sna: Replace the blt SYNC_COPY assert with a check
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
@@ -2957,11 +2957,12 @@ prefer_blt_copy(struct sna *sna, struct kgem_bo *dst_bo, unsigned flags) { +if (flags ©_SYNC ) +return false; +if (sna->kgem.mode ==KGEM_BLT )return true; -assert ((flags ©_SYNC ) ==0 ); -if (untiled_tlb_miss (src_bo) ||untiled_tlb_miss (dst_bo))return true;
src/sna/gen8_render.c +3 -2
@@ -2796,11 +2796,12 @@ prefer_blt_copy(struct sna *sna, struct kgem_bo *dst_bo, unsigned flags) { +if (flags ©_SYNC ) +return false; +if (sna->kgem.mode ==KGEM_BLT )return true; -assert ((flags ©_SYNC ) ==0 ); -if (untiled_tlb_miss (src_bo) ||untiled_tlb_miss (dst_bo))return true;
src/sna/gen9_render.c +3 -2
@@ -2859,11 +2859,12 @@ prefer_blt_copy(struct sna *sna, struct kgem_bo *dst_bo, unsigned flags) { +if (flags ©_SYNC ) +return false; +if (sna->kgem.mode ==KGEM_BLT )return true; -assert ((flags ©_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
Quoting Ville Syrjala (2018-06-18 18:39:43)
Quoted 7 lines
From: Ville Syrjälä <[email protected]>
My IVB hits the
SYNC_COPYassert inprefer_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 withsna_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 havesna_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