Re: [Qemu-devel] [RFC PATCH 1/5] suspend: add infrastructure

[Paolo Bonzini <pbonzini@redhat.com>]

On 01/11/2012 04:08 PM, Gerd Hoffmann wrote:
> +void qemu_system_suspend_request(qemu_irq wake_irq)
> +{
> +    if (suspend_wake_irq != NULL) {
> +        return;
> +    }
> +    cpu_stop_current();
> +    qemu_notify_event();
> +    suspend_wake_irq = wake_irq;
> +}
> +
> +void qemu_system_wakeup_request(void)
> +{
> +    if (suspend_wake_irq == NULL) {
> +        return;
> +    }
> +    reset_requested = 1;
> +    qemu_irq_raise(suspend_wake_irq);
> +    suspend_wake_irq = NULL;
> +}

The code in acpi.c is confusing, but it seems to me that IRQ is raised 
when the system _enters_ S3.  See especially xen-all.c.  It would be 
lowered by acpi_pm1_cnt_reset when the reset actually happens on the 
next main loop iteration.  However, acpi_pm1_cnt_reset has never been 
called since its introduction in commit eaba51c (acpi, acpi_piix, 
vt82c686: factor out PM1_CNT logic, 2011-03-25).

Overall, it seems to me that with your new infrastructure a global 
Notifier would be a better fit than a qemu_irq that the board would have 
to pass all the way.  A notifier would also remove the ugly Xen special 
casing.  However, this can be done separately.

Paolo