diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 122 |
1 files changed, 41 insertions, 81 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 9d568c16beb1..2b931e855abd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -188,7 +188,6 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, struct amdgpu_job *job; void *cpu_addr; uint64_t flags; - unsigned int i; int r; BUG_ON(adev->mman.buffer_funcs->copy_max_bytes < @@ -255,16 +254,9 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, dma_addr = &bo->ttm->dma_address[mm_cur->start >> PAGE_SHIFT]; amdgpu_gart_map(adev, 0, num_pages, dma_addr, flags, cpu_addr); } else { - dma_addr_t dma_address; - - dma_address = mm_cur->start; - dma_address += adev->vm_manager.vram_base_offset; + u64 pa = mm_cur->start + adev->vm_manager.vram_base_offset; - for (i = 0; i < num_pages; ++i) { - amdgpu_gart_map(adev, i << PAGE_SHIFT, 1, &dma_address, - flags, cpu_addr); - dma_address += PAGE_SIZE; - } + amdgpu_gart_map_vram_range(adev, pa, 0, num_pages, flags, cpu_addr); } dma_fence_put(amdgpu_job_submit(job)); @@ -286,12 +278,13 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, * move and different for a BO to BO copy. * */ -int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, - const struct amdgpu_copy_mem *src, - const struct amdgpu_copy_mem *dst, - uint64_t size, bool tmz, - struct dma_resv *resv, - struct dma_fence **f) +__attribute__((nonnull)) +static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, + const struct amdgpu_copy_mem *src, + const struct amdgpu_copy_mem *dst, + uint64_t size, bool tmz, + struct dma_resv *resv, + struct dma_fence **f) { struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; struct amdgpu_res_cursor src_mm, dst_mm; @@ -365,9 +358,7 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, } error: mutex_unlock(&adev->mman.gtt_window_lock); - if (f) - *f = dma_fence_get(fence); - dma_fence_put(fence); + *f = fence; return r; } @@ -706,10 +697,11 @@ struct amdgpu_ttm_tt { * memory and start HMM tracking CPU page table update * * Calling function must call amdgpu_ttm_tt_userptr_range_done() once and only - * once afterwards to stop HMM tracking + * once afterwards to stop HMM tracking. Its the caller responsibility to ensure + * that range is a valid memory and it is freed too. */ int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, - struct hmm_range **range) + struct amdgpu_hmm_range *range) { struct ttm_tt *ttm = bo->tbo.ttm; struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); @@ -719,9 +711,6 @@ int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo, bool readonly; int r = 0; - /* Make sure get_user_pages_done() can cleanup gracefully */ - *range = NULL; - mm = bo->notifier.mm; if (unlikely(!mm)) { DRM_DEBUG_DRIVER("BO is not registered?\n"); @@ -756,38 +745,6 @@ out_unlock: return r; } -/* amdgpu_ttm_tt_discard_user_pages - Discard range and pfn array allocations - */ -void amdgpu_ttm_tt_discard_user_pages(struct ttm_tt *ttm, - struct hmm_range *range) -{ - struct amdgpu_ttm_tt *gtt = (void *)ttm; - - if (gtt && gtt->userptr && range) - amdgpu_hmm_range_get_pages_done(range); -} - -/* - * amdgpu_ttm_tt_get_user_pages_done - stop HMM track the CPU page table change - * Check if the pages backing this ttm range have been invalidated - * - * Returns: true if pages are still valid - */ -bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm, - struct hmm_range *range) -{ - struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm); - - if (!gtt || !gtt->userptr || !range) - return false; - - DRM_DEBUG_DRIVER("user_pages_done 0x%llx pages 0x%x\n", - gtt->userptr, ttm->num_pages); - - WARN_ONCE(!range->hmm_pfns, "No user pages to check\n"); - - return !amdgpu_hmm_range_get_pages_done(range); -} #endif /* @@ -797,12 +754,12 @@ bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm, * that backs user memory and will ultimately be mapped into the device * address space. */ -void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct hmm_range *range) +void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct amdgpu_hmm_range *range) { unsigned long i; for (i = 0; i < ttm->num_pages; ++i) - ttm->pages[i] = range ? hmm_pfn_to_page(range->hmm_pfns[i]) : NULL; + ttm->pages[i] = range ? hmm_pfn_to_page(range->hmm_range.hmm_pfns[i]) : NULL; } /* @@ -1529,6 +1486,7 @@ static int amdgpu_ttm_access_memory_sdma(struct ttm_buffer_object *bo, if (r) goto out; + mutex_lock(&adev->mman.gtt_window_lock); amdgpu_res_first(abo->tbo.resource, offset, len, &src_mm); src_addr = amdgpu_ttm_domain_start(adev, bo->resource->mem_type) + src_mm.start; @@ -1543,6 +1501,7 @@ static int amdgpu_ttm_access_memory_sdma(struct ttm_buffer_object *bo, WARN_ON(job->ibs[0].length_dw > num_dw); fence = amdgpu_job_submit(job); + mutex_unlock(&adev->mman.gtt_window_lock); if (!dma_fence_wait_timeout(fence, false, adev->sdma_timeout)) r = -ETIMEDOUT; @@ -1804,18 +1763,14 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev) ctx->init = PSP_MEM_TRAIN_RESERVE_SUCCESS; } - if (!adev->gmc.is_app_apu) { - ret = amdgpu_bo_create_kernel_at( - adev, adev->gmc.real_vram_size - reserve_size, - reserve_size, &adev->mman.fw_reserved_memory, NULL); - if (ret) { - dev_err(adev->dev, "alloc tmr failed(%d)!\n", ret); - amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory, - NULL, NULL); - return ret; - } - } else { - DRM_DEBUG_DRIVER("backdoor fw loading path for PSP TMR, no reservation needed\n"); + ret = amdgpu_bo_create_kernel_at( + adev, adev->gmc.real_vram_size - reserve_size, reserve_size, + &adev->mman.fw_reserved_memory, NULL); + if (ret) { + dev_err(adev->dev, "alloc tmr failed(%d)!\n", ret); + amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory, NULL, + NULL); + return ret; } return 0; @@ -1837,7 +1792,7 @@ static int amdgpu_ttm_pools_init(struct amdgpu_device *adev) for (i = 0; i < adev->gmc.num_mem_partitions; i++) { ttm_pool_init(&adev->mman.ttm_pools[i], adev->dev, adev->gmc.mem_partitions[i].numa.node, - false, false); + TTM_ALLOCATION_POOL_BENEFICIAL_ORDER(get_order(SZ_2M))); } return 0; } @@ -1930,8 +1885,11 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) r = ttm_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)); + (adev->need_swiotlb ? + TTM_ALLOCATION_POOL_USE_DMA_ALLOC : 0) | + (dma_addressing_limited(adev->dev) ? + TTM_ALLOCATION_POOL_USE_DMA32 : 0) | + TTM_ALLOCATION_POOL_BENEFICIAL_ORDER(get_order(SZ_2M))); if (r) { dev_err(adev->dev, "failed initializing buffer object driver(%d).\n", r); @@ -1980,19 +1938,19 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) return r; /* - *The reserved vram for driver must be pinned to the specified - *place on the VRAM, so reserve it early. + * The reserved VRAM for the driver must be pinned to a specific + * location in VRAM, so reserve it early. */ r = amdgpu_ttm_drv_reserve_vram_init(adev); if (r) return r; /* - * only NAVI10 and onwards ASIC support for IP discovery. - * If IP discovery enabled, a block of memory should be - * reserved for IP discovey. + * only NAVI10 and later ASICs support IP discovery. + * If IP discovery is enabled, a block of memory should be + * reserved for it. */ - if (adev->mman.discovery_bin) { + if (adev->discovery.reserve_tmr) { r = amdgpu_ttm_reserve_tmr(adev); if (r) return r; @@ -2229,8 +2187,10 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) } else { drm_sched_entity_destroy(&adev->mman.high_pr); drm_sched_entity_destroy(&adev->mman.low_pr); - dma_fence_put(man->move); - man->move = NULL; + /* Drop all the old fences since re-creating the scheduler entities + * will allocate new contexts. + */ + ttm_resource_manager_cleanup(man); } /* this just adjusts TTM size idea, which sets lpfn to the correct value */ |