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