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         int err = 0, iter = 0;
  40
  41         if (npages <= RDS_MR_8K_MSG_SIZE)
  42                 pool = rds_ibdev->mr_8k_pool;
  43         else
  44                 pool = rds_ibdev->mr_1m_pool;
  45
  46         if (atomic_read(&pool->dirty_count) >= pool->max_items / 10)
  47                 queue_delayed_work(rds_ib_mr_wq, &pool->flush_worker, 10);
  48
  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;
  53                 else
  54                         pool = rds_ibdev->mr_8k_pool;
  55         }
  56
  57         while (1) {
  58                 ibmr = rds_ib_reuse_mr(pool);
  59                 if (ibmr)
  60                         return ibmr;
  61
  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.
  67                  *
  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
  71                  */
  72                 if (atomic_inc_return(&pool->item_count) <= pool->max_items)
  73                         break;
  74
  75                 atomic_dec(&pool->item_count);
  76
  77                 if (++iter > 2) {
  78                         if (pool->pool_type == RDS_IB_MR_8K_POOL)
  79                                 rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_depleted);
  80                         else
  81                                 rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_depleted);
  82                         return ERR_PTR(-EAGAIN);
  83                 }
  84
  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);
  88                 else
  89                         rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_wait);
  90                 rds_ib_flush_mr_pool(pool, 0, &ibmr);
  91                 if (ibmr)
  92                         return ibmr;
  93         }
  94
  95         ibmr = kzalloc_node(sizeof(*ibmr), GFP_KERNEL,
  96                             rdsibdev_to_node(rds_ibdev));
  97         if (!ibmr) {
  98                 err = -ENOMEM;
  99                 goto out_no_cigar;
 100         }
 101
 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),
 107                         &pool->fmr_attr);
 108         if (IS_ERR(ibmr->fmr)) {
 109                 err = PTR_ERR(ibmr->fmr);
 110                 ibmr->fmr = NULL;
 111                 pr_warn("RDS/IB: %s failed (err=%d)\n", __func__, err);
 112                 goto out_no_cigar;
 113         }
 114
 115         ibmr->pool = pool;
 116         if (pool->pool_type == RDS_IB_MR_8K_POOL)
 117                 rds_ib_stats_inc(s_ib_rdma_mr_8k_alloc);
 118         else
 119                 rds_ib_stats_inc(s_ib_rdma_mr_1m_alloc);
 120
 121         return ibmr;
 122
 123 out_no_cigar:
 124         if (ibmr) {
 125                 if (ibmr->fmr)
 126                         ib_dealloc_fmr(ibmr->fmr);
 127                 kfree(ibmr);
 128         }
 129         atomic_dec(&pool->item_count);
 130         return ERR_PTR(err);
 131 }
 132
 133 int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev, struct rds_ib_mr *ibmr,
 134                    struct scatterlist *sg, unsigned int nents)
 135 {
 136         struct ib_device *dev = rds_ibdev->dev;
 137         struct scatterlist *scat = sg;
 138         u64 io_addr = 0;
 139         u64 *dma_pages;
 140         u32 len;
 141         int page_cnt, sg_dma_len;
 142         int i, j;
 143         int ret;
 144
 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__);
 148                 return -EBUSY;
 149         }
 150
 151         len = 0;
 152         page_cnt = 0;
 153
 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]);
 157
 158                 if (dma_addr & ~PAGE_MASK) {
 159                         if (i > 0)
 160                                 return -EINVAL;
 161                         else
 162                                 ++page_cnt;
 163                 }
 164                 if ((dma_addr + dma_len) & ~PAGE_MASK) {
 165                         if (i < sg_dma_len - 1)
 166                                 return -EINVAL;
 167                         else
 168                                 ++page_cnt;
 169                 }
 170
 171                 len += dma_len;
 172         }
 173
 174         page_cnt += len >> PAGE_SHIFT;
 175         if (page_cnt > ibmr->pool->fmr_attr.max_pages)
 176                 return -EINVAL;
 177
 178         dma_pages = kmalloc_node(sizeof(u64) * page_cnt, GFP_ATOMIC,
 179                                  rdsibdev_to_node(rds_ibdev));
 180         if (!dma_pages)
 181                 return -ENOMEM;
 182
 183         page_cnt = 0;
 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]);
 187
 188                 for (j = 0; j < dma_len; j += PAGE_SIZE)
 189                         dma_pages[page_cnt++] =
 190                                 (dma_addr & PAGE_MASK) + j;
 191         }
 192
 193         ret = ib_map_phys_fmr(ibmr->fmr, dma_pages, page_cnt, io_addr);
 194         if (ret)
 195                 goto out;
 196
 197         /* Success - we successfully remapped the MR, so we can
 198          * safely tear down the old mapping.
 199          */
 200         rds_ib_teardown_mr(ibmr);
 201
 202         ibmr->sg = scat;
 203         ibmr->sg_len = nents;
 204         ibmr->sg_dma_len = sg_dma_len;
 205         ibmr->remap_count++;
 206
 207         if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
 208                 rds_ib_stats_inc(s_ib_rdma_mr_8k_used);
 209         else
 210                 rds_ib_stats_inc(s_ib_rdma_mr_1m_used);
 211         ret = 0;
 212
 213 out:
 214         kfree(dma_pages);
 215
 216         return ret;
 217 }