Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright 2007 Dave Airlied |
| 3 | * All Rights Reserved. |
| 4 | * |
| 5 | * Permission is hereby granted, free of charge, to any person obtaining a |
| 6 | * copy of this software and associated documentation files (the "Software"), |
| 7 | * to deal in the Software without restriction, including without limitation |
| 8 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 9 | * and/or sell copies of the Software, and to permit persons to whom the |
| 10 | * Software is furnished to do so, subject to the following conditions: |
| 11 | * |
| 12 | * The above copyright notice and this permission notice (including the next |
| 13 | * paragraph) shall be included in all copies or substantial portions of the |
| 14 | * 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 | * VA LINUX SYSTEMS AND/OR ITS SUPPLIERS 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 | /* |
| 25 | * Authors: Dave Airlied <airlied@linux.ie> |
| 26 | * Ben Skeggs <darktama@iinet.net.au> |
| 27 | * Jeremy Kolb <jkolb@brandeis.edu> |
| 28 | */ |
| 29 | |
| 30 | #include "drmP.h" |
| 31 | |
| 32 | #include "nouveau_drm.h" |
| 33 | #include "nouveau_drv.h" |
| 34 | #include "nouveau_dma.h" |
| 35 | |
| 36 | static void |
| 37 | nouveau_bo_del_ttm(struct ttm_buffer_object *bo) |
| 38 | { |
| 39 | struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 40 | struct drm_device *dev = dev_priv->dev; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 41 | struct nouveau_bo *nvbo = nouveau_bo(bo); |
| 42 | |
| 43 | ttm_bo_kunmap(&nvbo->kmap); |
| 44 | |
| 45 | if (unlikely(nvbo->gem)) |
| 46 | DRM_ERROR("bo %p still attached to GEM object\n", bo); |
| 47 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 48 | if (nvbo->tile) |
| 49 | nv10_mem_expire_tiling(dev, nvbo->tile, NULL); |
| 50 | |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 51 | spin_lock(&dev_priv->ttm.bo_list_lock); |
| 52 | list_del(&nvbo->head); |
| 53 | spin_unlock(&dev_priv->ttm.bo_list_lock); |
| 54 | kfree(nvbo); |
| 55 | } |
| 56 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 57 | static void |
| 58 | nouveau_bo_fixup_align(struct drm_device *dev, |
| 59 | uint32_t tile_mode, uint32_t tile_flags, |
| 60 | int *align, int *size) |
| 61 | { |
| 62 | struct drm_nouveau_private *dev_priv = dev->dev_private; |
| 63 | |
| 64 | /* |
| 65 | * Some of the tile_flags have a periodic structure of N*4096 bytes, |
| 66 | * align to to that as well as the page size. Overallocate memory to |
| 67 | * avoid corruption of other buffer objects. |
| 68 | */ |
| 69 | if (dev_priv->card_type == NV_50) { |
| 70 | switch (tile_flags) { |
| 71 | case 0x1800: |
| 72 | case 0x2800: |
| 73 | case 0x4800: |
| 74 | case 0x7a00: |
| 75 | if (dev_priv->chipset >= 0xA0) { |
Maarten Maathuis | 1c7059e | 2009-12-25 18:51:17 +0100 | [diff] [blame^] | 76 | *size = roundup(*size, 28672); |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 77 | /* This is based on high end cards with 448 bits |
| 78 | * memory bus, could be different elsewhere.*/ |
| 79 | *size += 6 * 28672; |
| 80 | /* 8 * 28672 is the actual alignment requirement |
| 81 | * but we must also align to page size. */ |
| 82 | *align = 2 * 8 * 28672; |
| 83 | } else if (dev_priv->chipset >= 0x90) { |
Maarten Maathuis | 1c7059e | 2009-12-25 18:51:17 +0100 | [diff] [blame^] | 84 | *size = roundup(*size, 16384); |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 85 | *size += 3 * 16384; |
| 86 | *align = 12 * 16384; |
| 87 | } else { |
Maarten Maathuis | 1c7059e | 2009-12-25 18:51:17 +0100 | [diff] [blame^] | 88 | *size = roundup(*size, 8192); |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 89 | *size += 3 * 8192; |
| 90 | /* 12 * 8192 is the actual alignment requirement |
| 91 | * but we must also align to page size. */ |
| 92 | *align = 2 * 12 * 8192; |
| 93 | } |
| 94 | break; |
| 95 | default: |
| 96 | break; |
| 97 | } |
| 98 | |
| 99 | } else { |
| 100 | if (tile_mode) { |
| 101 | if (dev_priv->chipset >= 0x40) { |
| 102 | *align = 65536; |
| 103 | *size = roundup(*size, 64 * tile_mode); |
| 104 | |
| 105 | } else if (dev_priv->chipset >= 0x30) { |
| 106 | *align = 32768; |
| 107 | *size = roundup(*size, 64 * tile_mode); |
| 108 | |
| 109 | } else if (dev_priv->chipset >= 0x20) { |
| 110 | *align = 16384; |
| 111 | *size = roundup(*size, 64 * tile_mode); |
| 112 | |
| 113 | } else if (dev_priv->chipset >= 0x10) { |
| 114 | *align = 16384; |
| 115 | *size = roundup(*size, 32 * tile_mode); |
| 116 | } |
| 117 | } |
| 118 | } |
| 119 | |
Maarten Maathuis | 1c7059e | 2009-12-25 18:51:17 +0100 | [diff] [blame^] | 120 | /* ALIGN works only on powers of two. */ |
| 121 | *size = roundup(*size, PAGE_SIZE); |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 122 | |
| 123 | if (dev_priv->card_type == NV_50) { |
Maarten Maathuis | 1c7059e | 2009-12-25 18:51:17 +0100 | [diff] [blame^] | 124 | *size = roundup(*size, 65536); |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 125 | *align = max(65536, *align); |
| 126 | } |
| 127 | } |
| 128 | |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 129 | int |
| 130 | nouveau_bo_new(struct drm_device *dev, struct nouveau_channel *chan, |
| 131 | int size, int align, uint32_t flags, uint32_t tile_mode, |
| 132 | uint32_t tile_flags, bool no_vm, bool mappable, |
| 133 | struct nouveau_bo **pnvbo) |
| 134 | { |
| 135 | struct drm_nouveau_private *dev_priv = dev->dev_private; |
| 136 | struct nouveau_bo *nvbo; |
Francisco Jerez | 8dea4a1 | 2009-12-16 19:03:28 +0100 | [diff] [blame] | 137 | int ret = 0; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 138 | |
| 139 | nvbo = kzalloc(sizeof(struct nouveau_bo), GFP_KERNEL); |
| 140 | if (!nvbo) |
| 141 | return -ENOMEM; |
| 142 | INIT_LIST_HEAD(&nvbo->head); |
| 143 | INIT_LIST_HEAD(&nvbo->entry); |
| 144 | nvbo->mappable = mappable; |
| 145 | nvbo->no_vm = no_vm; |
| 146 | nvbo->tile_mode = tile_mode; |
| 147 | nvbo->tile_flags = tile_flags; |
| 148 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 149 | nouveau_bo_fixup_align(dev, tile_mode, tile_flags, &align, &size); |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 150 | align >>= PAGE_SHIFT; |
| 151 | |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 152 | nvbo->placement.fpfn = 0; |
| 153 | nvbo->placement.lpfn = mappable ? dev_priv->fb_mappable_pages : 0; |
Francisco Jerez | 8dea4a1 | 2009-12-16 19:03:28 +0100 | [diff] [blame] | 154 | nouveau_bo_placement_set(nvbo, flags); |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 155 | |
| 156 | nvbo->channel = chan; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 157 | ret = ttm_bo_init(&dev_priv->ttm.bdev, &nvbo->bo, size, |
| 158 | ttm_bo_type_device, &nvbo->placement, align, 0, |
| 159 | false, NULL, size, nouveau_bo_del_ttm); |
| 160 | nvbo->channel = NULL; |
| 161 | if (ret) { |
| 162 | /* ttm will call nouveau_bo_del_ttm if it fails.. */ |
| 163 | return ret; |
| 164 | } |
| 165 | |
| 166 | spin_lock(&dev_priv->ttm.bo_list_lock); |
| 167 | list_add_tail(&nvbo->head, &dev_priv->ttm.bo_list); |
| 168 | spin_unlock(&dev_priv->ttm.bo_list_lock); |
| 169 | *pnvbo = nvbo; |
| 170 | return 0; |
| 171 | } |
| 172 | |
| 173 | void |
| 174 | nouveau_bo_placement_set(struct nouveau_bo *nvbo, uint32_t memtype) |
| 175 | { |
| 176 | int n = 0; |
| 177 | |
| 178 | if (memtype & TTM_PL_FLAG_VRAM) |
| 179 | nvbo->placements[n++] = TTM_PL_FLAG_VRAM | TTM_PL_MASK_CACHING; |
| 180 | if (memtype & TTM_PL_FLAG_TT) |
| 181 | nvbo->placements[n++] = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING; |
| 182 | if (memtype & TTM_PL_FLAG_SYSTEM) |
| 183 | nvbo->placements[n++] = TTM_PL_FLAG_SYSTEM | TTM_PL_MASK_CACHING; |
| 184 | nvbo->placement.placement = nvbo->placements; |
| 185 | nvbo->placement.busy_placement = nvbo->placements; |
| 186 | nvbo->placement.num_placement = n; |
| 187 | nvbo->placement.num_busy_placement = n; |
Ben Skeggs | 37cb3e08 | 2009-12-16 16:22:42 +1000 | [diff] [blame] | 188 | |
| 189 | if (nvbo->pin_refcnt) { |
| 190 | while (n--) |
| 191 | nvbo->placements[n] |= TTM_PL_FLAG_NO_EVICT; |
| 192 | } |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 193 | } |
| 194 | |
| 195 | int |
| 196 | nouveau_bo_pin(struct nouveau_bo *nvbo, uint32_t memtype) |
| 197 | { |
| 198 | struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev); |
| 199 | struct ttm_buffer_object *bo = &nvbo->bo; |
| 200 | int ret, i; |
| 201 | |
| 202 | if (nvbo->pin_refcnt && !(memtype & (1 << bo->mem.mem_type))) { |
| 203 | NV_ERROR(nouveau_bdev(bo->bdev)->dev, |
| 204 | "bo %p pinned elsewhere: 0x%08x vs 0x%08x\n", bo, |
| 205 | 1 << bo->mem.mem_type, memtype); |
| 206 | return -EINVAL; |
| 207 | } |
| 208 | |
| 209 | if (nvbo->pin_refcnt++) |
| 210 | return 0; |
| 211 | |
| 212 | ret = ttm_bo_reserve(bo, false, false, false, 0); |
| 213 | if (ret) |
| 214 | goto out; |
| 215 | |
| 216 | nouveau_bo_placement_set(nvbo, memtype); |
| 217 | for (i = 0; i < nvbo->placement.num_placement; i++) |
| 218 | nvbo->placements[i] |= TTM_PL_FLAG_NO_EVICT; |
| 219 | |
| 220 | ret = ttm_bo_validate(bo, &nvbo->placement, false, false); |
| 221 | if (ret == 0) { |
| 222 | switch (bo->mem.mem_type) { |
| 223 | case TTM_PL_VRAM: |
| 224 | dev_priv->fb_aper_free -= bo->mem.size; |
| 225 | break; |
| 226 | case TTM_PL_TT: |
| 227 | dev_priv->gart_info.aper_free -= bo->mem.size; |
| 228 | break; |
| 229 | default: |
| 230 | break; |
| 231 | } |
| 232 | } |
| 233 | ttm_bo_unreserve(bo); |
| 234 | out: |
| 235 | if (unlikely(ret)) |
| 236 | nvbo->pin_refcnt--; |
| 237 | return ret; |
| 238 | } |
| 239 | |
| 240 | int |
| 241 | nouveau_bo_unpin(struct nouveau_bo *nvbo) |
| 242 | { |
| 243 | struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev); |
| 244 | struct ttm_buffer_object *bo = &nvbo->bo; |
| 245 | int ret, i; |
| 246 | |
| 247 | if (--nvbo->pin_refcnt) |
| 248 | return 0; |
| 249 | |
| 250 | ret = ttm_bo_reserve(bo, false, false, false, 0); |
| 251 | if (ret) |
| 252 | return ret; |
| 253 | |
| 254 | for (i = 0; i < nvbo->placement.num_placement; i++) |
| 255 | nvbo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT; |
| 256 | |
| 257 | ret = ttm_bo_validate(bo, &nvbo->placement, false, false); |
| 258 | if (ret == 0) { |
| 259 | switch (bo->mem.mem_type) { |
| 260 | case TTM_PL_VRAM: |
| 261 | dev_priv->fb_aper_free += bo->mem.size; |
| 262 | break; |
| 263 | case TTM_PL_TT: |
| 264 | dev_priv->gart_info.aper_free += bo->mem.size; |
| 265 | break; |
| 266 | default: |
| 267 | break; |
| 268 | } |
| 269 | } |
| 270 | |
| 271 | ttm_bo_unreserve(bo); |
| 272 | return ret; |
| 273 | } |
| 274 | |
| 275 | int |
| 276 | nouveau_bo_map(struct nouveau_bo *nvbo) |
| 277 | { |
| 278 | int ret; |
| 279 | |
| 280 | ret = ttm_bo_reserve(&nvbo->bo, false, false, false, 0); |
| 281 | if (ret) |
| 282 | return ret; |
| 283 | |
| 284 | ret = ttm_bo_kmap(&nvbo->bo, 0, nvbo->bo.mem.num_pages, &nvbo->kmap); |
| 285 | ttm_bo_unreserve(&nvbo->bo); |
| 286 | return ret; |
| 287 | } |
| 288 | |
| 289 | void |
| 290 | nouveau_bo_unmap(struct nouveau_bo *nvbo) |
| 291 | { |
| 292 | ttm_bo_kunmap(&nvbo->kmap); |
| 293 | } |
| 294 | |
| 295 | u16 |
| 296 | nouveau_bo_rd16(struct nouveau_bo *nvbo, unsigned index) |
| 297 | { |
| 298 | bool is_iomem; |
| 299 | u16 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem); |
| 300 | mem = &mem[index]; |
| 301 | if (is_iomem) |
| 302 | return ioread16_native((void __force __iomem *)mem); |
| 303 | else |
| 304 | return *mem; |
| 305 | } |
| 306 | |
| 307 | void |
| 308 | nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val) |
| 309 | { |
| 310 | bool is_iomem; |
| 311 | u16 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem); |
| 312 | mem = &mem[index]; |
| 313 | if (is_iomem) |
| 314 | iowrite16_native(val, (void __force __iomem *)mem); |
| 315 | else |
| 316 | *mem = val; |
| 317 | } |
| 318 | |
| 319 | u32 |
| 320 | nouveau_bo_rd32(struct nouveau_bo *nvbo, unsigned index) |
| 321 | { |
| 322 | bool is_iomem; |
| 323 | u32 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem); |
| 324 | mem = &mem[index]; |
| 325 | if (is_iomem) |
| 326 | return ioread32_native((void __force __iomem *)mem); |
| 327 | else |
| 328 | return *mem; |
| 329 | } |
| 330 | |
| 331 | void |
| 332 | nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val) |
| 333 | { |
| 334 | bool is_iomem; |
| 335 | u32 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem); |
| 336 | mem = &mem[index]; |
| 337 | if (is_iomem) |
| 338 | iowrite32_native(val, (void __force __iomem *)mem); |
| 339 | else |
| 340 | *mem = val; |
| 341 | } |
| 342 | |
| 343 | static struct ttm_backend * |
| 344 | nouveau_bo_create_ttm_backend_entry(struct ttm_bo_device *bdev) |
| 345 | { |
| 346 | struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev); |
| 347 | struct drm_device *dev = dev_priv->dev; |
| 348 | |
| 349 | switch (dev_priv->gart_info.type) { |
Ben Skeggs | b694dfb | 2009-12-15 10:38:32 +1000 | [diff] [blame] | 350 | #if __OS_HAS_AGP |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 351 | case NOUVEAU_GART_AGP: |
| 352 | return ttm_agp_backend_init(bdev, dev->agp->bridge); |
Ben Skeggs | b694dfb | 2009-12-15 10:38:32 +1000 | [diff] [blame] | 353 | #endif |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 354 | case NOUVEAU_GART_SGDMA: |
| 355 | return nouveau_sgdma_init_ttm(dev); |
| 356 | default: |
| 357 | NV_ERROR(dev, "Unknown GART type %d\n", |
| 358 | dev_priv->gart_info.type); |
| 359 | break; |
| 360 | } |
| 361 | |
| 362 | return NULL; |
| 363 | } |
| 364 | |
| 365 | static int |
| 366 | nouveau_bo_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags) |
| 367 | { |
| 368 | /* We'll do this from user space. */ |
| 369 | return 0; |
| 370 | } |
| 371 | |
| 372 | static int |
| 373 | nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, |
| 374 | struct ttm_mem_type_manager *man) |
| 375 | { |
| 376 | struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev); |
| 377 | struct drm_device *dev = dev_priv->dev; |
| 378 | |
| 379 | switch (type) { |
| 380 | case TTM_PL_SYSTEM: |
| 381 | man->flags = TTM_MEMTYPE_FLAG_MAPPABLE; |
| 382 | man->available_caching = TTM_PL_MASK_CACHING; |
| 383 | man->default_caching = TTM_PL_FLAG_CACHED; |
| 384 | break; |
| 385 | case TTM_PL_VRAM: |
| 386 | man->flags = TTM_MEMTYPE_FLAG_FIXED | |
| 387 | TTM_MEMTYPE_FLAG_MAPPABLE | |
| 388 | TTM_MEMTYPE_FLAG_NEEDS_IOREMAP; |
| 389 | man->available_caching = TTM_PL_FLAG_UNCACHED | |
| 390 | TTM_PL_FLAG_WC; |
| 391 | man->default_caching = TTM_PL_FLAG_WC; |
| 392 | |
| 393 | man->io_addr = NULL; |
| 394 | man->io_offset = drm_get_resource_start(dev, 1); |
| 395 | man->io_size = drm_get_resource_len(dev, 1); |
| 396 | if (man->io_size > nouveau_mem_fb_amount(dev)) |
| 397 | man->io_size = nouveau_mem_fb_amount(dev); |
| 398 | |
| 399 | man->gpu_offset = dev_priv->vm_vram_base; |
| 400 | break; |
| 401 | case TTM_PL_TT: |
| 402 | switch (dev_priv->gart_info.type) { |
| 403 | case NOUVEAU_GART_AGP: |
| 404 | man->flags = TTM_MEMTYPE_FLAG_MAPPABLE | |
| 405 | TTM_MEMTYPE_FLAG_NEEDS_IOREMAP; |
| 406 | man->available_caching = TTM_PL_FLAG_UNCACHED; |
| 407 | man->default_caching = TTM_PL_FLAG_UNCACHED; |
| 408 | break; |
| 409 | case NOUVEAU_GART_SGDMA: |
| 410 | man->flags = TTM_MEMTYPE_FLAG_MAPPABLE | |
| 411 | TTM_MEMTYPE_FLAG_CMA; |
| 412 | man->available_caching = TTM_PL_MASK_CACHING; |
| 413 | man->default_caching = TTM_PL_FLAG_CACHED; |
| 414 | break; |
| 415 | default: |
| 416 | NV_ERROR(dev, "Unknown GART type: %d\n", |
| 417 | dev_priv->gart_info.type); |
| 418 | return -EINVAL; |
| 419 | } |
| 420 | |
| 421 | man->io_offset = dev_priv->gart_info.aper_base; |
| 422 | man->io_size = dev_priv->gart_info.aper_size; |
| 423 | man->io_addr = NULL; |
| 424 | man->gpu_offset = dev_priv->vm_gart_base; |
| 425 | break; |
| 426 | default: |
| 427 | NV_ERROR(dev, "Unsupported memory type %u\n", (unsigned)type); |
| 428 | return -EINVAL; |
| 429 | } |
| 430 | return 0; |
| 431 | } |
| 432 | |
| 433 | static void |
| 434 | nouveau_bo_evict_flags(struct ttm_buffer_object *bo, struct ttm_placement *pl) |
| 435 | { |
| 436 | struct nouveau_bo *nvbo = nouveau_bo(bo); |
| 437 | |
| 438 | switch (bo->mem.mem_type) { |
Francisco Jerez | 22fbd53 | 2009-12-11 18:40:17 +0100 | [diff] [blame] | 439 | case TTM_PL_VRAM: |
| 440 | nouveau_bo_placement_set(nvbo, TTM_PL_FLAG_TT | |
| 441 | TTM_PL_FLAG_SYSTEM); |
| 442 | break; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 443 | default: |
| 444 | nouveau_bo_placement_set(nvbo, TTM_PL_FLAG_SYSTEM); |
| 445 | break; |
| 446 | } |
Francisco Jerez | 22fbd53 | 2009-12-11 18:40:17 +0100 | [diff] [blame] | 447 | |
| 448 | *pl = nvbo->placement; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 449 | } |
| 450 | |
| 451 | |
| 452 | /* GPU-assisted copy using NV_MEMORY_TO_MEMORY_FORMAT, can access |
| 453 | * TTM_PL_{VRAM,TT} directly. |
| 454 | */ |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 455 | |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 456 | static int |
| 457 | nouveau_bo_move_accel_cleanup(struct nouveau_channel *chan, |
| 458 | struct nouveau_bo *nvbo, bool evict, bool no_wait, |
| 459 | struct ttm_mem_reg *new_mem) |
| 460 | { |
| 461 | struct nouveau_fence *fence = NULL; |
| 462 | int ret; |
| 463 | |
| 464 | ret = nouveau_fence_new(chan, &fence, true); |
| 465 | if (ret) |
| 466 | return ret; |
| 467 | |
| 468 | ret = ttm_bo_move_accel_cleanup(&nvbo->bo, fence, NULL, |
| 469 | evict, no_wait, new_mem); |
| 470 | nouveau_fence_unref((void *)&fence); |
| 471 | return ret; |
| 472 | } |
| 473 | |
| 474 | static inline uint32_t |
| 475 | nouveau_bo_mem_ctxdma(struct nouveau_bo *nvbo, struct nouveau_channel *chan, |
| 476 | struct ttm_mem_reg *mem) |
| 477 | { |
| 478 | if (chan == nouveau_bdev(nvbo->bo.bdev)->channel) { |
| 479 | if (mem->mem_type == TTM_PL_TT) |
| 480 | return NvDmaGART; |
| 481 | return NvDmaVRAM; |
| 482 | } |
| 483 | |
| 484 | if (mem->mem_type == TTM_PL_TT) |
| 485 | return chan->gart_handle; |
| 486 | return chan->vram_handle; |
| 487 | } |
| 488 | |
| 489 | static int |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 490 | nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr, |
| 491 | int no_wait, struct ttm_mem_reg *new_mem) |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 492 | { |
| 493 | struct nouveau_bo *nvbo = nouveau_bo(bo); |
| 494 | struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 495 | struct ttm_mem_reg *old_mem = &bo->mem; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 496 | struct nouveau_channel *chan; |
| 497 | uint64_t src_offset, dst_offset; |
| 498 | uint32_t page_count; |
| 499 | int ret; |
| 500 | |
| 501 | chan = nvbo->channel; |
Ben Skeggs | 0735f62 | 2009-12-16 14:28:55 +1000 | [diff] [blame] | 502 | if (!chan || nvbo->tile_flags || nvbo->no_vm) |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 503 | chan = dev_priv->channel; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 504 | |
| 505 | src_offset = old_mem->mm_node->start << PAGE_SHIFT; |
| 506 | dst_offset = new_mem->mm_node->start << PAGE_SHIFT; |
| 507 | if (chan != dev_priv->channel) { |
| 508 | if (old_mem->mem_type == TTM_PL_TT) |
| 509 | src_offset += dev_priv->vm_gart_base; |
| 510 | else |
| 511 | src_offset += dev_priv->vm_vram_base; |
| 512 | |
| 513 | if (new_mem->mem_type == TTM_PL_TT) |
| 514 | dst_offset += dev_priv->vm_gart_base; |
| 515 | else |
| 516 | dst_offset += dev_priv->vm_vram_base; |
| 517 | } |
| 518 | |
| 519 | ret = RING_SPACE(chan, 3); |
| 520 | if (ret) |
| 521 | return ret; |
| 522 | BEGIN_RING(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_DMA_SOURCE, 2); |
| 523 | OUT_RING(chan, nouveau_bo_mem_ctxdma(nvbo, chan, old_mem)); |
| 524 | OUT_RING(chan, nouveau_bo_mem_ctxdma(nvbo, chan, new_mem)); |
| 525 | |
| 526 | if (dev_priv->card_type >= NV_50) { |
| 527 | ret = RING_SPACE(chan, 4); |
| 528 | if (ret) |
| 529 | return ret; |
| 530 | BEGIN_RING(chan, NvSubM2MF, 0x0200, 1); |
| 531 | OUT_RING(chan, 1); |
| 532 | BEGIN_RING(chan, NvSubM2MF, 0x021c, 1); |
| 533 | OUT_RING(chan, 1); |
| 534 | } |
| 535 | |
| 536 | page_count = new_mem->num_pages; |
| 537 | while (page_count) { |
| 538 | int line_count = (page_count > 2047) ? 2047 : page_count; |
| 539 | |
| 540 | if (dev_priv->card_type >= NV_50) { |
| 541 | ret = RING_SPACE(chan, 3); |
| 542 | if (ret) |
| 543 | return ret; |
| 544 | BEGIN_RING(chan, NvSubM2MF, 0x0238, 2); |
| 545 | OUT_RING(chan, upper_32_bits(src_offset)); |
| 546 | OUT_RING(chan, upper_32_bits(dst_offset)); |
| 547 | } |
| 548 | ret = RING_SPACE(chan, 11); |
| 549 | if (ret) |
| 550 | return ret; |
| 551 | BEGIN_RING(chan, NvSubM2MF, |
| 552 | NV_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 8); |
| 553 | OUT_RING(chan, lower_32_bits(src_offset)); |
| 554 | OUT_RING(chan, lower_32_bits(dst_offset)); |
| 555 | OUT_RING(chan, PAGE_SIZE); /* src_pitch */ |
| 556 | OUT_RING(chan, PAGE_SIZE); /* dst_pitch */ |
| 557 | OUT_RING(chan, PAGE_SIZE); /* line_length */ |
| 558 | OUT_RING(chan, line_count); |
| 559 | OUT_RING(chan, (1<<8)|(1<<0)); |
| 560 | OUT_RING(chan, 0); |
| 561 | BEGIN_RING(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_NOP, 1); |
| 562 | OUT_RING(chan, 0); |
| 563 | |
| 564 | page_count -= line_count; |
| 565 | src_offset += (PAGE_SIZE * line_count); |
| 566 | dst_offset += (PAGE_SIZE * line_count); |
| 567 | } |
| 568 | |
| 569 | return nouveau_bo_move_accel_cleanup(chan, nvbo, evict, no_wait, new_mem); |
| 570 | } |
| 571 | |
| 572 | static int |
| 573 | nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr, |
| 574 | bool no_wait, struct ttm_mem_reg *new_mem) |
| 575 | { |
| 576 | u32 placement_memtype = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING; |
| 577 | struct ttm_placement placement; |
| 578 | struct ttm_mem_reg tmp_mem; |
| 579 | int ret; |
| 580 | |
| 581 | placement.fpfn = placement.lpfn = 0; |
| 582 | placement.num_placement = placement.num_busy_placement = 1; |
Francisco Jerez | 77e2b5e | 2009-12-16 19:05:00 +0100 | [diff] [blame] | 583 | placement.placement = placement.busy_placement = &placement_memtype; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 584 | |
| 585 | tmp_mem = *new_mem; |
| 586 | tmp_mem.mm_node = NULL; |
| 587 | ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait); |
| 588 | if (ret) |
| 589 | return ret; |
| 590 | |
| 591 | ret = ttm_tt_bind(bo->ttm, &tmp_mem); |
| 592 | if (ret) |
| 593 | goto out; |
| 594 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 595 | ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait, &tmp_mem); |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 596 | if (ret) |
| 597 | goto out; |
| 598 | |
| 599 | ret = ttm_bo_move_ttm(bo, evict, no_wait, new_mem); |
| 600 | out: |
| 601 | if (tmp_mem.mm_node) { |
| 602 | spin_lock(&bo->bdev->glob->lru_lock); |
| 603 | drm_mm_put_block(tmp_mem.mm_node); |
| 604 | spin_unlock(&bo->bdev->glob->lru_lock); |
| 605 | } |
| 606 | |
| 607 | return ret; |
| 608 | } |
| 609 | |
| 610 | static int |
| 611 | nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr, |
| 612 | bool no_wait, struct ttm_mem_reg *new_mem) |
| 613 | { |
| 614 | u32 placement_memtype = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING; |
| 615 | struct ttm_placement placement; |
| 616 | struct ttm_mem_reg tmp_mem; |
| 617 | int ret; |
| 618 | |
| 619 | placement.fpfn = placement.lpfn = 0; |
| 620 | placement.num_placement = placement.num_busy_placement = 1; |
Francisco Jerez | 77e2b5e | 2009-12-16 19:05:00 +0100 | [diff] [blame] | 621 | placement.placement = placement.busy_placement = &placement_memtype; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 622 | |
| 623 | tmp_mem = *new_mem; |
| 624 | tmp_mem.mm_node = NULL; |
| 625 | ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait); |
| 626 | if (ret) |
| 627 | return ret; |
| 628 | |
| 629 | ret = ttm_bo_move_ttm(bo, evict, no_wait, &tmp_mem); |
| 630 | if (ret) |
| 631 | goto out; |
| 632 | |
Francisco Jerez | 27f691a | 2009-12-16 19:05:38 +0100 | [diff] [blame] | 633 | ret = nouveau_bo_move_m2mf(bo, evict, intr, no_wait, new_mem); |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 634 | if (ret) |
| 635 | goto out; |
| 636 | |
| 637 | out: |
| 638 | if (tmp_mem.mm_node) { |
| 639 | spin_lock(&bo->bdev->glob->lru_lock); |
| 640 | drm_mm_put_block(tmp_mem.mm_node); |
| 641 | spin_unlock(&bo->bdev->glob->lru_lock); |
| 642 | } |
| 643 | |
| 644 | return ret; |
| 645 | } |
| 646 | |
| 647 | static int |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 648 | nouveau_bo_vm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem, |
| 649 | struct nouveau_tile_reg **new_tile) |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 650 | { |
| 651 | struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 652 | struct drm_device *dev = dev_priv->dev; |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 653 | struct nouveau_bo *nvbo = nouveau_bo(bo); |
| 654 | uint64_t offset; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 655 | int ret; |
| 656 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 657 | if (nvbo->no_vm || new_mem->mem_type != TTM_PL_VRAM) { |
| 658 | /* Nothing to do. */ |
| 659 | *new_tile = NULL; |
| 660 | return 0; |
| 661 | } |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 662 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 663 | offset = new_mem->mm_node->start << PAGE_SHIFT; |
| 664 | |
| 665 | if (dev_priv->card_type == NV_50) { |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 666 | ret = nv50_mem_vm_bind_linear(dev, |
| 667 | offset + dev_priv->vm_vram_base, |
| 668 | new_mem->size, nvbo->tile_flags, |
| 669 | offset); |
| 670 | if (ret) |
| 671 | return ret; |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 672 | |
| 673 | } else if (dev_priv->card_type >= NV_10) { |
| 674 | *new_tile = nv10_mem_set_tiling(dev, offset, new_mem->size, |
| 675 | nvbo->tile_mode); |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 676 | } |
| 677 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 678 | return 0; |
| 679 | } |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 680 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 681 | static void |
| 682 | nouveau_bo_vm_cleanup(struct ttm_buffer_object *bo, |
| 683 | struct nouveau_tile_reg *new_tile, |
| 684 | struct nouveau_tile_reg **old_tile) |
| 685 | { |
| 686 | struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); |
| 687 | struct drm_device *dev = dev_priv->dev; |
| 688 | |
| 689 | if (dev_priv->card_type >= NV_10 && |
| 690 | dev_priv->card_type < NV_50) { |
| 691 | if (*old_tile) |
| 692 | nv10_mem_expire_tiling(dev, *old_tile, bo->sync_obj); |
| 693 | |
| 694 | *old_tile = new_tile; |
| 695 | } |
| 696 | } |
| 697 | |
| 698 | static int |
| 699 | nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr, |
| 700 | bool no_wait, struct ttm_mem_reg *new_mem) |
| 701 | { |
| 702 | struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev); |
| 703 | struct nouveau_bo *nvbo = nouveau_bo(bo); |
| 704 | struct ttm_mem_reg *old_mem = &bo->mem; |
| 705 | struct nouveau_tile_reg *new_tile = NULL; |
| 706 | int ret = 0; |
| 707 | |
| 708 | ret = nouveau_bo_vm_bind(bo, new_mem, &new_tile); |
| 709 | if (ret) |
| 710 | return ret; |
| 711 | |
| 712 | /* Software copy if the card isn't up and running yet. */ |
| 713 | if (dev_priv->init_state != NOUVEAU_CARD_INIT_DONE || |
| 714 | !dev_priv->channel) { |
| 715 | ret = ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); |
| 716 | goto out; |
| 717 | } |
| 718 | |
| 719 | /* Fake bo copy. */ |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 720 | if (old_mem->mem_type == TTM_PL_SYSTEM && !bo->ttm) { |
| 721 | BUG_ON(bo->mem.mm_node != NULL); |
| 722 | bo->mem = *new_mem; |
| 723 | new_mem->mm_node = NULL; |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 724 | goto out; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 725 | } |
| 726 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 727 | /* Hardware assisted copy. */ |
| 728 | if (new_mem->mem_type == TTM_PL_SYSTEM) |
| 729 | ret = nouveau_bo_move_flipd(bo, evict, intr, no_wait, new_mem); |
| 730 | else if (old_mem->mem_type == TTM_PL_SYSTEM) |
| 731 | ret = nouveau_bo_move_flips(bo, evict, intr, no_wait, new_mem); |
| 732 | else |
| 733 | ret = nouveau_bo_move_m2mf(bo, evict, intr, no_wait, new_mem); |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 734 | |
Francisco Jerez | a0af9ad | 2009-12-11 16:51:09 +0100 | [diff] [blame] | 735 | if (!ret) |
| 736 | goto out; |
| 737 | |
| 738 | /* Fallback to software copy. */ |
| 739 | ret = ttm_bo_move_memcpy(bo, evict, no_wait, new_mem); |
| 740 | |
| 741 | out: |
| 742 | if (ret) |
| 743 | nouveau_bo_vm_cleanup(bo, NULL, &new_tile); |
| 744 | else |
| 745 | nouveau_bo_vm_cleanup(bo, new_tile, &nvbo->tile); |
| 746 | |
| 747 | return ret; |
Ben Skeggs | 6ee7386 | 2009-12-11 19:24:15 +1000 | [diff] [blame] | 748 | } |
| 749 | |
| 750 | static int |
| 751 | nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp) |
| 752 | { |
| 753 | return 0; |
| 754 | } |
| 755 | |
| 756 | struct ttm_bo_driver nouveau_bo_driver = { |
| 757 | .create_ttm_backend_entry = nouveau_bo_create_ttm_backend_entry, |
| 758 | .invalidate_caches = nouveau_bo_invalidate_caches, |
| 759 | .init_mem_type = nouveau_bo_init_mem_type, |
| 760 | .evict_flags = nouveau_bo_evict_flags, |
| 761 | .move = nouveau_bo_move, |
| 762 | .verify_access = nouveau_bo_verify_access, |
| 763 | .sync_obj_signaled = nouveau_fence_signalled, |
| 764 | .sync_obj_wait = nouveau_fence_wait, |
| 765 | .sync_obj_flush = nouveau_fence_flush, |
| 766 | .sync_obj_unref = nouveau_fence_unref, |
| 767 | .sync_obj_ref = nouveau_fence_ref, |
| 768 | }; |
| 769 | |