Re: [PATCH 1 of 4] mm: add a ptep_modify_prot transaction abstraction

[Jeremy Fitzhardinge <jeremy@goop.org>]

Ingo Molnar wrote:
> * Linus Torvalds <torvalds@linux-foundation.org> wrote:
>
>   
>> On Fri, 20 Jun 2008, Jeremy Fitzhardinge wrote:
>>     
>>> Blows up on "gcc version 3.4.4 20050314 (prerelease) (Debian 3.4.3-13)":
>>>       
>> Yeah, I was a bit worried about that. Gcc sometimes does insane 
>> things.
>>
>> We literally just tested that the asm should only _ever_ be generated 
>> with a constant value, but if some gcc dead-code removal thing doesn't 
>> work, it will then screw up and try to generate the asm even for a 
>> non-constant thing.
>>
>> The fairly trivial fix is probably to just change the "i" to "ir", 
>> safe in the knowledge that any _sane_ case will never use the "r" 
>> possibility. I suspect even your insane case will end up then killing 
>> the bad choice later.
>>     
>
> okay - Jeremy, could you try the fix below? (or tip/master, i just 
> pushed this out)
>
>   

Hm.  On 32-bit I get these, but they're warnings.  On 64-bit they're errors.

{standard input}: Assembler messages:
{standard input}:1609: Warning: using `%dl' instead of `%edx' due to `b' 
suffix

vs

{standard input}: Assembler messages:
{standard input}:20511: Error: Incorrect register `%eax' used with `b' 
suffix


> (i dont use gcc 3.x myself to build the kernel, had way too many 
> miscompilations in randconfig tests in the past.)
>   

I do it mainly to pick up these kinds of problems.


> 	Ingo
>
> -------------->
> commit b68b80b8ab39c42707dc126c41e87d46edc97c27
> Author: Ingo Molnar <mingo@elte.hu>
> Date:   Fri Jun 20 21:50:20 2008 +0200
>
>     x86, bitops: make constant-bit set/clear_bit ops faster, gcc workaround
>     
>     Jeremy Fitzhardinge reported this compiler bug:
>     
>     Suggestion from Linus: add "r" to the input constraint of the
>     set_bit()/clear_bit()'s constant 'nr' branch:
>     
>     Blows up on "gcc version 3.4.4 20050314 (prerelease) (Debian 3.4.3-13)":
>     
>      CC      init/main.o
>     include2/asm/bitops.h: In function `start_kernel':
>     include2/asm/bitops.h:59: warning: asm operand 1 probably doesn't match constraints
>     include2/asm/bitops.h:59: warning: asm operand 1 probably doesn't match constraints
>     include2/asm/bitops.h:59: warning: asm operand 1 probably doesn't match constraints
>     include2/asm/bitops.h:59: error: impossible constraint in `asm'
>     include2/asm/bitops.h:59: error: impossible constraint in `asm'
>     include2/asm/bitops.h:59: error: impossible constraint in `asm'
>     
>     Reported-by: Jeremy Fitzhardinge <jeremy@goop.org>
>     Signed-off-by: Ingo Molnar <mingo@elte.hu>
>
> diff --git a/include/asm-x86/bitops.h b/include/asm-x86/bitops.h
> index 6c50548..4575de4 100644
> --- a/include/asm-x86/bitops.h
> +++ b/include/asm-x86/bitops.h
> @@ -58,7 +58,7 @@ static inline void set_bit(unsigned int nr, volatile unsigned long *addr)
>  	if (IS_IMMEDIATE(nr)) {
>  		asm volatile(LOCK_PREFIX "orb %1,%0"
>  			: CONST_MASK_ADDR(nr, addr)
> -			: "i" (CONST_MASK(nr))
> +			: "ir" (CONST_MASK(nr))
>  			: "memory");
>  	} else {
>  		asm volatile(LOCK_PREFIX "bts %1,%0"
> @@ -95,7 +95,7 @@ static inline void clear_bit(int nr, volatile unsigned long *addr)
>  	if (IS_IMMEDIATE(nr)) {
>  		asm volatile(LOCK_PREFIX "andb %1,%0"
>  			: CONST_MASK_ADDR(nr, addr)
> -			: "i" (~CONST_MASK(nr)));
> +			: "ir" (~CONST_MASK(nr)));
>  	} else {
>  		asm volatile(LOCK_PREFIX "btr %1,%0"
>  			: BITOP_ADDR(addr)
>