Re: [PATCH] arm64: psci: Fix build breakage without PM_SLEEP

[Catalin Marinas <catalin.marinas@arm.com>]

On Tue, Dec 09, 2014 at 12:38:09PM +0000, Krzysztof Kozlowski wrote:
> On wto, 2014-12-09 at 13:29 +0100, Arnd Bergmann wrote:
> > On Tuesday 09 December 2014 12:48:36 Krzysztof Kozlowski wrote:
> > > Fix build failure of defconfig when PM_SLEEP is disabled (e.g. by
> > > disabling SUSPEND) and CPU_IDLE enabled:
> > > 
> > > arch/arm64/kernel/psci.c:543:2: error: unknown field ‘cpu_suspend’ specified in initializer
> > >   .cpu_suspend = cpu_psci_cpu_suspend,
> > >   ^
> > > arch/arm64/kernel/psci.c:543:2: warning: initialization from incompatible pointer type [enabled by default]
> > > arch/arm64/kernel/psci.c:543:2: warning: (near initialization for ‘cpu_psci_ops.cpu_prepare’) [enabled by default]
> > > make[1]: *** [arch/arm64/kernel/psci.o] Error 1
> > > 
> > > The cpu_operations.cpu_suspend field exists only if ARM64_CPU_SUSPEND is
> > > defined, not CPU_IDLE.
> > > 
> > > Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
> > > 
> > 
> > No objection to fixing this obvious build bug, but why do we even have
> > an ARM64_CPU_SUSPEND option? On ARM32 we only have the respective option
> > because we have a random collection of platform specific drivers that
> > use the symbols, but that's not the case on ARM64.
> 
> I believe because of cpuidle. It's the same as on ARM32: the cpu_suspend
> is used by both PM_SLEEP and CPU_IDLE.

I guess at some point we can replace (as a separate patch)
ARM64_CPU_SUSPEND with PM_SLEEP.

But what I don't fully understand, we can enable CPU_IDLE without
ARM64_CPU_SUSPEND. However, the cpuidle-arm64.c driver will fail to
link since it calls cpu_suspend(). Wouldn't it be better if
ARM64_CPU_SUSPEND depends on CPU_PM (or replaced by it) rather than
PM_SLEEP?

Can we allow deeper idle states when CONFIG_SUSPEND is disabled? I see
CONFIG_SUSPEND related to suspend-to-RAM (system standby) rather than
CPU idle, in which case we may want to allow cpu_suspend when only
CPU_IDLE is enabled (which implies CONFIG_CPU_PM).

-- 
Catalin