Re: [PATCH 1/4] ARC: add barriers to futex code

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Aug 06, 2015 at 03:48:26PM +0200, Peter Zijlstra wrote:
> On Thu, Aug 06, 2015 at 06:05:20PM +0530, Vineet Gupta wrote:
> > The atomic ops on futex need to provide the full barrier just like
> > regular atomics in kernel.
> > 
> > Also remove pagefault_enable/disable in futex_atomic_cmpxchg_inatomic()
> > as core code already does that
> 
> Urgh, and of course tglx just left for holidays :-)
> 
> > +++ b/arch/arc/include/asm/futex.h
> > @@ -20,6 +20,7 @@
> >  
> >  #define __futex_atomic_op(insn, ret, oldval, uaddr, oparg)\
> >  							\
> > +	smp_mb();					\
> >  	__asm__ __volatile__(				\
> >  	"1:	llock	%1, [%2]		\n"	\
> >  		insn				"\n"	\
> > @@ -40,12 +41,14 @@
> >  							\
> >  	: "=&r" (ret), "=&r" (oldval)			\
> >  	: "r" (uaddr), "r" (oparg), "ir" (-EFAULT)	\
> > -	: "cc", "memory")
> > +	: "cc", "memory");				\
> > +	smp_mb();					\
> >  
> 
> 
> So:
> 
>  - alhpa: only has the first smp_mb(), suggesting RELEASE
>  - arm: only has the first smp_mb(), suggesting RELEASE
>  - arm64: has store-release + smp_mb(), suggesting full barriers
>  - MIPS: has LLSC_MB after, suggesting ACQUIRE
>  - powerpc: lwsync before, sync after, full barrier
> 
> x86 is of course boring and fully ordered
> 
> Looking at the usage site of futex_atomic_op_inuser(), that's in
> futex_wake_op() which might suggest RELEASE is indeed sufficient.
> 
> Which leaves me puzzled on MIPS, but what do I know.

So I _think_ the MIPS code is broken. The mips futex code is from 2006
and the mips smp_mb__before_llsc bits are from 2010, so its well
possible this was missed.

Ralf, David, did I miss the obvious or does the below patch make sense?

---
Subject: MIPS: Fix __futex_atomic_op() for WEAK_REORDERING_BEYOND_LLSC

Current __futex_atomic_op() doesn't have smp_mb__before_llsc(), meaning
it would allow prior load/stores to escape out and re-order against the
ll/sc itself.

While the exact requirements on __futex_atomic_op() are a little unclear
at the moment it must be either RELEASE or fully ordered, without
smp_mb__before_llsc() the MIPS code is neither.

Therefore add smp_mb__before_llsc().

Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: David Daney <ddaney@caviumnetworks.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 arch/mips/include/asm/futex.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/arch/mips/include/asm/futex.h b/arch/mips/include/asm/futex.h
index 1de190bdfb9c..2b8023b9b661 100644
--- a/arch/mips/include/asm/futex.h
+++ b/arch/mips/include/asm/futex.h
@@ -20,6 +20,8 @@
 
 #define __futex_atomic_op(insn, ret, oldval, uaddr, oparg)		\
 {									\
+	smp_mb__before_llsc();						\
+									\
 	if (cpu_has_llsc && R10000_LLSC_WAR) {				\
 		__asm__ __volatile__(					\
 		"	.set	push				\n"	\