[PATCH v3] drm/amdgpu: Add EXT_COHERENT support for APU and NUMA systems

Felix Kuehling felix.kuehling at amd.com
Tue Oct 24 21:08:01 UTC 2023


On 2023-10-24 15:08, David Francis wrote:
> On gfx943 APU, EXT_COHERENT should give MTYPE_CC for local and
> MTYPE_UC for nonlocal memory.
>
> On NUMA systems, local memory gets the local mtype, set by an
> override callback. If EXT_COHERENT is set, memory will be set as
> MTYPE_UC by default, with local memory MTYPE_CC.
>
> Add an option in the override function for this case, and
> add a check to ensure it is not used on UNCACHED memory.
>
> V2: Combined APU and NUMA code into one patch
> V3: Fixed a potential nullptr in amdgpu_vm_bo_update
>
> Signed-off-by: David Francis <David.Francis at amd.com>

Reviewed-by: Felix Kuehling <Felix.Kuehling at amd.com>


> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c    | 17 +++++++-----
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h    |  8 +++++-
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c |  2 +-
>   drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c     | 33 +++++++++++++++--------
>   drivers/gpu/drm/amd/amdkfd/kfd_svm.c      |  8 +++---
>   5 files changed, 45 insertions(+), 23 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index d72daf15662f..155c04589753 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -761,6 +761,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
>    * @immediate: immediate submission in a page fault
>    * @unlocked: unlocked invalidation during MM callback
>    * @flush_tlb: trigger tlb invalidation after update completed
> + * @allow_override: change MTYPE for local NUMA nodes
>    * @resv: fences we need to sync to
>    * @start: start of mapped range
>    * @last: last mapped entry
> @@ -777,7 +778,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
>    * 0 for success, negative erro code for failure.
>    */
>   int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> -			   bool immediate, bool unlocked, bool flush_tlb,
> +			   bool immediate, bool unlocked, bool flush_tlb, bool allow_override,
>   			   struct dma_resv *resv, uint64_t start, uint64_t last,
>   			   uint64_t flags, uint64_t offset, uint64_t vram_base,
>   			   struct ttm_resource *res, dma_addr_t *pages_addr,
> @@ -815,6 +816,7 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
>   	params.immediate = immediate;
>   	params.pages_addr = pages_addr;
>   	params.unlocked = unlocked;
> +	params.allow_override = allow_override;
>   
>   	/* Implicitly sync to command submissions in the same VM before
>   	 * unmapping. Sync to moving fences before mapping.
> @@ -990,6 +992,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
>   	struct ttm_resource *mem;
>   	struct dma_fence **last_update;
>   	bool flush_tlb = clear;
> +	bool uncached;
>   	struct dma_resv *resv;
>   	uint64_t vram_base;
>   	uint64_t flags;
> @@ -1027,9 +1030,11 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
>   
>   		bo_adev = amdgpu_ttm_adev(bo->tbo.bdev);
>   		vram_base = bo_adev->vm_manager.vram_base_offset;
> +		uncached = (bo->flags & AMDGPU_GEM_CREATE_UNCACHED) != 0;
>   	} else {
>   		flags = 0x0;
>   		vram_base = 0;
> +		uncached = false;
>   	}
>   
>   	if (clear || (bo && bo->tbo.base.resv ==
> @@ -1063,7 +1068,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
>   		trace_amdgpu_vm_bo_update(mapping);
>   
>   		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb,
> -					   resv, mapping->start, mapping->last,
> +					   !uncached, resv, mapping->start, mapping->last,
>   					   update_flags, mapping->offset,
>   					   vram_base, mem, pages_addr,
>   					   last_update);
> @@ -1258,8 +1263,8 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
>   		    mapping->start < AMDGPU_GMC_HOLE_START)
>   			init_pte_value = AMDGPU_PTE_DEFAULT_ATC;
>   
> -		r = amdgpu_vm_update_range(adev, vm, false, false, true, resv,
> -					   mapping->start, mapping->last,
> +		r = amdgpu_vm_update_range(adev, vm, false, false, true, false,
> +					   resv, mapping->start, mapping->last,
>   					   init_pte_value, 0, 0, NULL, NULL,
>   					   &f);
>   		amdgpu_vm_free_mapping(adev, vm, mapping, f);
> @@ -2547,8 +2552,8 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
>   		goto error_unlock;
>   	}
>   
> -	r = amdgpu_vm_update_range(adev, vm, true, false, false, NULL, addr,
> -				   addr, flags, value, 0, NULL, NULL, NULL);
> +	r = amdgpu_vm_update_range(adev, vm, true, false, false, false,
> +				   NULL, addr, addr, flags, value, 0, NULL, NULL, NULL);
>   	if (r)
>   		goto error_unlock;
>   
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> index 6e71978db13f..9ea8b5b644e3 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> @@ -244,6 +244,12 @@ struct amdgpu_vm_update_params {
>   	 * @table_freed: return true if page table is freed when updating
>   	 */
>   	bool table_freed;
> +
> +	/**
> +	 * @allow_override: true for memory that is not uncached: allows MTYPE
> +	 * to be overridden for NUMA local memory.
> +	 */
> +	bool allow_override;
>   };
>   
>   struct amdgpu_vm_update_funcs {
> @@ -436,7 +442,7 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
>   void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
>   			    struct amdgpu_vm *vm, struct amdgpu_bo *bo);
>   int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> -			   bool immediate, bool unlocked, bool flush_tlb,
> +			   bool immediate, bool unlocked, bool flush_tlb, bool allow_override,
>   			   struct dma_resv *resv, uint64_t start, uint64_t last,
>   			   uint64_t flags, uint64_t offset, uint64_t vram_base,
>   			   struct ttm_resource *res, dma_addr_t *pages_addr,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
> index 9b025fd17b84..a2287bb25223 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
> @@ -843,7 +843,7 @@ static void amdgpu_vm_pte_update_flags(struct amdgpu_vm_update_params *params,
>   	 */
>   	if ((flags & AMDGPU_PTE_SYSTEM) && (adev->flags & AMD_IS_APU) &&
>   	    adev->gmc.gmc_funcs->override_vm_pte_flags &&
> -	    num_possible_nodes() > 1 && !params->pages_addr)
> +	    num_possible_nodes() > 1 && !params->pages_addr && params->allow_override)
>   		amdgpu_gmc_override_vm_pte_flags(adev, params->vm, addr, &flags);
>   
>   	params->vm->update_funcs->update(params, pt, pe, addr, count, incr,
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> index fee3141bb607..b66c5f7e1c56 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> @@ -1251,12 +1251,15 @@ static void gmc_v9_0_override_vm_pte_flags(struct amdgpu_device *adev,
>   		return;
>   	}
>   
> -	/* Only override mappings with MTYPE_NC, which is the safe default for
> -	 * cacheable memory.
> +	/* MTYPE_NC is the same default and can be overridden.
> +	 * MTYPE_UC will be present if the memory is extended-coherent
> +	 * and can also be overridden.
>   	 */
>   	if ((*flags & AMDGPU_PTE_MTYPE_VG10_MASK) !=
> -	    AMDGPU_PTE_MTYPE_VG10(MTYPE_NC)) {
> -		dev_dbg_ratelimited(adev->dev, "MTYPE is not NC\n");
> +	    AMDGPU_PTE_MTYPE_VG10(MTYPE_NC) &&
> +	    (*flags & AMDGPU_PTE_MTYPE_VG10_MASK) !=
> +	    AMDGPU_PTE_MTYPE_VG10(MTYPE_UC)) {
> +		dev_dbg_ratelimited(adev->dev, "MTYPE is not NC or UC\n");
>   		return;
>   	}
>   
> @@ -1283,15 +1286,23 @@ static void gmc_v9_0_override_vm_pte_flags(struct amdgpu_device *adev,
>   			    vm->mem_id, local_node, nid);
>   	if (nid == local_node) {
>   		uint64_t old_flags = *flags;
> -		unsigned int mtype_local = MTYPE_RW;
> +		if ((*flags & AMDGPU_PTE_MTYPE_VG10_MASK) ==
> +			AMDGPU_PTE_MTYPE_VG10(MTYPE_NC)) {
> +			unsigned int mtype_local = MTYPE_RW;
>   
> -		if (amdgpu_mtype_local == 1)
> -			mtype_local = MTYPE_NC;
> -		else if (amdgpu_mtype_local == 2)
> -			mtype_local = MTYPE_CC;
> +			if (amdgpu_mtype_local == 1)
> +				mtype_local = MTYPE_NC;
> +			else if (amdgpu_mtype_local == 2)
> +				mtype_local = MTYPE_CC;
> +
> +			*flags = (*flags & ~AMDGPU_PTE_MTYPE_VG10_MASK) |
> +				 AMDGPU_PTE_MTYPE_VG10(mtype_local);
> +		} else {
> +			/* MTYPE_UC case */
> +			*flags = (*flags & ~AMDGPU_PTE_MTYPE_VG10_MASK) |
> +				 AMDGPU_PTE_MTYPE_VG10(MTYPE_CC);
> +		}
>   
> -		*flags = (*flags & ~AMDGPU_PTE_MTYPE_VG10_MASK) |
> -			 AMDGPU_PTE_MTYPE_VG10(mtype_local);
>   		dev_dbg_ratelimited(adev->dev, "flags updated from %llx to %llx\n",
>   				    old_flags, *flags);
>   	}
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> index 640ada95e2c3..fe937670c927 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> @@ -1280,7 +1280,7 @@ svm_range_get_pte_flags(struct kfd_node *node,
>   			if (num_possible_nodes() <= 1)
>   				mapping_flags |= mtype_local;
>   			else
> -				mapping_flags |= AMDGPU_VM_MTYPE_NC;
> +				mapping_flags |= ext_coherent ? AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC;
>   		/* system memory accessed by the dGPU */
>   		} else {
>   			mapping_flags |= AMDGPU_VM_MTYPE_UC;
> @@ -1315,7 +1315,7 @@ svm_range_unmap_from_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm,
>   
>   	pr_debug("[0x%llx 0x%llx]\n", start, last);
>   
> -	return amdgpu_vm_update_range(adev, vm, false, true, true, NULL, start,
> +	return amdgpu_vm_update_range(adev, vm, false, true, true, false, NULL, start,
>   				      last, init_pte_value, 0, 0, NULL, NULL,
>   				      fence);
>   }
> @@ -1422,8 +1422,8 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
>   		 * different memory partition based on fpfn/lpfn, we should use
>   		 * same vm_manager.vram_base_offset regardless memory partition.
>   		 */
> -		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, NULL,
> -					   last_start, prange->start + i,
> +		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, true,
> +					   NULL, last_start, prange->start + i,
>   					   pte_flags,
>   					   (last_start - prange->start) << PAGE_SHIFT,
>   					   bo_adev ? bo_adev->vm_manager.vram_base_offset : 0,


More information about the amd-gfx mailing list