Re: [x86] Unify semaphore_32.S and rwlock_64.S

["H. Peter Anvin" <hpa@zytor.com>]

On 01/20/2010 04:46 PM, Linus Torvalds wrote:
> 
> 
> On Wed, 20 Jan 2010, Linus Torvalds wrote:
>>
>> 	#define RWSEM_UNLOCKED_VALUE            0x00000000
>> 	#define RWSEM_ACTIVE_BIAS               0x00000001
>> 	#define RWSEM_ACTIVE_MASK               0x3fffffff
>> 	#define RWSEM_WAITING_BIAS              (~RWSEM_ACTIVE_MASK)
>> 	#define RWSEM_ACTIVE_READ_BIAS          RWSEM_ACTIVE_BIAS
>> 	#define RWSEM_ACTIVE_WRITE_BIAS         (RWSEM_WAITING_BIAS + RWSEM_ACTIVE_BIAS)
> 
> Btw, doing that RWSEM_WAITING_BIAS cleanup (we currently have it as an 
> independent constant) means that now all constants are shared except for 
> that RWSEM_ACTIVE_MASK. So it ends up being something like this:
> 
> 	#ifdef CONFIG_RWSEM_64bit
> 	  typedef __s64 rwsem_count_t;
> 	  #define RWSEM_ACTIVE_MASK	0x3fffffff
> 	#else
> 	  typedef __s32 rwsem_count_t;
> 	  #define RWSEM_ACTIVE_MASK     0xffff
> 	#endif
> 
> 	#define RWSEM_UNLOCKED_VALUE	0x00000000
> 	#define RWSEM_ACTIVE_BIAS	0x00000001
> 	#define RWSEM_WAITING_BIAS	(~RWSEM_ACTIVE_MASK)
> 	#define RWSEM_ACTIVE_READ_BIAS	RWSEM_ACTIVE_BIAS   
> 	#define RWSEM_ACTIVE_WRITE_BIAS	(RWSEM_WAITING_BIAS + RWSEM_ACTIVE_BIAS)
> 
> with just that two-line difference for the 32-bit/64-bit case.
> 
> At least I _think_ so.
> 

Yes, I already had that change in my tree (or rather, I wrote it as
(-RWSEM_ACTIVE_MASK-1) to be consistent with what was previously there,
but (~RWSEM_ACTIVE_MASK) makes more sense.)

> And it's worth noting (again) that I didn't actually push the 
> twsem_count_t changes down into the slow-path code in lib/rwsem.c. There's 
> a few variables there that might need looking at too. I _think_ they are 
> all ok as-is (unlike the header file, lib/rwsem.c seems to consistently 
> use 'signed long' rather than mix 32-bit and 64-bit types), but it migh be 
> cleaner to make them rwsem_count_t's too.

Yes, if we have it we should it consistently.

	-hpa