By observing that we only use reloc-gtt on objects that are device coherent and idle, we can avoid the set-to-domain call. Then noting that our preferred partial GGTT mapping path automatically copes with tiling (it does not use a fence) and handles all the error cases of pinning, that dramatically simplifies that branch.
Signed-off-by: Chris Wilson <[email protected]> --- .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 21 +++++++------------ 1 file changed, 7 insertions(+), 14 deletions(-) diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c index d70ca36f74f6..fe170186dd42 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c @@ -1151,23 +1151,16 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj, struct i915_vma *vma; int err; - if (i915_gem_object_is_tiled(obj)) - return ERR_PTR(-EINVAL); - if (use_cpu_reloc(cache, obj)) return NULL; - err = i915_gem_object_set_to_gtt_domain(obj, true); - if (err) - return ERR_PTR(err); - - vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0, - PIN_MAPPABLE | - PIN_NONBLOCK /* NOWARN */ | - PIN_NOEVICT); - if (vma == ERR_PTR(-EDEADLK)) - return vma; - + vma = ERR_PTR(-ENODEV); + if (!i915_gem_object_is_tiled(obj)) + vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, + NULL, 0, 0, + PIN_MAPPABLE | + PIN_NONBLOCK /* NOWARN */ | + PIN_NOEVICT); if (IS_ERR(vma)) { memset(&cache->node, 0, sizeof(cache->node)); mutex_lock(&ggtt->vm.mutex); -- 2.20.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
