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/mm.h>
  37 #include <linux/bitmap.h>
  38 #include <linux/dma-mapping.h>
  39 #include <linux/vmalloc.h>
  40
  41 #include "mlx4.h"
  42
  43 u32 mlx4_bitmap_alloc(struct mlx4_bitmap *bitmap)
  44 {
  45         u32 obj;
  46
  47         spin_lock(&bitmap->lock);
  48
  49         obj = find_next_zero_bit(bitmap->table, bitmap->max, bitmap->last);
  50         if (obj >= bitmap->max) {
  51                 bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
  52                                 & bitmap->mask;
  53                 obj = find_first_zero_bit(bitmap->table, bitmap->max);
  54         }
  55
  56         if (obj < bitmap->max) {
  57                 set_bit(obj, bitmap->table);
  58                 bitmap->last = (obj + 1);
  59                 if (bitmap->last == bitmap->max)
  60                         bitmap->last = 0;
  61                 obj |= bitmap->top;
  62         } else
  63                 obj = -1;
  64
  65         if (obj != -1)
  66                 --bitmap->avail;
  67
  68         spin_unlock(&bitmap->lock);
  69
  70         return obj;
  71 }
  72
  73 void mlx4_bitmap_free(struct mlx4_bitmap *bitmap, u32 obj)
  74 {
  75         mlx4_bitmap_free_range(bitmap, obj, 1);
  76 }
  77
  78 u32 mlx4_bitmap_alloc_range(struct mlx4_bitmap *bitmap, int cnt, int align)
  79 {
  80         u32 obj;
  81
  82         if (likely(cnt == 1 && align == 1))
  83                 return mlx4_bitmap_alloc(bitmap);
  84
  85         spin_lock(&bitmap->lock);
  86
  87         obj = bitmap_find_next_zero_area(bitmap->table, bitmap->max,
  88                                 bitmap->last, cnt, align - 1);
  89         if (obj >= bitmap->max) {
  90                 bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
  91                                 & bitmap->mask;
  92                 obj = bitmap_find_next_zero_area(bitmap->table, bitmap->max,
  93                                                 0, cnt, align - 1);
  94         }
  95
  96         if (obj < bitmap->max) {
  97                 bitmap_set(bitmap->table, obj, cnt);
  98                 if (obj == bitmap->last) {
  99                         bitmap->last = (obj + cnt);
 100                         if (bitmap->last >= bitmap->max)
 101                                 bitmap->last = 0;
 102                 }
 103                 obj |= bitmap->top;
 104         } else
 105                 obj = -1;
 106
 107         if (obj != -1)
 108                 bitmap->avail -= cnt;
 109
 110         spin_unlock(&bitmap->lock);
 111
 112         return obj;
 113 }
 114
 115 u32 mlx4_bitmap_avail(struct mlx4_bitmap *bitmap)
 116 {
 117         return bitmap->avail;
 118 }
 119
 120 void mlx4_bitmap_free_range(struct mlx4_bitmap *bitmap, u32 obj, int cnt)
 121 {
 122         obj &= bitmap->max + bitmap->reserved_top - 1;
 123
 124         spin_lock(&bitmap->lock);
 125         bitmap_clear(bitmap->table, obj, cnt);
 126         bitmap->last = min(bitmap->last, obj);
 127         bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
 128                         & bitmap->mask;
 129         bitmap->avail += cnt;
 130         spin_unlock(&bitmap->lock);
 131 }
 132
 133 int mlx4_bitmap_init(struct mlx4_bitmap *bitmap, u32 num, u32 mask,
 134                      u32 reserved_bot, u32 reserved_top)
 135 {
 136         /* num must be a power of 2 */
 137         if (num != roundup_pow_of_two(num))
 138                 return -EINVAL;
 139
 140         bitmap->last = 0;
 141         bitmap->top  = 0;
 142         bitmap->max  = num - reserved_top;
 143         bitmap->mask = mask;
 144         bitmap->reserved_top = reserved_top;
 145         bitmap->avail = num - reserved_top - reserved_bot;
 146         spin_lock_init(&bitmap->lock);
 147         bitmap->table = kzalloc(BITS_TO_LONGS(bitmap->max) *
 148                                 sizeof (long), GFP_KERNEL);
 149         if (!bitmap->table)
 150                 return -ENOMEM;
 151
 152         bitmap_set(bitmap->table, 0, reserved_bot);
 153
 154         return 0;
 155 }
 156
 157 void mlx4_bitmap_cleanup(struct mlx4_bitmap *bitmap)
 158 {
 159         kfree(bitmap->table);
 160 }
 161
 162 /*
 163  * Handling for queue buffers -- we allocate a bunch of memory and
 164  * register it in a memory region at HCA virtual address 0.  If the
 165  * requested size is > max_direct, we split the allocation into
 166  * multiple pages, so we don't require too much contiguous memory.
 167  */
 168
 169 int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
 170                    struct mlx4_buf *buf)
 171 {
 172         dma_addr_t t;
 173
 174         if (size <= max_direct) {
 175                 buf->nbufs        = 1;
 176                 buf->npages       = 1;
 177                 buf->page_shift   = get_order(size) + PAGE_SHIFT;
 178                 buf->direct.buf   = dma_alloc_coherent(&dev->pdev->dev,
 179                                                        size, &t, GFP_KERNEL);
 180                 if (!buf->direct.buf)
 181                         return -ENOMEM;
 182
 183                 buf->direct.map = t;
 184
 185                 while (t & ((1 << buf->page_shift) - 1)) {
 186                         --buf->page_shift;
 187                         buf->npages *= 2;
 188                 }
 189
 190                 memset(buf->direct.buf, 0, size);
 191         } else {
 192                 int i;
 193
 194                 buf->direct.buf  = NULL;
 195                 buf->nbufs       = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 196                 buf->npages      = buf->nbufs;
 197                 buf->page_shift  = PAGE_SHIFT;
 198                 buf->page_list   = kcalloc(buf->nbufs, sizeof(*buf->page_list),
 199                                            GFP_KERNEL);
 200                 if (!buf->page_list)
 201                         return -ENOMEM;
 202
 203                 for (i = 0; i < buf->nbufs; ++i) {
 204                         buf->page_list[i].buf =
 205                                 dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
 206                                                    &t, GFP_KERNEL);
 207                         if (!buf->page_list[i].buf)
 208                                 goto err_free;
 209
 210                         buf->page_list[i].map = t;
 211
 212                         memset(buf->page_list[i].buf, 0, PAGE_SIZE);
 213                 }
 214
 215                 if (BITS_PER_LONG == 64) {
 216                         struct page **pages;
 217                         pages = kmalloc(sizeof *pages * buf->nbufs, GFP_KERNEL);
 218                         if (!pages)
 219                                 goto err_free;
 220                         for (i = 0; i < buf->nbufs; ++i)
 221                                 pages[i] = virt_to_page(buf->page_list[i].buf);
 222                         buf->direct.buf = vmap(pages, buf->nbufs, VM_MAP, PAGE_KERNEL);
 223                         kfree(pages);
 224                         if (!buf->direct.buf)
 225                                 goto err_free;
 226                 }
 227         }
 228
 229         return 0;
 230
 231 err_free:
 232         mlx4_buf_free(dev, size, buf);
 233
 234         return -ENOMEM;
 235 }
 236 EXPORT_SYMBOL_GPL(mlx4_buf_alloc);
 237
 238 void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
 239 {
 240         int i;
 241
 242         if (buf->nbufs == 1)
 243                 dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
 244                                   buf->direct.map);
 245         else {
 246                 if (BITS_PER_LONG == 64 && buf->direct.buf)
 247                         vunmap(buf->direct.buf);
 248
 249                 for (i = 0; i < buf->nbufs; ++i)
 250                         if (buf->page_list[i].buf)
 251                                 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
 252                                                   buf->page_list[i].buf,
 253                                                   buf->page_list[i].map);
 254                 kfree(buf->page_list);
 255         }
 256 }
 257 EXPORT_SYMBOL_GPL(mlx4_buf_free);
 258
 259 static struct mlx4_db_pgdir *mlx4_alloc_db_pgdir(struct device *dma_device)
 260 {
 261         struct mlx4_db_pgdir *pgdir;
 262
 263         pgdir = kzalloc(sizeof *pgdir, GFP_KERNEL);
 264         if (!pgdir)
 265                 return NULL;
 266
 267         bitmap_fill(pgdir->order1, MLX4_DB_PER_PAGE / 2);
 268         pgdir->bits[0] = pgdir->order0;
 269         pgdir->bits[1] = pgdir->order1;
 270         pgdir->db_page = dma_alloc_coherent(dma_device, PAGE_SIZE,
 271                                             &pgdir->db_dma, GFP_KERNEL);
 272         if (!pgdir->db_page) {
 273                 kfree(pgdir);
 274                 return NULL;
 275         }
 276
 277         return pgdir;
 278 }
 279
 280 static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir *pgdir,
 281                                     struct mlx4_db *db, int order)
 282 {
 283         int o;
 284         int i;
 285
 286         for (o = order; o <= 1; ++o) {
 287                 i = find_first_bit(pgdir->bits[o], MLX4_DB_PER_PAGE >> o);
 288                 if (i < MLX4_DB_PER_PAGE >> o)
 289                         goto found;
 290         }
 291
 292         return -ENOMEM;
 293
 294 found:
 295         clear_bit(i, pgdir->bits[o]);
 296
 297         i <<= o;
 298
 299         if (o > order)
 300                 set_bit(i ^ 1, pgdir->bits[order]);
 301
 302         db->u.pgdir = pgdir;
 303         db->index   = i;
 304         db->db      = pgdir->db_page + db->index;
 305         db->dma     = pgdir->db_dma  + db->index * 4;
 306         db->order   = order;
 307
 308         return 0;
 309 }
 310
 311 int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order)
 312 {
 313         struct mlx4_priv *priv = mlx4_priv(dev);
 314         struct mlx4_db_pgdir *pgdir;
 315         int ret = 0;
 316
 317         mutex_lock(&priv->pgdir_mutex);
 318
 319         list_for_each_entry(pgdir, &priv->pgdir_list, list)
 320                 if (!mlx4_alloc_db_from_pgdir(pgdir, db, order))
 321                         goto out;
 322
 323         pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev));
 324         if (!pgdir) {
 325                 ret = -ENOMEM;
 326                 goto out;
 327         }
 328
 329         list_add(&pgdir->list, &priv->pgdir_list);
 330
 331         /* This should never fail -- we just allocated an empty page: */
 332         WARN_ON(mlx4_alloc_db_from_pgdir(pgdir, db, order));
 333
 334 out:
 335         mutex_unlock(&priv->pgdir_mutex);
 336
 337         return ret;
 338 }
 339 EXPORT_SYMBOL_GPL(mlx4_db_alloc);
 340
 341 void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db)
 342 {
 343         struct mlx4_priv *priv = mlx4_priv(dev);
 344         int o;
 345         int i;
 346
 347         mutex_lock(&priv->pgdir_mutex);
 348
 349         o = db->order;
 350         i = db->index;
 351
 352         if (db->order == 0 && test_bit(i ^ 1, db->u.pgdir->order0)) {
 353                 clear_bit(i ^ 1, db->u.pgdir->order0);
 354                 ++o;
 355         }
 356         i >>= o;
 357         set_bit(i, db->u.pgdir->bits[o]);
 358
 359         if (bitmap_full(db->u.pgdir->order1, MLX4_DB_PER_PAGE / 2)) {
 360                 dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
 361                                   db->u.pgdir->db_page, db->u.pgdir->db_dma);
 362                 list_del(&db->u.pgdir->list);
 363                 kfree(db->u.pgdir);
 364         }
 365
 366         mutex_unlock(&priv->pgdir_mutex);
 367 }
 368 EXPORT_SYMBOL_GPL(mlx4_db_free);
 369
 370 int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
 371                        int size, int max_direct)
 372 {
 373         int err;
 374
 375         err = mlx4_db_alloc(dev, &wqres->db, 1);
 376         if (err)
 377                 return err;
 378
 379         *wqres->db.db = 0;
 380
 381         err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf);
 382         if (err)
 383                 goto err_db;
 384
 385         err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift,
 386                             &wqres->mtt);
 387         if (err)
 388                 goto err_buf;
 389
 390         err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf);
 391         if (err)
 392                 goto err_mtt;
 393
 394         return 0;
 395
 396 err_mtt:
 397         mlx4_mtt_cleanup(dev, &wqres->mtt);
 398 err_buf:
 399         mlx4_buf_free(dev, size, &wqres->buf);
 400 err_db:
 401         mlx4_db_free(dev, &wqres->db);
 402
 403         return err;
 404 }
 405 EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res);
 406
 407 void mlx4_free_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
 408                        int size)
 409 {
 410         mlx4_mtt_cleanup(dev, &wqres->mtt);
 411         mlx4_buf_free(dev, size, &wqres->buf);
 412         mlx4_db_free(dev, &wqres->db);
 413 }
 414 EXPORT_SYMBOL_GPL(mlx4_free_hwq_res);