[Intel-gfx] [PATCH v3 10/12] drm/i915/ttm: accelerated move implementation

[Thomas Hellström]

On 6/14/21 6:26 PM, Thomas Hellström wrote:
> From: Ramalingam C <ramalingam.c at intel.com>
>
> Invokes the pipelined page migration through blt, for
> i915_ttm_move requests of eviction and also obj clear.
>
> Signed-off-by: Ramalingam C <ramalingam.c at intel.com>
> ---
> v2:
>   - subfunction for accel_move (Thomas)
>   - engine_pm_get/put around context_move/clear (Thomas)
>   - Invalidation at accel_clear (Thomas)
> v3:
>   - conflict resolution s/&bo->mem/bo->resource/g
> ---
>   drivers/gpu/drm/i915/gem/i915_gem_ttm.c | 87 +++++++++++++++++++++----
>   1 file changed, 74 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c
> index bf33724bed5c..08b72c280cb5 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c
> @@ -15,6 +15,9 @@
>   #include "gem/i915_gem_ttm.h"
>   #include "gem/i915_gem_mman.h"
>   
> +#include "gt/intel_migrate.h"
> +#include "gt/intel_engine_pm.h"
> +
>   #define I915_PL_LMEM0 TTM_PL_PRIV
>   #define I915_PL_SYSTEM TTM_PL_SYSTEM
>   #define I915_PL_STOLEN TTM_PL_VRAM
> @@ -282,6 +285,61 @@ i915_ttm_resource_get_st(struct drm_i915_gem_object *obj,
>   	return intel_region_ttm_node_to_st(obj->mm.region, res);
>   }
>   
> +static int i915_ttm_accel_move(struct ttm_buffer_object *bo,
> +			       struct ttm_resource *dst_mem,
> +			       struct sg_table *dst_st)
> +{
> +	struct drm_i915_private *i915 = container_of(bo->bdev, typeof(*i915),
> +						     bdev);
> +	struct ttm_resource_manager *src_man =
> +		ttm_manager_type(bo->bdev, bo->resource->mem_type);
> +	struct drm_i915_gem_object *obj = i915_ttm_to_gem(bo);
> +	struct sg_table *src_st;
> +	struct i915_request *rq;
> +	int ret;
> +
> +	if (!i915->gt.migrate.context)
> +		return -EINVAL;
> +
> +	if (!bo->ttm || !ttm_tt_is_populated(bo->ttm)) {
> +		if (bo->type == ttm_bo_type_kernel)
> +			return -EINVAL;
> +
> +		if (bo->ttm &&
> +		    !(bo->ttm->page_flags & TTM_PAGE_FLAG_ZERO_ALLOC))
> +			return 0;
> +
> +		intel_engine_pm_get(i915->gt.migrate.context->engine);
> +		ret = intel_context_migrate_clear(i915->gt.migrate.context, NULL,
> +						  dst_st->sgl, I915_CACHE_NONE,
> +						  dst_mem->mem_type >= TTM_PL_PRIV,
Here we should probably use I915_PL_LMEM0 instead of TTM_PL_PRIV, but 
since this test  will replaced by gpu_binds_iomem() in an upcoming 
patch, doesn't matter really.
> +						  0, &rq);
> +
> +		if (!ret && rq) {
> +			i915_request_wait(rq, 0, HZ);
Could be a MAX_SCHEDULE_TIMEOUT here to avoid surprises in case the 
queue to the blitter is getting long?
> +			i915_request_put(rq);
> +		}
> +		intel_engine_pm_put(i915->gt.migrate.context->engine);
> +	} else {
> +		src_st = src_man->use_tt ? i915_ttm_tt_get_st(bo->ttm) :
> +						obj->ttm.cached_io_st;
> +
> +		intel_engine_pm_get(i915->gt.migrate.context->engine);
> +		ret = intel_context_migrate_copy(i915->gt.migrate.context,
> +						 NULL, src_st->sgl, I915_CACHE_NONE,
> +						 bo->resource->mem_type >= TTM_PL_PRIV,
> +						 dst_st->sgl, I915_CACHE_NONE,
> +						 dst_mem->mem_type >= TTM_PL_PRIV, &rq);
> +		if (!ret && rq) {
> +			i915_request_wait(rq, 0, HZ);
Same thing here.


With that fixed,

Reviewed-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>