[Intel-xe] [PATCH 2/2] drm/xe/dgfx: Release mmap mappings on rpm suspend

[Thomas Hellström]

On 12/6/23 14:34, Badal Nilawar wrote:
> Release all mmap mappings for all vram objects which are associated
> with userfault such that, while pcie function in D3hot, any access
> to memory mappings will raise a userfault.
>
> Upon userfault, in order to access memory mappings, if graphics
> function is in D3 then runtime resume of dgpu will be triggered to
> transition to D0.
>
> Previous commit has blocked the rpm but let's make sure that rpm
> does not get blocked for headed cards(client's parts).
> Above pagefault approach will be  useful for headed cards to save the
> power when display is off and there are active mmap mappings.
>
> v2:
>    - Add lock dep assertion before updating vram_userfault_count (Rodrigo)
>    - Avoid iomem check before bo migration check as bo can migrate
>      to system memory (Matthew Auld)
> v3:
>    - Apply pagefault approach for headed cards.
>
> Cc: Rodrigo Vivi <rodrigo.vivi at intel.com>
> Cc: Matthew Auld <matthew.auld at intel.com>
> Cc: Anshuman Gupta <anshuman.gupta at intel.com>
> Signed-off-by: Badal Nilawar <badal.nilawar at intel.com>
> ---
>   drivers/gpu/drm/xe/xe_bo.c           | 62 ++++++++++++++++++++++++++--
>   drivers/gpu/drm/xe/xe_bo.h           |  2 +
>   drivers/gpu/drm/xe/xe_bo_types.h     |  5 +++
>   drivers/gpu/drm/xe/xe_device_types.h | 20 +++++++++
>   drivers/gpu/drm/xe/xe_pm.c           |  7 ++++
>   5 files changed, 93 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> index 5741948a2a51..419bc5c55aa7 100644
> --- a/drivers/gpu/drm/xe/xe_bo.c
> +++ b/drivers/gpu/drm/xe/xe_bo.c
> @@ -797,6 +797,18 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
>   		dma_fence_put(fence);
>   	}
>   
> +	/*
> +	 * TTM has already nuked the mmap for us (see ttm_bo_unmap_virtual),
> +	 * so if we moved from VRAM make sure to unlink this from the userfault
> +	 * tracking.
> +	 */
> +	if (mem_type_is_vram(old_mem_type)) {
> +		spin_lock(&xe->mem_access.vram_userfault_lock);
> +		if (INTEL_DISPLAY_ENABLED(xe) && !list_empty(&bo->vram_userfault_link))
> +			list_del_init(&bo->vram_userfault_link);
> +		spin_unlock(&xe->mem_access.vram_userfault_lock);
> +	}
> +

Please move this block into move_notify() instead, if at all possible. 
That function is intended to release whatever various bindings we have 
set up to the backing memory in the old location.

Thanks,

Thomas