diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 69 |
1 files changed, 57 insertions, 12 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c index 73b8cca35bab..d2237ce9da70 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c @@ -34,6 +34,7 @@ #include <asm/set_memory.h> #endif #include "amdgpu.h" +#include "amdgpu_reset.h" #include <drm/drm_drv.h> #include <drm/ttm/ttm_tt.h> @@ -77,8 +78,9 @@ static int amdgpu_gart_dummy_page_init(struct amdgpu_device *adev) if (adev->dummy_page_addr) return 0; - adev->dummy_page_addr = dma_map_page(&adev->pdev->dev, dummy_page, 0, - PAGE_SIZE, DMA_BIDIRECTIONAL); + adev->dummy_page_addr = dma_map_page_attrs(&adev->pdev->dev, dummy_page, 0, + PAGE_SIZE, DMA_BIDIRECTIONAL, + DMA_ATTR_SKIP_CPU_SYNC); if (dma_mapping_error(&adev->pdev->dev, adev->dummy_page_addr)) { dev_err(&adev->pdev->dev, "Failed to DMA MAP the dummy page\n"); adev->dummy_page_addr = 0; @@ -98,8 +100,9 @@ void amdgpu_gart_dummy_page_fini(struct amdgpu_device *adev) { if (!adev->dummy_page_addr) return; - dma_unmap_page(&adev->pdev->dev, adev->dummy_page_addr, PAGE_SIZE, - DMA_BIDIRECTIONAL); + dma_unmap_page_attrs(&adev->pdev->dev, adev->dummy_page_addr, PAGE_SIZE, + DMA_BIDIRECTIONAL, + DMA_ATTR_SKIP_CPU_SYNC); adev->dummy_page_addr = 0; } @@ -121,6 +124,7 @@ int amdgpu_gart_table_ram_alloc(struct amdgpu_device *adev) struct amdgpu_bo_param bp; dma_addr_t dma_addr; struct page *p; + unsigned long x; int ret; if (adev->gart.bo != NULL) @@ -130,6 +134,10 @@ int amdgpu_gart_table_ram_alloc(struct amdgpu_device *adev) if (!p) return -ENOMEM; + /* assign pages to this device */ + for (x = 0; x < (1UL << order); x++) + p[x].mapping = adev->mman.bdev.dev_mapping; + /* If the hardware does not support UTCL2 snooping of the CPU caches * then set_memory_wc() could be used as a workaround to mark the pages * as write combine memory. @@ -223,6 +231,7 @@ void amdgpu_gart_table_ram_free(struct amdgpu_device *adev) unsigned int order = get_order(adev->gart.table_size); struct sg_table *sg = adev->gart.bo->tbo.sg; struct page *p; + unsigned long x; int ret; ret = amdgpu_bo_reserve(adev->gart.bo, false); @@ -234,6 +243,8 @@ void amdgpu_gart_table_ram_free(struct amdgpu_device *adev) sg_free_table(sg); kfree(sg); p = virt_to_page(adev->gart.ptr); + for (x = 0; x < (1UL << order); x++) + p[x].mapping = NULL; __free_pages(p, order); adev->gart.ptr = NULL; @@ -291,7 +302,6 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, int pages) { unsigned t; - unsigned p; int i, j; u64 page_base; /* Starting from VEGA10, system bit must be 0 to mean invalid. */ @@ -305,8 +315,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, return; t = offset / AMDGPU_GPU_PAGE_SIZE; - p = t / AMDGPU_GPU_PAGES_IN_CPU_PAGE; - for (i = 0; i < pages; i++, p++) { + for (i = 0; i < pages; i++) { page_base = adev->dummy_page_addr; if (!adev->gart.ptr) continue; @@ -317,10 +326,7 @@ void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset, page_base += AMDGPU_GPU_PAGE_SIZE; } } - mb(); - amdgpu_device_flush_hdp(adev, NULL); - for_each_set_bit(i, adev->vmhubs_mask, AMDGPU_MAX_VMHUBS) - amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0); + amdgpu_gart_invalidate_tlb(adev); drm_dev_exit(idx); } @@ -362,6 +368,42 @@ void amdgpu_gart_map(struct amdgpu_device *adev, uint64_t offset, } /** + * amdgpu_gart_map_vram_range - map VRAM pages into the GART page table + * + * @adev: amdgpu_device pointer + * @pa: physical address of the first page to be mapped + * @start_page: first page to map in the GART aperture + * @num_pages: number of pages to be mapped + * @flags: page table entry flags + * @dst: CPU address of the GART table + * + * Binds a BO that is allocated in VRAM to the GART page table + * (all ASICs). + * + * Useful when a kernel BO is located in VRAM but + * needs to be accessed from the GART address space. + */ +void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa, + uint64_t start_page, uint64_t num_pages, + uint64_t flags, void *dst) +{ + u32 i, idx; + + /* The SYSTEM flag indicates the pages aren't in VRAM. */ + WARN_ON_ONCE(flags & AMDGPU_PTE_SYSTEM); + + if (!drm_dev_enter(adev_to_drm(adev), &idx)) + return; + + for (i = 0; i < num_pages; ++i) { + amdgpu_gmc_set_pte_pde(adev, adev->gart.ptr, + start_page + i, pa + AMDGPU_GPU_PAGE_SIZE * i, flags); + } + + drm_dev_exit(idx); +} + +/** * amdgpu_gart_bind - bind pages into the gart page table * * @adev: amdgpu_device pointer @@ -400,7 +442,10 @@ void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev) return; mb(); - amdgpu_device_flush_hdp(adev, NULL); + if (down_read_trylock(&adev->reset_domain->sem)) { + amdgpu_device_flush_hdp(adev, NULL); + up_read(&adev->reset_domain->sem); + } for_each_set_bit(i, adev->vmhubs_mask, AMDGPU_MAX_VMHUBS) amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0); } |