[Intel-gfx] [PATCH v3] drm/i915/ttm: accelerated move implementation
Ramalingam C
ramalingam.c at intel.com
Tue Jun 15 10:06:21 UTC 2021
Invokes the pipelined page migration through blt, for
i915_ttm_move requests of eviction and also obj clear.
v2:
- subfunction for accel_move (Thomas)
- engine_pm_get/put around context_move/clear (Thomas)
- Invalidation at accel_clear (Thomas)
v3:
- Timeout is set for MAX_SCHEDULE_TIMEOUT (Thomas)
- s/TTM_PL_PRIV/I915_PL_LMEM0 (Thomas)
Signed-off-by: Ramalingam C <ramalingam.c at intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_ttm.c | 88 +++++++++++++++++++++----
1 file changed, 75 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 3748098b42d5..94571757fb42 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 +285,62 @@ i915_ttm_resource_get_st(struct drm_i915_gem_object *obj,
return intel_region_ttm_node_to_st(obj->mm.region, res->mm_node);
}
+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->mem.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 >= I915_PL_LMEM0,
+ 0, &rq);
+
+ if (!ret && rq) {
+ i915_request_wait(rq, 0, MAX_SCHEDULE_TIMEOUT);
+ 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->mem.mem_type >= I915_PL_LMEM0,
+ dst_st->sgl, I915_CACHE_NONE,
+ dst_mem->mem_type >= I915_PL_LMEM0,
+ &rq);
+ if (!ret && rq) {
+ i915_request_wait(rq, 0, MAX_SCHEDULE_TIMEOUT);
+ i915_request_put(rq);
+ }
+ intel_engine_pm_put(i915->gt.migrate.context->engine);
+ }
+
+ return ret;
+}
+
static int i915_ttm_move(struct ttm_buffer_object *bo, bool evict,
struct ttm_operation_ctx *ctx,
struct ttm_resource *dst_mem,
@@ -332,19 +391,22 @@ static int i915_ttm_move(struct ttm_buffer_object *bo, bool evict,
if (IS_ERR(dst_st))
return PTR_ERR(dst_st);
- /* If we start mapping GGTT, we can no longer use man::use_tt here. */
- dst_iter = dst_man->use_tt ?
- ttm_kmap_iter_tt_init(&_dst_iter.tt, bo->ttm) :
- ttm_kmap_iter_iomap_init(&_dst_iter.io, &dst_reg->iomap,
- dst_st, dst_reg->region.start);
-
- src_iter = src_man->use_tt ?
- ttm_kmap_iter_tt_init(&_src_iter.tt, bo->ttm) :
- ttm_kmap_iter_iomap_init(&_src_iter.io, &src_reg->iomap,
- obj->ttm.cached_io_st,
- src_reg->region.start);
-
- ttm_move_memcpy(bo, dst_mem->num_pages, dst_iter, src_iter);
+ ret = i915_ttm_accel_move(bo, dst_mem, dst_st);
+ if (ret) {
+ /* If we start mapping GGTT, we can no longer use man::use_tt here. */
+ dst_iter = dst_man->use_tt ?
+ ttm_kmap_iter_tt_init(&_dst_iter.tt, bo->ttm) :
+ ttm_kmap_iter_iomap_init(&_dst_iter.io, &dst_reg->iomap,
+ dst_st, dst_reg->region.start);
+
+ src_iter = src_man->use_tt ?
+ ttm_kmap_iter_tt_init(&_src_iter.tt, bo->ttm) :
+ ttm_kmap_iter_iomap_init(&_src_iter.io, &src_reg->iomap,
+ obj->ttm.cached_io_st,
+ src_reg->region.start);
+
+ ttm_move_memcpy(bo, dst_mem->num_pages, dst_iter, src_iter);
+ }
ttm_bo_move_sync_cleanup(bo, dst_mem);
i915_ttm_free_cached_io_st(obj);
--
2.20.1
More information about the Intel-gfx
mailing list