 209         return ibmr;
 210 }
 211
 212 void rds_ib_unreg_fmr(struct list_head *list, unsigned int *nfreed,
 213                       unsigned long *unpinned, unsigned int goal)
 214 {
 215         struct rds_ib_mr *ibmr, *next;
 216         struct rds_ib_fmr *fmr;
 217         LIST_HEAD(fmr_list);
 218         int ret = 0;
 219         unsigned int freed = *nfreed;
 220
 221         /* String all ib_mr's onto one list and hand them to  ib_unmap_fmr */
 222         list_for_each_entry(ibmr, list, unmap_list) {
 223                 fmr = &ibmr->u.fmr;
 224                 list_add(&fmr->fmr->list, &fmr_list);
 225         }
 226
 227         ret = ib_unmap_fmr(&fmr_list);
 228         if (ret)
 229                 pr_warn("RDS/IB: FMR invalidation failed (err=%d)\n", ret);
 230
 231         /* Now we can destroy the DMA mapping and unpin any pages */
 232         list_for_each_entry_safe(ibmr, next, list, unmap_list) {
 233                 fmr = &ibmr->u.fmr;
 234                 *unpinned += ibmr->sg_len;
 235                 __rds_ib_teardown_mr(ibmr);
 236                 if (freed < goal ||
 237                     ibmr->remap_count >= ibmr->pool->fmr_attr.max_maps) {
 238                         if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
 239                                 rds_ib_stats_inc(s_ib_rdma_mr_8k_free);
 240                         else
 241                                 rds_ib_stats_inc(s_ib_rdma_mr_1m_free);
 242                         list_del(&ibmr->unmap_list);
 243                         ib_dealloc_fmr(fmr->fmr);
 244                         kfree(ibmr);
 245                         freed++;
 246                 }
 247         }
 248         *nfreed = freed;
 249 }
 250
 251 void rds_ib_free_fmr_list(struct rds_ib_mr *ibmr)
 252 {
 253         struct rds_ib_mr_pool *pool = ibmr->pool;
 254
 255         if (ibmr->remap_count >= pool->fmr_attr.max_maps)
 256                 llist_add(&ibmr->llnode, &pool->drop_list);
 257         else
 258                 llist_add(&ibmr->llnode, &pool->free_list);
 259 }