Re: [PATCH v2] target/arm/hvf: Fix WFI halting to stop idle vCPU spinning

["Scott J. Goldman" <scottjgo@gmail.com>]

On Fri Apr 17, 2026 at 2:57 AM PDT, Philippe Mathieu-Daudé wrote:
> Hi Scott,
>
> On 16/4/26 23:20, Scott J. Goldman wrote:
>> Philippe-- apologies for the nag here but this is a fix for a unfortunate
>> regression. Any chance you could take a look? Thanks!
>> 
>> On Thu Apr 9, 2026 at 11:18 PM PDT, Mohamed Mediouni wrote:
>>>
>>>
>>>> On 10. Apr 2026, at 07:50, Scott J. Goldman <scottjgo@gmail.com> wrote:
>>>>
>>>> Commit b5f8f77271 ("accel/hvf: Implement WFI without using pselect()")
>>>> changed hvf_wfi() from blocking the vCPU thread with pselect() to
>>>> returning EXCP_HLT, intending QEMU's main event loop to handle the
>>>> idle wait. However, cpu->halted was never set, so cpu_thread_is_idle()
>>>> always returns false and the vCPU thread spins at 100% CPU per core
>>>> while the guest is idle.
>>>>
>>>> Fix this by:
>>>>
>>>> 1. Setting cpu->halted = 1 in hvf_wfi() so the vCPU thread sleeps on
>>>>    halt_cond in qemu_process_cpu_events().
>>>>
>>>> 2. Arming a host-side QEMU_CLOCK_HOST timer to fire when the guest's
>>>>    virtual timer (CNTV_CVAL_EL0) would expire. This is necessary
>>>>    because HVF only delivers HV_EXIT_REASON_VTIMER_ACTIVATED during
>>>>    hv_vcpu_run(), which is not called while the CPU is halted. The
>>>>    timer callback mirrors the VTIMER_ACTIVATED handler: it raises the
>>>>    vtimer IRQ through the GIC and marks vtimer_masked, causing the
>>>>    interrupt delivery chain to wake the vCPU via qemu_cpu_kick().
>>>>
>>>> 3. Clearing cpu->halted in hvf_arch_vcpu_exec() when cpu_has_work()
>>>>    indicates a pending interrupt, and cancelling the WFI timer.
>>>>
>>>> Fixes: b5f8f77271 ("accel/hvf: Implement WFI without using pselect()")
>>>> Signed-off-by: Scott J. Goldman <scottjgo@gmail.com>
>>>
>>> Reviewed-by: Mohamed Mediouni <mohamed@unpredictable.fr>
>>>> ---
>>>> include/system/hvf_int.h |  1 +
>>>> target/arm/hvf/hvf.c     | 65 +++++++++++++++++++++++++++++++++++++++-
>>>> 2 files changed, 65 insertions(+), 1 deletion(-)
>
> I haven't looked at your patch yet. While looking at commit
> b5f8f77271 I noticed the previous a14afa985e3 ("accel/hvf: Skip
> WFI if CPU has work to do") is different in my local tree, where
> I have:
>
> -- >8 --
> diff --git a/target/arm/hvf/hvf.c b/target/arm/hvf/hvf.c
> index 5fc8f6bbbd9..1a76c9cf402 100644
> --- a/target/arm/hvf/hvf.c
> +++ b/target/arm/hvf/hvf.c
> @@ -2037,6 +2037,11 @@ static int hvf_wfi(CPUState *cpu)
>           return 0;
>       }
>
> +    if (cpu_test_interrupt(cpu, CPU_INTERRUPT_HARD | CPU_INTERRUPT_FIQ)) {
> +        /* Interrupt pending, no need to wait */
> +        return EXCP_INTERRUPT;
> +    }
> +
>       return EXCP_HLT;
>   }
>
> ---
>
> Does that help? Now I'm really confused.


Hi Phillipe-- Appreciate your help here. The good news is the bug is
very easy to reproduce (start a linux VM with 1 cpu, watch the vCPU on
the host will always spin 100%). The bad news is I double checked my
bisect and you can see before b5f8f77271 there is no problem, and after
the commit is introduced, the problem appears. I also applied the
additional cpu_test_interrupt() patch you suggested and it did not
change the behavior.

thanks
-sjg