diff options
| author |   Chris Wilson <chris@chris-wilson.co.uk> | 2017-03-10 00:09:42 +0000 |
|---|---|---|
| committer | Chris Wilson <chris@chris-wilson.co.uk> | 2017-03-13 11:16:09 +0000 |
| commit | 7f5f95d8aca44661efb4c66d405c21ca703a8e44 (patch) | |
| tree | 53d4200162b2f002500345b4de4a6406b425332b /drivers/gpu/drm/i915/i915_gem.c | |
| parent | drm/i915: Rename REDIRECT_TO_GUC bit (diff) | |
| download | linux-dev-7f5f95d8aca44661efb4c66d405c21ca703a8e44.tar.xz linux-dev-7f5f95d8aca44661efb4c66d405c21ca703a8e44.zip | |
drm/i915: Move whole object to CPU domain for coherent shmem access
If the object is coherent, we can simply update the cache domain on the
whole object rather than calculate the before/after clflushes. The
advantage is that we then get correct tracking of ellided flushes when
changing coherency later.
Testcase: igt/gem_pwrite_snooped
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170310000942.11661-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Diffstat (limited to 'drivers/gpu/drm/i915/i915_gem.c')
| -rw-r--r-- | drivers/gpu/drm/i915/i915_gem.c | 45 |
1 files changed, 23 insertions, 22 deletions
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index aca1eaddafb4..202bb850f260 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -788,6 +788,15 @@ int i915_gem_obj_prepare_shmem_read(struct drm_i915_gem_object *obj, if (ret) return ret; + if (i915_gem_object_is_coherent(obj) || + !static_cpu_has(X86_FEATURE_CLFLUSH)) { + ret = i915_gem_object_set_to_cpu_domain(obj, false); + if (ret) + goto err_unpin; + else + goto out; + } + i915_gem_object_flush_gtt_write_domain(obj); /* If we're not in the cpu read domain, set ourself into the gtt @@ -796,16 +805,9 @@ int i915_gem_obj_prepare_shmem_read(struct drm_i915_gem_object *obj, * anyway again before the next pread happens. */ if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) - *needs_clflush = !i915_gem_object_is_coherent(obj); - - if (*needs_clflush && !static_cpu_has(X86_FEATURE_CLFLUSH)) { - ret = i915_gem_object_set_to_cpu_domain(obj, false); - if (ret) - goto err_unpin; - - *needs_clflush = 0; - } + *needs_clflush = CLFLUSH_BEFORE; +out: /* return with the pages pinned */ return 0; @@ -838,6 +840,15 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, if (ret) return ret; + if (i915_gem_object_is_coherent(obj) || + !static_cpu_has(X86_FEATURE_CLFLUSH)) { + ret = i915_gem_object_set_to_cpu_domain(obj, true); + if (ret) + goto err_unpin; + else + goto out; + } + i915_gem_object_flush_gtt_write_domain(obj); /* If we're not in the cpu write domain, set ourself into the @@ -846,25 +857,15 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, * right away and we therefore have to clflush anyway. */ if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) - *needs_clflush |= cpu_write_needs_clflush(obj) << 1; + *needs_clflush |= CLFLUSH_AFTER; /* Same trick applies to invalidate partially written cachelines read * before writing. */ if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) - *needs_clflush |= !i915_gem_object_is_coherent(obj); - - if (*needs_clflush && !static_cpu_has(X86_FEATURE_CLFLUSH)) { - ret = i915_gem_object_set_to_cpu_domain(obj, true); - if (ret) - goto err_unpin; - - *needs_clflush = 0; - } - - if ((*needs_clflush & CLFLUSH_AFTER) == 0) - obj->cache_dirty = true; + *needs_clflush |= CLFLUSH_BEFORE; +out: intel_fb_obj_invalidate(obj, ORIGIN_CPU); obj->mm.dirty = true; /* return with the pages pinned */ |