Re: [PATCH] arm64: KVM: export current vcpu->pause state via pseudo regs

From: Paolo Bonzini
Date: Thu Jul 31 2014 - 13:22:32 EST


Il 31/07/2014 19:04, Peter Maydell ha scritto:
> On 31 July 2014 17:57, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
>> Il 09/07/2014 15:55, Alex BennÃe ha scritto:
>>> To cleanly restore an SMP VM we need to ensure that the current pause
>>> state of each vcpu is correctly recorded. Things could get confused if
>>> the CPU starts running after migration restore completes when it was
>>> paused before it state was captured.
>>>
>>> I've done this by exposing a register (currently only 1 bit used) via
>>> the GET/SET_ONE_REG logic to pass the state between KVM and the VM
>>> controller (e.g. QEMU).
>>>
>>> Signed-off-by: Alex BennÃe <alex.bennee@xxxxxxxxxx>
>>> ---
>>> arch/arm64/include/uapi/asm/kvm.h | 8 +++++
>>> arch/arm64/kvm/guest.c | 61 ++++++++++++++++++++++++++++++++++++++-
>>> 2 files changed, 68 insertions(+), 1 deletion(-)
>>
>> Since it's a pseudo register anyway, would it make sense to use the
>> existing KVM_GET/SET_MP_STATE ioctl interface?
>
> That appears to be an x86-specific thing relating to
> IRQ chips.

No, it's not. It's just the state of the CPU, s390 will be using it too.

On x86 the states are uninitialized (UNINITIALIZED), stopped
(INIT_RECEIVED), running (RUNNABLE), halted (HALTED). CPU 0 starts in
RUNNABLE state, other CPUs start in UNINITIALIZED state. There are
x86-specific cases (uninitialized) and x86-isms (the INIT_RECEIVED
name), but the idea is widely applicable.

>> Also, how is KVM/ARM
>> representing (and passing to QEMU) the halted state of the
>> VCPU?
>
> We don't. In ARM the equivalent of x86 HLT (which is
> WFI, wait-for-interrupt) is allowed to resume at any time.
> So we don't need to care about saving and restoring
> whether we were sat in a WFI at point of migration.

What does ARM do if you have a WFI while interrupts are disabled? On
x86 after "cli;hlt" only an NMI will wake you up. With spurious
wakeups, it's pretty much guaranteed that you will break such "cli;hlt"
sequences.

Paolo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/