Re: [RFC][PATCH 1/3] locking: Introduce smp_acquire__after_ctrl_dep

[Will Deacon <will.deacon@arm.com>]

On Tue, Jun 07, 2016 at 08:23:15AM -0700, Paul E. McKenney wrote:
> On Tue, Jun 07, 2016 at 04:59:02PM +0200, Hannes Frederic Sowa wrote:
> > Sorry, to follow-up again on this. Will Deacon's comments were about
> > conditional-move instructions, which this compiler-option would prevent,
> > as far as I can see it.
> 
> According to this email thread, I believe that this works the other
> way around:
> 
> http://thread.gmane.org/gmane.linux.kernel/1721993
> 
> That parameter prevents the compiler from converting a conditional
> store into an unconditional store, which would be really problematic.
> Give the current kernel build, I believe that the compiler really is
> within its rights to use conditional-move instructions as shown above.
> But I again must defer to Will Deacon on the details.

A multi_v7_defconfig build of mainline certainly spits out conditional
store instructions, but I have no idea whether these correspond to
WRITE_ONCE or not:

$ objdump -d vmlinux | grep 'str\(eq\|ne\)' | wc -l
7326

At the end of the day, the ARM architecture says you can't rely on this
being ordered and I can see it happening in practice in the face of
conditional stores.

Will