Re: [Intel-gfx] [RFC PATCH 10/10] drm/i915: Use selective tlb invalidations where supported

[Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>]

On 10/10/2023 19:44, Jonathan Cavitt wrote:
> For platforms supporting selective tlb invalidations, we don't need to
> do a full tlb invalidation. Rather do a range based tlb invalidation for
> every unbind of purged vma belongs to an active vm.
> 
> Signed-off-by: Prathap Kumar Valsan <prathap.kumar.valsan@intel.com>
> Cc: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
> Cc: Fei Yang <fei.yang@intel.com>
> Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
> Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
> ---
>   drivers/gpu/drm/i915/gt/intel_ppgtt.c |  2 +-
>   drivers/gpu/drm/i915/i915_vma.c       | 14 +++++++++-----
>   drivers/gpu/drm/i915/i915_vma.h       |  3 ++-
>   3 files changed, 12 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/gt/intel_ppgtt.c b/drivers/gpu/drm/i915/gt/intel_ppgtt.c
> index d07a4f97b9434..b43dae3cbd59f 100644
> --- a/drivers/gpu/drm/i915/gt/intel_ppgtt.c
> +++ b/drivers/gpu/drm/i915/gt/intel_ppgtt.c
> @@ -211,7 +211,7 @@ void ppgtt_unbind_vma(struct i915_address_space *vm,
>   		return;
>   
>   	vm->clear_range(vm, vma_res->start, vma_res->vma_size);
> -	vma_invalidate_tlb(vm, vma_res->tlb);
> +	vma_invalidate_tlb(vm, vma_res->tlb, vma_res->start, vma_res->vma_size);
>   }
>   
>   static unsigned long pd_count(u64 size, int shift)
> diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
> index d09aad34ba37f..cb05d794f0d0f 100644
> --- a/drivers/gpu/drm/i915/i915_vma.c
> +++ b/drivers/gpu/drm/i915/i915_vma.c
> @@ -1339,7 +1339,8 @@ I915_SELFTEST_EXPORT int i915_vma_get_pages(struct i915_vma *vma)
>   	return err;
>   }
>   
> -void vma_invalidate_tlb(struct i915_address_space *vm, u32 *tlb)
> +void vma_invalidate_tlb(struct i915_address_space *vm, u32 *tlb
> +			u64 start, u64 size)
>   {
>   	struct intel_gt *gt;
>   	int id;
> @@ -1355,9 +1356,11 @@ void vma_invalidate_tlb(struct i915_address_space *vm, u32 *tlb)
>   	 * the most recent TLB invalidation seqno, and if we have not yet
>   	 * flushed the TLBs upon release, perform a full invalidation.
>   	 */
> -	for_each_gt(gt, vm->i915, id)
> -		WRITE_ONCE(tlb[id],
> -			   intel_gt_next_invalidate_tlb_full(gt));
> +	for_each_gt(gt, vm->i915, id) {
> +		if (!intel_gt_invalidate_tlb_range(gt, start, size))
> +			WRITE_ONCE(tlb[id],
> +				   intel_gt_next_invalidate_tlb_full(gt));
> +	}
>   }
>   
>   static void __vma_put_pages(struct i915_vma *vma, unsigned int count)
> @@ -2041,7 +2044,8 @@ struct dma_fence *__i915_vma_evict(struct i915_vma *vma, bool async)
>   			dma_fence_put(unbind_fence);
>   			unbind_fence = NULL;
>   		}
> -		vma_invalidate_tlb(vma->vm, vma->obj->mm.tlb);
> +		vma_invalidate_tlb(vma->vm, vma->obj->mm.tlb,
> +				   vma->node.start, vma->size);

The RFC looks like just what we needed so I'll drop an ack on the other 
patch series. Thanks for sending it out so quickly.

Regards,

Tvrtko

>   	}
>   
>   	/*
> diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
> index e356dfb883d34..5a604aad55dfe 100644
> --- a/drivers/gpu/drm/i915/i915_vma.h
> +++ b/drivers/gpu/drm/i915/i915_vma.h
> @@ -260,7 +260,8 @@ bool i915_vma_misplaced(const struct i915_vma *vma,
>   			u64 size, u64 alignment, u64 flags);
>   void __i915_vma_set_map_and_fenceable(struct i915_vma *vma);
>   void i915_vma_revoke_mmap(struct i915_vma *vma);
> -void vma_invalidate_tlb(struct i915_address_space *vm, u32 *tlb);
> +void vma_invalidate_tlb(struct i915_address_space *vm, u32 *tlb,
> +			u64 start, u64 size);
>   struct dma_fence *__i915_vma_evict(struct i915_vma *vma, bool async);
>   int __i915_vma_unbind(struct i915_vma *vma);
>   int __must_check i915_vma_unbind(struct i915_vma *vma);