[PATCH] arm64: Improve parking of stopped CPUs

[jnair@caviumnetworks.com (Jayachandran C)]

On Wed, Feb 01, 2017 at 03:07:56PM +0000, Will Deacon wrote:
> On Wed, Feb 01, 2017 at 02:31:38PM +0000, Suzuki K Poulose wrote:
> > On 01/02/17 14:16, Will Deacon wrote:
> > >On Wed, Feb 01, 2017 at 09:48:52AM +0000, Jayachandran C wrote:
> > >>The current code puts the stopped cpus in an 'yield' instruction loop.
> > >>Using a busy loop here is unnecessary, we can use the cpu_park_loop()
> > >>function here to do a wfi/wfe.
> > >>
> > >>Signed-off-by: Jayachandran C <jnair@caviumnetworks.com>
> > >>---
> > >> arch/arm64/kernel/smp.c | 3 +--
> > >> 1 file changed, 1 insertion(+), 2 deletions(-)
> > >>
> > >>diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> > >>index cbaab44..0691d2f 100644
> > >>--- a/arch/arm64/kernel/smp.c
> > >>+++ b/arch/arm64/kernel/smp.c
> > >>@@ -829,8 +829,7 @@ static void ipi_cpu_stop(unsigned int cpu)
> > >>
> > >> 	local_irq_disable();
> > >>
> > >>-	while (1)
> > >>-		cpu_relax();
> > >>+	cpu_park_loop();
> > >> }
> > >
> > >Hmm, so we actually added the yield for QEMU's benefit iirc, where QEMU
> > >will trap the yield and schedule a different vCPU. Should we be adding
> > >a yield to cpu_park_loop instead?
> > 
> > Wouldn't wfi/wfe trigger the same ? I don't know how yield affects a physical
> > CPU. The cpu_park_loop is also used by CPUs which cannot run due to the missing
> > capabilities on the system. As long as yield() doesn't affect the PCPUs, we
> > could do that. 
> 
> Yes, good point, it looks like WFE should do the same thing.

Agree, I am not sure if there is any case where 'yield' is a better option.
There are a few cases in kernel/process.c which ends in 'while (1);' where
the point about QEMU may be valid. These can be fixed up if needed.
 
> > Going another step further, we could also include local_irq_disable() in
> > cpu_park_loop().

There is a cpu_panic_kernel() which calls cpu_park_loop() without the irq
disable. It maybe OK to local_irq_disable for that too, but probably not
in the scope of this patch.

JC.