percpu: Avoid extra NOP in percpu_cmpxchg16b_double
percpu_cmpxchg16b_double() uses alternative_io() and looks like :
e8 .. .. .. .. call this_cpu_cmpxchg16b_emu
X bytes NOPX
or, once patched (if cpu supports native instruction) on SMP build :
65 48 0f c7 0e cmpxchg16b %gs:(%rsi)
0f 94 c0 sete %al
on !SMP build :
48 0f c7 0e cmpxchg16b (%rsi)
0f 94 c0 sete %al
Therefore, NOPX should be :
P6_NOP3 on SMP
P6_NOP2 on !SMP
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
diff --git a/arch/x86/include/asm/percpu.h b/arch/x86/include/asm/percpu.h
index d475b43..d68fca6 100644
--- a/arch/x86/include/asm/percpu.h
+++ b/arch/x86/include/asm/percpu.h
@@ -509,6 +509,11 @@
* it in software. The address used in the cmpxchg16 instruction must be
* aligned to a 16 byte boundary.
*/
+#ifdef CONFIG_SMP
+#define CMPXCHG16B_EMU_CALL "call this_cpu_cmpxchg16b_emu\n\t" P6_NOP3
+#else
+#define CMPXCHG16B_EMU_CALL "call this_cpu_cmpxchg16b_emu\n\t" P6_NOP2
+#endif
#define percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2) \
({ \
char __ret; \
@@ -517,7 +522,7 @@
typeof(o2) __o2 = o2; \
typeof(o2) __n2 = n2; \
typeof(o2) __dummy; \
- alternative_io("call this_cpu_cmpxchg16b_emu\n\t" P6_NOP4, \
+ alternative_io(CMPXCHG16B_EMU_CALL, \
"cmpxchg16b " __percpu_prefix "(%%rsi)\n\tsetz %0\n\t", \
X86_FEATURE_CX16, \
ASM_OUTPUT2("=a"(__ret), "=d"(__dummy)), \
