diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 316 |
1 files changed, 187 insertions, 129 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 69ebd07f3eee..12aa35ac0eb4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -47,7 +47,6 @@ #include <drm/ttm/ttm_bo_driver.h> #include <drm/ttm/ttm_placement.h> #include <drm/ttm/ttm_module.h> -#include <drm/ttm/ttm_page_alloc.h> #include <drm/drm_debugfs.h> #include <drm/amdgpu_drm.h> @@ -63,12 +62,17 @@ #define AMDGPU_TTM_VRAM_MAX_DW_READ (size_t)128 +static int amdgpu_ttm_backend_bind(struct ttm_bo_device *bdev, + struct ttm_tt *ttm, + struct ttm_resource *bo_mem); +static void amdgpu_ttm_backend_unbind(struct ttm_bo_device *bdev, + struct ttm_tt *ttm); + static int amdgpu_ttm_init_on_chip(struct amdgpu_device *adev, unsigned int type, uint64_t size_in_page) { return ttm_range_man_init(&adev->mman.bdev, type, - TTM_PL_FLAG_UNCACHED, TTM_PL_FLAG_UNCACHED, false, size_in_page); } @@ -88,7 +92,8 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo, static const struct ttm_place placements = { .fpfn = 0, .lpfn = 0, - .flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM + .mem_type = TTM_PL_SYSTEM, + .flags = 0 }; /* Don't handle scatter gather BOs */ @@ -175,24 +180,6 @@ static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp) } /** - * amdgpu_move_null - Register memory for a buffer object - * - * @bo: The bo to assign the memory to - * @new_mem: The memory to be assigned. - * - * Assign the memory from new_mem to the memory of the buffer object bo. - */ -static void amdgpu_move_null(struct ttm_buffer_object *bo, - struct ttm_resource *new_mem) -{ - struct ttm_resource *old_mem = &bo->mem; - - BUG_ON(old_mem->mm_node != NULL); - *old_mem = *new_mem; - new_mem->mm_node = NULL; -} - -/** * amdgpu_mm_node_addr - Compute the GPU relative offset of a GTT buffer. * * @bo: The bo to assign the memory to. @@ -306,11 +293,9 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, cpu_addr = &job->ibs[0].ptr[num_dw]; if (mem->mem_type == TTM_PL_TT) { - struct ttm_dma_tt *dma; dma_addr_t *dma_address; - dma = container_of(bo->ttm, struct ttm_dma_tt, ttm); - dma_address = &dma->dma_address[offset >> PAGE_SHIFT]; + dma_address = &bo->ttm->dma_address[offset >> PAGE_SHIFT]; r = amdgpu_gart_map(adev, 0, num_pages, dma_address, flags, cpu_addr); if (r) @@ -514,9 +499,9 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo, /* Always block for VM page tables before committing the new location */ if (bo->type == ttm_bo_type_kernel) - r = ttm_bo_move_accel_cleanup(bo, fence, true, new_mem); + r = ttm_bo_move_accel_cleanup(bo, fence, true, false, new_mem); else - r = ttm_bo_pipeline_move(bo, fence, evict, new_mem); + r = ttm_bo_move_accel_cleanup(bo, fence, evict, true, new_mem); dma_fence_put(fence); return r; @@ -551,21 +536,20 @@ static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict, placement.busy_placement = &placements; placements.fpfn = 0; placements.lpfn = 0; - placements.flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT; + placements.mem_type = TTM_PL_TT; + placements.flags = 0; r = ttm_bo_mem_space(bo, &placement, &tmp_mem, ctx); if (unlikely(r)) { pr_err("Failed to find GTT space for blit from VRAM\n"); return r; } - /* set caching flags */ - r = ttm_tt_set_placement_caching(bo->ttm, tmp_mem.placement); - if (unlikely(r)) { + r = ttm_tt_populate(bo->bdev, bo->ttm, ctx); + if (unlikely(r)) goto out_cleanup; - } /* Bind the memory to the GTT space */ - r = ttm_tt_bind(bo->ttm, &tmp_mem, ctx); + r = amdgpu_ttm_backend_bind(bo->bdev, bo->ttm, &tmp_mem); if (unlikely(r)) { goto out_cleanup; } @@ -576,8 +560,13 @@ static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict, goto out_cleanup; } - /* move BO (in tmp_mem) to new_mem */ - r = ttm_bo_move_ttm(bo, ctx, new_mem); + r = ttm_bo_wait_ctx(bo, ctx); + if (unlikely(r)) + goto out_cleanup; + + amdgpu_ttm_backend_unbind(bo->bdev, bo->ttm); + ttm_resource_free(bo, &bo->mem); + ttm_bo_assign_mem(bo, new_mem); out_cleanup: ttm_resource_free(bo, &tmp_mem); return r; @@ -607,7 +596,8 @@ static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict, placement.busy_placement = &placements; placements.fpfn = 0; placements.lpfn = 0; - placements.flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT; + placements.mem_type = TTM_PL_TT; + placements.flags = 0; r = ttm_bo_mem_space(bo, &placement, &tmp_mem, ctx); if (unlikely(r)) { pr_err("Failed to find GTT space for blit to VRAM\n"); @@ -615,11 +605,16 @@ static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict, } /* move/bind old memory to GTT space */ - r = ttm_bo_move_ttm(bo, ctx, &tmp_mem); + r = ttm_tt_populate(bo->bdev, bo->ttm, ctx); + if (unlikely(r)) + return r; + + r = amdgpu_ttm_backend_bind(bo->bdev, bo->ttm, &tmp_mem); if (unlikely(r)) { goto out_cleanup; } + ttm_bo_assign_mem(bo, &tmp_mem); /* copy to VRAM */ r = amdgpu_move_blit(bo, evict, new_mem, old_mem); if (unlikely(r)) { @@ -668,25 +663,43 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict, struct ttm_resource *old_mem = &bo->mem; int r; + if (new_mem->mem_type == TTM_PL_TT) { + r = amdgpu_ttm_backend_bind(bo->bdev, bo->ttm, new_mem); + if (r) + return r; + } + + amdgpu_bo_move_notify(bo, evict, new_mem); + /* Can't move a pinned BO */ abo = ttm_to_amdgpu_bo(bo); - if (WARN_ON_ONCE(abo->pin_count > 0)) + if (WARN_ON_ONCE(abo->tbo.pin_count > 0)) return -EINVAL; adev = amdgpu_ttm_adev(bo->bdev); if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) { - amdgpu_move_null(bo, new_mem); + ttm_bo_move_null(bo, new_mem); return 0; } - if ((old_mem->mem_type == TTM_PL_TT && - new_mem->mem_type == TTM_PL_SYSTEM) || - (old_mem->mem_type == TTM_PL_SYSTEM && - new_mem->mem_type == TTM_PL_TT)) { - /* bind is enough */ - amdgpu_move_null(bo, new_mem); + if (old_mem->mem_type == TTM_PL_SYSTEM && + new_mem->mem_type == TTM_PL_TT) { + ttm_bo_move_null(bo, new_mem); return 0; } + + if (old_mem->mem_type == TTM_PL_TT && + new_mem->mem_type == TTM_PL_SYSTEM) { + r = ttm_bo_wait_ctx(bo, ctx); + if (r) + goto fail; + + amdgpu_ttm_backend_unbind(bo->bdev, bo->ttm); + ttm_resource_free(bo, &bo->mem); + ttm_bo_assign_mem(bo, new_mem); + return 0; + } + if (old_mem->mem_type == AMDGPU_PL_GDS || old_mem->mem_type == AMDGPU_PL_GWS || old_mem->mem_type == AMDGPU_PL_OA || @@ -694,7 +707,7 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict, new_mem->mem_type == AMDGPU_PL_GWS || new_mem->mem_type == AMDGPU_PL_OA) { /* Nothing to save here */ - amdgpu_move_null(bo, new_mem); + ttm_bo_move_null(bo, new_mem); return 0; } @@ -720,12 +733,12 @@ memcpy: if (!amdgpu_mem_visible(adev, old_mem) || !amdgpu_mem_visible(adev, new_mem)) { pr_err("Move buffer fallback to memcpy unavailable\n"); - return r; + goto fail; } r = ttm_bo_move_memcpy(bo, ctx, new_mem); if (r) - return r; + goto fail; } if (bo->type == ttm_bo_type_device && @@ -740,6 +753,11 @@ memcpy: /* update statistics */ atomic64_add((u64)bo->num_pages << PAGE_SHIFT, &adev->num_bytes_moved); return 0; +fail: + swap(*new_mem, bo->mem); + amdgpu_bo_move_notify(bo, false, new_mem); + swap(*new_mem, bo->mem); + return r; } /** @@ -773,8 +791,9 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_reso mem->bus.addr = (u8 *)adev->mman.aper_base_kaddr + mem->bus.offset; - mem->bus.base = adev->gmc.aper_base; + mem->bus.offset += adev->gmc.aper_base; mem->bus.is_iomem = true; + mem->bus.caching = ttm_write_combined; break; default: return -EINVAL; @@ -785,12 +804,13 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_reso static unsigned long amdgpu_ttm_io_mem_pfn(struct ttm_buffer_object *bo, unsigned long page_offset) { + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev); uint64_t offset = (page_offset << PAGE_SHIFT); struct drm_mm_node *mm; mm = amdgpu_find_mm_node(&bo->mem, &offset); - return (bo->mem.bus.base >> PAGE_SHIFT) + mm->start + - (offset >> PAGE_SHIFT); + offset += adev->gmc.aper_base; + return mm->start + (offset >> PAGE_SHIFT); } /** @@ -818,12 +838,13 @@ uint64_t amdgpu_ttm_domain_start(struct amdgpu_device *adev, uint32_t type) * TTM backend functions. */ struct amdgpu_ttm_tt { - struct ttm_dma_tt ttm; + struct ttm_tt ttm; struct drm_gem_object *gobj; u64 offset; uint64_t userptr; struct task_struct *usertask; uint32_t userflags; + bool bound; #if IS_ENABLED(CONFIG_DRM_AMDGPU_USERPTR) struct hmm_range *range; #endif @@ -949,7 +970,7 @@ bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm) if (!gtt || !gtt->userptr) return false; - DRM_DEBUG_DRIVER("user_pages_done 0x%llx pages 0x%lx\n", + DRM_DEBUG_DRIVER("user_pages_done 0x%llx pages 0x%x\n", gtt->userptr, ttm->num_pages); WARN_ONCE(!gtt->range || !gtt->range->hmm_pfns, @@ -991,9 +1012,10 @@ void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages) * * Called by amdgpu_ttm_backend_bind() **/ -static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm) +static int amdgpu_ttm_tt_pin_userptr(struct ttm_bo_device *bdev, + struct ttm_tt *ttm) { - struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); + struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); struct amdgpu_ttm_tt *gtt = (void *)ttm; int r; @@ -1028,9 +1050,10 @@ release_sg: /** * amdgpu_ttm_tt_unpin_userptr - Unpin and unmap userptr pages */ -static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm) +static void amdgpu_ttm_tt_unpin_userptr(struct ttm_bo_device *bdev, + struct ttm_tt *ttm) { - struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); + struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); struct amdgpu_ttm_tt *gtt = (void *)ttm; int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY); @@ -1099,7 +1122,7 @@ static int amdgpu_ttm_gart_bind(struct amdgpu_device *adev, gart_bind_fail: if (r) - DRM_ERROR("failed to bind %lu pages at 0x%08llX\n", + DRM_ERROR("failed to bind %u pages at 0x%08llX\n", ttm->num_pages, gtt->offset); return r; @@ -1111,23 +1134,30 @@ gart_bind_fail: * Called by ttm_tt_bind() on behalf of ttm_bo_handle_move_mem(). * This handles binding GTT memory to the device address space. */ -static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm, +static int amdgpu_ttm_backend_bind(struct ttm_bo_device *bdev, + struct ttm_tt *ttm, struct ttm_resource *bo_mem) { - struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); + struct amdgpu_ttm_tt *gtt = (void*)ttm; uint64_t flags; int r = 0; + if (!bo_mem) + return -EINVAL; + + if (gtt->bound) + return 0; + if (gtt->userptr) { - r = amdgpu_ttm_tt_pin_userptr(ttm); + r = amdgpu_ttm_tt_pin_userptr(bdev, ttm); if (r) { DRM_ERROR("failed to pin userptr\n"); return r; } } if (!ttm->num_pages) { - WARN(1, "nothing to bind %lu pages for mreg %p back %p!\n", + WARN(1, "nothing to bind %u pages for mreg %p back %p!\n", ttm->num_pages, bo_mem, ttm); } @@ -1150,8 +1180,9 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm, ttm->pages, gtt->ttm.dma_address, flags); if (r) - DRM_ERROR("failed to bind %lu pages at 0x%08llX\n", + DRM_ERROR("failed to bind %u pages at 0x%08llX\n", ttm->num_pages, gtt->offset); + gtt->bound = true; return r; } @@ -1191,8 +1222,8 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo) placement.busy_placement = &placements; placements.fpfn = 0; placements.lpfn = adev->gmc.gart_size >> PAGE_SHIFT; - placements.flags = (bo->mem.placement & ~TTM_PL_MASK_MEM) | - TTM_PL_FLAG_TT; + placements.mem_type = TTM_PL_TT; + placements.flags = bo->mem.placement; r = ttm_bo_mem_space(bo, &placement, &tmp, &ctx); if (unlikely(r)) @@ -1243,15 +1274,19 @@ int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo) * Called by ttm_tt_unbind() on behalf of ttm_bo_move_ttm() and * ttm_tt_destroy(). */ -static void amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) +static void amdgpu_ttm_backend_unbind(struct ttm_bo_device *bdev, + struct ttm_tt *ttm) { - struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); + struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); struct amdgpu_ttm_tt *gtt = (void *)ttm; int r; + if (!gtt->bound) + return; + /* if the pages have userptr pinning then clear that first */ if (gtt->userptr) - amdgpu_ttm_tt_unpin_userptr(ttm); + amdgpu_ttm_tt_unpin_userptr(bdev, ttm); if (gtt->offset == AMDGPU_BO_INVALID_OFFSET) return; @@ -1259,27 +1294,25 @@ static void amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) /* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */ r = amdgpu_gart_unbind(adev, gtt->offset, ttm->num_pages); if (r) - DRM_ERROR("failed to unbind %lu pages at 0x%08llX\n", - gtt->ttm.ttm.num_pages, gtt->offset); + DRM_ERROR("failed to unbind %u pages at 0x%08llX\n", + gtt->ttm.num_pages, gtt->offset); + gtt->bound = false; } -static void amdgpu_ttm_backend_destroy(struct ttm_tt *ttm) +static void amdgpu_ttm_backend_destroy(struct ttm_bo_device *bdev, + struct ttm_tt *ttm) { struct amdgpu_ttm_tt *gtt = (void *)ttm; + amdgpu_ttm_backend_unbind(bdev, ttm); + ttm_tt_destroy_common(bdev, ttm); if (gtt->usertask) put_task_struct(gtt->usertask); - ttm_dma_tt_fini(>t->ttm); + ttm_tt_fini(>t->ttm); kfree(gtt); } -static struct ttm_backend_func amdgpu_backend_func = { - .bind = &amdgpu_ttm_backend_bind, - .unbind = &amdgpu_ttm_backend_unbind, - .destroy = &amdgpu_ttm_backend_destroy, -}; - /** * amdgpu_ttm_tt_create - Create a ttm_tt object for a given BO * @@ -1290,21 +1323,27 @@ static struct ttm_backend_func amdgpu_backend_func = { static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo, uint32_t page_flags) { + struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo); struct amdgpu_ttm_tt *gtt; + enum ttm_caching caching; gtt = kzalloc(sizeof(struct amdgpu_ttm_tt), GFP_KERNEL); if (gtt == NULL) { return NULL; } - gtt->ttm.ttm.func = &amdgpu_backend_func; gtt->gobj = &bo->base; + if (abo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) + caching = ttm_write_combined; + else + caching = ttm_cached; + /* allocate space for the uninitialized page entries */ - if (ttm_sg_tt_init(>t->ttm, bo, page_flags)) { + if (ttm_sg_tt_init(>t->ttm, bo, page_flags, caching)) { kfree(gtt); return NULL; } - return >t->ttm.ttm; + return >t->ttm; } /** @@ -1313,10 +1352,11 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo, * Map the pages of a ttm_tt object to an address space visible * to the underlying device. */ -static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm, - struct ttm_operation_ctx *ctx) +static int amdgpu_ttm_tt_populate(struct ttm_bo_device *bdev, + struct ttm_tt *ttm, + struct ttm_operation_ctx *ctx) { - struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); + struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); struct amdgpu_ttm_tt *gtt = (void *)ttm; /* user pages are bound by amdgpu_ttm_tt_pin_userptr() */ @@ -1326,7 +1366,6 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm, return -ENOMEM; ttm->page_flags |= TTM_PAGE_FLAG_SG; - ttm->state = tt_unbound; return 0; } @@ -1346,19 +1385,10 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm, drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages, gtt->ttm.dma_address, ttm->num_pages); - ttm->state = tt_unbound; return 0; } -#ifdef CONFIG_SWIOTLB - if (adev->need_swiotlb && swiotlb_nr_tbl()) { - return ttm_dma_populate(>t->ttm, adev->dev, ctx); - } -#endif - - /* fall back to generic helper to populate the page array - * and map them to the device */ - return ttm_populate_and_map_pages(adev->dev, >t->ttm, ctx); + return ttm_pool_alloc(&adev->mman.bdev.pool, ttm, ctx); } /** @@ -1367,7 +1397,8 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm, * Unmaps pages of a ttm_tt object from the device address space and * unpopulates the page array backing it. */ -static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm) +static void amdgpu_ttm_tt_unpopulate(struct ttm_bo_device *bdev, + struct ttm_tt *ttm) { struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_device *adev; @@ -1391,17 +1422,8 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm) if (ttm->page_flags & TTM_PAGE_FLAG_SG) return; - adev = amdgpu_ttm_adev(ttm->bdev); - -#ifdef CONFIG_SWIOTLB - if (adev->need_swiotlb && swiotlb_nr_tbl()) { - ttm_dma_unpopulate(>t->ttm, adev->dev); - return; - } -#endif - - /* fall back to generic helper to unmap and unpopulate array */ - ttm_unmap_and_unpopulate_pages(adev->dev, >t->ttm); + adev = amdgpu_ttm_adev(bdev); + return ttm_pool_free(&adev->mman.bdev.pool, ttm); } /** @@ -1472,7 +1494,7 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start, /* Return false if no part of the ttm_tt object lies within * the range */ - size = (unsigned long)gtt->ttm.ttm.num_pages * PAGE_SIZE; + size = (unsigned long)gtt->ttm.num_pages * PAGE_SIZE; if (gtt->userptr > end || gtt->userptr + size <= start) return false; @@ -1523,7 +1545,7 @@ uint64_t amdgpu_ttm_tt_pde_flags(struct ttm_tt *ttm, struct ttm_resource *mem) if (mem && mem->mem_type == TTM_PL_TT) { flags |= AMDGPU_PTE_SYSTEM; - if (ttm->caching_state == tt_cached) + if (ttm->caching == ttm_cached) flags |= AMDGPU_PTE_SNOOPED; } @@ -1693,17 +1715,23 @@ static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo, return ret; } +static void +amdgpu_bo_delete_mem_notify(struct ttm_buffer_object *bo) +{ + amdgpu_bo_move_notify(bo, false, NULL); +} + static struct ttm_bo_driver amdgpu_bo_driver = { .ttm_tt_create = &amdgpu_ttm_tt_create, .ttm_tt_populate = &amdgpu_ttm_tt_populate, .ttm_tt_unpopulate = &amdgpu_ttm_tt_unpopulate, + .ttm_tt_destroy = &amdgpu_ttm_backend_destroy, .eviction_valuable = amdgpu_ttm_bo_eviction_valuable, .evict_flags = &amdgpu_evict_flags, .move = &amdgpu_bo_move, .verify_access = &amdgpu_verify_access, - .move_notify = &amdgpu_bo_move_notify, + .delete_mem_notify = &amdgpu_bo_delete_mem_notify, .release_notify = &amdgpu_bo_release_notify, - .fault_reserve_notify = &amdgpu_bo_fault_reserve_notify, .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_pfn = amdgpu_ttm_io_mem_pfn, .access_memory = &amdgpu_ttm_access_memory, @@ -1875,10 +1903,10 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) mutex_init(&adev->mman.gtt_window_lock); /* No others user of address space so set it to 0 */ - r = ttm_bo_device_init(&adev->mman.bdev, - &amdgpu_bo_driver, + r = ttm_bo_device_init(&adev->mman.bdev, &amdgpu_bo_driver, adev->dev, adev_to_drm(adev)->anon_inode->i_mapping, adev_to_drm(adev)->vma_offset_manager, + adev->need_swiotlb, dma_addressing_limited(adev->dev)); if (r) { DRM_ERROR("failed initializing buffer object driver(%d).\n", r); @@ -1886,9 +1914,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) } adev->mman.initialized = true; - /* We opt to avoid OOM on system pages allocations */ - adev->mman.bdev.no_retry = true; - /* Initialize VRAM pool with all of VRAM divided into pages */ r = amdgpu_vram_mgr_init(adev); if (r) { @@ -2083,15 +2108,48 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) adev->mman.buffer_funcs_enabled = enable; } +static vm_fault_t amdgpu_ttm_fault(struct vm_fault *vmf) +{ + struct ttm_buffer_object *bo = vmf->vma->vm_private_data; + vm_fault_t ret; + + ret = ttm_bo_vm_reserve(bo, vmf); + if (ret) + return ret; + + ret = amdgpu_bo_fault_reserve_notify(bo); + if (ret) + goto unlock; + + ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot, + TTM_BO_VM_NUM_PREFAULT, 1); + if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) + return ret; + +unlock: + dma_resv_unlock(bo->base.resv); + return ret; +} + +static struct vm_operations_struct amdgpu_ttm_vm_ops = { + .fault = amdgpu_ttm_fault, + .open = ttm_bo_vm_open, + .close = ttm_bo_vm_close, + .access = ttm_bo_vm_access +}; + int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma) { struct drm_file *file_priv = filp->private_data; struct amdgpu_device *adev = drm_to_adev(file_priv->minor->dev); + int r; - if (adev == NULL) - return -EINVAL; + r = ttm_bo_mmap(filp, vma, &adev->mman.bdev); + if (unlikely(r != 0)) + return r; - return ttm_bo_mmap(filp, vma, &adev->mman.bdev); + vma->vm_ops = &amdgpu_ttm_vm_ops; + return 0; } int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, @@ -2275,16 +2333,22 @@ static int amdgpu_mm_dump_table(struct seq_file *m, void *data) return 0; } +static int amdgpu_ttm_pool_debugfs(struct seq_file *m, void *data) +{ + struct drm_info_node *node = (struct drm_info_node *)m->private; + struct drm_device *dev = node->minor->dev; + struct amdgpu_device *adev = drm_to_adev(dev); + + return ttm_pool_debugfs(&adev->mman.bdev.pool, m); +} + static const struct drm_info_list amdgpu_ttm_debugfs_list[] = { {"amdgpu_vram_mm", amdgpu_mm_dump_table, 0, (void *)TTM_PL_VRAM}, {"amdgpu_gtt_mm", amdgpu_mm_dump_table, 0, (void *)TTM_PL_TT}, {"amdgpu_gds_mm", amdgpu_mm_dump_table, 0, (void *)AMDGPU_PL_GDS}, {"amdgpu_gws_mm", amdgpu_mm_dump_table, 0, (void *)AMDGPU_PL_GWS}, {"amdgpu_oa_mm", amdgpu_mm_dump_table, 0, (void *)AMDGPU_PL_OA}, - {"ttm_page_pool", ttm_page_alloc_debugfs, 0, NULL}, -#ifdef CONFIG_SWIOTLB - {"ttm_dma_page_pool", ttm_dma_page_alloc_debugfs, 0, NULL} -#endif + {"ttm_page_pool", amdgpu_ttm_pool_debugfs, 0, NULL}, }; /** @@ -2577,12 +2641,6 @@ int amdgpu_ttm_debugfs_init(struct amdgpu_device *adev) } count = ARRAY_SIZE(amdgpu_ttm_debugfs_list); - -#ifdef CONFIG_SWIOTLB - if (!(adev->need_swiotlb && swiotlb_nr_tbl())) - --count; -#endif - return amdgpu_debugfs_add_files(adev, amdgpu_ttm_debugfs_list, count); #else return 0; |