diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
-rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 636 |
1 files changed, 338 insertions, 298 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index fb0d8bffdce2..57277b1cf183 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -43,12 +43,14 @@ #include <linux/sizes.h> #include <linux/module.h> +#include <drm/drm_drv.h> #include <drm/ttm/ttm_bo_api.h> #include <drm/ttm/ttm_bo_driver.h> #include <drm/ttm/ttm_placement.h> #include <drm/ttm/ttm_range_manager.h> #include <drm/amdgpu_drm.h> +#include <drm/drm_drv.h> #include "amdgpu.h" #include "amdgpu_object.h" @@ -115,7 +117,7 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo, } abo = ttm_to_amdgpu_bo(bo); - if (abo->flags & AMDGPU_AMDKFD_CREATE_SVM_BO) { + if (abo->flags & AMDGPU_GEM_CREATE_DISCARDABLE) { placement->num_placement = 0; placement->num_busy_placement = 0; return; @@ -169,10 +171,10 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo, * @bo: buffer object to map * @mem: memory object to map * @mm_cur: range to map - * @num_pages: number of pages to map * @window: which GART window to use * @ring: DMA ring to use for the copy * @tmz: if we should setup a TMZ enabled mapping + * @size: in number of bytes to map, out number of bytes mapped * @addr: resulting address inside the MC address space * * Setup one of the GART windows to access a specific piece of memory or return @@ -181,15 +183,14 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo, static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, struct ttm_resource *mem, struct amdgpu_res_cursor *mm_cur, - unsigned num_pages, unsigned window, - struct amdgpu_ring *ring, bool tmz, - uint64_t *addr) + unsigned window, struct amdgpu_ring *ring, + bool tmz, uint64_t *size, uint64_t *addr) { struct amdgpu_device *adev = ring->adev; - struct amdgpu_job *job; - unsigned num_dw, num_bytes; - struct dma_fence *fence; + unsigned offset, num_pages, num_dw, num_bytes; uint64_t src_addr, dst_addr; + struct dma_fence *fence; + struct amdgpu_job *job; void *cpu_addr; uint64_t flags; unsigned int i; @@ -197,7 +198,9 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, BUG_ON(adev->mman.buffer_funcs->copy_max_bytes < AMDGPU_GTT_MAX_TRANSFER_SIZE * 8); - BUG_ON(mem->mem_type == AMDGPU_PL_PREEMPT); + + if (WARN_ON(mem->mem_type == AMDGPU_PL_PREEMPT)) + return -EINVAL; /* Map only what can't be accessed directly */ if (!tmz && mem->start != AMDGPU_BO_INVALID_OFFSET) { @@ -206,10 +209,22 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, return 0; } + + /* + * If start begins at an offset inside the page, then adjust the size + * and addr accordingly + */ + offset = mm_cur->start & ~PAGE_MASK; + + num_pages = PFN_UP(*size + offset); + num_pages = min_t(uint32_t, num_pages, AMDGPU_GTT_MAX_TRANSFER_SIZE); + + *size = min(*size, (uint64_t)num_pages * PAGE_SIZE - offset); + *addr = adev->gmc.gart_start; *addr += (u64)window * AMDGPU_GTT_MAX_TRANSFER_SIZE * AMDGPU_GPU_PAGE_SIZE; - *addr += mm_cur->start & ~PAGE_MASK; + *addr += offset; num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8); num_bytes = num_pages * 8 * AMDGPU_GPU_PAGES_IN_CPU_PAGE; @@ -240,10 +255,7 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, dma_addr_t *dma_addr; dma_addr = &bo->ttm->dma_address[mm_cur->start >> PAGE_SHIFT]; - r = amdgpu_gart_map(adev, 0, num_pages, dma_addr, flags, - cpu_addr); - if (r) - goto error_free; + amdgpu_gart_map(adev, 0, num_pages, dma_addr, flags, cpu_addr); } else { dma_addr_t dma_address; @@ -251,11 +263,8 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, dma_address += adev->vm_manager.vram_base_offset; for (i = 0; i < num_pages; ++i) { - r = amdgpu_gart_map(adev, i << PAGE_SHIFT, 1, - &dma_address, flags, cpu_addr); - if (r) - goto error_free; - + amdgpu_gart_map(adev, i << PAGE_SHIFT, 1, &dma_address, + flags, cpu_addr); dma_address += PAGE_SIZE; } } @@ -296,9 +305,6 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, struct dma_resv *resv, struct dma_fence **f) { - const uint32_t GTT_MAX_BYTES = (AMDGPU_GTT_MAX_TRANSFER_SIZE * - AMDGPU_GPU_PAGE_SIZE); - struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; struct amdgpu_res_cursor src_mm, dst_mm; struct dma_fence *fence = NULL; @@ -314,29 +320,20 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, mutex_lock(&adev->mman.gtt_window_lock); while (src_mm.remaining) { - uint32_t src_page_offset = src_mm.start & ~PAGE_MASK; - uint32_t dst_page_offset = dst_mm.start & ~PAGE_MASK; + uint64_t from, to, cur_size; struct dma_fence *next; - uint32_t cur_size; - uint64_t from, to; - /* Copy size cannot exceed GTT_MAX_BYTES. So if src or dst - * begins at an offset, then adjust the size accordingly - */ - cur_size = max(src_page_offset, dst_page_offset); - cur_size = min(min3(src_mm.size, dst_mm.size, size), - (uint64_t)(GTT_MAX_BYTES - cur_size)); + /* Never copy more than 256MiB at once to avoid a timeout */ + cur_size = min3(src_mm.size, dst_mm.size, 256ULL << 20); /* Map src to window 0 and dst to window 1. */ r = amdgpu_ttm_map_buffer(src->bo, src->mem, &src_mm, - PFN_UP(cur_size + src_page_offset), - 0, ring, tmz, &from); + 0, ring, tmz, &cur_size, &from); if (r) goto error; r = amdgpu_ttm_map_buffer(dst->bo, dst->mem, &dst_mm, - PFN_UP(cur_size + dst_page_offset), - 1, ring, tmz, &to); + 1, ring, tmz, &cur_size, &to); if (r) goto error; @@ -395,8 +392,7 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo, (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) { struct dma_fence *wipe_fence = NULL; - r = amdgpu_fill_buffer(ttm_to_amdgpu_bo(bo), AMDGPU_POISON, - NULL, &wipe_fence); + r = amdgpu_fill_buffer(abo, AMDGPU_POISON, NULL, &wipe_fence); if (r) { goto error; } else if (wipe_fence) { @@ -428,8 +424,9 @@ error: static bool amdgpu_mem_visible(struct amdgpu_device *adev, struct ttm_resource *mem) { - uint64_t mem_size = (u64)mem->num_pages << PAGE_SHIFT; + u64 mem_size = (u64)mem->num_pages << PAGE_SHIFT; struct amdgpu_res_cursor cursor; + u64 end; if (mem->mem_type == TTM_PL_SYSTEM || mem->mem_type == TTM_PL_TT) @@ -438,12 +435,21 @@ static bool amdgpu_mem_visible(struct amdgpu_device *adev, return false; amdgpu_res_first(mem, 0, mem_size, &cursor); + end = cursor.start + cursor.size; + while (cursor.remaining) { + amdgpu_res_next(&cursor, cursor.size); - /* ttm_resource_ioremap only supports contiguous memory */ - if (cursor.size != mem_size) - return false; + if (!cursor.remaining) + break; + + /* ttm_resource_ioremap only supports contiguous memory */ + if (end != cursor.start) + return false; + + end = cursor.start + cursor.size; + } - return cursor.start + cursor.size <= adev->gmc.visible_vram_size; + return end <= adev->gmc.visible_vram_size; } /* @@ -475,7 +481,8 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict, adev = amdgpu_ttm_adev(bo->bdev); - if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) { + if (!old_mem || (old_mem->mem_type == TTM_PL_SYSTEM && + bo->ttm == NULL)) { ttm_bo_move_null(bo, new_mem); goto out; } @@ -641,6 +648,8 @@ struct amdgpu_ttm_tt { #endif }; +#define ttm_to_amdgpu_ttm_tt(ptr) container_of(ptr, struct amdgpu_ttm_tt, ttm) + #ifdef CONFIG_DRM_AMDGPU_USERPTR /* * amdgpu_ttm_tt_get_user_pages - get device accessible pages that back user @@ -652,7 +661,7 @@ struct amdgpu_ttm_tt { int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, struct page **pages) { struct ttm_tt *ttm = bo->tbo.ttm; - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); unsigned long start = gtt->userptr; struct vm_area_struct *vma; struct mm_struct *mm; @@ -706,7 +715,7 @@ out_unlock: */ bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm) { - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); bool r = false; if (!gtt || !gtt->userptr) @@ -755,7 +764,7 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_device *bdev, struct ttm_tt *ttm) { struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY); enum dma_data_direction direction = write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE; @@ -792,7 +801,7 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_device *bdev, struct ttm_tt *ttm) { struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY); enum dma_data_direction direction = write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE; @@ -820,14 +829,13 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_device *bdev, #endif } -static int amdgpu_ttm_gart_bind(struct amdgpu_device *adev, - struct ttm_buffer_object *tbo, - uint64_t flags) +static void amdgpu_ttm_gart_bind(struct amdgpu_device *adev, + struct ttm_buffer_object *tbo, + uint64_t flags) { struct amdgpu_bo *abo = ttm_to_amdgpu_bo(tbo); struct ttm_tt *ttm = tbo->ttm; - struct amdgpu_ttm_tt *gtt = (void *)ttm; - int r; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); if (amdgpu_bo_encrypted(abo)) flags |= AMDGPU_PTE_TMZ; @@ -835,10 +843,8 @@ static int amdgpu_ttm_gart_bind(struct amdgpu_device *adev, if (abo->flags & AMDGPU_GEM_CREATE_CP_MQD_GFX9) { uint64_t page_idx = 1; - r = amdgpu_gart_bind(adev, gtt->offset, page_idx, - gtt->ttm.dma_address, flags); - if (r) - goto gart_bind_fail; + amdgpu_gart_bind(adev, gtt->offset, page_idx, + gtt->ttm.dma_address, flags); /* The memory type of the first page defaults to UC. Now * modify the memory type to NC from the second page of @@ -847,21 +853,13 @@ static int amdgpu_ttm_gart_bind(struct amdgpu_device *adev, flags &= ~AMDGPU_PTE_MTYPE_VG10_MASK; flags |= AMDGPU_PTE_MTYPE_VG10(AMDGPU_MTYPE_NC); - r = amdgpu_gart_bind(adev, - gtt->offset + (page_idx << PAGE_SHIFT), - ttm->num_pages - page_idx, - &(gtt->ttm.dma_address[page_idx]), flags); + amdgpu_gart_bind(adev, gtt->offset + (page_idx << PAGE_SHIFT), + ttm->num_pages - page_idx, + &(gtt->ttm.dma_address[page_idx]), flags); } else { - r = amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages, - gtt->ttm.dma_address, flags); + amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages, + gtt->ttm.dma_address, flags); } - -gart_bind_fail: - if (r) - DRM_ERROR("failed to bind %u pages at 0x%08llX\n", - ttm->num_pages, gtt->offset); - - return r; } /* @@ -875,9 +873,9 @@ static int amdgpu_ttm_backend_bind(struct ttm_device *bdev, struct ttm_resource *bo_mem) { struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); - struct amdgpu_ttm_tt *gtt = (void*)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); uint64_t flags; - int r = 0; + int r; if (!bo_mem) return -EINVAL; @@ -924,14 +922,10 @@ static int amdgpu_ttm_backend_bind(struct ttm_device *bdev, /* bind pages into GART page tables */ gtt->offset = (u64)bo_mem->start << PAGE_SHIFT; - r = amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages, - gtt->ttm.dma_address, flags); - - if (r) - DRM_ERROR("failed to bind %u pages at 0x%08llX\n", - ttm->num_pages, gtt->offset); + amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages, + gtt->ttm.dma_address, flags); gtt->bound = true; - return r; + return 0; } /* @@ -946,7 +940,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo) { struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev); struct ttm_operation_ctx ctx = { false, false }; - struct amdgpu_ttm_tt *gtt = (void *)bo->ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(bo->ttm); struct ttm_placement placement; struct ttm_place placements; struct ttm_resource *tmp; @@ -981,12 +975,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo) /* Bind pages */ gtt->offset = (u64)tmp->start << PAGE_SHIFT; - r = amdgpu_ttm_gart_bind(adev, bo, flags); - if (unlikely(r)) { - ttm_resource_free(bo, &tmp); - return r; - } - + amdgpu_ttm_gart_bind(adev, bo, flags); amdgpu_gart_invalidate_tlb(adev); ttm_resource_free(bo, &bo->resource); ttm_bo_assign_mem(bo, tmp); @@ -1000,19 +989,16 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo) * Called by amdgpu_gtt_mgr_recover() from amdgpu_device_reset() to * rebind GTT pages during a GPU reset. */ -int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo) +void amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo) { struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev); uint64_t flags; - int r; if (!tbo->ttm) - return 0; + return; flags = amdgpu_ttm_tt_pte_flags(adev, tbo->ttm, tbo->resource); - r = amdgpu_ttm_gart_bind(adev, tbo, flags); - - return r; + amdgpu_ttm_gart_bind(adev, tbo, flags); } /* @@ -1025,8 +1011,7 @@ static void amdgpu_ttm_backend_unbind(struct ttm_device *bdev, struct ttm_tt *ttm) { struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); - struct amdgpu_ttm_tt *gtt = (void *)ttm; - int r; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); /* if the pages have userptr pinning then clear that first */ if (gtt->userptr) { @@ -1046,17 +1031,14 @@ static void amdgpu_ttm_backend_unbind(struct ttm_device *bdev, return; /* 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 %u pages at 0x%08llX\n", - gtt->ttm.num_pages, gtt->offset); + amdgpu_gart_unbind(adev, gtt->offset, ttm->num_pages); gtt->bound = false; } static void amdgpu_ttm_backend_destroy(struct ttm_device *bdev, struct ttm_tt *ttm) { - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); if (gtt->usertask) put_task_struct(gtt->usertask); @@ -1110,7 +1092,7 @@ static int amdgpu_ttm_tt_populate(struct ttm_device *bdev, struct ttm_operation_ctx *ctx) { struct amdgpu_device *adev = amdgpu_ttm_adev(bdev); - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); pgoff_t i; int ret; @@ -1144,7 +1126,7 @@ static int amdgpu_ttm_tt_populate(struct ttm_device *bdev, static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev, struct ttm_tt *ttm) { - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); struct amdgpu_device *adev; pgoff_t i; @@ -1168,6 +1150,26 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev, } /** + * amdgpu_ttm_tt_get_userptr - Return the userptr GTT ttm_tt for the current + * task + * + * @tbo: The ttm_buffer_object that contains the userptr + * @user_addr: The returned value + */ +int amdgpu_ttm_tt_get_userptr(const struct ttm_buffer_object *tbo, + uint64_t *user_addr) +{ + struct amdgpu_ttm_tt *gtt; + + if (!tbo->ttm) + return -EINVAL; + + gtt = (void *)tbo->ttm; + *user_addr = gtt->userptr; + return 0; +} + +/** * amdgpu_ttm_tt_set_userptr - Initialize userptr GTT ttm_tt for the current * task * @@ -1193,7 +1195,7 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_buffer_object *bo, /* Set TTM_TT_FLAG_EXTERNAL before populate but after create. */ bo->ttm->page_flags |= TTM_TT_FLAG_EXTERNAL; - gtt = (void *)bo->ttm; + gtt = ttm_to_amdgpu_ttm_tt(bo->ttm); gtt->userptr = addr; gtt->userflags = flags; @@ -1210,7 +1212,7 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_buffer_object *bo, */ struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm) { - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); if (gtt == NULL) return NULL; @@ -1229,7 +1231,7 @@ struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm) bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start, unsigned long end, unsigned long *userptr) { - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); unsigned long size; if (gtt == NULL || !gtt->userptr) @@ -1252,7 +1254,7 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start, */ bool amdgpu_ttm_tt_is_userptr(struct ttm_tt *ttm) { - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); if (gtt == NULL || !gtt->userptr) return false; @@ -1265,7 +1267,7 @@ bool amdgpu_ttm_tt_is_userptr(struct ttm_tt *ttm) */ bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm) { - struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); if (gtt == NULL) return false; @@ -1338,11 +1340,12 @@ uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm, static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo, const struct ttm_place *place) { - unsigned long num_pages = bo->resource->num_pages; struct dma_resv_iter resv_cursor; - struct amdgpu_res_cursor cursor; struct dma_fence *f; + if (!amdgpu_bo_is_amdgpu_bo(bo)) + return ttm_bo_eviction_valuable(bo, place); + /* Swapout? */ if (bo->resource->mem_type == TTM_PL_SYSTEM) return true; @@ -1355,45 +1358,26 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo, * If true, then return false as any KFD process needs all its BOs to * be resident to run successfully */ - dma_resv_for_each_fence(&resv_cursor, bo->base.resv, true, f) { + dma_resv_for_each_fence(&resv_cursor, bo->base.resv, + DMA_RESV_USAGE_BOOKKEEP, f) { if (amdkfd_fence_check_mm(f, current->mm)) return false; } - switch (bo->resource->mem_type) { - case AMDGPU_PL_PREEMPT: - /* Preemptible BOs don't own system resources managed by the - * driver (pages, VRAM, GART space). They point to resources - * owned by someone else (e.g. pageable memory in user mode - * or a DMABuf). They are used in a preemptible context so we - * can guarantee no deadlocks and good QoS in case of MMU - * notifiers or DMABuf move notifiers from the resource owner. - */ + /* Preemptible BOs don't own system resources managed by the + * driver (pages, VRAM, GART space). They point to resources + * owned by someone else (e.g. pageable memory in user mode + * or a DMABuf). They are used in a preemptible context so we + * can guarantee no deadlocks and good QoS in case of MMU + * notifiers or DMABuf move notifiers from the resource owner. + */ + if (bo->resource->mem_type == AMDGPU_PL_PREEMPT) return false; - case TTM_PL_TT: - if (amdgpu_bo_is_amdgpu_bo(bo) && - amdgpu_bo_encrypted(ttm_to_amdgpu_bo(bo))) - return false; - return true; - case TTM_PL_VRAM: - /* Check each drm MM node individually */ - amdgpu_res_first(bo->resource, 0, (u64)num_pages << PAGE_SHIFT, - &cursor); - while (cursor.remaining) { - if (place->fpfn < PFN_DOWN(cursor.start + cursor.size) - && !(place->lpfn && - place->lpfn <= PFN_DOWN(cursor.start))) - return true; - - amdgpu_res_next(&cursor, cursor.size); - } + if (bo->resource->mem_type == TTM_PL_TT && + amdgpu_bo_encrypted(ttm_to_amdgpu_bo(bo))) return false; - default: - break; - } - return ttm_bo_eviction_valuable(bo, place); } @@ -1432,6 +1416,63 @@ static void amdgpu_ttm_vram_mm_access(struct amdgpu_device *adev, loff_t pos, } } +static int amdgpu_ttm_access_memory_sdma(struct ttm_buffer_object *bo, + unsigned long offset, void *buf, int len, int write) +{ + struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo); + struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev); + struct amdgpu_res_cursor src_mm; + struct amdgpu_job *job; + struct dma_fence *fence; + uint64_t src_addr, dst_addr; + unsigned int num_dw; + int r, idx; + + if (len != PAGE_SIZE) + return -EINVAL; + + if (!adev->mman.sdma_access_ptr) + return -EACCES; + + if (!drm_dev_enter(adev_to_drm(adev), &idx)) + return -ENODEV; + + if (write) + memcpy(adev->mman.sdma_access_ptr, buf, len); + + num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8); + r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, AMDGPU_IB_POOL_DELAYED, &job); + if (r) + goto out; + + amdgpu_res_first(abo->tbo.resource, offset, len, &src_mm); + src_addr = amdgpu_ttm_domain_start(adev, bo->resource->mem_type) + src_mm.start; + dst_addr = amdgpu_bo_gpu_offset(adev->mman.sdma_access_bo); + if (write) + swap(src_addr, dst_addr); + + amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_addr, dst_addr, PAGE_SIZE, false); + + amdgpu_ring_pad_ib(adev->mman.buffer_funcs_ring, &job->ibs[0]); + WARN_ON(job->ibs[0].length_dw > num_dw); + + r = amdgpu_job_submit(job, &adev->mman.entity, AMDGPU_FENCE_OWNER_UNDEFINED, &fence); + if (r) { + amdgpu_job_free(job); + goto out; + } + + if (!dma_fence_wait_timeout(fence, false, adev->sdma_timeout)) + r = -ETIMEDOUT; + dma_fence_put(fence); + + if (!(r || write)) + memcpy(buf, adev->mman.sdma_access_ptr, len); +out: + drm_dev_exit(idx); + return r; +} + /** * amdgpu_ttm_access_memory - Read or Write memory that backs a buffer object. * @@ -1456,6 +1497,10 @@ static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo, if (bo->resource->mem_type != TTM_PL_VRAM) return -EIO; + if (amdgpu_device_has_timeouts_enabled(adev) && + !amdgpu_ttm_access_memory_sdma(bo, offset, buf, len, write)) + return len; + amdgpu_res_first(bo->resource, offset, len, &cursor); while (cursor.remaining) { size_t count, size = cursor.size; @@ -1497,7 +1542,6 @@ static struct ttm_device_funcs amdgpu_bo_driver = { .io_mem_reserve = &amdgpu_ttm_io_mem_reserve, .io_mem_pfn = amdgpu_ttm_io_mem_pfn, .access_memory = &amdgpu_ttm_access_memory, - .del_from_lru_notify = &amdgpu_vm_del_from_lru_notify }; /* @@ -1748,18 +1792,26 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) DRM_INFO("amdgpu: %uM of VRAM memory ready\n", (unsigned) (adev->gmc.real_vram_size / (1024 * 1024))); - /* Compute GTT size, either bsaed on 3/4th the size of RAM size + /* Compute GTT size, either based on 1/2 the size of RAM size * or whatever the user passed on module init */ if (amdgpu_gtt_size == -1) { struct sysinfo si; si_meminfo(&si); - gtt_size = min(max((AMDGPU_DEFAULT_GTT_SIZE_MB << 20), - adev->gmc.mc_vram_size), - ((uint64_t)si.totalram * si.mem_unit * 3/4)); - } - else + /* Certain GL unit tests for large textures can cause problems + * with the OOM killer since there is no way to link this memory + * to a process. This was originally mitigated (but not necessarily + * eliminated) by limiting the GTT size. The problem is this limit + * is often too low for many modern games so just make the limit 1/2 + * of system memory which aligns with TTM. The OOM accounting needs + * to be addressed, but we shouldn't prevent common 3D applications + * from being usable just to potentially mitigate that corner case. + */ + gtt_size = max((AMDGPU_DEFAULT_GTT_SIZE_MB << 20), + (u64)si.totalram * si.mem_unit / 2); + } else { gtt_size = (uint64_t)amdgpu_gtt_size << 20; + } /* Initialize GTT memory pool */ r = amdgpu_gtt_mgr_init(adev, gtt_size); @@ -1796,6 +1848,12 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) return r; } + if (amdgpu_bo_create_kernel(adev, PAGE_SIZE, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_GTT, + &adev->mman.sdma_access_bo, NULL, + &adev->mman.sdma_access_ptr)) + DRM_WARN("Debug VRAM access will use slowpath MM access\n"); + return 0; } @@ -1804,6 +1862,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) */ void amdgpu_ttm_fini(struct amdgpu_device *adev) { + int idx; if (!adev->mman.initialized) return; @@ -1816,8 +1875,19 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev) if (adev->mman.stolen_reserved_size) amdgpu_bo_free_kernel(&adev->mman.stolen_reserved_memory, NULL, NULL); + amdgpu_bo_free_kernel(&adev->mman.sdma_access_bo, NULL, + &adev->mman.sdma_access_ptr); amdgpu_ttm_fw_reserve_vram_fini(adev); + if (drm_dev_enter(adev_to_drm(adev), &idx)) { + + if (adev->mman.aper_base_kaddr) + iounmap(adev->mman.aper_base_kaddr); + adev->mman.aper_base_kaddr = NULL; + + drm_dev_exit(idx); + } + amdgpu_vram_mgr_fini(adev); amdgpu_gtt_mgr_fini(adev); amdgpu_preempt_mgr_fini(adev); @@ -1873,27 +1943,59 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) size = adev->gmc.real_vram_size; else size = adev->gmc.visible_vram_size; - man->size = size >> PAGE_SHIFT; + man->size = size; adev->mman.buffer_funcs_enabled = enable; } +static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev, + bool direct_submit, + unsigned int num_dw, + struct dma_resv *resv, + bool vm_needs_flush, + struct amdgpu_job **job) +{ + enum amdgpu_ib_pool_type pool = direct_submit ? + AMDGPU_IB_POOL_DIRECT : + AMDGPU_IB_POOL_DELAYED; + int r; + + r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, pool, job); + if (r) + return r; + + if (vm_needs_flush) { + (*job)->vm_pd_addr = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo ? + adev->gmc.pdb0_bo : + adev->gart.bo); + (*job)->vm_needs_flush = true; + } + if (resv) { + r = amdgpu_sync_resv(adev, &(*job)->sync, resv, + AMDGPU_SYNC_ALWAYS, + AMDGPU_FENCE_OWNER_UNDEFINED); + if (r) { + DRM_ERROR("sync failed (%d).\n", r); + amdgpu_job_free(*job); + return r; + } + } + return 0; +} + int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, uint64_t dst_offset, uint32_t byte_count, struct dma_resv *resv, struct dma_fence **fence, bool direct_submit, bool vm_needs_flush, bool tmz) { - enum amdgpu_ib_pool_type pool = direct_submit ? AMDGPU_IB_POOL_DIRECT : - AMDGPU_IB_POOL_DELAYED; struct amdgpu_device *adev = ring->adev; + unsigned num_loops, num_dw; struct amdgpu_job *job; - uint32_t max_bytes; - unsigned num_loops, num_dw; unsigned i; int r; - if (direct_submit && !ring->sched.ready) { + if (!direct_submit && !ring->sched.ready) { DRM_ERROR("Trying to move memory with ring turned off.\n"); return -EINVAL; } @@ -1901,26 +2003,11 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, max_bytes = adev->mman.buffer_funcs->copy_max_bytes; num_loops = DIV_ROUND_UP(byte_count, max_bytes); num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->copy_num_dw, 8); - - r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, pool, &job); + r = amdgpu_ttm_prepare_job(adev, direct_submit, num_dw, + resv, vm_needs_flush, &job); if (r) return r; - if (vm_needs_flush) { - job->vm_pd_addr = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo ? - adev->gmc.pdb0_bo : adev->gart.bo); - job->vm_needs_flush = true; - } - if (resv) { - r = amdgpu_sync_resv(adev, &job->sync, resv, - AMDGPU_SYNC_ALWAYS, - AMDGPU_FENCE_OWNER_UNDEFINED); - if (r) { - DRM_ERROR("sync failed (%d).\n", r); - goto error_free; - } - } - for (i = 0; i < num_loops; i++) { uint32_t cur_size_in_bytes = min(byte_count, max_bytes); @@ -1950,77 +2037,35 @@ error_free: return r; } -int amdgpu_fill_buffer(struct amdgpu_bo *bo, - uint32_t src_data, - struct dma_resv *resv, - struct dma_fence **fence) +static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data, + uint64_t dst_addr, uint32_t byte_count, + struct dma_resv *resv, + struct dma_fence **fence, + bool vm_needs_flush) { - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - uint32_t max_bytes = adev->mman.buffer_funcs->fill_max_bytes; - struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; - - struct amdgpu_res_cursor cursor; + struct amdgpu_device *adev = ring->adev; unsigned int num_loops, num_dw; - uint64_t num_bytes; - struct amdgpu_job *job; + uint32_t max_bytes; + unsigned int i; int r; - if (!adev->mman.buffer_funcs_enabled) { - DRM_ERROR("Trying to clear memory with ring turned off.\n"); - return -EINVAL; - } - - if (bo->tbo.resource->mem_type == AMDGPU_PL_PREEMPT) { - DRM_ERROR("Trying to clear preemptible memory.\n"); - return -EINVAL; - } - - if (bo->tbo.resource->mem_type == TTM_PL_TT) { - r = amdgpu_ttm_alloc_gart(&bo->tbo); - if (r) - return r; - } - - num_bytes = bo->tbo.resource->num_pages << PAGE_SHIFT; - num_loops = 0; - - amdgpu_res_first(bo->tbo.resource, 0, num_bytes, &cursor); - while (cursor.remaining) { - num_loops += DIV_ROUND_UP_ULL(cursor.size, max_bytes); - amdgpu_res_next(&cursor, cursor.size); - } - num_dw = num_loops * adev->mman.buffer_funcs->fill_num_dw; - - /* for IB padding */ - num_dw += 64; - - r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, AMDGPU_IB_POOL_DELAYED, - &job); + max_bytes = adev->mman.buffer_funcs->fill_max_bytes; + num_loops = DIV_ROUND_UP_ULL(byte_count, max_bytes); + num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->fill_num_dw, 8); + r = amdgpu_ttm_prepare_job(adev, false, num_dw, resv, vm_needs_flush, + &job); if (r) return r; - if (resv) { - r = amdgpu_sync_resv(adev, &job->sync, resv, - AMDGPU_SYNC_ALWAYS, - AMDGPU_FENCE_OWNER_UNDEFINED); - if (r) { - DRM_ERROR("sync failed (%d).\n", r); - goto error_free; - } - } - - amdgpu_res_first(bo->tbo.resource, 0, num_bytes, &cursor); - while (cursor.remaining) { - uint32_t cur_size = min_t(uint64_t, cursor.size, max_bytes); - uint64_t dst_addr = cursor.start; + for (i = 0; i < num_loops; i++) { + uint32_t cur_size = min(byte_count, max_bytes); - dst_addr += amdgpu_ttm_domain_start(adev, - bo->tbo.resource->mem_type); amdgpu_emit_fill_buffer(adev, &job->ibs[0], src_data, dst_addr, cur_size); - amdgpu_res_next(&cursor, cur_size); + dst_addr += cur_size; + byte_count -= cur_size; } amdgpu_ring_pad_ib(ring, &job->ibs[0]); @@ -2037,6 +2082,55 @@ error_free: return r; } +int amdgpu_fill_buffer(struct amdgpu_bo *bo, + uint32_t src_data, + struct dma_resv *resv, + struct dma_fence **f) +{ + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); + struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; + struct dma_fence *fence = NULL; + struct amdgpu_res_cursor dst; + int r; + + if (!adev->mman.buffer_funcs_enabled) { + DRM_ERROR("Trying to clear memory with ring turned off.\n"); + return -EINVAL; + } + + amdgpu_res_first(bo->tbo.resource, 0, amdgpu_bo_size(bo), &dst); + + mutex_lock(&adev->mman.gtt_window_lock); + while (dst.remaining) { + struct dma_fence *next; + uint64_t cur_size, to; + + /* Never fill more than 256MiB at once to avoid timeouts */ + cur_size = min(dst.size, 256ULL << 20); + + r = amdgpu_ttm_map_buffer(&bo->tbo, bo->tbo.resource, &dst, + 1, ring, false, &cur_size, &to); + if (r) + goto error; + + r = amdgpu_ttm_fill_mem(ring, src_data, to, cur_size, resv, + &next, true); + if (r) + goto error; + + dma_fence_put(fence); + fence = next; + + amdgpu_res_next(&dst, cur_size); + } +error: + mutex_unlock(&adev->mman.gtt_window_lock); + if (f) + *f = dma_fence_get(fence); + dma_fence_put(fence); + return r; +} + /** * amdgpu_ttm_evict_resources - evict memory buffers * @adev: amdgpu device object @@ -2069,17 +2163,6 @@ int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type) #if defined(CONFIG_DEBUG_FS) -static int amdgpu_mm_vram_table_show(struct seq_file *m, void *unused) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)m->private; - struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, - TTM_PL_VRAM); - struct drm_printer p = drm_seq_file_printer(m); - - man->func->debug(man, &p); - return 0; -} - static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused) { struct amdgpu_device *adev = (struct amdgpu_device *)m->private; @@ -2087,55 +2170,6 @@ static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused) return ttm_pool_debugfs(&adev->mman.bdev.pool, m); } -static int amdgpu_mm_tt_table_show(struct seq_file *m, void *unused) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)m->private; - struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, - TTM_PL_TT); - struct drm_printer p = drm_seq_file_printer(m); - - man->func->debug(man, &p); - return 0; -} - -static int amdgpu_mm_gds_table_show(struct seq_file *m, void *unused) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)m->private; - struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, - AMDGPU_PL_GDS); - struct drm_printer p = drm_seq_file_printer(m); - - man->func->debug(man, &p); - return 0; -} - -static int amdgpu_mm_gws_table_show(struct seq_file *m, void *unused) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)m->private; - struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, - AMDGPU_PL_GWS); - struct drm_printer p = drm_seq_file_printer(m); - - man->func->debug(man, &p); - return 0; -} - -static int amdgpu_mm_oa_table_show(struct seq_file *m, void *unused) -{ - struct amdgpu_device *adev = (struct amdgpu_device *)m->private; - struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, - AMDGPU_PL_OA); - struct drm_printer p = drm_seq_file_printer(m); - - man->func->debug(man, &p); - return 0; -} - -DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_vram_table); -DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_tt_table); -DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_gds_table); -DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_gws_table); -DEFINE_SHOW_ATTRIBUTE(amdgpu_mm_oa_table); DEFINE_SHOW_ATTRIBUTE(amdgpu_ttm_page_pool); /* @@ -2345,17 +2379,23 @@ void amdgpu_ttm_debugfs_init(struct amdgpu_device *adev) &amdgpu_ttm_vram_fops, adev->gmc.mc_vram_size); debugfs_create_file("amdgpu_iomem", 0444, root, adev, &amdgpu_ttm_iomem_fops); - debugfs_create_file("amdgpu_vram_mm", 0444, root, adev, - &amdgpu_mm_vram_table_fops); - debugfs_create_file("amdgpu_gtt_mm", 0444, root, adev, - &amdgpu_mm_tt_table_fops); - debugfs_create_file("amdgpu_gds_mm", 0444, root, adev, - &amdgpu_mm_gds_table_fops); - debugfs_create_file("amdgpu_gws_mm", 0444, root, adev, - &amdgpu_mm_gws_table_fops); - debugfs_create_file("amdgpu_oa_mm", 0444, root, adev, - &amdgpu_mm_oa_table_fops); debugfs_create_file("ttm_page_pool", 0444, root, adev, &amdgpu_ttm_page_pool_fops); + ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev, + TTM_PL_VRAM), + root, "amdgpu_vram_mm"); + ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev, + TTM_PL_TT), + root, "amdgpu_gtt_mm"); + ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev, + AMDGPU_PL_GDS), + root, "amdgpu_gds_mm"); + ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev, + AMDGPU_PL_GWS), + root, "amdgpu_gws_mm"); + ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev, + AMDGPU_PL_OA), + root, "amdgpu_oa_mm"); + #endif } |