Skip to contentSkip to conversation

sna/kgem: Gen8 blt broken when dest base offset has bit 4 set

PATCH In review Mika Kuoppala 2 participants 3 messages Nov 19, 2014

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sna/kgem: Gen8 blt broken when dest base offset has bit 4 set

With bit 4 set in destination base address, the Gen8 blitter fails and blits errorneously into area before destination (dest - 16 bytes), corrupting memory.

Broken hw is suspect.

v2: Update the destination base offset pattern as revealed
by igt/tests/gem_userptr_blits/destination-bo-align

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79053

---
 src/sna/kgem.h | 5 +++++
 1 file changed, 5 insertions(+)
src/sna/kgem.h +5 -0
index 6adae3b..344dcea 100644
@@ -551,6 +551,11 @@ static inline bool kgem_bo_blt_pitch_is_ok(struct kgem *kgem,
 					   struct kgem_bo *bo)
 {
 	int pitch = bo->pitch;
+
+	/* bdw is broken with blit dst align */
+	if (kgem->gen >= 0100 && pitch & (1 << 4))
+		return false;
+
 	if (kgem->gen >= 040 && bo->tiling)
 		pitch /= 4;
 	if (pitch > MAXSHORT) {
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw
Quoted 23 lines · Mika Kuoppala

With bit 4 set in destination base address, the Gen8 blitter fails and blits errorneously into area before destination (dest - 16 bytes), corrupting memory.

Broken hw is suspect.

v2: Update the destination base offset pattern as revealed
by igt/tests/gem_userptr_blits/destination-bo-align

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79053

---
 src/sna/kgem.h | 5 +++++
 1 file changed, 5 insertions(+)
src/sna/kgem.h +0 -0
index 6adae3b..344dcea 100644
@@ -551,6 +551,11 @@ static inline bool kgem_bo_blt_pitch_is_ok(struct kgem *kgem,
 					   struct kgem_bo *bo)

src/sna/kgem.h static inline bool kgem_bo_blt_pitch_is_ok(struct kgem *kgem, · line 561

Quoted 10 lines
src/sna/kgem.h +5 -0
 {
 	int pitch = bo->pitch;
+
+	/* bdw is broken with blit dst align */
+	if (kgem->gen >= 0100 && pitch & (1 << 4))
+		return false;
+
 	if (kgem->gen >= 040 && bo->tiling)
 		pitch /= 4;
 	if (pitch > MAXSHORT) {

I added a chunk:

@@ -573,6 +580,12 @@ static inline bool kgem_bo_can_blt(struct kgem *kgem,

                return false;
        }

+ if (kgem->gen >= 0100 && bo->proxy && bo->delta & (1 << 4)) {
+ DBG(("%s: can not blt to handle=%d, delta=%d\n",
+ __FUNCTION__, bo->handle, bo->delta));
+ return false;
+ }
+

        return kgem_bo_blt_pitch_is_ok(kgem, bo);

}

In case we had a 32-byte pitch but starting at a 16-byte offset, which should also trigger the corruption, right?
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw
Quoted 50 lines · Chris Wilson
Quoted 23 lines · Mika Kuoppala

With bit 4 set in destination base address, the Gen8 blitter fails and blits errorneously into area before destination (dest - 16 bytes), corrupting memory.

Broken hw is suspect.

v2: Update the destination base offset pattern as revealed
by igt/tests/gem_userptr_blits/destination-bo-align

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79053

---
 src/sna/kgem.h | 5 +++++
 1 file changed, 5 insertions(+)
src/sna/kgem.h +0 -0
index 6adae3b..344dcea 100644
@@ -551,6 +551,11 @@ static inline bool kgem_bo_blt_pitch_is_ok(struct kgem *kgem,
 					   struct kgem_bo *bo)

src/sna/kgem.h static inline bool kgem_bo_blt_pitch_is_ok(struct kgem *kgem, · line 561

Quoted 10 lines
src/sna/kgem.h +5 -0
 {
 	int pitch = bo->pitch;
+
+	/* bdw is broken with blit dst align */
+	if (kgem->gen >= 0100 && pitch & (1 << 4))
+		return false;
+
 	if (kgem->gen >= 040 && bo->tiling)
 		pitch /= 4;
 	if (pitch > MAXSHORT) {

I added a chunk:

(quoted diff) +6 -0
@@ -573,6 +580,12 @@ static inline bool kgem_bo_can_blt(struct kgem *kgem,
                return false;
        }
 
+       if (kgem->gen >= 0100 && bo->proxy && bo->delta & (1 << 4)) {
+               DBG(("%s: can not blt to handle=%d, delta=%d\n",
+                    __FUNCTION__, bo->handle, bo->delta));
+               return false;
+       }
+
        return kgem_bo_blt_pitch_is_ok(kgem, bo);
 }
 
Quoted 3 lines

In case we had a 32-byte pitch but starting at a 16-byte offset, which should also trigger the corruption, right?
-Chris

So far the story is:

in XY_SRC_COPY_BLT, if Destination base address bit 4 is set, it will blit into wrong destination

If bit 4 is set in Source base address, you get blit into correct spot but with bad data.

I have used 4 * 4 dwords as a pattern. I will do more experiments with the pitch tomorrow.

-Mika

Quoted 3 lines
Signature · 2 lines
-- 
Chris Wilson, Intel Open Source Technology Centre

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx