| Message ID | 20240529133446.28446-18-Jonathan.Cameron@huawei.com |
|---|---|
| State | Accepted |
| Commit | 9d0873892f4d4bf62a351897e91e5169b4c6f4aa |
| Headers | show |
| Series | ACPI/arm64: add support for virtual cpu hotplug | expand |
Hi Jonathan and Catalin, On 5/29/24 11:34 PM, Jonathan Cameron wrote: > In order to move arch_register_cpu() to be called via the same path > for initially present CPUs described by ACPI and hotplugged CPUs > ACPI_HOTPLUG_CPU needs to be enabled. > > The protection against invalid IDs in acpi_map_cpu() is needed as > at least one production BIOS is in the wild which reports entries > in DSDT (with no _STA method, so assumed enabled and present) > that don't match MADT. > > Tested-by: Miguel Luis <miguel.luis@oracle.com> > Reviewed-by: Gavin Shan <gshan@redhat.com> > Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> > --- > arch/arm64/Kconfig | 1 + > arch/arm64/kernel/acpi.c | 22 ++++++++++++++++++++++ > 2 files changed, 23 insertions(+) > > diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig > index 5d91259ee7b5..e8f2ef2312db 100644 > --- a/arch/arm64/Kconfig > +++ b/arch/arm64/Kconfig > @@ -5,6 +5,7 @@ config ARM64 > select ACPI_CCA_REQUIRED if ACPI > select ACPI_GENERIC_GSI if ACPI > select ACPI_GTDT if ACPI > + select ACPI_HOTPLUG_CPU if ACPI_PROCESSOR ACPI_HOTPLUG_CPU depends on (ACPI_PROCESSOR && HOTPLUG_CPU). It needs to be: select ACPI_HOTPLUG_CPU if ACPI_PROCESSOR && HOTPLUG_CPU Otherwise, we can have compiling error with the following configurations. CONFIG_ACPI_PROCESSOR=y CONFIG_HOTPLUG_CPU=n CONFIG_ACPI_HOTPLUG_CPU=y arch/arm64/kernel/smp.c: In function ‘arch_unregister_cpu’: arch/arm64/kernel/smp.c:563:9: error: implicit declaration of function ‘unregister_cpu’; did you mean ‘register_cpu’? [-Werror=implicit-function-declaration] 563 | unregister_cpu(c); | ^~~~~~~~~~~~~~ | register_cpu Since the series has been queued to Catalin's "for-next/vcpu-hotplug" branch, I guess the easiest way would be to fix it in place with Catalin's help. Thanks, Gavin > select ACPI_IORT if ACPI > select ACPI_REDUCED_HARDWARE_ONLY if ACPI > select ACPI_MCFG if (ACPI && PCI) > diff --git a/arch/arm64/kernel/acpi.c b/arch/arm64/kernel/acpi.c > index e0e7b93c16cc..9360ba86678b 100644 > --- a/arch/arm64/kernel/acpi.c > +++ b/arch/arm64/kernel/acpi.c > @@ -30,6 +30,7 @@ > #include <linux/pgtable.h> > > #include <acpi/ghes.h> > +#include <acpi/processor.h> > #include <asm/cputype.h> > #include <asm/cpu_ops.h> > #include <asm/daifflags.h> > @@ -423,6 +424,27 @@ void arch_reserve_mem_area(acpi_physical_address addr, size_t size) > memblock_mark_nomap(addr, size); > } > > +#ifdef CONFIG_ACPI_HOTPLUG_CPU > +int acpi_map_cpu(acpi_handle handle, phys_cpuid_t physid, u32 apci_id, > + int *pcpu) > +{ > + /* If an error code is passed in this stub can't fix it */ > + if (*pcpu < 0) { > + pr_warn_once("Unable to map CPU to valid ID\n"); > + return *pcpu; > + } > + > + return 0; > +} > +EXPORT_SYMBOL(acpi_map_cpu); > + > +int acpi_unmap_cpu(int cpu) > +{ > + return 0; > +} > +EXPORT_SYMBOL(acpi_unmap_cpu); > +#endif /* CONFIG_ACPI_HOTPLUG_CPU */ > + > #ifdef CONFIG_ACPI_FFH > /* > * Implements ARM64 specific callbacks to support ACPI FFH Operation Region as
Hi Gavin, On Sun, Jun 30, 2024 at 10:39:04AM +1000, Gavin Shan wrote: > On 5/29/24 11:34 PM, Jonathan Cameron wrote: > > In order to move arch_register_cpu() to be called via the same path > > for initially present CPUs described by ACPI and hotplugged CPUs > > ACPI_HOTPLUG_CPU needs to be enabled. > > > > The protection against invalid IDs in acpi_map_cpu() is needed as > > at least one production BIOS is in the wild which reports entries > > in DSDT (with no _STA method, so assumed enabled and present) > > that don't match MADT. > > > > Tested-by: Miguel Luis <miguel.luis@oracle.com> > > Reviewed-by: Gavin Shan <gshan@redhat.com> > > Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> > > --- > > arch/arm64/Kconfig | 1 + > > arch/arm64/kernel/acpi.c | 22 ++++++++++++++++++++++ > > 2 files changed, 23 insertions(+) > > > > diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig > > index 5d91259ee7b5..e8f2ef2312db 100644 > > --- a/arch/arm64/Kconfig > > +++ b/arch/arm64/Kconfig > > @@ -5,6 +5,7 @@ config ARM64 > > select ACPI_CCA_REQUIRED if ACPI > > select ACPI_GENERIC_GSI if ACPI > > select ACPI_GTDT if ACPI > > + select ACPI_HOTPLUG_CPU if ACPI_PROCESSOR > > ACPI_HOTPLUG_CPU depends on (ACPI_PROCESSOR && HOTPLUG_CPU). It needs to be: > > select ACPI_HOTPLUG_CPU if ACPI_PROCESSOR && HOTPLUG_CPU > > Otherwise, we can have compiling error with the following configurations. > > CONFIG_ACPI_PROCESSOR=y > CONFIG_HOTPLUG_CPU=n > CONFIG_ACPI_HOTPLUG_CPU=y > > arch/arm64/kernel/smp.c: In function ‘arch_unregister_cpu’: > arch/arm64/kernel/smp.c:563:9: error: implicit declaration of function ‘unregister_cpu’; did you mean ‘register_cpu’? [-Werror=implicit-function-declaration] > 563 | unregister_cpu(c); > | ^~~~~~~~~~~~~~ > | register_cpu Ah, I thought that in addition to the warning for unmet dependencies, kbuild would also leave the option off. I need to add SUSPEND=n and HIBERNATE=n to my build scripts. The fix matches what x86 does, so I'm ok with it. > Since the series has been queued to Catalin's "for-next/vcpu-hotplug" branch, I > guess the easiest way would be to fix it in place with Catalin's help. I wasn't planning to rebase the branch unless there's something major. Since the doesn't happen with defconfig, it's doesn't affect bisection that much, so my preference would be for a fix on top of this branch (and with a Fixes: tag since the branch is stable). Thanks.
Hi Catalin, On 6/30/24 7:26 PM, Catalin Marinas wrote: > On Sun, Jun 30, 2024 at 10:39:04AM +1000, Gavin Shan wrote: >> On 5/29/24 11:34 PM, Jonathan Cameron wrote: >>> In order to move arch_register_cpu() to be called via the same path >>> for initially present CPUs described by ACPI and hotplugged CPUs >>> ACPI_HOTPLUG_CPU needs to be enabled. >>> >>> The protection against invalid IDs in acpi_map_cpu() is needed as >>> at least one production BIOS is in the wild which reports entries >>> in DSDT (with no _STA method, so assumed enabled and present) >>> that don't match MADT. >>> >>> Tested-by: Miguel Luis <miguel.luis@oracle.com> >>> Reviewed-by: Gavin Shan <gshan@redhat.com> >>> Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> >>> --- >>> arch/arm64/Kconfig | 1 + >>> arch/arm64/kernel/acpi.c | 22 ++++++++++++++++++++++ >>> 2 files changed, 23 insertions(+) >>> >>> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig >>> index 5d91259ee7b5..e8f2ef2312db 100644 >>> --- a/arch/arm64/Kconfig >>> +++ b/arch/arm64/Kconfig >>> @@ -5,6 +5,7 @@ config ARM64 >>> select ACPI_CCA_REQUIRED if ACPI >>> select ACPI_GENERIC_GSI if ACPI >>> select ACPI_GTDT if ACPI >>> + select ACPI_HOTPLUG_CPU if ACPI_PROCESSOR >> >> ACPI_HOTPLUG_CPU depends on (ACPI_PROCESSOR && HOTPLUG_CPU). It needs to be: >> >> select ACPI_HOTPLUG_CPU if ACPI_PROCESSOR && HOTPLUG_CPU >> >> Otherwise, we can have compiling error with the following configurations. >> >> CONFIG_ACPI_PROCESSOR=y >> CONFIG_HOTPLUG_CPU=n >> CONFIG_ACPI_HOTPLUG_CPU=y >> >> arch/arm64/kernel/smp.c: In function ‘arch_unregister_cpu’: >> arch/arm64/kernel/smp.c:563:9: error: implicit declaration of function ‘unregister_cpu’; did you mean ‘register_cpu’? [-Werror=implicit-function-declaration] >> 563 | unregister_cpu(c); >> | ^~~~~~~~~~~~~~ >> | register_cpu > > Ah, I thought that in addition to the warning for unmet dependencies, > kbuild would also leave the option off. I need to add SUSPEND=n and > HIBERNATE=n to my build scripts. > > The fix matches what x86 does, so I'm ok with it. > Ok, thanks for your confirm. >> Since the series has been queued to Catalin's "for-next/vcpu-hotplug" branch, I >> guess the easiest way would be to fix it in place with Catalin's help. > > I wasn't planning to rebase the branch unless there's something major. > Since the doesn't happen with defconfig, it's doesn't affect bisection > that much, so my preference would be for a fix on top of this branch > (and with a Fixes: tag since the branch is stable). > Sure, I've posted a fix, applicable to branch 'for-next/vcpu-hotplug'. https://lists.infradead.org/pipermail/linux-arm-kernel/2024-June/939690.html Thanks, Gavin
diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig index 5d91259ee7b5..e8f2ef2312db 100644 --- a/arch/arm64/Kconfig +++ b/arch/arm64/Kconfig @@ -5,6 +5,7 @@ config ARM64 select ACPI_CCA_REQUIRED if ACPI select ACPI_GENERIC_GSI if ACPI select ACPI_GTDT if ACPI + select ACPI_HOTPLUG_CPU if ACPI_PROCESSOR select ACPI_IORT if ACPI select ACPI_REDUCED_HARDWARE_ONLY if ACPI select ACPI_MCFG if (ACPI && PCI) diff --git a/arch/arm64/kernel/acpi.c b/arch/arm64/kernel/acpi.c index e0e7b93c16cc..9360ba86678b 100644 --- a/arch/arm64/kernel/acpi.c +++ b/arch/arm64/kernel/acpi.c @@ -30,6 +30,7 @@ #include <linux/pgtable.h> #include <acpi/ghes.h> +#include <acpi/processor.h> #include <asm/cputype.h> #include <asm/cpu_ops.h> #include <asm/daifflags.h> @@ -423,6 +424,27 @@ void arch_reserve_mem_area(acpi_physical_address addr, size_t size) memblock_mark_nomap(addr, size); } +#ifdef CONFIG_ACPI_HOTPLUG_CPU +int acpi_map_cpu(acpi_handle handle, phys_cpuid_t physid, u32 apci_id, + int *pcpu) +{ + /* If an error code is passed in this stub can't fix it */ + if (*pcpu < 0) { + pr_warn_once("Unable to map CPU to valid ID\n"); + return *pcpu; + } + + return 0; +} +EXPORT_SYMBOL(acpi_map_cpu); + +int acpi_unmap_cpu(int cpu) +{ + return 0; +} +EXPORT_SYMBOL(acpi_unmap_cpu); +#endif /* CONFIG_ACPI_HOTPLUG_CPU */ + #ifdef CONFIG_ACPI_FFH /* * Implements ARM64 specific callbacks to support ACPI FFH Operation Region as