mbox series

[RFC,00/32] ACPI/arm64: add support for virtual cpuhotplug

Message ID 20230203135043.409192-1-james.morse@arm.com
Headers show
Series ACPI/arm64: add support for virtual cpuhotplug | expand

Message

James Morse Feb. 3, 2023, 1:50 p.m. UTC
Hello!

This series adds what looks like cpuhotplug support to arm64 for use in
virtual machines. It does this by moving the cpu_register() calls for
architectures that support ACPI out of the arch code by using
GENERIC_CPU_DEVICES, then into the ACPI processor driver.

The kubernetes folk really want to be able to add CPUs to an existing VM,
in exactly the same way they do on x86. The use-case is pre-booting guests
with one CPU, then adding the number that were actually needed when the
workload is provisioned.

Wait? Doesn't arm64 support cpuhotplug already!?
In the arm world, cpuhotplug gets used to mean removing the power from a CPU.
The CPU is offline, and remains present. For x86, and ACPI, cpuhotplug
has the additional step of physically removing the CPU, so that it isn't
present anymore.

Arm64 doesn't support this, and can't support it: CPUs are really a slice
of the SoC, and there is not enough information in the existing ACPI tables
to describe which bits of the slice also got removed. Without a reference
machine: adding this support to the spec is a wild goose chase.

Critically: everything described in the firmware tables must remain present.

For a virtual machine this is easy as all the other bits of 'virtual SoC'
are emulated, so they can (and do) remain present when a vCPU is 'removed'.

On a system that supports cpuhotplug the MADT has to describe every possible
CPU at boot. Under KVM, the vGIC needs to know about every possible vCPU before
the guest is started.
With these constraints, virtual-cpuhotplug is really just a hypervisor/firmware
policy about which CPUs can be brought online.

This series adds support for virtual-cpuhotplug as exactly that: firmware
policy. This may even work on a physical machine too; for a guest the part of
firmware is played by the VMM. (typically Qemu).

PSCI support is modified to return 'DENIED' if the CPU can't be brought
online/enabled yet. The CPU object's _STA method's enabled bit is used to
indicate firmware's current disposition. If the CPU has its enabled bit clear,
it will not be registered with sysfs, and attempts to bring it online will
fail. The notifications that _STA has changed its value then work in the same
way as physical hotplug, and firmware can cause the CPU to be registered some
time later, allowing it to be brought online.

This creates something that looks like cpuhotplug to user-space, as the sysfs
files appear and disappear, and the udev notifications look the same.

One notable difference is the CPU present mask, which is exposed via sysfs.
Because the CPUs remain present throughout, they can still be seen in that mask.
This value does get used by webbrowsers to estimate the number of CPUs
as the CPU online mask is constantly changed on mobile phones.

Linux is tolerant of PSCI returning errors, as its always been allowed to do
that. To avoid confusing OS that can't tolerate this, we needed an additional
bit in the MADT GICC flags. This series copies ACPI_MADT_ONLINE_CAPABLE, which
appears to be for this purpose, but calls it ACPI_MADT_GICC_CPU_CAPABLE as it
has a different bit position in the GICC.

This code is unconditionally enabled for all ACPI architectures.
If there are problems with firmware tables on some devices, the CPUs will
already be online by the time the acpi_processor_make_enabled() is called.
A mismatch here causes a firmware-bug message and kernel taint. This should
only affect people with broken firmware who also boot with maxcpus=1, and
bring CPUs online later.

I had a go at switching the remaining architectures over to GENERIC_CPU_DEVICES,
so that the Kconfig symbol can be removed, but I got stuck with powerpc
and s390.


The first patch has already been posted as a fix here:
https://www.spinics.net/lists/linux-ia64/msg21920.html
I've only build tested Loongarch and ia64.


If folk want to play along at home, you'll need a copy of Qemu that supports this.
https://github.com/salil-mehta/qemu.git salil/virt-cpuhp-armv8/rfc-v1-port29092022.psci.present

You'll need to fix the numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
to match your host kernel. Replace your '-smp' argument with something like:
| -smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1

then feed the following to the Qemu montior;
| (qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
| (qemu) device_del cpu1


This series is based on v6.2-rc3, and can be retrieved from:
https://git.kernel.org/pub/scm/linux/kernel/git/morse/linux.git/ virtual_cpu_hotplug/rfc/v1


Thanks,

James Morse (29):
  ia64: Fix build error due to switch case label appearing next to
    declaration
  ACPI: Move ACPI_HOTPLUG_CPU to be enabled per architecture
  drivers: base: Use present CPUs in GENERIC_CPU_DEVICES
  drivers: base: Allow parts of GENERIC_CPU_DEVICES to be overridden
  drivers: base: Move cpu_dev_init() after node_dev_init()
  arm64: setup: Switch over to GENERIC_CPU_DEVICES using
    arch_register_cpu()
  ia64/topology: Switch over to GENERIC_CPU_DEVICES
  x86/topology: Switch over to GENERIC_CPU_DEVICES
  LoongArch: Switch over to GENERIC_CPU_DEVICES
  arch_topology: Make register_cpu_capacity_sysctl() tolerant to late
    CPUs
  ACPI: processor: Add support for processors described as container
    packages
  ACPI: processor: Register CPUs that are online, but not described in
    the DSDT
  ACPI: processor: Register all CPUs from acpi_processor_get_info()
  ACPI: Rename ACPI_HOTPLUG_CPU to include 'present'
  ACPI: Move acpi_bus_trim_one() before acpi_scan_hot_remove()
  ACPI: Rename acpi_processor_hotadd_init and remove pre-processor
    guards
  ACPI: Add post_eject to struct acpi_scan_handler for cpu hotplug
  ACPI: Check _STA present bit before making CPUs not present
  ACPI: Warn when the present bit changes but the feature is not enabled
  drivers: base: Implement weak arch_unregister_cpu()
  LoongArch: Use the __weak version of arch_unregister_cpu()
  arm64: acpi: Move get_cpu_for_acpi_id() to a header
  ACPICA: Add new MADT GICC flags fields [code first?]
  arm64, irqchip/gic-v3, ACPI: Move MADT GICC enabled check into a
    helper
  irqchip/gic-v3: Don't return errors from gic_acpi_match_gicc()
  irqchip/gic-v3: Add support for ACPI's disabled but 'online capable'
    CPUs
  ACPI: add support to register CPUs based on the _STA enabled bit
  arm64: document virtual CPU hotplug's expectations
  cpumask: Add enabled cpumask for present CPUs that can be brought
    online

Jean-Philippe Brucker (3):
  arm64: psci: Ignore DENIED CPUs
  KVM: arm64: Pass hypercalls to userspace
  KVM: arm64: Pass PSCI calls to userspace

 Documentation/arm64/cpu-hotplug.rst       |  79 ++++++++++++
 Documentation/arm64/index.rst             |   1 +
 Documentation/virt/kvm/api.rst            |  31 ++++-
 Documentation/virt/kvm/arm/hypercalls.rst |   1 +
 arch/arm64/Kconfig                        |   1 +
 arch/arm64/include/asm/acpi.h             |  11 ++
 arch/arm64/include/asm/cpu.h              |   1 -
 arch/arm64/include/asm/kvm_host.h         |   2 +
 arch/arm64/kernel/acpi_numa.c             |  11 --
 arch/arm64/kernel/psci.c                  |   2 +-
 arch/arm64/kernel/setup.c                 |  13 +-
 arch/arm64/kernel/smp.c                   |   5 +-
 arch/arm64/kvm/arm.c                      |  15 ++-
 arch/arm64/kvm/hypercalls.c               |  28 ++++-
 arch/arm64/kvm/psci.c                     |  13 ++
 arch/ia64/Kconfig                         |   2 +
 arch/ia64/include/asm/acpi.h              |   2 +-
 arch/ia64/include/asm/cpu.h               |  11 --
 arch/ia64/kernel/acpi.c                   |   6 +-
 arch/ia64/kernel/setup.c                  |   2 +-
 arch/ia64/kernel/sys_ia64.c               |   7 +-
 arch/ia64/kernel/topology.c               |  35 +-----
 arch/loongarch/Kconfig                    |   2 +
 arch/loongarch/kernel/topology.c          |  31 +----
 arch/x86/Kconfig                          |   2 +
 arch/x86/include/asm/cpu.h                |   6 -
 arch/x86/kernel/acpi/boot.c               |   4 +-
 arch/x86/kernel/topology.c                |  19 +--
 drivers/acpi/Kconfig                      |   5 +-
 drivers/acpi/acpi_processor.c             | 146 +++++++++++++++++-----
 drivers/acpi/processor_core.c             |   2 +-
 drivers/acpi/scan.c                       | 116 +++++++++++------
 drivers/base/arch_topology.c              |  38 ++++--
 drivers/base/cpu.c                        |  31 ++++-
 drivers/base/init.c                       |   2 +-
 drivers/firmware/psci/psci.c              |   2 +
 drivers/irqchip/irq-gic-v3.c              |  38 +++---
 include/acpi/acpi_bus.h                   |   1 +
 include/acpi/actbl2.h                     |   1 +
 include/kvm/arm_hypercalls.h              |   1 +
 include/kvm/arm_psci.h                    |   4 +
 include/linux/acpi.h                      |  10 +-
 include/linux/cpu.h                       |   6 +
 include/linux/cpumask.h                   |  25 ++++
 include/uapi/linux/kvm.h                  |   2 +
 kernel/cpu.c                              |   3 +
 46 files changed, 532 insertions(+), 244 deletions(-)
 create mode 100644 Documentation/arm64/cpu-hotplug.rst

Comments

Jonathan Cameron March 7, 2023, noon UTC | #1
On Fri,  3 Feb 2023 13:50:11 +0000
James Morse <james.morse@arm.com> wrote:


...

> On a system that supports cpuhotplug the MADT has to describe every possible
> CPU at boot. Under KVM, the vGIC needs to know about every possible vCPU before
> the guest is started.
> With these constraints, virtual-cpuhotplug is really just a hypervisor/firmware
> policy about which CPUs can be brought online.
> 
> This series adds support for virtual-cpuhotplug as exactly that: firmware
> policy. This may even work on a physical machine too; for a guest the part of
> firmware is played by the VMM. (typically Qemu).
> 
> PSCI support is modified to return 'DENIED' if the CPU can't be brought
> online/enabled yet. The CPU object's _STA method's enabled bit is used to
> indicate firmware's current disposition. If the CPU has its enabled bit clear,
> it will not be registered with sysfs, and attempts to bring it online will
> fail. The notifications that _STA has changed its value then work in the same
> way as physical hotplug, and firmware can cause the CPU to be registered some
> time later, allowing it to be brought online.

Hi James,

As we discussed on an LOD call a while back, I think that we need some path to
find out if the guest supports vCPU HP or not so that info can be queried by
an orchestrator / libvirt etc.  In general the entity responsible for allocating
extra vCPUs may not know what support the VM has for this feature.

There are various ways we could get this information into the VMM.
My immediate thought is to use one of the ACPI interfaces that lets us write
AML that can set an emulated register. A query to the VMM can check if this
register is set.

So options.

_OSI() - Deprecated on ARM64 so lets not use that ;)
_OSC() - Could add a bit to Table 6.13 Platform-Wide Capabilites in ACPI 6.5 spec.
         Given x86 has a similar online capable bit perhaps this is the best option
         though it is the one that requires a formal code first proposal to ASWG.
_OSC() - Could add a new UUID and put it under a suitable device - maybe all CPUs?
         You could definitely argue this feature is an operating system property.
_DSM() - Similar to OSC but always under a device.
         Whilst can be used for this I'm not sure it really matches intended usecase.

Assuming everyone agrees this bit of introspection is useful,
Rafael / other ACPI specialists: Any suggestions on how best to do this?

Jonathan





> 
> This creates something that looks like cpuhotplug to user-space, as the sysfs
> files appear and disappear, and the udev notifications look the same.
> 
> One notable difference is the CPU present mask, which is exposed via sysfs.
> Because the CPUs remain present throughout, they can still be seen in that mask.
> This value does get used by webbrowsers to estimate the number of CPUs
> as the CPU online mask is constantly changed on mobile phones.
> 
> Linux is tolerant of PSCI returning errors, as its always been allowed to do
> that. To avoid confusing OS that can't tolerate this, we needed an additional
> bit in the MADT GICC flags. This series copies ACPI_MADT_ONLINE_CAPABLE, which
> appears to be for this purpose, but calls it ACPI_MADT_GICC_CPU_CAPABLE as it
> has a different bit position in the GICC.
> 
> This code is unconditionally enabled for all ACPI architectures.
> If there are problems with firmware tables on some devices, the CPUs will
> already be online by the time the acpi_processor_make_enabled() is called.
> A mismatch here causes a firmware-bug message and kernel taint. This should
> only affect people with broken firmware who also boot with maxcpus=1, and
> bring CPUs online later.
> 
> I had a go at switching the remaining architectures over to GENERIC_CPU_DEVICES,
> so that the Kconfig symbol can be removed, but I got stuck with powerpc
> and s390.
> 
> 
> The first patch has already been posted as a fix here:
> https://www.spinics.net/lists/linux-ia64/msg21920.html
> I've only build tested Loongarch and ia64.
> 
> 
> If folk want to play along at home, you'll need a copy of Qemu that supports this.
> https://github.com/salil-mehta/qemu.git salil/virt-cpuhp-armv8/rfc-v1-port29092022.psci.present
> 
> You'll need to fix the numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
> to match your host kernel. Replace your '-smp' argument with something like:
> | -smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1
> 
> then feed the following to the Qemu montior;
> | (qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
> | (qemu) device_del cpu1
> 
> 
> This series is based on v6.2-rc3, and can be retrieved from:
> https://git.kernel.org/pub/scm/linux/kernel/git/morse/linux.git/ virtual_cpu_hotplug/rfc/v1
> 
> 
> Thanks,
> 
> James Morse (29):
>   ia64: Fix build error due to switch case label appearing next to
>     declaration
>   ACPI: Move ACPI_HOTPLUG_CPU to be enabled per architecture
>   drivers: base: Use present CPUs in GENERIC_CPU_DEVICES
>   drivers: base: Allow parts of GENERIC_CPU_DEVICES to be overridden
>   drivers: base: Move cpu_dev_init() after node_dev_init()
>   arm64: setup: Switch over to GENERIC_CPU_DEVICES using
>     arch_register_cpu()
>   ia64/topology: Switch over to GENERIC_CPU_DEVICES
>   x86/topology: Switch over to GENERIC_CPU_DEVICES
>   LoongArch: Switch over to GENERIC_CPU_DEVICES
>   arch_topology: Make register_cpu_capacity_sysctl() tolerant to late
>     CPUs
>   ACPI: processor: Add support for processors described as container
>     packages
>   ACPI: processor: Register CPUs that are online, but not described in
>     the DSDT
>   ACPI: processor: Register all CPUs from acpi_processor_get_info()
>   ACPI: Rename ACPI_HOTPLUG_CPU to include 'present'
>   ACPI: Move acpi_bus_trim_one() before acpi_scan_hot_remove()
>   ACPI: Rename acpi_processor_hotadd_init and remove pre-processor
>     guards
>   ACPI: Add post_eject to struct acpi_scan_handler for cpu hotplug
>   ACPI: Check _STA present bit before making CPUs not present
>   ACPI: Warn when the present bit changes but the feature is not enabled
>   drivers: base: Implement weak arch_unregister_cpu()
>   LoongArch: Use the __weak version of arch_unregister_cpu()
>   arm64: acpi: Move get_cpu_for_acpi_id() to a header
>   ACPICA: Add new MADT GICC flags fields [code first?]
>   arm64, irqchip/gic-v3, ACPI: Move MADT GICC enabled check into a
>     helper
>   irqchip/gic-v3: Don't return errors from gic_acpi_match_gicc()
>   irqchip/gic-v3: Add support for ACPI's disabled but 'online capable'
>     CPUs
>   ACPI: add support to register CPUs based on the _STA enabled bit
>   arm64: document virtual CPU hotplug's expectations
>   cpumask: Add enabled cpumask for present CPUs that can be brought
>     online
> 
> Jean-Philippe Brucker (3):
>   arm64: psci: Ignore DENIED CPUs
>   KVM: arm64: Pass hypercalls to userspace
>   KVM: arm64: Pass PSCI calls to userspace
> 
>  Documentation/arm64/cpu-hotplug.rst       |  79 ++++++++++++
>  Documentation/arm64/index.rst             |   1 +
>  Documentation/virt/kvm/api.rst            |  31 ++++-
>  Documentation/virt/kvm/arm/hypercalls.rst |   1 +
>  arch/arm64/Kconfig                        |   1 +
>  arch/arm64/include/asm/acpi.h             |  11 ++
>  arch/arm64/include/asm/cpu.h              |   1 -
>  arch/arm64/include/asm/kvm_host.h         |   2 +
>  arch/arm64/kernel/acpi_numa.c             |  11 --
>  arch/arm64/kernel/psci.c                  |   2 +-
>  arch/arm64/kernel/setup.c                 |  13 +-
>  arch/arm64/kernel/smp.c                   |   5 +-
>  arch/arm64/kvm/arm.c                      |  15 ++-
>  arch/arm64/kvm/hypercalls.c               |  28 ++++-
>  arch/arm64/kvm/psci.c                     |  13 ++
>  arch/ia64/Kconfig                         |   2 +
>  arch/ia64/include/asm/acpi.h              |   2 +-
>  arch/ia64/include/asm/cpu.h               |  11 --
>  arch/ia64/kernel/acpi.c                   |   6 +-
>  arch/ia64/kernel/setup.c                  |   2 +-
>  arch/ia64/kernel/sys_ia64.c               |   7 +-
>  arch/ia64/kernel/topology.c               |  35 +-----
>  arch/loongarch/Kconfig                    |   2 +
>  arch/loongarch/kernel/topology.c          |  31 +----
>  arch/x86/Kconfig                          |   2 +
>  arch/x86/include/asm/cpu.h                |   6 -
>  arch/x86/kernel/acpi/boot.c               |   4 +-
>  arch/x86/kernel/topology.c                |  19 +--
>  drivers/acpi/Kconfig                      |   5 +-
>  drivers/acpi/acpi_processor.c             | 146 +++++++++++++++++-----
>  drivers/acpi/processor_core.c             |   2 +-
>  drivers/acpi/scan.c                       | 116 +++++++++++------
>  drivers/base/arch_topology.c              |  38 ++++--
>  drivers/base/cpu.c                        |  31 ++++-
>  drivers/base/init.c                       |   2 +-
>  drivers/firmware/psci/psci.c              |   2 +
>  drivers/irqchip/irq-gic-v3.c              |  38 +++---
>  include/acpi/acpi_bus.h                   |   1 +
>  include/acpi/actbl2.h                     |   1 +
>  include/kvm/arm_hypercalls.h              |   1 +
>  include/kvm/arm_psci.h                    |   4 +
>  include/linux/acpi.h                      |  10 +-
>  include/linux/cpu.h                       |   6 +
>  include/linux/cpumask.h                   |  25 ++++
>  include/uapi/linux/kvm.h                  |   2 +
>  kernel/cpu.c                              |   3 +
>  46 files changed, 532 insertions(+), 244 deletions(-)
>  create mode 100644 Documentation/arm64/cpu-hotplug.rst
>
James Morse March 13, 2023, 3:50 p.m. UTC | #2
Hi Jonathan,

On 07/03/2023 12:00, Jonathan Cameron wrote:
> On Fri,  3 Feb 2023 13:50:11 +0000
> James Morse <james.morse@arm.com> wrote:

>> On a system that supports cpuhotplug the MADT has to describe every possible
>> CPU at boot. Under KVM, the vGIC needs to know about every possible vCPU before
>> the guest is started.
>> With these constraints, virtual-cpuhotplug is really just a hypervisor/firmware
>> policy about which CPUs can be brought online.
>>
>> This series adds support for virtual-cpuhotplug as exactly that: firmware
>> policy. This may even work on a physical machine too; for a guest the part of
>> firmware is played by the VMM. (typically Qemu).
>>
>> PSCI support is modified to return 'DENIED' if the CPU can't be brought
>> online/enabled yet. The CPU object's _STA method's enabled bit is used to
>> indicate firmware's current disposition. If the CPU has its enabled bit clear,
>> it will not be registered with sysfs, and attempts to bring it online will
>> fail. The notifications that _STA has changed its value then work in the same
>> way as physical hotplug, and firmware can cause the CPU to be registered some
>> time later, allowing it to be brought online.

> As we discussed on an LOD call a while back, I think that we need some path to
> find out if the guest supports vCPU HP or not so that info can be queried by
> an orchestrator / libvirt etc.  In general the entity responsible for allocating
> extra vCPUs may not know what support the VM has for this feature.

I agree. For arm64 this is going to be important if/when there are machines that do
physical hotplug of CPUs too.


> There are various ways we could get this information into the VMM.
> My immediate thought is to use one of the ACPI interfaces that lets us write
> AML that can set an emulated register. A query to the VMM can check if this
> register is set.
> 
> So options.
> 
> _OSI() - Deprecated on ARM64 so lets not use that ;)

News to me, I've only just discovered it!


> _OSC() - Could add a bit to Table 6.13 Platform-Wide Capabilites in ACPI 6.5 spec.
>          Given x86 has a similar online capable bit perhaps this is the best option
>          though it is the one that requires a formal code first proposal to ASWG.

I've had a go at writing this one:
https://gitlab.arm.com/linux-arm/linux-jm/-/commit/220b0d8b0261d7467c8705e6f614d57325798859

It'll appear in the v1 of the series once the kernel and qemu bits are all lined up again.


Thanks,

James


> _OSC() - Could add a new UUID and put it under a suitable device - maybe all CPUs?
>          You could definitely argue this feature is an operating system property.
> _DSM() - Similar to OSC but always under a device.
>          Whilst can be used for this I'm not sure it really matches intended usecase.
> 
> Assuming everyone agrees this bit of introspection is useful,
> Rafael / other ACPI specialists: Any suggestions on how best to do this?
Jonathan Cameron March 14, 2023, 11:02 a.m. UTC | #3
On Mon, 13 Mar 2023 15:50:52 +0000
James Morse <james.morse@arm.com> wrote:

> Hi Jonathan,
> 
> On 07/03/2023 12:00, Jonathan Cameron wrote:
> > On Fri,  3 Feb 2023 13:50:11 +0000
> > James Morse <james.morse@arm.com> wrote:  
> 
> >> On a system that supports cpuhotplug the MADT has to describe every possible
> >> CPU at boot. Under KVM, the vGIC needs to know about every possible vCPU before
> >> the guest is started.
> >> With these constraints, virtual-cpuhotplug is really just a hypervisor/firmware
> >> policy about which CPUs can be brought online.
> >>
> >> This series adds support for virtual-cpuhotplug as exactly that: firmware
> >> policy. This may even work on a physical machine too; for a guest the part of
> >> firmware is played by the VMM. (typically Qemu).
> >>
> >> PSCI support is modified to return 'DENIED' if the CPU can't be brought
> >> online/enabled yet. The CPU object's _STA method's enabled bit is used to
> >> indicate firmware's current disposition. If the CPU has its enabled bit clear,
> >> it will not be registered with sysfs, and attempts to bring it online will
> >> fail. The notifications that _STA has changed its value then work in the same
> >> way as physical hotplug, and firmware can cause the CPU to be registered some
> >> time later, allowing it to be brought online.  
> 
> > As we discussed on an LOD call a while back, I think that we need some path to
> > find out if the guest supports vCPU HP or not so that info can be queried by
> > an orchestrator / libvirt etc.  In general the entity responsible for allocating
> > extra vCPUs may not know what support the VM has for this feature.  
> 
> I agree. For arm64 this is going to be important if/when there are machines that do
> physical hotplug of CPUs too.
> 
> 
> > There are various ways we could get this information into the VMM.
> > My immediate thought is to use one of the ACPI interfaces that lets us write
> > AML that can set an emulated register. A query to the VMM can check if this
> > register is set.
> > 
> > So options.
> > 
> > _OSI() - Deprecated on ARM64 so lets not use that ;)  
> 
> News to me, I've only just discovered it!
> 
> 
> > _OSC() - Could add a bit to Table 6.13 Platform-Wide Capabilites in ACPI 6.5 spec.
> >          Given x86 has a similar online capable bit perhaps this is the best option
> >          though it is the one that requires a formal code first proposal to ASWG.  
> 
> I've had a go at writing this one:
> https://gitlab.arm.com/linux-arm/linux-jm/-/commit/220b0d8b0261d7467c8705e6f614d57325798859
Gavin Shan March 29, 2023, 2:35 a.m. UTC | #4
Hi James,

On 2/3/23 9:50 PM, James Morse wrote:

[...]

> 
> 
> The first patch has already been posted as a fix here:
> https://www.spinics.net/lists/linux-ia64/msg21920.html
> I've only build tested Loongarch and ia64.
> 

It has been merged to upstream.

> 
> If folk want to play along at home, you'll need a copy of Qemu that supports this.
> https://github.com/salil-mehta/qemu.git salil/virt-cpuhp-armv8/rfc-v1-port29092022.psci.present
> 
> You'll need to fix the numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
> to match your host kernel. Replace your '-smp' argument with something like:
> | -smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1
> 
> then feed the following to the Qemu montior;
> | (qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
> | (qemu) device_del cpu1
> 
> 
> This series is based on v6.2-rc3, and can be retrieved from:
> https://git.kernel.org/pub/scm/linux/kernel/git/morse/linux.git/ virtual_cpu_hotplug/rfc/v1
> 

I give it a try, but the hot-added CPU needs to be put into online
state manually. I'm not sure if it's expected or not.

     /home/gavin/sandbox/qemu.main/build/qemu-system-aarch64                \
     -accel kvm,dirty-ring-size=65536                                       \
     -machine virt,gic-version=host,nvdimm=on                               \
     -cpu host -smp maxcpus=8,cpus=1,sockets=1,clusters=1,cores=8,threads=1 \
     -m 1024M,slots=16,maxmem=64G                                           \
     -object memory-backend-ram,id=mem0,size=1024M                          \
     -numa node,nodeid=0,memdev=mem0                                        \
     -L /home/gavin/sandbox/qemu.main/build/pc-bios                         \
     -monitor none -serial mon:stdio -nographic -gdb tcp::1234              \
     -bios /home/gavin/sandbox/qemu.main/build/pc-bios/edk2-aarch64-code.fd \
     -kernel /home/gavin/sandbox/linux.guest/arch/arm64/boot/Image          \
     -initrd /home/gavin/sandbox/images/rootfs.cpio.xz                      \
     -append memhp_default_state=online_movable                             \
        :
        :
     guest# cat /proc/cpuinfo | grep "CPU implementer" | wc -l
     1
     (qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
     guest# cat /proc/cpuinfo | grep "CPU implementer" | wc -l
     1
     guest# echo 1 > /sys/devices/system/cpu/cpu1/online
     guest# cat /proc/cpuinfo | grep "CPU implementer" | wc -l
     2
     (qemu) device_del cpu1
     guest# cat /proc/cpuinfo | grep "CPU implementer" | wc -l
     1

Note that the QEMU binary is directly built from Salil's repository and
the kernel image is built from v6.3-rc4, plus this patchset excluding the
first patch since it has been merged.

Thanks,
Gavin
Shaoqin Huang March 29, 2023, 5:52 a.m. UTC | #5
Hi James,

On 2/3/23 21:50, James Morse wrote:
> Hello!
> 
> This series adds what looks like cpuhotplug support to arm64 for use in
> virtual machines. It does this by moving the cpu_register() calls for
> architectures that support ACPI out of the arch code by using
> GENERIC_CPU_DEVICES, then into the ACPI processor driver.
> 
> The kubernetes folk really want to be able to add CPUs to an existing VM,
> in exactly the same way they do on x86. The use-case is pre-booting guests
> with one CPU, then adding the number that were actually needed when the
> workload is provisioned.
> 
> Wait? Doesn't arm64 support cpuhotplug already!?
> In the arm world, cpuhotplug gets used to mean removing the power from a CPU.
> The CPU is offline, and remains present. For x86, and ACPI, cpuhotplug
> has the additional step of physically removing the CPU, so that it isn't
> present anymore.
> 
> Arm64 doesn't support this, and can't support it: CPUs are really a slice
> of the SoC, and there is not enough information in the existing ACPI tables
> to describe which bits of the slice also got removed. Without a reference
> machine: adding this support to the spec is a wild goose chase.
> 
> Critically: everything described in the firmware tables must remain present.
> 
> For a virtual machine this is easy as all the other bits of 'virtual SoC'
> are emulated, so they can (and do) remain present when a vCPU is 'removed'.
> 
> On a system that supports cpuhotplug the MADT has to describe every possible
> CPU at boot. Under KVM, the vGIC needs to know about every possible vCPU before
> the guest is started.
> With these constraints, virtual-cpuhotplug is really just a hypervisor/firmware
> policy about which CPUs can be brought online.
> 
> This series adds support for virtual-cpuhotplug as exactly that: firmware
> policy. This may even work on a physical machine too; for a guest the part of
> firmware is played by the VMM. (typically Qemu).
> 
> PSCI support is modified to return 'DENIED' if the CPU can't be brought
> online/enabled yet. The CPU object's _STA method's enabled bit is used to
> indicate firmware's current disposition. If the CPU has its enabled bit clear,
> it will not be registered with sysfs, and attempts to bring it online will
> fail. The notifications that _STA has changed its value then work in the same
> way as physical hotplug, and firmware can cause the CPU to be registered some
> time later, allowing it to be brought online.
> 
> This creates something that looks like cpuhotplug to user-space, as the sysfs
> files appear and disappear, and the udev notifications look the same.
> 
> One notable difference is the CPU present mask, which is exposed via sysfs.
> Because the CPUs remain present throughout, they can still be seen in that mask.
> This value does get used by webbrowsers to estimate the number of CPUs
> as the CPU online mask is constantly changed on mobile phones.
> 
> Linux is tolerant of PSCI returning errors, as its always been allowed to do
> that. To avoid confusing OS that can't tolerate this, we needed an additional
> bit in the MADT GICC flags. This series copies ACPI_MADT_ONLINE_CAPABLE, which
> appears to be for this purpose, but calls it ACPI_MADT_GICC_CPU_CAPABLE as it
> has a different bit position in the GICC.
> 
> This code is unconditionally enabled for all ACPI architectures.
> If there are problems with firmware tables on some devices, the CPUs will
> already be online by the time the acpi_processor_make_enabled() is called.
> A mismatch here causes a firmware-bug message and kernel taint. This should
> only affect people with broken firmware who also boot with maxcpus=1, and
> bring CPUs online later.
> 
> I had a go at switching the remaining architectures over to GENERIC_CPU_DEVICES,
> so that the Kconfig symbol can be removed, but I got stuck with powerpc
> and s390.
> 
> 
> The first patch has already been posted as a fix here:
> https://www.spinics.net/lists/linux-ia64/msg21920.html
> I've only build tested Loongarch and ia64.
> 
> 
> If folk want to play along at home, you'll need a copy of Qemu that supports this.
> https://github.com/salil-mehta/qemu.git salil/virt-cpuhp-armv8/rfc-v1-port29092022.psci.present
> 
> You'll need to fix the numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
> to match your host kernel. Replace your '-smp' argument with something like:
> | -smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1
> 
> then feed the following to the Qemu montior;
> | (qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
> | (qemu) device_del cpu1
> 
> 
> This series is based on v6.2-rc3, and can be retrieved from:
> https://git.kernel.org/pub/scm/linux/kernel/git/morse/linux.git/ virtual_cpu_hotplug/rfc/v1

I applied this patch series on v6.2-rc3 and using the QEMU cloned from 
the salil-mehta/qemu.git repo. But when I try to run the QEMU, it shows:

$ qemu-system-aarch64: -accel kvm: Failed to enable 
KVM_CAP_ARM_PSCI_TO_USER cap.

Here is the command I use:

$ qemu-system-aarch64
-machine virt
-bios /usr/share/qemu-efi-aarch64/QEMU_EFI.fd
-accel kvm
-m 4096
-smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1
-cpu host
-qmp unix:./src.socket,server,nowait
-hda ./XXX.qcow2
-serial unix:./src.serial,server,nowait
-monitor stdio

It seems something related to your notice: You'll need to fix the 
numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
to match your host kernel.

But I'm not actually understand what should I fix, since I haven't 
review the patch series. Could you give me some more information? Maybe 
I'm doing something wrong.

Thanks,

> 
> 
> Thanks,
> 
> James Morse (29):
>    ia64: Fix build error due to switch case label appearing next to
>      declaration
>    ACPI: Move ACPI_HOTPLUG_CPU to be enabled per architecture
>    drivers: base: Use present CPUs in GENERIC_CPU_DEVICES
>    drivers: base: Allow parts of GENERIC_CPU_DEVICES to be overridden
>    drivers: base: Move cpu_dev_init() after node_dev_init()
>    arm64: setup: Switch over to GENERIC_CPU_DEVICES using
>      arch_register_cpu()
>    ia64/topology: Switch over to GENERIC_CPU_DEVICES
>    x86/topology: Switch over to GENERIC_CPU_DEVICES
>    LoongArch: Switch over to GENERIC_CPU_DEVICES
>    arch_topology: Make register_cpu_capacity_sysctl() tolerant to late
>      CPUs
>    ACPI: processor: Add support for processors described as container
>      packages
>    ACPI: processor: Register CPUs that are online, but not described in
>      the DSDT
>    ACPI: processor: Register all CPUs from acpi_processor_get_info()
>    ACPI: Rename ACPI_HOTPLUG_CPU to include 'present'
>    ACPI: Move acpi_bus_trim_one() before acpi_scan_hot_remove()
>    ACPI: Rename acpi_processor_hotadd_init and remove pre-processor
>      guards
>    ACPI: Add post_eject to struct acpi_scan_handler for cpu hotplug
>    ACPI: Check _STA present bit before making CPUs not present
>    ACPI: Warn when the present bit changes but the feature is not enabled
>    drivers: base: Implement weak arch_unregister_cpu()
>    LoongArch: Use the __weak version of arch_unregister_cpu()
>    arm64: acpi: Move get_cpu_for_acpi_id() to a header
>    ACPICA: Add new MADT GICC flags fields [code first?]
>    arm64, irqchip/gic-v3, ACPI: Move MADT GICC enabled check into a
>      helper
>    irqchip/gic-v3: Don't return errors from gic_acpi_match_gicc()
>    irqchip/gic-v3: Add support for ACPI's disabled but 'online capable'
>      CPUs
>    ACPI: add support to register CPUs based on the _STA enabled bit
>    arm64: document virtual CPU hotplug's expectations
>    cpumask: Add enabled cpumask for present CPUs that can be brought
>      online
> 
> Jean-Philippe Brucker (3):
>    arm64: psci: Ignore DENIED CPUs
>    KVM: arm64: Pass hypercalls to userspace
>    KVM: arm64: Pass PSCI calls to userspace
> 
>   Documentation/arm64/cpu-hotplug.rst       |  79 ++++++++++++
>   Documentation/arm64/index.rst             |   1 +
>   Documentation/virt/kvm/api.rst            |  31 ++++-
>   Documentation/virt/kvm/arm/hypercalls.rst |   1 +
>   arch/arm64/Kconfig                        |   1 +
>   arch/arm64/include/asm/acpi.h             |  11 ++
>   arch/arm64/include/asm/cpu.h              |   1 -
>   arch/arm64/include/asm/kvm_host.h         |   2 +
>   arch/arm64/kernel/acpi_numa.c             |  11 --
>   arch/arm64/kernel/psci.c                  |   2 +-
>   arch/arm64/kernel/setup.c                 |  13 +-
>   arch/arm64/kernel/smp.c                   |   5 +-
>   arch/arm64/kvm/arm.c                      |  15 ++-
>   arch/arm64/kvm/hypercalls.c               |  28 ++++-
>   arch/arm64/kvm/psci.c                     |  13 ++
>   arch/ia64/Kconfig                         |   2 +
>   arch/ia64/include/asm/acpi.h              |   2 +-
>   arch/ia64/include/asm/cpu.h               |  11 --
>   arch/ia64/kernel/acpi.c                   |   6 +-
>   arch/ia64/kernel/setup.c                  |   2 +-
>   arch/ia64/kernel/sys_ia64.c               |   7 +-
>   arch/ia64/kernel/topology.c               |  35 +-----
>   arch/loongarch/Kconfig                    |   2 +
>   arch/loongarch/kernel/topology.c          |  31 +----
>   arch/x86/Kconfig                          |   2 +
>   arch/x86/include/asm/cpu.h                |   6 -
>   arch/x86/kernel/acpi/boot.c               |   4 +-
>   arch/x86/kernel/topology.c                |  19 +--
>   drivers/acpi/Kconfig                      |   5 +-
>   drivers/acpi/acpi_processor.c             | 146 +++++++++++++++++-----
>   drivers/acpi/processor_core.c             |   2 +-
>   drivers/acpi/scan.c                       | 116 +++++++++++------
>   drivers/base/arch_topology.c              |  38 ++++--
>   drivers/base/cpu.c                        |  31 ++++-
>   drivers/base/init.c                       |   2 +-
>   drivers/firmware/psci/psci.c              |   2 +
>   drivers/irqchip/irq-gic-v3.c              |  38 +++---
>   include/acpi/acpi_bus.h                   |   1 +
>   include/acpi/actbl2.h                     |   1 +
>   include/kvm/arm_hypercalls.h              |   1 +
>   include/kvm/arm_psci.h                    |   4 +
>   include/linux/acpi.h                      |  10 +-
>   include/linux/cpu.h                       |   6 +
>   include/linux/cpumask.h                   |  25 ++++
>   include/uapi/linux/kvm.h                  |   2 +
>   kernel/cpu.c                              |   3 +
>   46 files changed, 532 insertions(+), 244 deletions(-)
>   create mode 100644 Documentation/arm64/cpu-hotplug.rst
>
Gavin Shan April 3, 2023, 6:25 a.m. UTC | #6
Hi Shaoqin,

On 3/29/23 1:52 PM, Shaoqin Huang wrote:
> On 2/3/23 21:50, James Morse wrote:

[...]

>>
>> The first patch has already been posted as a fix here:
>> https://www.spinics.net/lists/linux-ia64/msg21920.html
>> I've only build tested Loongarch and ia64.
>>
>>
>> If folk want to play along at home, you'll need a copy of Qemu that supports this.
>> https://github.com/salil-mehta/qemu.git salil/virt-cpuhp-armv8/rfc-v1-port29092022.psci.present
>>
>> You'll need to fix the numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
>> to match your host kernel. Replace your '-smp' argument with something like:
>> | -smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1
>>
>> then feed the following to the Qemu montior;
>> | (qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
>> | (qemu) device_del cpu1
>>
>>
>> This series is based on v6.2-rc3, and can be retrieved from:
>> https://git.kernel.org/pub/scm/linux/kernel/git/morse/linux.git/ virtual_cpu_hotplug/rfc/v1
> 
> I applied this patch series on v6.2-rc3 and using the QEMU cloned from the salil-mehta/qemu.git repo. But when I try to run the QEMU, it shows:
> 
> $ qemu-system-aarch64: -accel kvm: Failed to enable KVM_CAP_ARM_PSCI_TO_USER cap.
> 
> Here is the command I use:
> 
> $ qemu-system-aarch64
> -machine virt
> -bios /usr/share/qemu-efi-aarch64/QEMU_EFI.fd
> -accel kvm
> -m 4096
> -smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1
> -cpu host
> -qmp unix:./src.socket,server,nowait
> -hda ./XXX.qcow2
> -serial unix:./src.serial,server,nowait
> -monitor stdio
> 
> It seems something related to your notice: You'll need to fix the numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
> to match your host kernel.
> 
> But I'm not actually understand what should I fix, since I haven't review the patch series. Could you give me some more information? Maybe I'm doing something wrong.
> 

When the kernel is rebased to v6.2.rc3, the two capabilities are conflictsing
between QEMU and host kernel. Please adjust them like below and have a try:

In qemu/linux-headers/linux/kvm.h

#define KVM_CAP_ARM_HVC_TO_USER 250 /* TODO: as per linux 6.1-rc2 */
#define KVM_CAP_ARM_PSCI_TO_USER 251 /* TODO: as per linux 6.1-rc2 */

In linux/include/uapi/linux/kvm.h

#define KVM_CAP_ARM_HVC_TO_USER 250
#define KVM_CAP_ARM_PSCI_TO_USER 251

Thanks,
Gavin
James Morse Sept. 12, 2023, 5:01 p.m. UTC | #7
Hi Gavin,

On 29/03/2023 03:35, Gavin Shan wrote:
> On 2/3/23 9:50 PM, James Morse wrote:

>> If folk want to play along at home, you'll need a copy of Qemu that supports this.
>> https://github.com/salil-mehta/qemu.git
>> salil/virt-cpuhp-armv8/rfc-v1-port29092022.psci.present
>>
>> You'll need to fix the numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
>> to match your host kernel. Replace your '-smp' argument with something like:
>> | -smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1
>>
>> then feed the following to the Qemu montior;
>> | (qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
>> | (qemu) device_del cpu1
>>
>>
>> This series is based on v6.2-rc3, and can be retrieved from:
>> https://git.kernel.org/pub/scm/linux/kernel/git/morse/linux.git/ virtual_cpu_hotplug/rfc/v1

> I give it a try, but the hot-added CPU needs to be put into online
> state manually. I'm not sure if it's expected or not.

This is expected. If you want the CPUs to be brought online automatically, you can add
udev rules to do that.


Thanks,

James
Gavin Shan Sept. 12, 2023, 10:38 p.m. UTC | #8
Hi James,

On 9/13/23 03:01, James Morse wrote:
> On 29/03/2023 03:35, Gavin Shan wrote:
>> On 2/3/23 9:50 PM, James Morse wrote:
> 
>>> If folk want to play along at home, you'll need a copy of Qemu that supports this.
>>> https://github.com/salil-mehta/qemu.git
>>> salil/virt-cpuhp-armv8/rfc-v1-port29092022.psci.present
>>>
>>> You'll need to fix the numbers of KVM_CAP_ARM_HVC_TO_USER and KVM_CAP_ARM_PSCI_TO_USER
>>> to match your host kernel. Replace your '-smp' argument with something like:
>>> | -smp cpus=1,maxcpus=3,cores=3,threads=1,sockets=1
>>>
>>> then feed the following to the Qemu montior;
>>> | (qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
>>> | (qemu) device_del cpu1
>>>
>>>
>>> This series is based on v6.2-rc3, and can be retrieved from:
>>> https://git.kernel.org/pub/scm/linux/kernel/git/morse/linux.git/ virtual_cpu_hotplug/rfc/v1
> 
>> I give it a try, but the hot-added CPU needs to be put into online
>> state manually. I'm not sure if it's expected or not.
> 
> This is expected. If you want the CPUs to be brought online automatically, you can add
> udev rules to do that.
> 

Yeah, I usually execute the following command to bring the CPU into online state,
after the vCPU is hot added by QMP command.

(qemu) device_add driver=host-arm-cpu,core-id=1,id=cpu1
guest# echo 1 > /sys/devices/system/cpu/cpux/online

James, the series was posted a while ago and do you have plan to respin
and post RFCv2 in near future? :)

Thanks,
Gavin