On Tue, 23 Dec 2025 at 12:02, Ashish Anand <[email protected]> wrote:
>
> From: "ashish.a6" <[email protected]>
>
> Currently, QEMU implements the 'Wait For Event' (WFE) instruction as a
> a simple yield. This causes high host CPU usage because Guest
> RTOS idle loops effectively become busy-wait loops.
>
> To improve efficiency, this patch transitions WFE to use the architectural
> 'Halt' state (EXCP_HLT) for M-profile CPUs. This allows the host thread
> to sleep when the guest is idle.
> diff --git a/target/arm/cpu.h b/target/arm/cpu.h
> index 39f2b2e54d..44433a444c 100644
> --- a/target/arm/cpu.h
> +++ b/target/arm/cpu.h
> @@ -639,6 +639,7 @@ typedef struct CPUArchState {
> uint32_t nsacr;
> uint32_t ltpsize;
> uint32_t vpr;
> + uint32_t event_register;
> } v7m;
One more small thing that I thought of -- although we're only
implementing WFE support for M-profile here, the concept of the
event register is the same for A-profile. So we should put the
field in the top level of the CPU state struct, not inside the v7m
sub-struct. That way it's available for us to use and share
if and when we ever do it for A-profile.
> +static bool m_event_needed(void *opaque)
> +{
> + ARMCPU *cpu = opaque;
> + /* Only save the state if the event register is set (non-zero) */
> + return cpu->env.v7m.event_register != 0;
> +}
> +
> +static const VMStateDescription vmstate_m_event = {
> + .name = "cpu/m/event",
> + .version_id = 1,
> + .minimum_version_id = 1,
> + .needed = m_event_needed,
> + .fields = (const VMStateField[]) {
> + VMSTATE_UINT32(env.v7m.event_register, ARMCPU),
> + VMSTATE_END_OF_LIST()
> + }
> +};
...and so similarly the migration handling should not have
anything M-profile specific to it.
thanks
-- PMM
