drivers/infiniband/hw/mlx5/mem.c

   1 /*
   2  * Copyright (c) 2013-2015, Mellanox Technologies. 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 <linux/module.h>
  34 #include <rdma/ib_umem.h>
  35 #include <rdma/ib_umem_odp.h>
  36 #include "mlx5_ib.h"
  37
  38 /* @umem: umem object to scan
  39  * @addr: ib virtual address requested by the user
  40  * @max_page_shift: high limit for page_shift - 0 means no limit
  41  * @count: number of PAGE_SIZE pages covered by umem
  42  * @shift: page shift for the compound pages found in the region
  43  * @ncont: number of compund pages
  44  * @order: log2 of the number of compound pages
  45  */
  46 void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr,
  47                         unsigned long max_page_shift,
  48                         int *count, int *shift,
  49                         int *ncont, int *order)
  50 {
  51         unsigned long tmp;
  52         unsigned long m;
  53         int i, k;
  54         u64 base = 0;
  55         int p = 0;
  56         int skip;
  57         int mask;
  58         u64 len;
  59         u64 pfn;
  60         struct scatterlist *sg;
  61         int entry;
  62         unsigned long page_shift = umem->page_shift;
  63
  64         if (umem->odp_data) {
  65                 *ncont = ib_umem_page_count(umem);
  66                 *count = *ncont << (page_shift - PAGE_SHIFT);
  67                 *shift = page_shift;
  68                 if (order)
  69                         *order = ilog2(roundup_pow_of_two(*ncont));
  70
  71                 return;
  72         }
  73
  74         addr = addr >> page_shift;
  75         tmp = (unsigned long)addr;
  76         m = find_first_bit(&tmp, BITS_PER_LONG);
  77         if (max_page_shift)
  78                 m = min_t(unsigned long, max_page_shift - page_shift, m);
  79         skip = 1 << m;
  80         mask = skip - 1;
  81         i = 0;
  82         for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
  83                 len = sg_dma_len(sg) >> page_shift;
  84                 pfn = sg_dma_address(sg) >> page_shift;
  85                 for (k = 0; k < len; k++) {
  86                         if (!(i & mask)) {
  87                                 tmp = (unsigned long)pfn;
  88                                 m = min_t(unsigned long, m, find_first_bit(&tmp, BITS_PER_LONG));
  89                                 skip = 1 << m;
  90                                 mask = skip - 1;
  91                                 base = pfn;
  92                                 p = 0;
  93                         } else {
  94                                 if (base + p != pfn) {
  95                                         tmp = (unsigned long)p;
  96                                         m = find_first_bit(&tmp, BITS_PER_LONG);
  97                                         skip = 1 << m;
  98                                         mask = skip - 1;
  99                                         base = pfn;
 100                                         p = 0;
 101                                 }
 102                         }
 103                         p++;
 104                         i++;
 105                 }
 106         }
 107
 108         if (i) {
 109                 m = min_t(unsigned long, ilog2(roundup_pow_of_two(i)), m);
 110
 111                 if (order)
 112                         *order = ilog2(roundup_pow_of_two(i) >> m);
 113
 114                 *ncont = DIV_ROUND_UP(i, (1 << m));
 115         } else {
 116                 m  = 0;
 117
 118                 if (order)
 119                         *order = 0;
 120
 121                 *ncont = 0;
 122         }
 123         *shift = page_shift + m;
 124         *count = i;
 125 }
 126
 127 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
 128 static u64 umem_dma_to_mtt(dma_addr_t umem_dma)
 129 {
 130         u64 mtt_entry = umem_dma & ODP_DMA_ADDR_MASK;
 131
 132         if (umem_dma & ODP_READ_ALLOWED_BIT)
 133                 mtt_entry |= MLX5_IB_MTT_READ;
 134         if (umem_dma & ODP_WRITE_ALLOWED_BIT)
 135                 mtt_entry |= MLX5_IB_MTT_WRITE;
 136
 137         return mtt_entry;
 138 }
 139 #endif
 140
 141 /*
 142  * Populate the given array with bus addresses from the umem.
 143  *
 144  * dev - mlx5_ib device
 145  * umem - umem to use to fill the pages
 146  * page_shift - determines the page size used in the resulting array
 147  * offset - offset into the umem to start from,
 148  *          only implemented for ODP umems
 149  * num_pages - total number of pages to fill
 150  * pas - bus addresses array to fill
 151  * access_flags - access flags to set on all present pages.
 152                   use enum mlx5_ib_mtt_access_flags for this.
 153  */
 154 void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
 155                             int page_shift, size_t offset, size_t num_pages,
 156                             __be64 *pas, int access_flags)
 157 {
 158         unsigned long umem_page_shift = umem->page_shift;
 159         int shift = page_shift - umem_page_shift;
 160         int mask = (1 << shift) - 1;
 161         int i, k, idx;
 162         u64 cur = 0;
 163         u64 base;
 164         int len;
 165         struct scatterlist *sg;
 166         int entry;
 167 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
 168         const bool odp = umem->odp_data != NULL;
 169
 170         if (odp) {
 171                 WARN_ON(shift != 0);
 172                 WARN_ON(access_flags != (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE));
 173
 174                 for (i = 0; i < num_pages; ++i) {
 175                         dma_addr_t pa = umem->odp_data->dma_list[offset + i];
 176
 177                         pas[i] = cpu_to_be64(umem_dma_to_mtt(pa));
 178                 }
 179                 return;
 180         }
 181 #endif
 182
 183         i = 0;
 184         for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
 185                 len = sg_dma_len(sg) >> umem_page_shift;
 186                 base = sg_dma_address(sg);
 187
 188                 /* Skip elements below offset */
 189                 if (i + len < offset << shift) {
 190                         i += len;
 191                         continue;
 192                 }
 193
 194                 /* Skip pages below offset */
 195                 if (i < offset << shift) {
 196                         k = (offset << shift) - i;
 197                         i = offset << shift;
 198                 } else {
 199                         k = 0;
 200                 }
 201
 202                 for (; k < len; k++) {
 203                         if (!(i & mask)) {
 204                                 cur = base + (k << umem_page_shift);
 205                                 cur |= access_flags;
 206                                 idx = (i >> shift) - offset;
 207
 208                                 pas[idx] = cpu_to_be64(cur);
 209                                 mlx5_ib_dbg(dev, "pas[%d] 0x%llx\n",
 210                                             i >> shift, be64_to_cpu(pas[idx]));
 211                         }
 212                         i++;
 213
 214                         /* Stop after num_pages reached */
 215                         if (i >> shift >= offset + num_pages)
 216                                 return;
 217                 }
 218         }
 219 }
 220
 221 void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
 222                           int page_shift, __be64 *pas, int access_flags)
 223 {
 224         return __mlx5_ib_populate_pas(dev, umem, page_shift, 0,
 225                                       ib_umem_num_pages(umem), pas,
 226                                       access_flags);
 227 }
 228 int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset)
 229 {
 230         u64 page_size;
 231         u64 page_mask;
 232         u64 off_size;
 233         u64 off_mask;
 234         u64 buf_off;
 235
 236         page_size = (u64)1 << page_shift;
 237         page_mask = page_size - 1;
 238         buf_off = addr & page_mask;
 239         off_size = page_size >> 6;
 240         off_mask = off_size - 1;
 241
 242         if (buf_off & off_mask)
 243                 return -EINVAL;
 244
 245         *offset = buf_off >> ilog2(off_size);
 246         return 0;
 247 }