On Thu, Mar 26, 2015 at 10:34:42AM +0000, Peter Zijlstra wrote:
> On Thu, Mar 26, 2015 at 07:31:12PM +1100, Stephen Rothwell wrote:
> > Hi Christian,
> >
> > After merging the access_once tree, today's linux-next build (arm
> > multi_v7_defconfig) produced lots of this warning:
> >
> > In file included from include/linux/linkage.h:4:0,
> > from include/linux/preempt.h:9,
> > from include/linux/spinlock.h:50,
> > from include/linux/lockref.h:17,
> > from lib/lockref.c:2:
> > In function '__read_once_size',
> > inlined from 'lockref_get' at lib/lockref.c:50:2:
> > include/linux/compiler.h:216:3: warning: call to 'data_access_exceeds_word_size' declared with attribute warning: data access exceeds word size and won't be atomic
> > data_access_exceeds_word_size();
> > ^
> >
> > Introduced by commit 6becd6bd5e89 ("compiler.h: Fix word size check for
> > READ/WRITE_ONCE") presumably interacting with commit 4d3199e4ca8e
> > ("locking: Remove ACCESS_ONCE() usage") from the tip tree.
>
> Hmm, valid warning though, ARM is a 32bit arch and therefore it will
> 'have' to load a u64 in two goes, which violates the
> ACCESS_ONCE/READ_ONCE 'promise'.
>
> Now ARM can indeed to the cmpxchg64 thing, but I'm not sure what to do
> here, I suspect the code is fine, seeing how the cmpxchg64 will fail if
> the split loads got it wrong, but I've not overly thought about it.
Yeah, I think it's fine because, as you point out, the cmpxchg can only
succeed if the 64-bit load appeared to be single-copy atomic (amongst other
things).
Have fun getting ACCESS_ONCE to figure that out, though...
Will
--
To unsubscribe from this list: send the line "unsubscribe linux-next" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
