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 * $Id: mthca_allocator.c 1349 2004-12-16 21:09:43Z roland $ 33 */ 34 35 #include <linux/errno.h> 36 #include <linux/slab.h> 37 #include <linux/bitmap.h> 38 39 #include "mthca_dev.h" 40 41 /* Trivial bitmap-based allocator */ 42 u32 mthca_alloc(struct mthca_alloc *alloc) 43 { 44 u32 obj; 45 46 spin_lock(&alloc->lock); 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(&alloc->lock); 60 61 return obj; 62 } 63 64 void mthca_free(struct mthca_alloc *alloc, u32 obj) 65 { 66 obj &= alloc->max - 1; 67 spin_lock(&alloc->lock); 68 clear_bit(obj, alloc->table); 69 alloc->last = min(alloc->last, obj); 70 alloc->top = (alloc->top + alloc->max) & alloc->mask; 71 spin_unlock(&alloc->lock); 72 } 73 74 int mthca_alloc_init(struct mthca_alloc *alloc, u32 num, u32 mask, 75 u32 reserved) 76 { 77 int i; 78 79 /* num must be a power of 2 */ 80 if (num != 1 << (ffs(num) - 1)) 81 return -EINVAL; 82 83 alloc->last = 0; 84 alloc->top = 0; 85 alloc->max = num; 86 alloc->mask = mask; 87 spin_lock_init(&alloc->lock); 88 alloc->table = kmalloc(BITS_TO_LONGS(num) * sizeof (long), 89 GFP_KERNEL); 90 if (!alloc->table) 91 return -ENOMEM; 92 93 bitmap_zero(alloc->table, num); 94 for (i = 0; i < reserved; ++i) 95 set_bit(i, alloc->table); 96 97 return 0; 98 } 99 100 void mthca_alloc_cleanup(struct mthca_alloc *alloc) 101 { 102 kfree(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 void *mthca_array_get(struct mthca_array *array, int index) 112 { 113 int p = (index * sizeof (void *)) >> PAGE_SHIFT; 114 115 if (array->page_list[p].page) { 116 int i = index & (PAGE_SIZE / sizeof (void *) - 1); 117 return array->page_list[p].page[i]; 118 } else 119 return NULL; 120 } 121 122 int mthca_array_set(struct mthca_array *array, int index, void *value) 123 { 124 int p = (index * sizeof (void *)) >> PAGE_SHIFT; 125 126 /* Allocate with GFP_ATOMIC because we'll be called with locks held. */ 127 if (!array->page_list[p].page) 128 array->page_list[p].page = (void **) get_zeroed_page(GFP_ATOMIC); 129 130 if (!array->page_list[p].page) 131 return -ENOMEM; 132 133 array->page_list[p].page[index & (PAGE_SIZE / sizeof (void *) - 1)] = 134 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 } 148 149 if (array->page_list[p].used < 0) 150 pr_debug("Array %p index %d page %d with ref count %d < 0\n", 151 array, index, p, array->page_list[p].used); 152 } 153 154 int mthca_array_init(struct mthca_array *array, int nent) 155 { 156 int npage = (nent * sizeof (void *) + PAGE_SIZE - 1) / PAGE_SIZE; 157 int i; 158 159 array->page_list = kmalloc(npage * sizeof *array->page_list, GFP_KERNEL); 160 if (!array->page_list) 161 return -ENOMEM; 162 163 for (i = 0; i < npage; ++i) { 164 array->page_list[i].page = NULL; 165 array->page_list[i].used = 0; 166 } 167 168 return 0; 169 } 170 171 void mthca_array_cleanup(struct mthca_array *array, int nent) 172 { 173 int i; 174 175 for (i = 0; i < (nent * sizeof (void *) + PAGE_SIZE - 1) / PAGE_SIZE; ++i) 176 free_page((unsigned long) array->page_list[i].page); 177 178 kfree(array->page_list); 179 } 180 181 /* 182 * Handling for queue buffers -- we allocate a bunch of memory and 183 * register it in a memory region at HCA virtual address 0. If the 184 * requested size is > max_direct, we split the allocation into 185 * multiple pages, so we don't require too much contiguous memory. 186 */ 187 188 int mthca_buf_alloc(struct mthca_dev *dev, int size, int max_direct, 189 union mthca_buf *buf, int *is_direct, struct mthca_pd *pd, 190 int hca_write, struct mthca_mr *mr) 191 { 192 int err = -ENOMEM; 193 int npages, shift; 194 u64 *dma_list = NULL; 195 dma_addr_t t; 196 int i; 197 198 if (size <= max_direct) { 199 *is_direct = 1; 200 npages = 1; 201 shift = get_order(size) + PAGE_SHIFT; 202 203 buf->direct.buf = dma_alloc_coherent(&dev->pdev->dev, 204 size, &t, GFP_KERNEL); 205 if (!buf->direct.buf) 206 return -ENOMEM; 207 208 pci_unmap_addr_set(&buf->direct, mapping, t); 209 210 memset(buf->direct.buf, 0, size); 211 212 while (t & ((1 << shift) - 1)) { 213 --shift; 214 npages *= 2; 215 } 216 217 dma_list = kmalloc(npages * sizeof *dma_list, 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(npages * sizeof *dma_list, GFP_KERNEL); 229 if (!dma_list) 230 return -ENOMEM; 231 232 buf->page_list = kmalloc(npages * sizeof *buf->page_list, 233 GFP_KERNEL); 234 if (!buf->page_list) 235 goto err_out; 236 237 for (i = 0; i < npages; ++i) 238 buf->page_list[i].buf = NULL; 239 240 for (i = 0; i < npages; ++i) { 241 buf->page_list[i].buf = 242 dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE, 243 &t, GFP_KERNEL); 244 if (!buf->page_list[i].buf) 245 goto err_free; 246 247 dma_list[i] = t; 248 pci_unmap_addr_set(&buf->page_list[i], mapping, t); 249 250 memset(buf->page_list[i].buf, 0, PAGE_SIZE); 251 } 252 } 253 254 err = mthca_mr_alloc_phys(dev, pd->pd_num, 255 dma_list, shift, npages, 256 0, size, 257 MTHCA_MPT_FLAG_LOCAL_READ | 258 (hca_write ? MTHCA_MPT_FLAG_LOCAL_WRITE : 0), 259 mr); 260 if (err) 261 goto err_free; 262 263 kfree(dma_list); 264 265 return 0; 266 267 err_free: 268 mthca_buf_free(dev, size, buf, *is_direct, NULL); 269 270 err_out: 271 kfree(dma_list); 272 273 return err; 274 } 275 276 void mthca_buf_free(struct mthca_dev *dev, int size, union mthca_buf *buf, 277 int is_direct, struct mthca_mr *mr) 278 { 279 int i; 280 281 if (mr) 282 mthca_free_mr(dev, mr); 283 284 if (is_direct) 285 dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf, 286 pci_unmap_addr(&buf->direct, mapping)); 287 else { 288 for (i = 0; i < (size + PAGE_SIZE - 1) / PAGE_SIZE; ++i) 289 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE, 290 buf->page_list[i].buf, 291 pci_unmap_addr(&buf->page_list[i], 292 mapping)); 293 kfree(buf->page_list); 294 } 295 } 296