1 /* 2 * Copyright (c) 2004 Topspin Communications. 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/errno.h> 34 #include <linux/slab.h> 35 #include <linux/bitmap.h> 36 37 #include "mthca_dev.h" 38 39 /* Trivial bitmap-based allocator */ 40 u32 mthca_alloc(struct mthca_alloc *alloc) 41 { 42 unsigned long flags; 43 u32 obj; 44 45 spin_lock_irqsave(&alloc->lock, flags); 46 47 obj = find_next_zero_bit(alloc->table, alloc->max, alloc->last); 48 if (obj >= alloc->max) { 49 alloc->top = (alloc->top + alloc->max) & alloc->mask; 50 obj = find_first_zero_bit(alloc->table, alloc->max); 51 } 52 53 if (obj < alloc->max) { 54 __set_bit(obj, alloc->table); 55 obj |= alloc->top; 56 } else 57 obj = -1; 58 59 spin_unlock_irqrestore(&alloc->lock, flags); 60 61 return obj; 62 } 63 64 void mthca_free(struct mthca_alloc *alloc, u32 obj) 65 { 66 unsigned long flags; 67 68 obj &= alloc->max - 1; 69 70 spin_lock_irqsave(&alloc->lock, flags); 71 72 __clear_bit(obj, alloc->table); 73 alloc->last = min(alloc->last, obj); 74 alloc->top = (alloc->top + alloc->max) & alloc->mask; 75 76 spin_unlock_irqrestore(&alloc->lock, flags); 77 } 78 79 int mthca_alloc_init(struct mthca_alloc *alloc, u32 num, u32 mask, 80 u32 reserved) 81 { 82 /* num must be a power of 2 */ 83 if (num != 1 << (ffs(num) - 1)) 84 return -EINVAL; 85 86 alloc->last = 0; 87 alloc->top = 0; 88 alloc->max = num; 89 alloc->mask = mask; 90 spin_lock_init(&alloc->lock); 91 alloc->table = bitmap_zalloc(num, GFP_KERNEL); 92 if (!alloc->table) 93 return -ENOMEM; 94 95 bitmap_set(alloc->table, 0, reserved); 96 97 return 0; 98 } 99 100 void mthca_alloc_cleanup(struct mthca_alloc *alloc) 101 { 102 bitmap_free(alloc->table); 103 } 104 105 /* 106 * Array of pointers with lazy allocation of leaf pages. Callers of 107 * _get, _set and _clear methods must use a lock or otherwise 108 * serialize access to the array. 109 */ 110 111 #define MTHCA_ARRAY_MASK (PAGE_SIZE / sizeof (void *) - 1) 112 113 void *mthca_array_get(struct mthca_array *array, int index) 114 { 115 int p = (index * sizeof (void *)) >> PAGE_SHIFT; 116 117 if (array->page_list[p].page) 118 return array->page_list[p].page[index & MTHCA_ARRAY_MASK]; 119 else 120 return NULL; 121 } 122 123 int mthca_array_set(struct mthca_array *array, int index, void *value) 124 { 125 int p = (index * sizeof (void *)) >> PAGE_SHIFT; 126 127 /* Allocate with GFP_ATOMIC because we'll be called with locks held. */ 128 if (!array->page_list[p].page) 129 array->page_list[p].page = (void **) get_zeroed_page(GFP_ATOMIC); 130 131 if (!array->page_list[p].page) 132 return -ENOMEM; 133 134 array->page_list[p].page[index & MTHCA_ARRAY_MASK] = value; 135 ++array->page_list[p].used; 136 137 return 0; 138 } 139 140 void mthca_array_clear(struct mthca_array *array, int index) 141 { 142 int p = (index * sizeof (void *)) >> PAGE_SHIFT; 143 144 if (--array->page_list[p].used == 0) { 145 free_page((unsigned long) array->page_list[p].page); 146 array->page_list[p].page = NULL; 147 } else 148 array->page_list[p].page[index & MTHCA_ARRAY_MASK] = NULL; 149 150 if (array->page_list[p].used < 0) 151 pr_debug("Array %p index %d page %d with ref count %d < 0\n", 152 array, index, p, array->page_list[p].used); 153 } 154 155 int mthca_array_init(struct mthca_array *array, int nent) 156 { 157 int npage = (nent * sizeof (void *) + PAGE_SIZE - 1) / PAGE_SIZE; 158 int i; 159 160 array->page_list = kmalloc_objs(*array->page_list, npage); 161 if (!array->page_list) 162 return -ENOMEM; 163 164 for (i = 0; i < npage; ++i) { 165 array->page_list[i].page = NULL; 166 array->page_list[i].used = 0; 167 } 168 169 return 0; 170 } 171 172 void mthca_array_cleanup(struct mthca_array *array, int nent) 173 { 174 int i; 175 176 for (i = 0; i < (nent * sizeof (void *) + PAGE_SIZE - 1) / PAGE_SIZE; ++i) 177 free_page((unsigned long) array->page_list[i].page); 178 179 kfree(array->page_list); 180 } 181 182 /* 183 * Handling for queue buffers -- we allocate a bunch of memory and 184 * register it in a memory region at HCA virtual address 0. If the 185 * requested size is > max_direct, we split the allocation into 186 * multiple pages, so we don't require too much contiguous memory. 187 */ 188 189 int mthca_buf_alloc(struct mthca_dev *dev, int size, int max_direct, 190 union mthca_buf *buf, int *is_direct, struct mthca_pd *pd, 191 int hca_write, struct mthca_mr *mr) 192 { 193 int err = -ENOMEM; 194 int npages, shift; 195 u64 *dma_list = NULL; 196 dma_addr_t t; 197 int i; 198 199 if (size <= max_direct) { 200 *is_direct = 1; 201 npages = 1; 202 shift = get_order(size) + PAGE_SHIFT; 203 204 buf->direct.buf = dma_alloc_coherent(&dev->pdev->dev, 205 size, &t, GFP_KERNEL); 206 if (!buf->direct.buf) 207 return -ENOMEM; 208 209 dma_unmap_addr_set(&buf->direct, mapping, t); 210 211 while (t & ((1 << shift) - 1)) { 212 --shift; 213 npages *= 2; 214 } 215 216 dma_list = kmalloc_array(npages, sizeof(*dma_list), 217 GFP_KERNEL); 218 if (!dma_list) 219 goto err_free; 220 221 for (i = 0; i < npages; ++i) 222 dma_list[i] = t + i * (1 << shift); 223 } else { 224 *is_direct = 0; 225 npages = (size + PAGE_SIZE - 1) / PAGE_SIZE; 226 shift = PAGE_SHIFT; 227 228 dma_list = kmalloc_array(npages, sizeof(*dma_list), 229 GFP_KERNEL); 230 if (!dma_list) 231 return -ENOMEM; 232 233 buf->page_list = kmalloc_objs(*buf->page_list, npages, 234 GFP_KERNEL); 235 if (!buf->page_list) 236 goto err_out; 237 238 for (i = 0; i < npages; ++i) 239 buf->page_list[i].buf = NULL; 240 241 for (i = 0; i < npages; ++i) { 242 buf->page_list[i].buf = 243 dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE, 244 &t, GFP_KERNEL); 245 if (!buf->page_list[i].buf) 246 goto err_free; 247 248 dma_list[i] = t; 249 dma_unmap_addr_set(&buf->page_list[i], mapping, t); 250 251 clear_page(buf->page_list[i].buf); 252 } 253 } 254 255 err = mthca_mr_alloc_phys(dev, pd->pd_num, 256 dma_list, shift, npages, 257 0, size, 258 MTHCA_MPT_FLAG_LOCAL_READ | 259 (hca_write ? MTHCA_MPT_FLAG_LOCAL_WRITE : 0), 260 mr); 261 if (err) 262 goto err_free; 263 264 kfree(dma_list); 265 266 return 0; 267 268 err_free: 269 mthca_buf_free(dev, size, buf, *is_direct, NULL); 270 271 err_out: 272 kfree(dma_list); 273 274 return err; 275 } 276 277 void mthca_buf_free(struct mthca_dev *dev, int size, union mthca_buf *buf, 278 int is_direct, struct mthca_mr *mr) 279 { 280 int i; 281 282 if (mr) 283 mthca_free_mr(dev, mr); 284 285 if (is_direct) 286 dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf, 287 dma_unmap_addr(&buf->direct, mapping)); 288 else { 289 for (i = 0; i < (size + PAGE_SIZE - 1) / PAGE_SIZE; ++i) 290 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE, 291 buf->page_list[i].buf, 292 dma_unmap_addr(&buf->page_list[i], 293 mapping)); 294 kfree(buf->page_list); 295 } 296 } 297