Re: [PATCH] arch: tile: include: asm: add cmpxchg64() definition

[Chen Gang <gang.chen@asianux.com>]

Firstly thank you very much for replying quickly with details.

On 06/28/2013 11:09 PM, Ralf Baechle wrote:
> On Fri, Jun 28, 2013 at 09:51:35AM +0800, Chen Gang wrote:
> 
>> Need add cmpxchg64(), or will cause compiling issue.
>>
>> Just define it as cmpxchg(), since cmpxchg() can support 8 bytes.
>>
>> The related error (with allmodconfig):
>>
>>   drivers/block/blockconsole.c: In function ‘bcon_advance_console_bytes’:
>>   drivers/block/blockconsole.c:164:2: error: implicit declaration of function ‘cmpxchg64’ [-Werror=implicit-function-declaration]
>>
>> Signed-off-by: Chen Gang <gang.chen@asianux.com>
>> ---
>>  arch/tile/include/asm/cmpxchg.h |    2 ++
>>  1 files changed, 2 insertions(+), 0 deletions(-)
>>
>> diff --git a/arch/tile/include/asm/cmpxchg.h b/arch/tile/include/asm/cmpxchg.h
>> index 276f067..7688c28 100644
>> --- a/arch/tile/include/asm/cmpxchg.h
>> +++ b/arch/tile/include/asm/cmpxchg.h
>> @@ -68,6 +68,8 @@ extern unsigned long __cmpxchg_called_with_bad_pointer(void);
>>  
>>  #define tas(ptr) (xchg((ptr), 1))
>>  
>> +#define cmpxchg64(ptr, o, n)		cmpxchg((ptr), (o), (n))
> 
> that's broken.  cmpxchg64 is suposed to work on 64 bit operands ONLY.  This
> definition (used by MIPS and Alpha) will work properly:
> 

Oh, really it is. thanks.

> #define cmpxchg64(ptr, o, n)                                            \
> ({                                                                      \
> 	BUILD_BUG_ON(sizeof(*(ptr)) != 8);                              \
> 	cmpxchg((ptr), (o), (n));                                       \
> })
> 

I should follow it to send patch v2, thanks.

> This should cure blockconsole on Tile but not on MIPS.
> 
> struct blockconsole {
> ..
>            u64 console_bytes;
> ..
> };
> 
> static void bcon_advance_console_bytes(struct blockconsole *bc, int bytes)
> {
> 	u64 old, new;
> 	...
> 	} while (cmpxchg64(&bc->console_bytes, old, new) != old);
> }
> 
> So it's using cmpxchg64() to operate on 64 bit objects - but that's not
> going to work on every architecture.  Many 32 bit architectures only
> provide atomic operations that operate on 32 bit quantities.
> 

I guess your meaning is:

  cmpxchg64() should be an atomic operation.
  but on some of 32-bit architectures, they don't implement it as atomic
    (no lock protected, at least).
  it may cause issue on these 32-bit architectures.

Is it correct ?

If it is correct, I should continue to try to fix it.


Thanks.
-- 
Chen Gang