| Message ID | 52A95CB3.2030805@linaro.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Dec 12, 2013 at 06:50:27AM +0000, Hanjun Guo wrote: > On 2013-12-11 22:27, Catalin Marinas wrote: > > On Wed, Dec 11, 2013 at 02:23:08PM +0000, Mark Brown wrote: > >> On Wed, Dec 11, 2013 at 02:10:19PM +0000, Catalin Marinas wrote: > >>> On Wed, Dec 11, 2013 at 01:13:21PM +0000, Mark Brown wrote: > >> > >>>> +struct cpuinfo_arm { > >>>> + struct cpu cpu; > >>>> + u64 cpuid; > >>>> +#ifdef CONFIG_SMP > >>>> + unsigned int loops_per_jiffy; > >>>> +#endif > >>>> +}; > >> > >>> How is this structure used? I haven't seen the ACPI code doing anything > >>> with struct cpu (though I haven't dug deep enough). Also loops_per_jiffy > >>> is useless, that's related to the delay loop based on the generic timer. > >> > >> Now I look again we can probably drop this for the toplogy work - it had > >> been pulled in as part of pulling things in from pre-v8 and the cpuid > >> was used in the code while I was working on it but isn't any more unless > >> I'm misreading the code. > >> > >> I don't know what the ACPI guys are doing wtih it, I just saw they added > >> the same thing. > > > > It looked to me that for ACPI and empty asm/cpu.h file would do for now. > > There are a couple of prototypes for CPU hotplug I think but haven't > > seen the structure used (not even the x86_cpu one). > > In ACPI code, only struct cpu is used for ACPI based CPU hotplug, here > is the code will send to upstream when the ACPI core for ARM is ready. > > --- > arch/arm64/include/asm/cpu.h | 5 +++++ > arch/arm64/kernel/topology.c | 26 ++++++++++++++++++++++++++ > 2 files changed, 31 insertions(+) > > diff --git a/arch/arm64/include/asm/cpu.h b/arch/arm64/include/asm/cpu.h > index dbeb98d..5613e09 100644 > --- a/arch/arm64/include/asm/cpu.h > +++ b/arch/arm64/include/asm/cpu.h > @@ -20,6 +20,11 @@ struct cpuinfo_arm { > #endif > }; > > +#ifdef CONFIG_HOTPLUG_CPU > +extern int arch_register_cpu(int cpu); > +extern void arch_unregister_cpu(int cpu); > +#endif > + > DECLARE_PER_CPU(struct cpuinfo_arm, cpu_data); > > #endif > diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c > index cb548f1..5c8e69c 100644 > --- a/arch/arm64/kernel/topology.c > +++ b/arch/arm64/kernel/topology.c > @@ -18,3 +18,29 @@ void arch_fix_phys_package_id(int num, u32 slot) > } > EXPORT_SYMBOL_GPL(arch_fix_phys_package_id); > > +#ifdef CONFIG_HOTPLUG_CPU > +int __ref arch_register_cpu(int cpu) > +{ > + struct cpuinfo_arm *cpuinfo = &per_cpu(cpu_data, cpu); > + > + /* BSP cann't be taken down on arm */ > + if (cpu) > + cpuinfo->cpu.hotpluggable = 1; _why_ does the ACPI standard prohibit hotplugging the boot CPU? In non-ACPI systems we can hotplug out the boot CPU (we can do so under KVM using PSCI). I note that the x86 arch_register_cpu allows CPU0 to be hotpluggable on Intel systems as long as there are no dependencies on CPU0 being active. Surely we can test for something more fine-grained rather than disallowing CPU0 hotplug outright? How does this interact with the existing arm64 hotplug code? Thanks, Mark.
On 2013-12-12 18:36, Mark Rutland wrote: > On Thu, Dec 12, 2013 at 06:50:27AM +0000, Hanjun Guo wrote: [...] >> #endif >> diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c >> index cb548f1..5c8e69c 100644 >> --- a/arch/arm64/kernel/topology.c >> +++ b/arch/arm64/kernel/topology.c >> @@ -18,3 +18,29 @@ void arch_fix_phys_package_id(int num, u32 slot) >> } >> EXPORT_SYMBOL_GPL(arch_fix_phys_package_id); >> >> +#ifdef CONFIG_HOTPLUG_CPU >> +int __ref arch_register_cpu(int cpu) >> +{ >> + struct cpuinfo_arm *cpuinfo = &per_cpu(cpu_data, cpu); >> + >> + /* BSP cann't be taken down on arm */ >> + if (cpu) >> + cpuinfo->cpu.hotpluggable = 1; > > _why_ does the ACPI standard prohibit hotplugging the boot CPU? ACPI spec has not limitations to hotplug the boot CPU, it depends on the arch. > > In non-ACPI systems we can hotplug out the boot CPU (we can do so under > KVM using PSCI). If all the things (interrupts and etc) running on CPU0 can be migrated to other CPUs when offline a CPU, it can be hotplugable in ACPI way too. > > I note that the x86 arch_register_cpu allows CPU0 to be hotpluggable on > Intel systems as long as there are no dependencies on CPU0 being > active. Surely we can test for something more fine-grained rather than > disallowing CPU0 hotplug outright? Ok, will update it when I formally send this patch out. > > How does this interact with the existing arm64 hotplug code? Some other patches are needed for ACPI based CPU hotplug,will send out when the ACPI core for ARM is ready, I will cc you and then you will know what's going on :) Thanks Hanjun
diff --git a/arch/arm64/include/asm/cpu.h b/arch/arm64/include/asm/cpu.h index dbeb98d..5613e09 100644 --- a/arch/arm64/include/asm/cpu.h +++ b/arch/arm64/include/asm/cpu.h @@ -20,6 +20,11 @@ struct cpuinfo_arm { #endif }; +#ifdef CONFIG_HOTPLUG_CPU +extern int arch_register_cpu(int cpu); +extern void arch_unregister_cpu(int cpu); +#endif + DECLARE_PER_CPU(struct cpuinfo_arm, cpu_data); #endif diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c index cb548f1..5c8e69c 100644 --- a/arch/arm64/kernel/topology.c +++ b/arch/arm64/kernel/topology.c @@ -18,3 +18,29 @@ void arch_fix_phys_package_id(int num, u32 slot) } EXPORT_SYMBOL_GPL(arch_fix_phys_package_id); +#ifdef CONFIG_HOTPLUG_CPU +int __ref arch_register_cpu(int cpu) +{ + struct cpuinfo_arm *cpuinfo = &per_cpu(cpu_data, cpu); + + /* BSP cann't be taken down on arm */ + if (cpu) + cpuinfo->cpu.hotpluggable = 1; + + return register_cpu(&cpuinfo->cpu, cpu); +} +EXPORT_SYMBOL(arch_register_cpu); + +void arch_unregister_cpu(int cpu) +{ + unregister_cpu(&per_cpu(cpu_data, cpu).cpu); +} +EXPORT_SYMBOL(arch_unregister_cpu); +#else /* CONFIG_HOTPLUG_CPU */ + +static int __init arch_register_cpu(int cpu) +{ + return register_cpu(&per_cpu(cpu_data, cpu).cpu, cpu); +} +#endif /* CONFIG_HOTPLUG_CPU */ +