Skip to contentSkip to conversation

sna: Replace the blt SYNC_COPY assert with a check

PATCH In review Ville Syrjala 2 participants 2 messages Jun 18, 2018 – Jun 19, 2018

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sna: Replace the blt SYNC_COPY assert with a check

From: Ville Syrjälä <[email protected]>

My IVB hits the SYNC_COPY assert in prefer_blt_copy() when I force the use of the software cursor and I move the cursor on top of a dri2 window. Looks like any platform with sna_wait_for_scanline() implemented should be capable of tripping this assert.

Let's just replace the assert with a check, which is what gen6 already does. IVB/HSW/BDW have sna_wait_for_scanline() so we'll have to change the gen7 and gen8 code at least. gen9+ doesn't have sna_wait_for_scanline() so in theory we could leave that one be, but I like consistency so let's change that one too.

---
 src/sna/gen7_render.c | 5 +++--
 src/sna/gen8_render.c | 5 +++--
 src/sna/gen9_render.c | 5 +++--
 3 files changed, 9 insertions(+), 6 deletions(-)
src/sna/gen7_render.c +3 -2
index 0a3bda768100..68002409a73d 100644
@@ -2957,11 +2957,12 @@ prefer_blt_copy(struct sna *sna,
 		struct kgem_bo *dst_bo,
 		unsigned flags)
 {
+	if (flags & COPY_SYNC)
+		return false;
+
 	if (sna->kgem.mode == KGEM_BLT)
 		return true;
 
-	assert((flags & COPY_SYNC) == 0);
-
 	if (untiled_tlb_miss(src_bo) ||
 	    untiled_tlb_miss(dst_bo))
 		return true;
src/sna/gen8_render.c +3 -2
index 69617da561b6..69c29bc6d901 100644
@@ -2796,11 +2796,12 @@ prefer_blt_copy(struct sna *sna,
 		struct kgem_bo *dst_bo,
 		unsigned flags)
 {
+	if (flags & COPY_SYNC)
+		return false;
+
 	if (sna->kgem.mode == KGEM_BLT)
 		return true;
 
-	assert((flags & COPY_SYNC) == 0);
-
 	if (untiled_tlb_miss(src_bo) ||
 	    untiled_tlb_miss(dst_bo))
 		return true;
src/sna/gen9_render.c +3 -2
index 505b98afa38c..b9622437897a 100644
@@ -2859,11 +2859,12 @@ prefer_blt_copy(struct sna *sna,
 		struct kgem_bo *dst_bo,
 		unsigned flags)
 {
+	if (flags & COPY_SYNC)
+		return false;
+
 	if (sna->kgem.mode == KGEM_BLT)
 		return true;
 
-	assert((flags & COPY_SYNC) == 0);
-
 	if (untiled_tlb_miss(src_bo) ||
 	    untiled_tlb_miss(dst_bo))
 		return true;
Signature · 6 lines
2.16.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw

Quoting Ville Syrjala (2018-06-18 18:39:43)

Quoted 7 lines

From: Ville Syrjälä <[email protected]>

My IVB hits the SYNC_COPY assert in prefer_blt_copy() when I force the use of the software cursor and I move the cursor on top of a dri2 window. Looks like any platform with sna_wait_for_scanline() implemented should be capable of tripping this assert.

Quoted 5 lines

Let's just replace the assert with a check, which is what gen6 already does. IVB/HSW/BDW have sna_wait_for_scanline() so we'll have to change the gen7 and gen8 code at least. gen9+ doesn't have sna_wait_for_scanline() so in theory we could leave that one be, but I like consistency so let's change that one too.

Hmm, my memory says that the assertion was concerned with making sure the wait-for-event + copy operation was in the same batch. At any rate it seems like the swcursor is breaking a few of my assumptions, so before acting I'd like to tidy up the validation logic to check if it is not working as intended.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx