Re: [tip:core/locking] locking, x86: Slightly shorten __ticket_spin_trylock()

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

On 12/02/2009 09:48 AM, Linus Torvalds wrote:
> 
> Hmm. Odd. I just checked:
> 
> 	_Bool myfunction(char val)
> 	{
> 		return val;
> 	}
> 
> and compiling it with
> 
> 	gcc -O2 -S -m32 -mregparm=3 -fomit-frame-pointer t.c
> 
> I get
> 
> 	myfunction:
> 		testb	%al, %al
> 		setne	%al
> 		ret
> 
> which only sets the low 8 bits. So my gcc actually seems to think that 
> _Bool is just 8 bits, at least for return values, and then upper 24 bits 
> are undefined. It also generates 'testb' for a test of a return value.
> 

Damn.  I stand corrected :-/  I just tested it on x86-64, and gcc 4.4.1
actually *violates the documented ABI* for x86-64.

> So it so happens that I think Jan's patch would have worked - except for 
> the PV_OPS mess. _Bool does act like a 'char' on x86 at least with gcc. I 
> still think that it's fundamentally wrong to use 'bool' because of how 
> subtly it can act.

I personally think using "bool" for C values is a good thing -- people
have a very nasty tendency to come up with the clever idea of "oh, there
is this flag which is 'int'... well, in this special case let's set it
to -1 or 2", and of course there is absolutely no way to know, globally,
that this value once in a blue moon gets set to a bizarre value.  I have
seen this a number of times in the kernel.  It doesn't mean one should
pass it to assembly code.

	-hpa
-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.