[PATCH v2 1/4] drm/i915/clflush: fixup handling of cache_dirty

[PATCH v2 1/4] drm/i915/clflush: fixup handling of cache_dirty

[Matthew Auld]

In theory if clflush_work_create() somehow fails here, and we don't yet
have mm.pages populated then we end up resetting cache_dirty, which is
likely wrong, since that will potentially skip the flush-on-acquire, if
it was needed.

It looks like intel_user_framebuffer_dirty() can arrive here before the
pages are populated.

v2(Thomas):
  - Move setting cache_dirty out of the async portion, also add a
    comment for why that should still be safe.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
---
 drivers/gpu/drm/i915/gem/i915_gem_clflush.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_clflush.c b/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
index f0435c6feb68..47586a8a1b73 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
@@ -109,12 +109,20 @@ bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
 						I915_FENCE_GFP);
 		dma_resv_add_excl_fence(obj->base.resv, &clflush->base.dma);
 		dma_fence_work_commit(&clflush->base);
+		/*
+		 * We must have successfully populated the pages(since we are
+		 * holding a pin on the pages as per the flush worker) to reach
+		 * this point, which must mean we have already done the required
+		 * flush-on-acquire, hence resetting cache_dirty here should be
+		 * safe.
+		 */
+		obj->cache_dirty = false;
 	} else if (obj->mm.pages) {
 		__do_clflush(obj);
+		obj->cache_dirty = false;
 	} else {
 		GEM_BUG_ON(obj->write_domain != I915_GEM_DOMAIN_CPU);
 	}
 
-	obj->cache_dirty = false;
 	return true;
 }
-- 
2.26.3

[PATCH v2 2/4] drm/i915/clflush: disallow on discrete

[Matthew Auld]

We seem to have an unfortunate issue where we arrive from:

    i915_gem_object_flush_if_display+0x86/0xd0 [i915]
    intel_user_framebuffer_dirty+0x1a/0x50 [i915]
    drm_mode_dirtyfb_ioctl+0xfb/0x1b0

which can be before the pages are populated(and pinned for display), and
so i915_gem_object_has_struct_page() might still return true, as per the
ttm backend. We could re-order the later get_pages() call here, but
since on discrete everything should already be coherent, with the
exception of the display engine, and even there display surfaces must be
allocated in device local-memory anyway, so there should in theory be no
conceivable reason to ever call i915_gem_clflush_object() on discrete.

References: https://gitlab.freedesktop.org/drm/intel/-/issues/4320
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
---
 drivers/gpu/drm/i915/gem/i915_gem_clflush.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_clflush.c b/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
index 47586a8a1b73..aa91252c9324 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
@@ -69,6 +69,8 @@ static struct clflush *clflush_work_create(struct drm_i915_gem_object *obj)
 bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
 			     unsigned int flags)
 {
+
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
 	struct clflush *clflush;
 
 	assert_object_held(obj);
@@ -80,7 +82,7 @@ bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
 	 * anything not backed by physical memory we consider to be always
 	 * coherent and not need clflushing.
 	 */
-	if (!i915_gem_object_has_struct_page(obj)) {
+	if (!i915_gem_object_has_struct_page(obj) || IS_DGFX(i915)) {
 		obj->cache_dirty = false;
 		return false;
 	}
@@ -105,7 +107,7 @@ bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
 	if (clflush) {
 		i915_sw_fence_await_reservation(&clflush->base.chain,
 						obj->base.resv, NULL, true,
-						i915_fence_timeout(to_i915(obj->base.dev)),
+						i915_fence_timeout(i915),
 						I915_FENCE_GFP);
 		dma_resv_add_excl_fence(obj->base.resv, &clflush->base.dma);
 		dma_fence_work_commit(&clflush->base);
-- 
2.26.3

[PATCH v2 3/4] drm/i915: move cpu_write_needs_clflush

[Matthew Auld]

Move it next to its partner in crime; gpu_write_needs_clflush. For
better readability lets keep gpu vs cpu at least in the same file.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
---
 drivers/gpu/drm/i915/gem/i915_gem_domain.c | 12 ++++++++++++
 drivers/gpu/drm/i915/gem/i915_gem_object.h | 15 ++-------------
 drivers/gpu/drm/i915/i915_gem.c            |  2 +-
 3 files changed, 15 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_domain.c b/drivers/gpu/drm/i915/gem/i915_gem_domain.c
index b684a62bf3b0..d30d5a699788 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_domain.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_domain.c
@@ -22,6 +22,18 @@ static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj)
 		 obj->cache_level == I915_CACHE_WT);
 }
 
+bool i915_gem_cpu_write_needs_clflush(struct drm_i915_gem_object *obj)
+{
+	if (obj->cache_dirty)
+		return false;
+
+	if (!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE))
+		return true;
+
+	/* Currently in use by HW (display engine)? Keep flushed. */
+	return i915_gem_object_is_framebuffer(obj);
+}
+
 static void
 flush_write_domain(struct drm_i915_gem_object *obj, unsigned int flush_domains)
 {
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h
index a5479ac7a4ad..de0c8145cc4c 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
@@ -522,6 +522,7 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,
 bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj);
 void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj);
 void i915_gem_object_flush_if_display_locked(struct drm_i915_gem_object *obj);
+bool i915_gem_cpu_write_needs_clflush(struct drm_i915_gem_object *obj);
 
 int __must_check
 i915_gem_object_set_to_wc_domain(struct drm_i915_gem_object *obj, bool write);
@@ -542,23 +543,11 @@ void __i915_gem_object_make_shrinkable(struct drm_i915_gem_object *obj);
 void __i915_gem_object_make_purgeable(struct drm_i915_gem_object *obj);
 void i915_gem_object_make_purgeable(struct drm_i915_gem_object *obj);
 
-static inline bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj)
-{
-	if (obj->cache_dirty)
-		return false;
-
-	if (!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE))
-		return true;
-
-	/* Currently in use by HW (display engine)? Keep flushed. */
-	return i915_gem_object_is_framebuffer(obj);
-}
-
 static inline void __start_cpu_write(struct drm_i915_gem_object *obj)
 {
 	obj->read_domains = I915_GEM_DOMAIN_CPU;
 	obj->write_domain = I915_GEM_DOMAIN_CPU;
-	if (cpu_write_needs_clflush(obj))
+	if (i915_gem_cpu_write_needs_clflush(obj))
 		obj->cache_dirty = true;
 }
 
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 981e383d1a5d..d0e642c82064 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -764,7 +764,7 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
 	 * perspective, requiring manual detiling by the client.
 	 */
 	if (!i915_gem_object_has_struct_page(obj) ||
-	    cpu_write_needs_clflush(obj))
+	    i915_gem_cpu_write_needs_clflush(obj))
 		/* Note that the gtt paths might fail with non-page-backed user
 		 * pointers (e.g. gtt mappings when moving data between
 		 * textures). Fallback to the shmem path in that case.
-- 
2.26.3

[PATCH v2 4/4] drm/i915: stop setting cache_dirty on discrete

[Matthew Auld]

Should not be needed. Even with non-coherent display, we should be using
device local-memory there, and not system memory.

v2: also add a warning in i915_gem_clflush_object

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> #v1
---
 drivers/gpu/drm/i915/gem/i915_gem_clflush.c |  7 ++++++-
 drivers/gpu/drm/i915/gem/i915_gem_domain.c  | 10 ++++++++++
 drivers/gpu/drm/i915/gem/i915_gem_object.c  |  7 +++++--
 drivers/gpu/drm/i915/gem/i915_gem_pages.c   |  1 +
 4 files changed, 22 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_clflush.c b/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
index aa91252c9324..539b2116b2d5 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_clflush.c
@@ -75,6 +75,11 @@ bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
 
 	assert_object_held(obj);
 
+	if (IS_DGFX(i915)) {
+		WARN_ON_ONCE(obj->cache_dirty);
+		return false;
+	}
+
 	/*
 	 * Stolen memory is always coherent with the GPU as it is explicitly
 	 * marked as wc by the system, or the system is cache-coherent.
@@ -82,7 +87,7 @@ bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
 	 * anything not backed by physical memory we consider to be always
 	 * coherent and not need clflushing.
 	 */
-	if (!i915_gem_object_has_struct_page(obj) || IS_DGFX(i915)) {
+	if (!i915_gem_object_has_struct_page(obj)) {
 		obj->cache_dirty = false;
 		return false;
 	}
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_domain.c b/drivers/gpu/drm/i915/gem/i915_gem_domain.c
index d30d5a699788..26532c07d467 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_domain.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_domain.c
@@ -18,18 +18,28 @@
 
 static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj)
 {
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+
+	if (IS_DGFX(i915))
+		return false;
+
 	return !(obj->cache_level == I915_CACHE_NONE ||
 		 obj->cache_level == I915_CACHE_WT);
 }
 
 bool i915_gem_cpu_write_needs_clflush(struct drm_i915_gem_object *obj)
 {
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+
 	if (obj->cache_dirty)
 		return false;
 
 	if (!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE))
 		return true;
 
+	if (IS_DGFX(i915))
+		return false;
+
 	/* Currently in use by HW (display engine)? Keep flushed. */
 	return i915_gem_object_is_framebuffer(obj);
 }
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index 55b0f1df3192..591ee3cb7275 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -114,18 +114,21 @@ void __i915_gem_object_fini(struct drm_i915_gem_object *obj)
 void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,
 					 unsigned int cache_level)
 {
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+
 	obj->cache_level = cache_level;
 
 	if (cache_level != I915_CACHE_NONE)
 		obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ |
 				       I915_BO_CACHE_COHERENT_FOR_WRITE);
-	else if (HAS_LLC(to_i915(obj->base.dev)))
+	else if (HAS_LLC(i915))
 		obj->cache_coherent = I915_BO_CACHE_COHERENT_FOR_READ;
 	else
 		obj->cache_coherent = 0;
 
 	obj->cache_dirty =
-		!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE);
+		!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE) &&
+		!IS_DGFX(i915);
 }
 
 bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
index 308e22a80af4..c4f684b7cc51 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
@@ -26,6 +26,7 @@ void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
 
 	/* Make the pages coherent with the GPU (flushing any swapin). */
 	if (obj->cache_dirty) {
+		WARN_ON_ONCE(IS_DGFX(i915));
 		obj->write_domain = 0;
 		if (i915_gem_object_has_struct_page(obj))
 			drm_clflush_sg(pages);
-- 
2.26.3