On 6/14/21 6:26 PM, Thomas Hellström wrote:
From: Ramalingam C ramalingam.c@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@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@linux.intel.com