Re: [PATCH v4] cpuidle: Init cpuidle only for present CPUs

[Sudeep Holla <sudeep.holla@arm.com>]

On Fri, Mar 07, 2025 at 10:10:50AM +0000, Jacky Bai wrote:
> > Subject: Re: [PATCH v4] cpuidle: Init cpuidle only for present CPUs
> > 
> > On Fri, Mar 07, 2025 at 10:02:14AM +0000, Jacky Bai wrote:
> > > Hi Sudeep,
> > >
> > > > Subject: Re: [PATCH v4] cpuidle: Init cpuidle only for present CPUs
> > > >
> > > > On Fri, Mar 07, 2025 at 04:03:03PM +0800, Jacky Bai wrote:
> > > > > for_each_possible_cpu() is currently used to initialize cpuidle in
> > > > > below cpuidle drivers:
> > > > >   drivers/cpuidle/cpuidle-arm.c
> > > > >   drivers/cpuidle/cpuidle-big_little.c
> > > > >   drivers/cpuidle/cpuidle-psci.c
> > > > >   drivers/cpuidle/cpuidle-riscv-sbi.c
> > > > >
> > > > > However, in cpu_dev_register_generic(), for_each_present_cpu() is
> > > > > used to register CPU devices which means the CPU devices are only
> > > > > registered for present CPUs and not all possible CPUs.
> > > > >
> > > > > With nosmp or maxcpus=0, only the boot CPU is present, lead to the
> > > > > failure:
> > > > >
> > > > >   |  Failed to register cpuidle device for cpu1
> > > > >
> > > > > Then rollback to cancel all CPUs' cpuidle registration.
> > > > >
> > > > > Change for_each_possible_cpu() to for_each_present_cpu() in the
> > > > > above cpuidle drivers to ensure it only registers cpuidle devices
> > > > > for CPUs that are actually present.
> > > > >
> > > > > Fixes: b0c69e1214bc ("drivers: base: Use present CPUs in
> > > > > GENERIC_CPU_DEVICES")
> > > > > Reviewed-by: Dhruva Gole <d-gole@ti.com>
> > > > > Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
> > > > > Tested-by: Yuanjie Yang <quic_yuanjiey@quicinc.com>
> > > > > Signed-off-by: Jacky Bai <ping.bai@nxp.com>
> > > > > ---
> > > > >  - v4 changes:
> > > > >   - add changes for other cpuidle driver that has the similar issue
> > > > >     as cpuidle-pcsi driver.
> > > > >
> > > > >  - v3 changes:
> > > > >   - improve the changelog as suggested by Sudeep
> > > > > ---
> > > > >  drivers/cpuidle/cpuidle-arm.c        | 8 ++++----
> > > > >  drivers/cpuidle/cpuidle-big_little.c | 2 +-
> > > > >  drivers/cpuidle/cpuidle-psci.c       | 4 ++--
> > > > >  drivers/cpuidle/cpuidle-riscv-sbi.c  | 4 ++--
> > > >
> > > >
> > > > Why have you spared drivers/cpuidle/cpuidle-qcom-spm.c ? IIUC the
> > > > issue exists there as well.
> > > >
> > >
> > > For qcom-spm driver, it has below code logic to handle no cpu device
> > > case, and no rollback to cancel the whole cpuidle registration. So I just leave
> > it as it is.
> > > Do we need to update it?
> > >
> > > for_each_possible_cpu(cpu) {
> > >         ret = spm_cpuidle_register(&pdev->dev, cpu);
> > 
> > Did you look into this function ?
> 
> Yes, at the very beginning of this function it will check if the cpu device
> is available, if not, directly return -ENODEV, something I misunderstood?
> 

So why do you think spm_cpuidle_register() does anything different than
psci_idle_init_cpu(). They do exactly same check and yet you apply the
change for psci_idle_init_cpu() but not for spm_cpuidle_register().
What am I missing ?

-- 
Regards,
Sudeep