]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
x86: Fix and improve percpu_cmpxchg{8,16}b_double()
authorJan Beulich <JBeulich@suse.com>
Wed, 14 Dec 2011 08:33:25 +0000 (08:33 +0000)
committerIngo Molnar <mingo@elte.hu>
Thu, 15 Dec 2011 07:17:14 +0000 (08:17 +0100)
They had several problems/shortcomings:

Only the first memory operand was mentioned in the 2x32bit asm()
operands, and 2x64-bit version had a memory clobber. The first
allowed the compiler to not recognize the need to re-load the
data in case it had it cached in some register, and the second
was overly destructive.

The memory operand in the 2x32-bit asm() was declared to only be
an output.

The types of the local copies of the old and new values were
incorrect (as in other per-CPU ops, the types of the per-CPU
variables accessed should be used here, to make sure the
respective types are compatible).

The __dummy variable was pointless (and needlessly initialized
in the 2x32-bit case), given that local copies of the inputs
already exist.

The 2x64-bit variant forced the address of the first object into
%rsi, even though this is needed only for the call to the
emulation function. The real cmpxchg16b can operate on an
memory.

At once also change the return value type to what it really is -
'bool'.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/4EE86D6502000078000679FE@nat28.tlf.novell.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/include/asm/percpu.h

index 3470c9d0ebba4b67a03eb293ed689c905e097598..529bf07e8067fa21b40f823d58064427def4c814 100644 (file)
@@ -451,23 +451,20 @@ do {                                                                      \
 #endif /* !CONFIG_M386 */
 
 #ifdef CONFIG_X86_CMPXCHG64
-#define percpu_cmpxchg8b_double(pcp1, o1, o2, n1, n2)                  \
+#define percpu_cmpxchg8b_double(pcp1, pcp2, o1, o2, n1, n2)            \
 ({                                                                     \
-       char __ret;                                                     \
-       typeof(o1) __o1 = o1;                                           \
-       typeof(o1) __n1 = n1;                                           \
-       typeof(o2) __o2 = o2;                                           \
-       typeof(o2) __n2 = n2;                                           \
-       typeof(o2) __dummy = n2;                                        \
+       bool __ret;                                                     \
+       typeof(pcp1) __o1 = (o1), __n1 = (n1);                          \
+       typeof(pcp2) __o2 = (o2), __n2 = (n2);                          \
        asm volatile("cmpxchg8b "__percpu_arg(1)"\n\tsetz %0\n\t"       \
-                   : "=a"(__ret), "=m" (pcp1), "=d"(__dummy)           \
-                   :  "b"(__n1), "c"(__n2), "a"(__o1), "d"(__o2));     \
+                   : "=a" (__ret), "+m" (pcp1), "+m" (pcp2), "+d" (__o2) \
+                   :  "b" (__n1), "c" (__n2), "a" (__o1));             \
        __ret;                                                          \
 })
 
-#define __this_cpu_cmpxchg_double_4(pcp1, pcp2, o1, o2, n1, n2)                percpu_cmpxchg8b_double(pcp1, o1, o2, n1, n2)
-#define this_cpu_cmpxchg_double_4(pcp1, pcp2, o1, o2, n1, n2)          percpu_cmpxchg8b_double(pcp1, o1, o2, n1, n2)
-#define irqsafe_cpu_cmpxchg_double_4(pcp1, pcp2, o1, o2, n1, n2)       percpu_cmpxchg8b_double(pcp1, o1, o2, n1, n2)
+#define __this_cpu_cmpxchg_double_4    percpu_cmpxchg8b_double
+#define this_cpu_cmpxchg_double_4      percpu_cmpxchg8b_double
+#define irqsafe_cpu_cmpxchg_double_4   percpu_cmpxchg8b_double
 #endif /* CONFIG_X86_CMPXCHG64 */
 
 /*
@@ -508,31 +505,23 @@ do {                                                                      \
  * 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" ASM_NOP3
-#else
-#define CMPXCHG16B_EMU_CALL "call this_cpu_cmpxchg16b_emu\n\t" ASM_NOP2
-#endif
-#define percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2)                 \
+#define percpu_cmpxchg16b_double(pcp1, pcp2, o1, o2, n1, n2)           \
 ({                                                                     \
-       char __ret;                                                     \
-       typeof(o1) __o1 = o1;                                           \
-       typeof(o1) __n1 = n1;                                           \
-       typeof(o2) __o2 = o2;                                           \
-       typeof(o2) __n2 = n2;                                           \
-       typeof(o2) __dummy;                                             \
-       alternative_io(CMPXCHG16B_EMU_CALL,                             \
-                      "cmpxchg16b " __percpu_prefix "(%%rsi)\n\tsetz %0\n\t",  \
+       bool __ret;                                                     \
+       typeof(pcp1) __o1 = (o1), __n1 = (n1);                          \
+       typeof(pcp2) __o2 = (o2), __n2 = (n2);                          \
+       alternative_io("leaq %P1,%%rsi\n\tcall this_cpu_cmpxchg16b_emu\n\t", \
+                      "cmpxchg16b " __percpu_arg(1) "\n\tsetz %0\n\t", \
                       X86_FEATURE_CX16,                                \
-                      ASM_OUTPUT2("=a"(__ret), "=d"(__dummy)),         \
-                      "S" (&pcp1), "b"(__n1), "c"(__n2),               \
-                      "a"(__o1), "d"(__o2) : "memory");                \
+                      ASM_OUTPUT2("=a" (__ret), "+m" (pcp1),           \
+                                  "+m" (pcp2), "+d" (__o2)),           \
+                      "b" (__n1), "c" (__n2), "a" (__o1) : "rsi");     \
        __ret;                                                          \
 })
 
-#define __this_cpu_cmpxchg_double_8(pcp1, pcp2, o1, o2, n1, n2)                percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2)
-#define this_cpu_cmpxchg_double_8(pcp1, pcp2, o1, o2, n1, n2)          percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2)
-#define irqsafe_cpu_cmpxchg_double_8(pcp1, pcp2, o1, o2, n1, n2)       percpu_cmpxchg16b_double(pcp1, o1, o2, n1, n2)
+#define __this_cpu_cmpxchg_double_8    percpu_cmpxchg16b_double
+#define this_cpu_cmpxchg_double_8      percpu_cmpxchg16b_double
+#define irqsafe_cpu_cmpxchg_double_8   percpu_cmpxchg16b_double
 
 #endif