| 1 | /* |
| 2 | * Copyright 2008 Advanced Micro Devices, Inc. |
| 3 | * Copyright 2008 Red Hat Inc. |
| 4 | * Copyright 2009 Jerome Glisse. |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a |
| 7 | * copy of this software and associated documentation files (the "Software"), |
| 8 | * to deal in the Software without restriction, including without limitation |
| 9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 10 | * and/or sell copies of the Software, and to permit persons to whom the |
| 11 | * Software is furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in |
| 14 | * all copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 19 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
| 20 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
| 21 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
| 22 | * OTHER DEALINGS IN THE SOFTWARE. |
| 23 | * |
| 24 | * Authors: Dave Airlie |
| 25 | * Alex Deucher |
| 26 | * Jerome Glisse |
| 27 | */ |
| 28 | #include <drm/drmP.h> |
| 29 | #include <drm/radeon_drm.h> |
| 30 | #include "radeon.h" |
| 31 | |
| 32 | /* |
| 33 | * GART |
| 34 | * The GART (Graphics Aperture Remapping Table) is an aperture |
| 35 | * in the GPU's address space. System pages can be mapped into |
| 36 | * the aperture and look like contiguous pages from the GPU's |
| 37 | * perspective. A page table maps the pages in the aperture |
| 38 | * to the actual backing pages in system memory. |
| 39 | * |
| 40 | * Radeon GPUs support both an internal GART, as described above, |
| 41 | * and AGP. AGP works similarly, but the GART table is configured |
| 42 | * and maintained by the northbridge rather than the driver. |
| 43 | * Radeon hw has a separate AGP aperture that is programmed to |
| 44 | * point to the AGP aperture provided by the northbridge and the |
| 45 | * requests are passed through to the northbridge aperture. |
| 46 | * Both AGP and internal GART can be used at the same time, however |
| 47 | * that is not currently supported by the driver. |
| 48 | * |
| 49 | * This file handles the common internal GART management. |
| 50 | */ |
| 51 | |
| 52 | /* |
| 53 | * Common GART table functions. |
| 54 | */ |
| 55 | /** |
| 56 | * radeon_gart_table_ram_alloc - allocate system ram for gart page table |
| 57 | * |
| 58 | * @rdev: radeon_device pointer |
| 59 | * |
| 60 | * Allocate system memory for GART page table |
| 61 | * (r1xx-r3xx, non-pcie r4xx, rs400). These asics require the |
| 62 | * gart table to be in system memory. |
| 63 | * Returns 0 for success, -ENOMEM for failure. |
| 64 | */ |
| 65 | int radeon_gart_table_ram_alloc(struct radeon_device *rdev) |
| 66 | { |
| 67 | void *ptr; |
| 68 | |
| 69 | ptr = pci_alloc_consistent(rdev->pdev, rdev->gart.table_size, |
| 70 | &rdev->gart.table_addr); |
| 71 | if (ptr == NULL) { |
| 72 | return -ENOMEM; |
| 73 | } |
| 74 | #ifdef CONFIG_X86 |
| 75 | if (rdev->family == CHIP_RS400 || rdev->family == CHIP_RS480 || |
| 76 | rdev->family == CHIP_RS690 || rdev->family == CHIP_RS740) { |
| 77 | set_memory_uc((unsigned long)ptr, |
| 78 | rdev->gart.table_size >> PAGE_SHIFT); |
| 79 | } |
| 80 | #endif |
| 81 | rdev->gart.ptr = ptr; |
| 82 | memset((void *)rdev->gart.ptr, 0, rdev->gart.table_size); |
| 83 | return 0; |
| 84 | } |
| 85 | |
| 86 | /** |
| 87 | * radeon_gart_table_ram_free - free system ram for gart page table |
| 88 | * |
| 89 | * @rdev: radeon_device pointer |
| 90 | * |
| 91 | * Free system memory for GART page table |
| 92 | * (r1xx-r3xx, non-pcie r4xx, rs400). These asics require the |
| 93 | * gart table to be in system memory. |
| 94 | */ |
| 95 | void radeon_gart_table_ram_free(struct radeon_device *rdev) |
| 96 | { |
| 97 | if (rdev->gart.ptr == NULL) { |
| 98 | return; |
| 99 | } |
| 100 | #ifdef CONFIG_X86 |
| 101 | if (rdev->family == CHIP_RS400 || rdev->family == CHIP_RS480 || |
| 102 | rdev->family == CHIP_RS690 || rdev->family == CHIP_RS740) { |
| 103 | set_memory_wb((unsigned long)rdev->gart.ptr, |
| 104 | rdev->gart.table_size >> PAGE_SHIFT); |
| 105 | } |
| 106 | #endif |
| 107 | pci_free_consistent(rdev->pdev, rdev->gart.table_size, |
| 108 | (void *)rdev->gart.ptr, |
| 109 | rdev->gart.table_addr); |
| 110 | rdev->gart.ptr = NULL; |
| 111 | rdev->gart.table_addr = 0; |
| 112 | } |
| 113 | |
| 114 | /** |
| 115 | * radeon_gart_table_vram_alloc - allocate vram for gart page table |
| 116 | * |
| 117 | * @rdev: radeon_device pointer |
| 118 | * |
| 119 | * Allocate video memory for GART page table |
| 120 | * (pcie r4xx, r5xx+). These asics require the |
| 121 | * gart table to be in video memory. |
| 122 | * Returns 0 for success, error for failure. |
| 123 | */ |
| 124 | int radeon_gart_table_vram_alloc(struct radeon_device *rdev) |
| 125 | { |
| 126 | int r; |
| 127 | |
| 128 | if (rdev->gart.robj == NULL) { |
| 129 | r = radeon_bo_create(rdev, rdev->gart.table_size, |
| 130 | PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM, |
| 131 | NULL, &rdev->gart.robj); |
| 132 | if (r) { |
| 133 | return r; |
| 134 | } |
| 135 | } |
| 136 | return 0; |
| 137 | } |
| 138 | |
| 139 | /** |
| 140 | * radeon_gart_table_vram_pin - pin gart page table in vram |
| 141 | * |
| 142 | * @rdev: radeon_device pointer |
| 143 | * |
| 144 | * Pin the GART page table in vram so it will not be moved |
| 145 | * by the memory manager (pcie r4xx, r5xx+). These asics require the |
| 146 | * gart table to be in video memory. |
| 147 | * Returns 0 for success, error for failure. |
| 148 | */ |
| 149 | int radeon_gart_table_vram_pin(struct radeon_device *rdev) |
| 150 | { |
| 151 | uint64_t gpu_addr; |
| 152 | int r; |
| 153 | |
| 154 | r = radeon_bo_reserve(rdev->gart.robj, false); |
| 155 | if (unlikely(r != 0)) |
| 156 | return r; |
| 157 | r = radeon_bo_pin(rdev->gart.robj, |
| 158 | RADEON_GEM_DOMAIN_VRAM, &gpu_addr); |
| 159 | if (r) { |
| 160 | radeon_bo_unreserve(rdev->gart.robj); |
| 161 | return r; |
| 162 | } |
| 163 | r = radeon_bo_kmap(rdev->gart.robj, &rdev->gart.ptr); |
| 164 | if (r) |
| 165 | radeon_bo_unpin(rdev->gart.robj); |
| 166 | radeon_bo_unreserve(rdev->gart.robj); |
| 167 | rdev->gart.table_addr = gpu_addr; |
| 168 | return r; |
| 169 | } |
| 170 | |
| 171 | /** |
| 172 | * radeon_gart_table_vram_unpin - unpin gart page table in vram |
| 173 | * |
| 174 | * @rdev: radeon_device pointer |
| 175 | * |
| 176 | * Unpin the GART page table in vram (pcie r4xx, r5xx+). |
| 177 | * These asics require the gart table to be in video memory. |
| 178 | */ |
| 179 | void radeon_gart_table_vram_unpin(struct radeon_device *rdev) |
| 180 | { |
| 181 | int r; |
| 182 | |
| 183 | if (rdev->gart.robj == NULL) { |
| 184 | return; |
| 185 | } |
| 186 | r = radeon_bo_reserve(rdev->gart.robj, false); |
| 187 | if (likely(r == 0)) { |
| 188 | radeon_bo_kunmap(rdev->gart.robj); |
| 189 | radeon_bo_unpin(rdev->gart.robj); |
| 190 | radeon_bo_unreserve(rdev->gart.robj); |
| 191 | rdev->gart.ptr = NULL; |
| 192 | } |
| 193 | } |
| 194 | |
| 195 | /** |
| 196 | * radeon_gart_table_vram_free - free gart page table vram |
| 197 | * |
| 198 | * @rdev: radeon_device pointer |
| 199 | * |
| 200 | * Free the video memory used for the GART page table |
| 201 | * (pcie r4xx, r5xx+). These asics require the gart table to |
| 202 | * be in video memory. |
| 203 | */ |
| 204 | void radeon_gart_table_vram_free(struct radeon_device *rdev) |
| 205 | { |
| 206 | if (rdev->gart.robj == NULL) { |
| 207 | return; |
| 208 | } |
| 209 | radeon_bo_unref(&rdev->gart.robj); |
| 210 | } |
| 211 | |
| 212 | /* |
| 213 | * Common gart functions. |
| 214 | */ |
| 215 | /** |
| 216 | * radeon_gart_unbind - unbind pages from the gart page table |
| 217 | * |
| 218 | * @rdev: radeon_device pointer |
| 219 | * @offset: offset into the GPU's gart aperture |
| 220 | * @pages: number of pages to unbind |
| 221 | * |
| 222 | * Unbinds the requested pages from the gart page table and |
| 223 | * replaces them with the dummy page (all asics). |
| 224 | */ |
| 225 | void radeon_gart_unbind(struct radeon_device *rdev, unsigned offset, |
| 226 | int pages) |
| 227 | { |
| 228 | unsigned t; |
| 229 | unsigned p; |
| 230 | int i, j; |
| 231 | u64 page_base; |
| 232 | |
| 233 | if (!rdev->gart.ready) { |
| 234 | WARN(1, "trying to unbind memory from uninitialized GART !\n"); |
| 235 | return; |
| 236 | } |
| 237 | t = offset / RADEON_GPU_PAGE_SIZE; |
| 238 | p = t / (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); |
| 239 | for (i = 0; i < pages; i++, p++) { |
| 240 | if (rdev->gart.pages[p]) { |
| 241 | rdev->gart.pages[p] = NULL; |
| 242 | rdev->gart.pages_addr[p] = rdev->dummy_page.addr; |
| 243 | page_base = rdev->gart.pages_addr[p]; |
| 244 | for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) { |
| 245 | if (rdev->gart.ptr) { |
| 246 | radeon_gart_set_page(rdev, t, page_base); |
| 247 | } |
| 248 | page_base += RADEON_GPU_PAGE_SIZE; |
| 249 | } |
| 250 | } |
| 251 | } |
| 252 | mb(); |
| 253 | radeon_gart_tlb_flush(rdev); |
| 254 | } |
| 255 | |
| 256 | /** |
| 257 | * radeon_gart_bind - bind pages into the gart page table |
| 258 | * |
| 259 | * @rdev: radeon_device pointer |
| 260 | * @offset: offset into the GPU's gart aperture |
| 261 | * @pages: number of pages to bind |
| 262 | * @pagelist: pages to bind |
| 263 | * @dma_addr: DMA addresses of pages |
| 264 | * |
| 265 | * Binds the requested pages to the gart page table |
| 266 | * (all asics). |
| 267 | * Returns 0 for success, -EINVAL for failure. |
| 268 | */ |
| 269 | int radeon_gart_bind(struct radeon_device *rdev, unsigned offset, |
| 270 | int pages, struct page **pagelist, dma_addr_t *dma_addr) |
| 271 | { |
| 272 | unsigned t; |
| 273 | unsigned p; |
| 274 | uint64_t page_base; |
| 275 | int i, j; |
| 276 | |
| 277 | if (!rdev->gart.ready) { |
| 278 | WARN(1, "trying to bind memory to uninitialized GART !\n"); |
| 279 | return -EINVAL; |
| 280 | } |
| 281 | t = offset / RADEON_GPU_PAGE_SIZE; |
| 282 | p = t / (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); |
| 283 | |
| 284 | for (i = 0; i < pages; i++, p++) { |
| 285 | rdev->gart.pages_addr[p] = dma_addr[i]; |
| 286 | rdev->gart.pages[p] = pagelist[i]; |
| 287 | if (rdev->gart.ptr) { |
| 288 | page_base = rdev->gart.pages_addr[p]; |
| 289 | for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) { |
| 290 | radeon_gart_set_page(rdev, t, page_base); |
| 291 | page_base += RADEON_GPU_PAGE_SIZE; |
| 292 | } |
| 293 | } |
| 294 | } |
| 295 | mb(); |
| 296 | radeon_gart_tlb_flush(rdev); |
| 297 | return 0; |
| 298 | } |
| 299 | |
| 300 | /** |
| 301 | * radeon_gart_restore - bind all pages in the gart page table |
| 302 | * |
| 303 | * @rdev: radeon_device pointer |
| 304 | * |
| 305 | * Binds all pages in the gart page table (all asics). |
| 306 | * Used to rebuild the gart table on device startup or resume. |
| 307 | */ |
| 308 | void radeon_gart_restore(struct radeon_device *rdev) |
| 309 | { |
| 310 | int i, j, t; |
| 311 | u64 page_base; |
| 312 | |
| 313 | if (!rdev->gart.ptr) { |
| 314 | return; |
| 315 | } |
| 316 | for (i = 0, t = 0; i < rdev->gart.num_cpu_pages; i++) { |
| 317 | page_base = rdev->gart.pages_addr[i]; |
| 318 | for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) { |
| 319 | radeon_gart_set_page(rdev, t, page_base); |
| 320 | page_base += RADEON_GPU_PAGE_SIZE; |
| 321 | } |
| 322 | } |
| 323 | mb(); |
| 324 | radeon_gart_tlb_flush(rdev); |
| 325 | } |
| 326 | |
| 327 | /** |
| 328 | * radeon_gart_init - init the driver info for managing the gart |
| 329 | * |
| 330 | * @rdev: radeon_device pointer |
| 331 | * |
| 332 | * Allocate the dummy page and init the gart driver info (all asics). |
| 333 | * Returns 0 for success, error for failure. |
| 334 | */ |
| 335 | int radeon_gart_init(struct radeon_device *rdev) |
| 336 | { |
| 337 | int r, i; |
| 338 | |
| 339 | if (rdev->gart.pages) { |
| 340 | return 0; |
| 341 | } |
| 342 | /* We need PAGE_SIZE >= RADEON_GPU_PAGE_SIZE */ |
| 343 | if (PAGE_SIZE < RADEON_GPU_PAGE_SIZE) { |
| 344 | DRM_ERROR("Page size is smaller than GPU page size!\n"); |
| 345 | return -EINVAL; |
| 346 | } |
| 347 | r = radeon_dummy_page_init(rdev); |
| 348 | if (r) |
| 349 | return r; |
| 350 | /* Compute table size */ |
| 351 | rdev->gart.num_cpu_pages = rdev->mc.gtt_size / PAGE_SIZE; |
| 352 | rdev->gart.num_gpu_pages = rdev->mc.gtt_size / RADEON_GPU_PAGE_SIZE; |
| 353 | DRM_INFO("GART: num cpu pages %u, num gpu pages %u\n", |
| 354 | rdev->gart.num_cpu_pages, rdev->gart.num_gpu_pages); |
| 355 | /* Allocate pages table */ |
| 356 | rdev->gart.pages = vzalloc(sizeof(void *) * rdev->gart.num_cpu_pages); |
| 357 | if (rdev->gart.pages == NULL) { |
| 358 | radeon_gart_fini(rdev); |
| 359 | return -ENOMEM; |
| 360 | } |
| 361 | rdev->gart.pages_addr = vzalloc(sizeof(dma_addr_t) * |
| 362 | rdev->gart.num_cpu_pages); |
| 363 | if (rdev->gart.pages_addr == NULL) { |
| 364 | radeon_gart_fini(rdev); |
| 365 | return -ENOMEM; |
| 366 | } |
| 367 | /* set GART entry to point to the dummy page by default */ |
| 368 | for (i = 0; i < rdev->gart.num_cpu_pages; i++) { |
| 369 | rdev->gart.pages_addr[i] = rdev->dummy_page.addr; |
| 370 | } |
| 371 | return 0; |
| 372 | } |
| 373 | |
| 374 | /** |
| 375 | * radeon_gart_fini - tear down the driver info for managing the gart |
| 376 | * |
| 377 | * @rdev: radeon_device pointer |
| 378 | * |
| 379 | * Tear down the gart driver info and free the dummy page (all asics). |
| 380 | */ |
| 381 | void radeon_gart_fini(struct radeon_device *rdev) |
| 382 | { |
| 383 | if (rdev->gart.pages && rdev->gart.pages_addr && rdev->gart.ready) { |
| 384 | /* unbind pages */ |
| 385 | radeon_gart_unbind(rdev, 0, rdev->gart.num_cpu_pages); |
| 386 | } |
| 387 | rdev->gart.ready = false; |
| 388 | vfree(rdev->gart.pages); |
| 389 | vfree(rdev->gart.pages_addr); |
| 390 | rdev->gart.pages = NULL; |
| 391 | rdev->gart.pages_addr = NULL; |
| 392 | |
| 393 | radeon_dummy_page_fini(rdev); |
| 394 | } |