Re: [gcv v3 27/35] arm: Replace __get_cpu_var uses

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

Hi Christoph,

On Tue, Sep 03, 2013 at 03:39:57PM +0100, Christoph Lameter wrote:
> On Fri, 30 Aug 2013, Will Deacon wrote:
> > ...so I don't think this is quite right, and indeed, we get a bunch of errors
> > from GCC:
> >
> > arch/arm/kernel/hw_breakpoint.c: In function ‘arch_install_hw_breakpoint’:
> > arch/arm/kernel/hw_breakpoint.c:347:33: error: incompatible types when assigning to type ‘struct perf_event *[16]’ from type ‘struct perf_event **’
> > arch/arm/kernel/hw_breakpoint.c:347:1: error: incompatible types when assigning to type ‘struct perf_event *[16]’ from type ‘struct perf_event **’
> > arch/arm/kernel/hw_breakpoint.c:347:1: error: incompatible types when assigning to type ‘struct perf_event *[16]’ from type ‘struct perf_event **’
> > arch/arm/kernel/hw_breakpoint.c:347:1: error: incompatible types when assigning to type ‘struct perf_event *[16]’ from type ‘struct perf_event **’
> 
> Did you apply the first patch of this series which is a bug fix?

No, sorry, I didn't see that. Do you have a branch anywhere that I can play
with?

> > changing to match your recipe still doesn't work, however:
> >
> > arch/arm/kernel/hw_breakpoint.c: In function ‘arch_install_hw_breakpoint’:
> > arch/arm/kernel/hw_breakpoint.c:347:33: error: cast specifies array type
> 
> Yep that is the macro bug that was fixed in the first patch.

Ok. Sorry for the noise.

> > >
> > >         WARN_ON(preemptible());
> > >
> > > -       if (local_inc_return(&__get_cpu_var(mde_ref_count)) == 1)
> > > +       if (this_cpu_inc_return(mde_ref_count) == 1)
> > >                 enable = DBG_MDSCR_MDE;
> >
> > I'm not sure that this is safe. We rely on local_inc_return to be atomic
> > with respect to the current CPU, which will end up being a wrapper around
> > atomic64_inc_return. However, this_cpu_inc_return simply uses a lock, so
> > other people accessing the count in a different manner (local_dec_and_test
> > below) may break local atomicity unless we start disabling interrupts or
> > something horrible like that.
> 
> I do not see any special code for ARM for this_cpu_inc_return. The
> fallback solution in the core code is to disable interrupts for the
> inc_return and arch/arm/include/asm/percpu.h includes
> asm-generic/percpu.h.
> 
> Where did you see it using a lock?

God knows! You're completely right, and we simply disable interrupts which I
somehow misread as taking a lock. However, is it guaranteed that mixing
an atomic64_* access with a this_cpu_inc_return will retain atomicity
between the two? E.g. if you get interrupted during an atomic64_xchg
operation, the interrupt handler issues this_cpu_inc_return, then on return
to the xchg operation it must reissue any reads that had been executed
prior to the interrupt. This should work on ARM/ARM64 (returning from the
interrupt will clear the exclusive monitor) but I don't know about other
architectures.

Will