drm/amdgpu: Map all visible VRAM at startup
Subsystems (2)
PATCH v2 1/1 drm/amdgpu: Map all visible VRAM at startup
When using CPU to update page table, we need to kmap all the PDs/PTs after they are allocated and that requires a TLB shot down on each CPU, which is quite heavy.
Instead, we map the whole visible VRAM to a kernel address at once. Pages can be obtained from the offset.
v2: move the mapping base from gmc to amdgpu_mman structure, and the
implementation in amdgpu_ttm_* functions
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 17 +++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 + 2 files changed, 18 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +17 -0
@@ -621,6 +621,7 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_ {struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];struct amdgpu_device *adev =amdgpu_ttm_adev (bdev); +struct drm_mm_node *mm_node = mem->mm_node; mem->bus.addr =NULL ; mem->bus.offset =0 ;
@@ -640,6 +641,15 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_/* check if it's visible */ if ((mem->bus.offset + mem->bus.size) > adev->gmc.visible_vram_size)return -EINVAL ; +/* Only physically contiguous buffers apply. In a contiguous +* buffer, size of the first mm_node would match the number of +* pages in ttm_mem_reg. +*/ +if (adev->mman.aper_base_kaddr && + (mm_node->size == mem->num_pages)) + mem->bus.addr = (u8 *)adev->mman.aper_base_kaddr + + mem->bus.offset; + mem->bus.base = adev->gmc.aper_base; mem->bus.is_iomem = true;break ;
@@ -1402,6 +1412,10 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)/* Change the size here instead of the init above so only lpfn is affected */ amdgpu_ttm_set_active_vram_size (adev, adev->gmc.visible_vram_size); +#ifdef CONFIG_64BIT + adev->mman.aper_base_kaddr =ioremap_wc (adev->gmc.aper_base, + adev->gmc.visible_vram_size); +#endif /* *The reserved vram for firmware must be pinned to the specified
@@ -1494,6 +1508,9 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)amdgpu_ttm_debugfs_fini (adev);amdgpu_bo_free_kernel (&adev->stolen_vga_memory,NULL ,NULL );amdgpu_ttm_fw_reserve_vram_fini (adev); +if (adev->mman.aper_base_kaddr) +iounmap (adev->mman.aper_base_kaddr); + adev->mman.aper_base_kaddr =NULL ;ttm_bo_clean_mm (&adev->mman.bdev,TTM_PL_VRAM );ttm_bo_clean_mm (&adev->mman.bdev,TTM_PL_TT );
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +1 -0
@@ -44,6 +44,7 @@ struct amdgpu_mman {struct ttm_bo_device bdev;bool mem_global_referenced;bool initialized; +void __iomem *aper_base_kaddr;#if defined(CONFIG_DEBUG_FS )struct dentry *debugfs_entries[8 ];
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
A message that is not in the archive (sent off-list or not yet received): 1519756593-11409-1-git-send-email-Amber.Lin-5C7GfCeVMHo@public.gmane.org
Christian König
Reviewed-byAm 27.02.2018 um 19:36 schrieb Amber Lin:
Quoted 4 lines
When using CPU to update page table, we need to kmap all the PDs/PTs after they are allocated and that requires a TLB shot down on each CPU, which is quite heavy.
Quoted 8 lines
Instead, we map the whole visible VRAM to a kernel address at once. Pages can be obtained from the offset.
v2: move the mapping base from gmc to
amdgpu_mmanstructure, and the
implementation in amdgpu_ttm_* functionsChange-Id I56574bd544dae273da50e8b5dd6894cd5d9454bdSigned-off-by Amber Lin <[email protected]>
Quoted 59 lines
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 17 +++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 + 2 files changed, 18 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +17
-0@@ -621,6 +621,7 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_ {struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];struct amdgpu_device *adev =amdgpu_ttm_adev (bdev); +struct drm_mm_node *mm_node = mem->mm_node; mem->bus.addr =NULL ; mem->bus.offset =0 ;@@ -640,6 +641,15 @@ static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_/* check if it's visible */ if ((mem->bus.offset + mem->bus.size) > adev->gmc.visible_vram_size)return -EINVAL ; +/* Only physically contiguous buffers apply. In a contiguous +* buffer, size of the first mm_node would match the number of +* pages in ttm_mem_reg. +*/ +if (adev->mman.aper_base_kaddr && + (mm_node->size == mem->num_pages)) + mem->bus.addr = (u8 *)adev->mman.aper_base_kaddr + + mem->bus.offset; + mem->bus.base = adev->gmc.aper_base; mem->bus.is_iomem = true;break ;@@ -1402,6 +1412,10 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)/* Change the size here instead of the init above so only lpfn is affected */ amdgpu_ttm_set_active_vram_size (adev, adev->gmc.visible_vram_size); +#ifdef CONFIG_64BIT + adev->mman.aper_base_kaddr =ioremap_wc (adev->gmc.aper_base, + adev->gmc.visible_vram_size); +#endif /* *The reserved vram for firmware must be pinned to the specified @@ -1494,6 +1508,9 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)amdgpu_ttm_debugfs_fini (adev);amdgpu_bo_free_kernel (&adev->stolen_vga_memory,NULL ,NULL );amdgpu_ttm_fw_reserve_vram_fini (adev); +if (adev->mman.aper_base_kaddr) +iounmap (adev->mman.aper_base_kaddr); + adev->mman.aper_base_kaddr =NULL ;ttm_bo_clean_mm (&adev->mman.bdev,TTM_PL_VRAM );ttm_bo_clean_mm (&adev->mman.bdev,TTM_PL_TT );drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h struct amdgpu_mman { · line 50
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +1
-0@@ -44,6 +44,7 @@ struct amdgpu_mman {struct ttm_bo_device bdev;bool mem_global_referenced;bool initialized; +void __iomem *aper_base_kaddr;#if defined(CONFIG_DEBUG_FS )struct dentry *debugfs_entries[8 ];
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx