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 = ilog2(umem->page_size);
  63
  64         /* With ODP we must always match OS page size. */
  65         if (umem->odp_data) {
  66                 *count = ib_umem_page_count(umem);
  67                 *shift = PAGE_SHIFT;
  68                 *ncont = *count;
  69                 if (order)
  70                         *order = ilog2(roundup_pow_of_two(*count));
  71
  72                 return;
  73         }
  74
  75         addr = addr >> page_shift;
  76         tmp = (unsigned long)addr;
  77         m = find_first_bit(&tmp, BITS_PER_LONG);
  78         if (max_page_shift)
  79                 m = min_t(unsigned long, max_page_shift - page_shift, m);
  80         skip = 1 << m;
  81         mask = skip - 1;
  82         i = 0;
  83         for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
  84                 len = sg_dma_len(sg) >> page_shift;
  85                 pfn = sg_dma_address(sg) >> page_shift;
  86                 for (k = 0; k < len; k++) {
  87                         if (!(i & mask)) {
  88                                 tmp = (unsigned long)pfn;
  89                                 m = min_t(unsigned long, m, find_first_bit(&tmp, BITS_PER_LONG));
  90                                 skip = 1 << m;
  91                                 mask = skip - 1;
  92                                 base = pfn;
  93                                 p = 0;
  94                         } else {
  95                                 if (base + p != pfn) {
  96                                         tmp = (unsigned long)p;
  97                                         m = find_first_bit(&tmp, BITS_PER_LONG);
  98                                         skip = 1 << m;
  99                                         mask = skip - 1;
 100                                         base = pfn;
 101                                         p = 0;
 102                                 }
 103                         }
 104                         p++;
 105                         i++;
 106                 }
 107         }
 108
 109         if (i) {
 110                 m = min_t(unsigned long, ilog2(roundup_pow_of_two(i)), m);
 111
 112                 if (order)
 113                         *order = ilog2(roundup_pow_of_two(i) >> m);
 114
 115                 *ncont = DIV_ROUND_UP(i, (1 << m));
 116         } else {
 117                 m  = 0;
 118
 119                 if (order)
 120                         *order = 0;
 121
 122                 *ncont = 0;
 123         }
 124         *shift = page_shift + m;
 125         *count = i;
 126 }
 127
 128 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
 129 static u64 umem_dma_to_mtt(dma_addr_t umem_dma)
 130 {
 131         u64 mtt_entry = umem_dma & ODP_DMA_ADDR_MASK;
 132
 133         if (umem_dma & ODP_READ_ALLOWED_BIT)
 134                 mtt_entry |= MLX5_IB_MTT_READ;
 135         if (umem_dma & ODP_WRITE_ALLOWED_BIT)
 136                 mtt_entry |= MLX5_IB_MTT_WRITE;
 137
 138         return mtt_entry;
 139 }
 140 #endif
 141
 142 /*
 143  * Populate the given array with bus addresses from the umem.
 144  *
 145  * dev - mlx5_ib device
 146  * umem - umem to use to fill the pages
 147  * page_shift - determines the page size used in the resulting array
 148  * offset - offset into the umem to start from,
 149  *          only implemented for ODP umems
 150  * num_pages - total number of pages to fill
 151  * pas - bus addresses array to fill
 152  * access_flags - access flags to set on all present pages.
 153                   use enum mlx5_ib_mtt_access_flags for this.
 154  */
 155 void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
 156                             int page_shift, size_t offset, size_t num_pages,
 157                             __be64 *pas, int access_flags)
 158 {
 159         unsigned long umem_page_shift = ilog2(umem->page_size);
 160         int shift = page_shift - umem_page_shift;
 161         int mask = (1 << shift) - 1;
 162         int i, k, idx;
 163         u64 cur = 0;
 164         u64 base;
 165         int len;
 166         struct scatterlist *sg;
 167         int entry;
 168 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
 169         const bool odp = umem->odp_data != NULL;
 170
 171         if (odp) {
 172                 WARN_ON(shift != 0);
 173                 WARN_ON(access_flags != (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE));
 174
 175                 for (i = 0; i < num_pages; ++i) {
 176                         dma_addr_t pa = umem->odp_data->dma_list[offset + i];
 177
 178                         pas[i] = cpu_to_be64(umem_dma_to_mtt(pa));
 179                 }
 180                 return;
 181         }
 182 #endif
 183
 184         i = 0;
 185         for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
 186                 len = sg_dma_len(sg) >> umem_page_shift;
 187                 base = sg_dma_address(sg);
 188
 189                 /* Skip elements below offset */
 190                 if (i + len < offset << shift) {
 191                         i += len;
 192                         continue;
 193                 }
 194
 195                 /* Skip pages below offset */
 196                 if (i < offset << shift) {
 197                         k = (offset << shift) - i;
 198                         i = offset << shift;
 199                 } else {
 200                         k = 0;
 201                 }
 202
 203                 for (; k < len; k++) {
 204                         if (!(i & mask)) {
 205                                 cur = base + (k << umem_page_shift);
 206                                 cur |= access_flags;
 207                                 idx = (i >> shift) - offset;
 208
 209                                 pas[idx] = cpu_to_be64(cur);
 210                                 mlx5_ib_dbg(dev, "pas[%d] 0x%llx\n",
 211                                             i >> shift, be64_to_cpu(pas[idx]));
 212                         }
 213                         i++;
 214
 215                         /* Stop after num_pages reached */
 216                         if (i >> shift >= offset + num_pages)
 217                                 return;
 218                 }
 219         }
 220 }
 221
 222 void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
 223                           int page_shift, __be64 *pas, int access_flags)
 224 {
 225         return __mlx5_ib_populate_pas(dev, umem, page_shift, 0,
 226                                       ib_umem_num_pages(umem), pas,
 227                                       access_flags);
 228 }
 229 int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset)
 230 {
 231         u64 page_size;
 232         u64 page_mask;
 233         u64 off_size;
 234         u64 off_mask;
 235         u64 buf_off;
 236
 237         page_size = (u64)1 << page_shift;
 238         page_mask = page_size - 1;
 239         buf_off = addr & page_mask;
 240         off_size = page_size >> 6;
 241         off_mask = off_size - 1;
 242
 243         if (buf_off & off_mask)
 244                 return -EINVAL;
 245
 246         *offset = buf_off >> ilog2(off_size);
 247         return 0;
 248 }