Prev: [PATCH] [x86] NUMA implied when DISCONTIGMEM
Next: x86/atomic changes for v2.6.35
From: David Howells on 19 May 2010 07:50
Michel Lespinasse <walken(a)google.com> wrote:

> The only functional change here is that __up_write won't call
> call_rwsem_wake anymore if the new rwsem value is >0. This makes
> no real difference since call_rwsem_wake would have noticed the
> active count being nonzero and done nothing anyway.

Given that you describe this first, this would suggest that the subject of the
patch should be this. I'm not sure I'd count this as a minor cleanup. I think
I'd split it into its own patch.

> Besides that, I clarified a few comments.

Mostly okay, except where you said "expects old value in %edx" - that's only
true on i386, not x86_64. On the latter it would be %rdi. However, I can live
with that: it's true enough.

> - rwsem_count_t tmp;
> + rwsem_count_t tmp = -RWSEM_ACTIVE_WRITE_BIAS;
> ...
> : "+m" (sem->count), "=d" (tmp)
> - : "a" (sem), "1" (-RWSEM_ACTIVE_WRITE_BIAS)
> + : "a" (sem), "1" (tmp)

If you're going to put the initialisation of EDX/RDI on tmp (which isn't really
necessary), rather than directly on the asm statement, you could change the
'"=d" (tmp)' output constraint to be '"+d" (tmp)' and drop the '"1" (tmp)'
constraint entirely.

However, apart from that, feel free to add my Acked-by to this patch or its
split resultant patches.

David
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Michel Lespinasse on 20 May 2010 17:40
On Wed, May 19, 2010 at 4:47 AM, David Howells <dhowells(a)redhat.com> wrote:
> Given that you describe this first, this would suggest that the subject of the
> patch should be this. �I'm not sure I'd count this as a minor cleanup. �I think
> I'd split it into its own patch.

OK, split it in two.

> Mostly okay, except where you said "expects old value in %edx" - that's only
> true on i386, not x86_64. �On the latter it would be %rdi. �However, I can live
> with that: it's true enough.

It's actually still %edx in x86_64 - we're calling into
arch/x86/lib/rwsem_64.S which has its own unusual conventions.

>> - � � rwsem_count_t tmp;
>> + � � rwsem_count_t tmp = -RWSEM_ACTIVE_WRITE_BIAS;
>> ...
>> � � � � � � � � � �: "+m" (sem->count), "=d" (tmp)
>> - � � � � � � � � �: "a" (sem), "1" (-RWSEM_ACTIVE_WRITE_BIAS)
>> + � � � � � � � � �: "a" (sem), "1" (tmp)
>
> If you're going to put the initialisation of EDX/RDI on tmp (which isn't really
> necessary), rather than directly on the asm statement, you could change the
> '"=d" (tmp)' output constraint to be '"+d" (tmp)' and drop the '"1" (tmp)'
> constraint entirely.

I agree it'd be nicer, but I wondered if all gcc versions would handle
the constraints change fine and then I chickened out. Instead I moved
the initialization on the constraints list as was already done in
__up_write(). All I was really shooting for here is consistency
accross __down_write_nested, __up_read and __up_write functions.

> However, apart from that, feel free to add my Acked-by to this patch or its
> split resultant patches.

Thanks. I'll send a V4 series soon integrating your feedback & mark
the two splitted patches resulting from this one as Acked-by.

--
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
 | 
Pages: 1
Prev: [PATCH] [x86] NUMA implied when DISCONTIGMEM
Next: x86/atomic changes for v2.6.35