drivers/net/mlx4/alloc.c

   1 /*
   2  * Copyright (c) 2006, 2007 Cisco Systems, Inc.  All rights reserved.
   3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <linux/errno.h>
  35 #include <linux/slab.h>
  36 #include <linux/bitmap.h>
  37 #include <linux/dma-mapping.h>
  38 #include <linux/vmalloc.h>
  39
  40 #include "mlx4.h"
  41
  42 u32 mlx4_bitmap_alloc(struct mlx4_bitmap *bitmap)
  43 {
  44         u32 obj;
  45
  46         spin_lock(&bitmap->lock);
  47
  48         obj = find_next_zero_bit(bitmap->table, bitmap->max, bitmap->last);
  49         if (obj >= bitmap->max) {
  50                 bitmap->top = (bitmap->top + bitmap->max) & bitmap->mask;
  51                 obj = find_first_zero_bit(bitmap->table, bitmap->max);
  52         }
  53
  54         if (obj < bitmap->max) {
  55                 set_bit(obj, bitmap->table);
  56                 bitmap->last = (obj + 1) & (bitmap->max - 1);
  57                 obj |= bitmap->top;
  58         } else
  59                 obj = -1;
  60
  61         spin_unlock(&bitmap->lock);
  62
  63         return obj;
  64 }
  65
  66 void mlx4_bitmap_free(struct mlx4_bitmap *bitmap, u32 obj)
  67 {
  68         obj &= bitmap->max - 1;
  69
  70         spin_lock(&bitmap->lock);
  71         clear_bit(obj, bitmap->table);
  72         bitmap->last = min(bitmap->last, obj);
  73         bitmap->top = (bitmap->top + bitmap->max) & bitmap->mask;
  74         spin_unlock(&bitmap->lock);
  75 }
  76
  77 int mlx4_bitmap_init(struct mlx4_bitmap *bitmap, u32 num, u32 mask, u32 reserved)
  78 {
  79         int i;
  80
  81         /* num must be a power of 2 */
  82         if (num != roundup_pow_of_two(num))
  83                 return -EINVAL;
  84
  85         bitmap->last = 0;
  86         bitmap->top  = 0;
  87         bitmap->max  = num;
  88         bitmap->mask = mask;
  89         spin_lock_init(&bitmap->lock);
  90         bitmap->table = kzalloc(BITS_TO_LONGS(num) * sizeof (long), GFP_KERNEL);
  91         if (!bitmap->table)
  92                 return -ENOMEM;
  93
  94         for (i = 0; i < reserved; ++i)
  95                 set_bit(i, bitmap->table);
  96
  97         return 0;
  98 }
  99
 100 void mlx4_bitmap_cleanup(struct mlx4_bitmap *bitmap)
 101 {
 102         kfree(bitmap->table);
 103 }
 104
 105 /*
 106  * Handling for queue buffers -- we allocate a bunch of memory and
 107  * register it in a memory region at HCA virtual address 0.  If the
 108  * requested size is > max_direct, we split the allocation into
 109  * multiple pages, so we don't require too much contiguous memory.
 110  */
 111
 112 int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
 113                    struct mlx4_buf *buf)
 114 {
 115         dma_addr_t t;
 116
 117         if (size <= max_direct) {
 118                 buf->nbufs        = 1;
 119                 buf->npages       = 1;
 120                 buf->page_shift   = get_order(size) + PAGE_SHIFT;
 121                 buf->direct.buf   = dma_alloc_coherent(&dev->pdev->dev,
 122                                                        size, &t, GFP_KERNEL);
 123                 if (!buf->direct.buf)
 124                         return -ENOMEM;
 125
 126                 buf->direct.map = t;
 127
 128                 while (t & ((1 << buf->page_shift) - 1)) {
 129                         --buf->page_shift;
 130                         buf->npages *= 2;
 131                 }
 132
 133                 memset(buf->direct.buf, 0, size);
 134         } else {
 135                 int i;
 136
 137                 buf->nbufs       = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 138                 buf->npages      = buf->nbufs;
 139                 buf->page_shift  = PAGE_SHIFT;
 140                 buf->page_list   = kzalloc(buf->nbufs * sizeof *buf->page_list,
 141                                            GFP_KERNEL);
 142                 if (!buf->page_list)
 143                         return -ENOMEM;
 144
 145                 for (i = 0; i < buf->nbufs; ++i) {
 146                         buf->page_list[i].buf =
 147                                 dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
 148                                                    &t, GFP_KERNEL);
 149                         if (!buf->page_list[i].buf)
 150                                 goto err_free;
 151
 152                         buf->page_list[i].map = t;
 153
 154                         memset(buf->page_list[i].buf, 0, PAGE_SIZE);
 155                 }
 156
 157                 if (BITS_PER_LONG == 64) {
 158                         struct page **pages;
 159                         pages = kmalloc(sizeof *pages * buf->nbufs, GFP_KERNEL);
 160                         if (!pages)
 161                                 goto err_free;
 162                         for (i = 0; i < buf->nbufs; ++i)
 163                                 pages[i] = virt_to_page(buf->page_list[i].buf);
 164                         buf->direct.buf = vmap(pages, buf->nbufs, VM_MAP, PAGE_KERNEL);
 165                         kfree(pages);
 166                         if (!buf->direct.buf)
 167                                 goto err_free;
 168                 }
 169         }
 170
 171         return 0;
 172
 173 err_free:
 174         mlx4_buf_free(dev, size, buf);
 175
 176         return -ENOMEM;
 177 }
 178 EXPORT_SYMBOL_GPL(mlx4_buf_alloc);
 179
 180 void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
 181 {
 182         int i;
 183
 184         if (buf->nbufs == 1)
 185                 dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
 186                                   buf->direct.map);
 187         else {
 188                 if (BITS_PER_LONG == 64)
 189                         vunmap(buf->direct.buf);
 190
 191                 for (i = 0; i < buf->nbufs; ++i)
 192                         if (buf->page_list[i].buf)
 193                                 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
 194                                                   buf->page_list[i].buf,
 195                                                   buf->page_list[i].map);
 196                 kfree(buf->page_list);
 197         }
 198 }
 199 EXPORT_SYMBOL_GPL(mlx4_buf_free);
 200
 201 static struct mlx4_db_pgdir *mlx4_alloc_db_pgdir(struct device *dma_device)
 202 {
 203         struct mlx4_db_pgdir *pgdir;
 204
 205         pgdir = kzalloc(sizeof *pgdir, GFP_KERNEL);
 206         if (!pgdir)
 207                 return NULL;
 208
 209         bitmap_fill(pgdir->order1, MLX4_DB_PER_PAGE / 2);
 210         pgdir->bits[0] = pgdir->order0;
 211         pgdir->bits[1] = pgdir->order1;
 212         pgdir->db_page = dma_alloc_coherent(dma_device, PAGE_SIZE,
 213                                             &pgdir->db_dma, GFP_KERNEL);
 214         if (!pgdir->db_page) {
 215                 kfree(pgdir);
 216                 return NULL;
 217         }
 218
 219         return pgdir;
 220 }
 221
 222 static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir *pgdir,
 223                                     struct mlx4_db *db, int order)
 224 {
 225         int o;
 226         int i;
 227
 228         for (o = order; o <= 1; ++o) {
 229                 i = find_first_bit(pgdir->bits[o], MLX4_DB_PER_PAGE >> o);
 230                 if (i < MLX4_DB_PER_PAGE >> o)
 231                         goto found;
 232         }
 233
 234         return -ENOMEM;
 235
 236 found:
 237         clear_bit(i, pgdir->bits[o]);
 238
 239         i <<= o;
 240
 241         if (o > order)
 242                 set_bit(i ^ 1, pgdir->bits[order]);
 243
 244         db->u.pgdir = pgdir;
 245         db->index   = i;
 246         db->db      = pgdir->db_page + db->index;
 247         db->dma     = pgdir->db_dma  + db->index * 4;
 248         db->order   = order;
 249
 250         return 0;
 251 }
 252
 253 int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order)
 254 {
 255         struct mlx4_priv *priv = mlx4_priv(dev);
 256         struct mlx4_db_pgdir *pgdir;
 257         int ret = 0;
 258
 259         mutex_lock(&priv->pgdir_mutex);
 260
 261         list_for_each_entry(pgdir, &priv->pgdir_list, list)
 262                 if (!mlx4_alloc_db_from_pgdir(pgdir, db, order))
 263                         goto out;
 264
 265         pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev));
 266         if (!pgdir) {
 267                 ret = -ENOMEM;
 268                 goto out;
 269         }
 270
 271         list_add(&pgdir->list, &priv->pgdir_list);
 272
 273         /* This should never fail -- we just allocated an empty page: */
 274         WARN_ON(mlx4_alloc_db_from_pgdir(pgdir, db, order));
 275
 276 out:
 277         mutex_unlock(&priv->pgdir_mutex);
 278
 279         return ret;
 280 }
 281 EXPORT_SYMBOL_GPL(mlx4_db_alloc);
 282
 283 void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db)
 284 {
 285         struct mlx4_priv *priv = mlx4_priv(dev);
 286         int o;
 287         int i;
 288
 289         mutex_lock(&priv->pgdir_mutex);
 290
 291         o = db->order;
 292         i = db->index;
 293
 294         if (db->order == 0 && test_bit(i ^ 1, db->u.pgdir->order0)) {
 295                 clear_bit(i ^ 1, db->u.pgdir->order0);
 296                 ++o;
 297         }
 298         i >>= o;
 299         set_bit(i, db->u.pgdir->bits[o]);
 300
 301         if (bitmap_full(db->u.pgdir->order1, MLX4_DB_PER_PAGE / 2)) {
 302                 dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
 303                                   db->u.pgdir->db_page, db->u.pgdir->db_dma);
 304                 list_del(&db->u.pgdir->list);
 305                 kfree(db->u.pgdir);
 306         }
 307
 308         mutex_unlock(&priv->pgdir_mutex);
 309 }
 310 EXPORT_SYMBOL_GPL(mlx4_db_free);
 311
 312 int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
 313                        int size, int max_direct)
 314 {
 315         int err;
 316
 317         err = mlx4_db_alloc(dev, &wqres->db, 1);
 318         if (err)
 319                 return err;
 320
 321         *wqres->db.db = 0;
 322
 323         err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf);
 324         if (err)
 325                 goto err_db;
 326
 327         err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift,
 328                             &wqres->mtt);
 329         if (err)
 330                 goto err_buf;
 331
 332         err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf);
 333         if (err)
 334                 goto err_mtt;
 335
 336         return 0;
 337
 338 err_mtt:
 339         mlx4_mtt_cleanup(dev, &wqres->mtt);
 340 err_buf:
 341         mlx4_buf_free(dev, size, &wqres->buf);
 342 err_db:
 343         mlx4_db_free(dev, &wqres->db);
 344
 345         return err;
 346 }
 347 EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res);
 348
 349 void mlx4_free_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
 350                        int size)
 351 {
 352         mlx4_mtt_cleanup(dev, &wqres->mtt);
 353         mlx4_buf_free(dev, size, &wqres->buf);
 354         mlx4_db_free(dev, &wqres->db);
 355 }
 356 EXPORT_SYMBOL_GPL(mlx4_free_hwq_res);