Re: [patch 04/15] Generic Mutex Subsystem, add-atomic-call-func-x86_64.patch

[Nick Piggin <nickpiggin@yahoo.com.au>]

Nicolas Pitre wrote:
> On Tue, 20 Dec 2005, Ingo Molnar wrote:
> 
> 
>>* David Woodhouse <dwmw2@infradead.org> wrote:
>>
>>
>>>On Mon, 2005-12-19 at 09:49 -0800, Zwane Mwaikambo wrote:
>>>
>>>>Hi Ingo,
>>>>        Doesn't this corrupt caller saved registers?
>>>
>>>Looks like it. I _really_ don't like calling functions from inline 
>>>asm. It's not nice. Can't we use atomic_dec_return() for this?
>>
>>we can use atomic_dec_return(), but that will add one more instruction 
>>to the fastpath. OTOH, atomic_dec_return() is available on every 
>>architecture, so it's a really tempting thing. I'll experiment with it.
> 
> 
> Please consider using (a variant of) xchg() instead.  Although 
> atomic_dec() is available on all architectures, its implementation is 
> far from being the most efficient thing to do for them all.  For 
> example, see my discussion about swp on ARM:
> 

Considering that on UP, the arm should not need to disable interrupts
for this function (or has someone refuted Linus?), how about:

#ifndef CONFIG_SMP
typedef struct { volatile int counter; } mutex_counter_t;

static inline int mutex_counter_dec_return(mutex_counter *v)
{
     return --v->counter;
}

...
#else
#define mutex_counter_t atomic_t
...
#endif

Or does that get too bulky or have other problems?

MP ARMs should have an adequate atomic_dec_return.

-- 
SUSE Labs, Novell Inc.

Send instant messages to your online friends http://au.messenger.yahoo.com