From 819165fb34b9777f852429f2c6d6f79fbb71b9eb Mon Sep 17 00:00:00 2001 From: Jan Beulich Date: Fri, 20 Jan 2012 16:21:41 +0000 Subject: x86: Adjust asm constraints in atomic64 wrappers Eric pointed out overly restrictive constraints in atomic64_set(), but there are issues throughout the file. In the cited case, %ebx and %ecx are inputs only (don't get changed by either of the two low level implementations). This was also the case elsewhere. Further in many cases early-clobber indicators were missing. Finally, the previous implementation rolled a custom alternative instruction macro from scratch, rather than using alternative_call() (which was introduced with the commit that the description of the change in question actually refers to). Adjusting has the benefit of not hiding referenced symbols from the compiler, which however requires them to be declared not just in the exporting source file (which, as a desirable side effect, in turn allows that exporting file to become a real 5-line stub). This patch does not eliminate the overly restrictive memory clobbers, however: Doing so would occasionally make the compiler set up a second register for accessing the memory object (to satisfy the added "m" constraint), and it's not clear which of the two non-optimal alternatives is better. v2: Re-do the declaration and exporting of the internal symbols. Reported-by: Eric Dumazet Signed-off-by: Jan Beulich Link: http://lkml.kernel.org/r/4F19A2A5020000780006E0D9@nat28.tlf.novell.com Cc: Luca Barbieri Signed-off-by: H. Peter Anvin --- arch/x86/lib/atomic64_32.c | 59 ++-------------------------------------------- 1 file changed, 2 insertions(+), 57 deletions(-) (limited to 'arch/x86/lib') diff --git a/arch/x86/lib/atomic64_32.c b/arch/x86/lib/atomic64_32.c index 042f682..a0b4a35 100644 --- a/arch/x86/lib/atomic64_32.c +++ b/arch/x86/lib/atomic64_32.c @@ -1,59 +1,4 @@ -#include -#include -#include +#define ATOMIC64_EXPORT EXPORT_SYMBOL -#include -#include +#include #include - -long long atomic64_read_cx8(long long, const atomic64_t *v); -EXPORT_SYMBOL(atomic64_read_cx8); -long long atomic64_set_cx8(long long, const atomic64_t *v); -EXPORT_SYMBOL(atomic64_set_cx8); -long long atomic64_xchg_cx8(long long, unsigned high); -EXPORT_SYMBOL(atomic64_xchg_cx8); -long long atomic64_add_return_cx8(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_add_return_cx8); -long long atomic64_sub_return_cx8(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_sub_return_cx8); -long long atomic64_inc_return_cx8(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_inc_return_cx8); -long long atomic64_dec_return_cx8(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_dec_return_cx8); -long long atomic64_dec_if_positive_cx8(atomic64_t *v); -EXPORT_SYMBOL(atomic64_dec_if_positive_cx8); -int atomic64_inc_not_zero_cx8(atomic64_t *v); -EXPORT_SYMBOL(atomic64_inc_not_zero_cx8); -int atomic64_add_unless_cx8(atomic64_t *v, long long a, long long u); -EXPORT_SYMBOL(atomic64_add_unless_cx8); - -#ifndef CONFIG_X86_CMPXCHG64 -long long atomic64_read_386(long long, const atomic64_t *v); -EXPORT_SYMBOL(atomic64_read_386); -long long atomic64_set_386(long long, const atomic64_t *v); -EXPORT_SYMBOL(atomic64_set_386); -long long atomic64_xchg_386(long long, unsigned high); -EXPORT_SYMBOL(atomic64_xchg_386); -long long atomic64_add_return_386(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_add_return_386); -long long atomic64_sub_return_386(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_sub_return_386); -long long atomic64_inc_return_386(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_inc_return_386); -long long atomic64_dec_return_386(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_dec_return_386); -long long atomic64_add_386(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_add_386); -long long atomic64_sub_386(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_sub_386); -long long atomic64_inc_386(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_inc_386); -long long atomic64_dec_386(long long a, atomic64_t *v); -EXPORT_SYMBOL(atomic64_dec_386); -long long atomic64_dec_if_positive_386(atomic64_t *v); -EXPORT_SYMBOL(atomic64_dec_if_positive_386); -int atomic64_inc_not_zero_386(atomic64_t *v); -EXPORT_SYMBOL(atomic64_inc_not_zero_386); -int atomic64_add_unless_386(atomic64_t *v, long long a, long long u); -EXPORT_SYMBOL(atomic64_add_unless_386); -#endif -- cgit v1.1 From cb8095bba6d24118135a5683a956f4f4fb5f17bb Mon Sep 17 00:00:00 2001 From: Jan Beulich Date: Fri, 20 Jan 2012 16:22:04 +0000 Subject: x86: atomic64 assembly improvements In the "xchg" implementation, %ebx and %ecx don't need to be copied into %eax and %edx respectively (this is only necessary when desiring to only read the stored value). In the "add_unless" implementation, swapping the use of %ecx and %esi for passing arguments allows %esi to become an input only (i.e. permitting the register to be re-used to address the same object without reload). In "{add,sub}_return", doing the initial read64 through the passed in %ecx decreases a register dependency. In "inc_not_zero", a branch can be eliminated by or-ing together the two halves of the current (64-bit) value, and code size can be further reduced by adjusting the arithmetic slightly. v2: Undo the folding of "xchg" and "set". Signed-off-by: Jan Beulich Link: http://lkml.kernel.org/r/4F19A2BC020000780006E0DC@nat28.tlf.novell.com Cc: Luca Barbieri Cc: Eric Dumazet Signed-off-by: H. Peter Anvin --- arch/x86/lib/atomic64_386_32.S | 6 +++--- arch/x86/lib/atomic64_cx8_32.S | 29 +++++++++++------------------ 2 files changed, 14 insertions(+), 21 deletions(-) (limited to 'arch/x86/lib') diff --git a/arch/x86/lib/atomic64_386_32.S b/arch/x86/lib/atomic64_386_32.S index e8e7e0d..00933d5 100644 --- a/arch/x86/lib/atomic64_386_32.S +++ b/arch/x86/lib/atomic64_386_32.S @@ -137,13 +137,13 @@ BEGIN(dec_return) RET_ENDP #undef v -#define v %ecx +#define v %esi BEGIN(add_unless) - addl %eax, %esi + addl %eax, %ecx adcl %edx, %edi addl (v), %eax adcl 4(v), %edx - cmpl %eax, %esi + cmpl %eax, %ecx je 3f 1: movl %eax, (v) diff --git a/arch/x86/lib/atomic64_cx8_32.S b/arch/x86/lib/atomic64_cx8_32.S index 391a083..f5cc9eb 100644 --- a/arch/x86/lib/atomic64_cx8_32.S +++ b/arch/x86/lib/atomic64_cx8_32.S @@ -55,8 +55,6 @@ ENDPROC(atomic64_set_cx8) ENTRY(atomic64_xchg_cx8) CFI_STARTPROC - movl %ebx, %eax - movl %ecx, %edx 1: LOCK_PREFIX cmpxchg8b (%esi) @@ -78,7 +76,7 @@ ENTRY(atomic64_\func\()_return_cx8) movl %edx, %edi movl %ecx, %ebp - read64 %ebp + read64 %ecx 1: movl %eax, %ebx movl %edx, %ecx @@ -159,23 +157,22 @@ ENTRY(atomic64_add_unless_cx8) SAVE ebx /* these just push these two parameters on the stack */ SAVE edi - SAVE esi + SAVE ecx - movl %ecx, %ebp - movl %eax, %esi + movl %eax, %ebp movl %edx, %edi - read64 %ebp + read64 %esi 1: cmpl %eax, 0(%esp) je 4f 2: movl %eax, %ebx movl %edx, %ecx - addl %esi, %ebx + addl %ebp, %ebx adcl %edi, %ecx LOCK_PREFIX - cmpxchg8b (%ebp) + cmpxchg8b (%esi) jne 1b movl $1, %eax @@ -199,13 +196,13 @@ ENTRY(atomic64_inc_not_zero_cx8) read64 %esi 1: - testl %eax, %eax - je 4f -2: + movl %eax, %ecx + orl %edx, %ecx + jz 3f movl %eax, %ebx - movl %edx, %ecx + xorl %ecx, %ecx addl $1, %ebx - adcl $0, %ecx + adcl %edx, %ecx LOCK_PREFIX cmpxchg8b (%esi) jne 1b @@ -214,9 +211,5 @@ ENTRY(atomic64_inc_not_zero_cx8) 3: RESTORE ebx ret -4: - testl %edx, %edx - jne 2b - jmp 3b CFI_ENDPROC ENDPROC(atomic64_inc_not_zero_cx8) -- cgit v1.1