Re: [PATCH v2 11/32] drm/xe/svm: Add function to determine if range needs VRAM migration

[Matthew Brost <matthew.brost@intel.com>]

On Mon, Apr 07, 2025 at 03:46:58PM +0530, Himal Prasad Ghimiray wrote:
> xe_svm_range_needs_migrate_to_vram() determines whether range needs
> migration to vram or not, for pagefault try at least once.
> 

So I pulled this patch into requested series to minimally enable atomic
on the existing upstream code here [1]. I suspect we will get my series
in first, perhaps even in the 6.15 cycle as I think it could reasonably
justified as a fixes series as the compute UMD doesn't really work
without it. Just a heads up.

[1] https://patchwork.freedesktop.org/patch/647159/?series=146290&rev=4

> Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_svm.c | 49 +++++++++++++++++++++++++++++++++++--
>  drivers/gpu/drm/xe/xe_svm.h | 10 ++++++++
>  2 files changed, 57 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_svm.c b/drivers/gpu/drm/xe/xe_svm.c
> index 8cd35553a927..f4ae3feaf9d3 100644
> --- a/drivers/gpu/drm/xe/xe_svm.c
> +++ b/drivers/gpu/drm/xe/xe_svm.c
> @@ -709,6 +709,51 @@ int xe_svm_alloc_vram(struct xe_vm *vm, struct xe_tile *tile,
>  }
>  #endif
>  
> +static bool supports_4K_migration(struct xe_device *xe)
> +{
> +	if (xe->info.platform == XE_BATTLEMAGE)
> +		return true;
> +
> +	return false;

	if (xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K)
		return false;

	return true;

> +}
> +
> +/**
> + * xe_svm_range_needs_migrate_to_vram() - SVM range needs migrate to VRAM or not
> + * @range: SVM range for which migration needs to be decided
> + * @vma: vma which has range
> + * @region: default placement for range
> + *
> + * Return: True for range needing migration and migration is supported else false
> + */
> +bool xe_svm_range_needs_migrate_to_vram(struct xe_svm_range *range, struct xe_vma *vma,
> +					u32 region)
> +{
> +	struct xe_vm *vm = range_to_vm(&range->base);
> +	u64 range_size = xe_svm_range_size(range);
> +	bool needs_migrate = false;
> +
> +	if (!range->base.flags.migrate_devmem)
> +		return false;
> +
> +	needs_migrate = region;
> +
> +	if (needs_migrate && !IS_DGFX(vm->xe)) {
> +		drm_warn(&vm->xe->drm, "Platform doesn't support VRAM\n");
> +		return false;
> +	}

I'm not sure if this warrents a drm_warn, I think an assert would be
better here as it shouldn't happen unless we have an internal
programming error.

Matt

> +
> +	if (needs_migrate && xe_svm_range_in_vram(range)) {
> +		drm_info(&vm->xe->drm, "Range is already in VRAM\n");
> +		return false;
> +	}
> +
> +	if (needs_migrate && range_size <= SZ_64K && !supports_4K_migration(vm->xe)) {
> +		drm_warn(&vm->xe->drm, "Platform doesn't support SZ_4K range migration\n");
> +		return false;
> +	}
> +
> +	return needs_migrate;
> +}
>  
>  /**
>   * xe_svm_handle_pagefault() - SVM handle page fault
> @@ -763,8 +808,8 @@ int xe_svm_handle_pagefault(struct xe_vm *vm, struct xe_vma *vma,
>  	range_debug(range, "PAGE FAULT");
>  
>  	/* XXX: Add migration policy, for now migrate range once */
> -	if (!range->skip_migrate && range->base.flags.migrate_devmem &&
> -	    xe_svm_range_size(range) >= SZ_64K) {
> +	if (!range->skip_migrate &&
> +	    xe_svm_range_needs_migrate_to_vram(range, vma, IS_DGFX(vm->xe))) {

A little odd to pass IS_DGFX(vm->xe) as the region...

>  		range->skip_migrate = true;
>  
>  		err = xe_svm_alloc_vram(vm, tile, range, &ctx);
> diff --git a/drivers/gpu/drm/xe/xe_svm.h b/drivers/gpu/drm/xe/xe_svm.h
> index 9c4c3aeacc6c..d5be8229ca7e 100644
> --- a/drivers/gpu/drm/xe/xe_svm.h
> +++ b/drivers/gpu/drm/xe/xe_svm.h
> @@ -95,6 +95,9 @@ struct xe_svm_range *xe_svm_range_find_or_insert(struct xe_vm *vm, u64 addr,
>  int xe_svm_range_get_pages(struct xe_vm *vm, struct xe_svm_range *range,
>  			   struct drm_gpusvm_ctx *ctx);
>  
> +bool xe_svm_range_needs_migrate_to_vram(struct xe_svm_range *range, struct xe_vma *vma,
> +					u32 region);
> +
>  /**
>   * xe_svm_range_has_dma_mapping() - SVM range has DMA mapping
>   * @range: SVM range
> @@ -281,6 +284,13 @@ static inline unsigned long xe_svm_range_size(struct xe_svm_range *range)
>  	return 0;
>  }
>  
> +static inline
> +bool xe_svm_range_needs_migrate_to_vram(struct xe_svm_range *range, struct xe_vma *vma,
> +					u32 region)
> +{
> +	return false;
> +}
> +
>  #define xe_svm_assert_in_notifier(...) do {} while (0)
>  #define xe_svm_range_has_dma_mapping(...) false
>  
> -- 
> 2.34.1
>