arm64: kvm + RT

arm64: kvm + RT

[Jan Kiszka]

Hi all,

what's the state of getting kvm working (again?) along RT? Seems
HAVE_POSIX_CPU_TIMERS_TASK_WORK is in conflict with KVM but a
precondition for RT on recent kernels. What's the story behind that?

Thanks,
Jan

-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux

Re: arm64: kvm + RT

[Thomas Gleixner]

Jan,

On Mon, Oct 04 2021 at 19:54, Jan Kiszka wrote:
> what's the state of getting kvm working (again?) along RT? Seems
> HAVE_POSIX_CPU_TIMERS_TASK_WORK is in conflict with KVM but a
> precondition for RT on recent kernels. What's the story behind that?

the ARM64 KVM implementation is incorrect vs. TIF handling before
VMENTER which breaks the posix cpu timers task work assumptions and also
violates a few other expectations. AFAICT the arm64 folks are working on
that, but you have to ask them directly what the state of this is.

Thanks,

        tglx

Re: arm64: kvm + RT

[Jan Kiszka]

On 05.10.21 11:52, Thomas Gleixner wrote:
> Jan,
> 
> On Mon, Oct 04 2021 at 19:54, Jan Kiszka wrote:
>> what's the state of getting kvm working (again?) along RT? Seems
>> HAVE_POSIX_CPU_TIMERS_TASK_WORK is in conflict with KVM but a
>> precondition for RT on recent kernels. What's the story behind that?
> 
> the ARM64 KVM implementation is incorrect vs. TIF handling before
> VMENTER which breaks the posix cpu timers task work assumptions and also
> violates a few other expectations. AFAICT the arm64 folks are working on
> that, but you have to ask them directly what the state of this is.
> 

Thanks, then let's try this: Marc, what is the status of this?

Jan

-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux

Re: arm64: kvm + RT

[Mark Rutland]

On Tue, Oct 05, 2021 at 11:52:52AM +0200, Thomas Gleixner wrote:
> Jan,
> 
> On Mon, Oct 04 2021 at 19:54, Jan Kiszka wrote:
> > what's the state of getting kvm working (again?) along RT? Seems
> > HAVE_POSIX_CPU_TIMERS_TASK_WORK is in conflict with KVM but a
> > precondition for RT on recent kernels. What's the story behind that?
> 
> the ARM64 KVM implementation is incorrect vs. TIF handling before
> VMENTER which breaks the posix cpu timers task work assumptions and also
> violates a few other expectations. AFAICT the arm64 folks are working on
> that, but you have to ask them directly what the state of this is.

I think that was dealt with as of mainline commits:

* e1c6b9e1669e44fb ("entry: KVM: Allow use of generic KVM entry w/o full generic support")
  https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e1c6b9e1669e44fb7f9688e34e460b759e3b9187

* 6caa5812e2d126a0 ("KVM: arm64: Use generic KVM xfer to guest work function")
  https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6caa5812e2d126a0aa8a17816c1ba6f0a0c2b309

... unless I'm missing something.

Mark.

Re: arm64: kvm + RT

[Thomas Gleixner]

On Tue, Oct 05 2021 at 11:54, Mark Rutland wrote:
> On Tue, Oct 05, 2021 at 11:52:52AM +0200, Thomas Gleixner wrote:
>> Jan,
>> 
>> On Mon, Oct 04 2021 at 19:54, Jan Kiszka wrote:
>> > what's the state of getting kvm working (again?) along RT? Seems
>> > HAVE_POSIX_CPU_TIMERS_TASK_WORK is in conflict with KVM but a
>> > precondition for RT on recent kernels. What's the story behind that?
>> 
>> the ARM64 KVM implementation is incorrect vs. TIF handling before
>> VMENTER which breaks the posix cpu timers task work assumptions and also
>> violates a few other expectations. AFAICT the arm64 folks are working on
>> that, but you have to ask them directly what the state of this is.
>
> I think that was dealt with as of mainline commits:
>
> * e1c6b9e1669e44fb ("entry: KVM: Allow use of generic KVM entry w/o full generic support")
>   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e1c6b9e1669e44fb7f9688e34e460b759e3b9187
>
> * 6caa5812e2d126a0 ("KVM: arm64: Use generic KVM xfer to guest work function")
>   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6caa5812e2d126a0aa8a17816c1ba6f0a0c2b309
>
> ... unless I'm missing something.

Just a minor detail:

$ cd linus
$ git grep HAVE_POSIX_CPU_TIMERS_TASK_WORK
arch/x86/Kconfig:       select HAVE_POSIX_CPU_TIMERS_TASK_WORK
kernel/time/Kconfig:config HAVE_POSIX_CPU_TIMERS_TASK_WORK
kernel/time/Kconfig:    default y if POSIX_TIMERS && HAVE_POSIX_CPU_TIMERS_TASK_WORK

$cd rt
$ git grep HAVE_POSIX_CPU_TIMERS_TASK_WORK
arch/arm/Kconfig:       select ARCH_SUPPORTS_RT if HAVE_POSIX_CPU_TIMERS_TASK_WORK
arch/arm/Kconfig:       select HAVE_POSIX_CPU_TIMERS_TASK_WORK if !KVM
arch/arm64/Kconfig:     select ARCH_SUPPORTS_RT if HAVE_POSIX_CPU_TIMERS_TASK_WORK
arch/arm64/Kconfig:     select HAVE_POSIX_CPU_TIMERS_TASK_WORK if !KVM
arch/powerpc/Kconfig:   select ARCH_SUPPORTS_RT                 if HAVE_POSIX_CPU_TIMERS_TASK_WORK
arch/powerpc/Kconfig:   select HAVE_POSIX_CPU_TIMERS_TASK_WORK  if !KVM
arch/x86/Kconfig:       select HAVE_POSIX_CPU_TIMERS_TASK_WORK

Which is simply because arm64 does not yet have:

arch/arm64/Kconfig:     select HAVE_POSIX_CPU_TIMERS_TASK_WORK

set in mainline to indicate that it does support it unconditionally
including in KVM.

See?

Thanks,

        tglx