[SPARC64]: Avoid membar instructions in delay slots.
In particular, avoid membar instructions in the delay
slot of a jmpl instruction.
UltraSPARC-I, II, IIi, and IIe have a bug, documented in
the UltraSPARC-IIi User's Manual, Appendix K, Erratum 51
The long and short of it is that if the IMU unit misses
on a branch or jmpl, and there is a store buffer synchronizing
membar in the delay slot, the chip can stop fetching instructions.
If interrupts are enabled or some other trap is enabled, the
chip will unwedge itself, but performance will suffer.
We already had a workaround for this bug in a few spots, but
it's better to have the entire tree sanitized for this rule.
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/arch/sparc64/lib/U1memcpy.S b/arch/sparc64/lib/U1memcpy.S
index da9b520..bafd2fc 100644
--- a/arch/sparc64/lib/U1memcpy.S
+++ b/arch/sparc64/lib/U1memcpy.S
@@ -87,14 +87,17 @@
#define LOOP_CHUNK3(src, dest, len, branch_dest) \
MAIN_LOOP_CHUNK(src, dest, f32, f48, len, branch_dest)
+#define DO_SYNC membar #Sync;
#define STORE_SYNC(dest, fsrc) \
EX_ST(STORE_BLK(%fsrc, %dest)); \
- add %dest, 0x40, %dest;
+ add %dest, 0x40, %dest; \
+ DO_SYNC
#define STORE_JUMP(dest, fsrc, target) \
EX_ST(STORE_BLK(%fsrc, %dest)); \
add %dest, 0x40, %dest; \
- ba,pt %xcc, target;
+ ba,pt %xcc, target; \
+ nop;
#define FINISH_VISCHUNK(dest, f0, f1, left) \
subcc %left, 8, %left;\
@@ -239,17 +242,17 @@
ba,pt %xcc, 1b+4
faligndata %f0, %f2, %f48
1: FREG_FROB(f16,f18,f20,f22,f24,f26,f28,f30,f32)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f32,f34,f36,f38,f40,f42,f44,f46,f0)
- STORE_JUMP(o0, f48, 40f) membar #Sync
+ STORE_JUMP(o0, f48, 40f)
2: FREG_FROB(f32,f34,f36,f38,f40,f42,f44,f46,f0)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f0, f2, f4, f6, f8, f10,f12,f14,f16)
- STORE_JUMP(o0, f48, 48f) membar #Sync
+ STORE_JUMP(o0, f48, 48f)
3: FREG_FROB(f0, f2, f4, f6, f8, f10,f12,f14,f16)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f16,f18,f20,f22,f24,f26,f28,f30,f32)
- STORE_JUMP(o0, f48, 56f) membar #Sync
+ STORE_JUMP(o0, f48, 56f)
1: FREG_FROB(f2, f4, f6, f8, f10,f12,f14,f16,f18)
LOOP_CHUNK1(o1, o0, GLOBAL_SPARE, 1f)
@@ -260,17 +263,17 @@
ba,pt %xcc, 1b+4
faligndata %f2, %f4, %f48
1: FREG_FROB(f18,f20,f22,f24,f26,f28,f30,f32,f34)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f34,f36,f38,f40,f42,f44,f46,f0, f2)
- STORE_JUMP(o0, f48, 41f) membar #Sync
+ STORE_JUMP(o0, f48, 41f)
2: FREG_FROB(f34,f36,f38,f40,f42,f44,f46,f0, f2)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f2, f4, f6, f8, f10,f12,f14,f16,f18)
- STORE_JUMP(o0, f48, 49f) membar #Sync
+ STORE_JUMP(o0, f48, 49f)
3: FREG_FROB(f2, f4, f6, f8, f10,f12,f14,f16,f18)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f18,f20,f22,f24,f26,f28,f30,f32,f34)
- STORE_JUMP(o0, f48, 57f) membar #Sync
+ STORE_JUMP(o0, f48, 57f)
1: FREG_FROB(f4, f6, f8, f10,f12,f14,f16,f18,f20)
LOOP_CHUNK1(o1, o0, GLOBAL_SPARE, 1f)
@@ -281,17 +284,17 @@
ba,pt %xcc, 1b+4
faligndata %f4, %f6, %f48
1: FREG_FROB(f20,f22,f24,f26,f28,f30,f32,f34,f36)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f36,f38,f40,f42,f44,f46,f0, f2, f4)
- STORE_JUMP(o0, f48, 42f) membar #Sync
+ STORE_JUMP(o0, f48, 42f)
2: FREG_FROB(f36,f38,f40,f42,f44,f46,f0, f2, f4)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f4, f6, f8, f10,f12,f14,f16,f18,f20)
- STORE_JUMP(o0, f48, 50f) membar #Sync
+ STORE_JUMP(o0, f48, 50f)
3: FREG_FROB(f4, f6, f8, f10,f12,f14,f16,f18,f20)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f20,f22,f24,f26,f28,f30,f32,f34,f36)
- STORE_JUMP(o0, f48, 58f) membar #Sync
+ STORE_JUMP(o0, f48, 58f)
1: FREG_FROB(f6, f8, f10,f12,f14,f16,f18,f20,f22)
LOOP_CHUNK1(o1, o0, GLOBAL_SPARE, 1f)
@@ -302,17 +305,17 @@
ba,pt %xcc, 1b+4
faligndata %f6, %f8, %f48
1: FREG_FROB(f22,f24,f26,f28,f30,f32,f34,f36,f38)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f38,f40,f42,f44,f46,f0, f2, f4, f6)
- STORE_JUMP(o0, f48, 43f) membar #Sync
+ STORE_JUMP(o0, f48, 43f)
2: FREG_FROB(f38,f40,f42,f44,f46,f0, f2, f4, f6)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f6, f8, f10,f12,f14,f16,f18,f20,f22)
- STORE_JUMP(o0, f48, 51f) membar #Sync
+ STORE_JUMP(o0, f48, 51f)
3: FREG_FROB(f6, f8, f10,f12,f14,f16,f18,f20,f22)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f22,f24,f26,f28,f30,f32,f34,f36,f38)
- STORE_JUMP(o0, f48, 59f) membar #Sync
+ STORE_JUMP(o0, f48, 59f)
1: FREG_FROB(f8, f10,f12,f14,f16,f18,f20,f22,f24)
LOOP_CHUNK1(o1, o0, GLOBAL_SPARE, 1f)
@@ -323,17 +326,17 @@
ba,pt %xcc, 1b+4
faligndata %f8, %f10, %f48
1: FREG_FROB(f24,f26,f28,f30,f32,f34,f36,f38,f40)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f40,f42,f44,f46,f0, f2, f4, f6, f8)
- STORE_JUMP(o0, f48, 44f) membar #Sync
+ STORE_JUMP(o0, f48, 44f)
2: FREG_FROB(f40,f42,f44,f46,f0, f2, f4, f6, f8)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f8, f10,f12,f14,f16,f18,f20,f22,f24)
- STORE_JUMP(o0, f48, 52f) membar #Sync
+ STORE_JUMP(o0, f48, 52f)
3: FREG_FROB(f8, f10,f12,f14,f16,f18,f20,f22,f24)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f24,f26,f28,f30,f32,f34,f36,f38,f40)
- STORE_JUMP(o0, f48, 60f) membar #Sync
+ STORE_JUMP(o0, f48, 60f)
1: FREG_FROB(f10,f12,f14,f16,f18,f20,f22,f24,f26)
LOOP_CHUNK1(o1, o0, GLOBAL_SPARE, 1f)
@@ -344,17 +347,17 @@
ba,pt %xcc, 1b+4
faligndata %f10, %f12, %f48
1: FREG_FROB(f26,f28,f30,f32,f34,f36,f38,f40,f42)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f42,f44,f46,f0, f2, f4, f6, f8, f10)
- STORE_JUMP(o0, f48, 45f) membar #Sync
+ STORE_JUMP(o0, f48, 45f)
2: FREG_FROB(f42,f44,f46,f0, f2, f4, f6, f8, f10)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f10,f12,f14,f16,f18,f20,f22,f24,f26)
- STORE_JUMP(o0, f48, 53f) membar #Sync
+ STORE_JUMP(o0, f48, 53f)
3: FREG_FROB(f10,f12,f14,f16,f18,f20,f22,f24,f26)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f26,f28,f30,f32,f34,f36,f38,f40,f42)
- STORE_JUMP(o0, f48, 61f) membar #Sync
+ STORE_JUMP(o0, f48, 61f)
1: FREG_FROB(f12,f14,f16,f18,f20,f22,f24,f26,f28)
LOOP_CHUNK1(o1, o0, GLOBAL_SPARE, 1f)
@@ -365,17 +368,17 @@
ba,pt %xcc, 1b+4
faligndata %f12, %f14, %f48
1: FREG_FROB(f28,f30,f32,f34,f36,f38,f40,f42,f44)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f44,f46,f0, f2, f4, f6, f8, f10,f12)
- STORE_JUMP(o0, f48, 46f) membar #Sync
+ STORE_JUMP(o0, f48, 46f)
2: FREG_FROB(f44,f46,f0, f2, f4, f6, f8, f10,f12)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f12,f14,f16,f18,f20,f22,f24,f26,f28)
- STORE_JUMP(o0, f48, 54f) membar #Sync
+ STORE_JUMP(o0, f48, 54f)
3: FREG_FROB(f12,f14,f16,f18,f20,f22,f24,f26,f28)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f28,f30,f32,f34,f36,f38,f40,f42,f44)
- STORE_JUMP(o0, f48, 62f) membar #Sync
+ STORE_JUMP(o0, f48, 62f)
1: FREG_FROB(f14,f16,f18,f20,f22,f24,f26,f28,f30)
LOOP_CHUNK1(o1, o0, GLOBAL_SPARE, 1f)
@@ -386,17 +389,17 @@
ba,pt %xcc, 1b+4
faligndata %f14, %f16, %f48
1: FREG_FROB(f30,f32,f34,f36,f38,f40,f42,f44,f46)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f46,f0, f2, f4, f6, f8, f10,f12,f14)
- STORE_JUMP(o0, f48, 47f) membar #Sync
+ STORE_JUMP(o0, f48, 47f)
2: FREG_FROB(f46,f0, f2, f4, f6, f8, f10,f12,f14)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f14,f16,f18,f20,f22,f24,f26,f28,f30)
- STORE_JUMP(o0, f48, 55f) membar #Sync
+ STORE_JUMP(o0, f48, 55f)
3: FREG_FROB(f14,f16,f18,f20,f22,f24,f26,f28,f30)
- STORE_SYNC(o0, f48) membar #Sync
+ STORE_SYNC(o0, f48)
FREG_FROB(f30,f32,f34,f36,f38,f40,f42,f44,f46)
- STORE_JUMP(o0, f48, 63f) membar #Sync
+ STORE_JUMP(o0, f48, 63f)
40: FINISH_VISCHUNK(o0, f0, f2, g3)
41: FINISH_VISCHUNK(o0, f2, f4, g3)