Re: [PATCH 1/9] ARM: get rid of arch_cpu_idle_prepare()

[Russell King - ARM Linux <linux@arm.linux.org.uk>]

On Mon, Jan 27, 2014 at 10:45:59AM -0500, Nicolas Pitre wrote:
> On Mon, 27 Jan 2014, Russell King - ARM Linux wrote:
> 
> > On Mon, Jan 27, 2014 at 01:08:16AM -0500, Nicolas Pitre wrote:
> > > ARM and ARM64 are the only two architectures implementing
> > > arch_cpu_idle_prepare() simply to call local_fiq_enable().
> > > 
> > > We have secondary_start_kernel() already calling local_fiq_enable() and
> > > this is done a second time in arch_cpu_idle_prepare() in that case. And
> > > enabling FIQs has nothing to do with idling the CPU to start with.
> > > 
> > > So let's introduce init_fiq_boot_cpu() to take care of FIQs on the boot
> > > CPU and remove arch_cpu_idle_prepare(). This is now done a bit earlier
> > > at late_initcall time but this shouldn't make a difference in practice
> > > i.e. when FIQs are actually used.
> > > 
> > > Signed-off-by: Nicolas Pitre <nico@linaro.org>
> > > ---
> > >  arch/arm/kernel/process.c | 5 -----
> > >  arch/arm/kernel/setup.c   | 7 +++++++
> > >  2 files changed, 7 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
> > > index 92f7b15dd2..725b8c95e0 100644
> > > --- a/arch/arm/kernel/process.c
> > > +++ b/arch/arm/kernel/process.c
> > > @@ -142,11 +142,6 @@ static void default_idle(void)
> > >  	local_irq_enable();
> > >  }
> > >  
> > > -void arch_cpu_idle_prepare(void)
> > > -{
> > > -	local_fiq_enable();
> > > -}
> > > -
> > >  void arch_cpu_idle_enter(void)
> > >  {
> > >  	ledtrig_cpu(CPU_LED_IDLE_START);
> > > diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c
> > > index 987a7f5bce..d027b1a6fe 100644
> > > --- a/arch/arm/kernel/setup.c
> > > +++ b/arch/arm/kernel/setup.c
> > > @@ -789,6 +789,13 @@ static int __init init_machine_late(void)
> > >  }
> > >  late_initcall(init_machine_late);
> > >  
> > > +static int __init init_fiq_boot_cpu(void)
> > > +{
> > > +	local_fiq_enable();
> > > +	return 0;
> > > +}
> > > +late_initcall(init_fiq_boot_cpu);
> > 
> > arch_cpu_idle_prepare() gets called from the swapper thread, and changes
> > the swapper thread's CPSR. init_fiq_boot_cpu() gets called from PID1, the
> > init thread, and changes the init thread's CPSR, which will already have
> > FIQs enabled by way of how kernel threads are created.
> > 
> > Hence, the above code fragment has no effect what so ever, and those
> > platforms using FIQs will not have FIQs delivered if they're idle
> > (because the swapper will have FIQs masked at the CPU.)
> 
> You're right.
> 
> What about moving local_fiq_enable() to trap_init() then?

That's potentially unsafe, as we haven't touched any of the IRQ
controllers at that point - we can't guarantee what state they'd be
in.  Given that the default FIQ is to just return, a FIQ being raised
at that point will end up with an infinite loop re-entering the FIQ
handler.

-- 
FTTC broadband for 0.8mile line: 5.8Mbps down 500kbps up.  Estimation
in database were 13.1 to 19Mbit for a good line, about 7.5+ for a bad.
Estimate before purchase was "up to 13.2Mbit".