Re: [PATCH] mc146818rtc: add a way to generate RTC interrupts via QMP

[Markus Armbruster]

Philippe Mathieu-Daudé <philmd@linaro.org> writes:

> Hi Daniil, Markus,
>
> On 26/4/24 10:39, Markus Armbruster wrote:
>> Daniil Tatianin <d-tatianin@yandex-team.ru> writes:
>> 
>>> This can be used to force-synchronize the time in guest after a long
>>> stop-cont pause, which can be useful for serverless-type workload.
>>>
>>> Signed-off-by: Daniil Tatianin <d-tatianin@yandex-team.ru>
>>> ---
>>>   hw/rtc/mc146818rtc.c         | 15 +++++++++++++++
>>>   include/hw/rtc/mc146818rtc.h |  1 +
>>>   qapi/misc-target.json        | 16 ++++++++++++++++
>>>   3 files changed, 32 insertions(+)
>>>
>>> diff --git a/hw/rtc/mc146818rtc.c b/hw/rtc/mc146818rtc.c
>>> index f4c1869232..6980a78d5f 100644
>>> --- a/hw/rtc/mc146818rtc.c
>>> +++ b/hw/rtc/mc146818rtc.c
>>> @@ -116,6 +116,21 @@ void qmp_rtc_reset_reinjection(Error **errp)
>>>       }
>>>   }
>>>   +void qmp_rtc_notify(Error **errp)
>>> +{
>>> +    MC146818RtcState *s;
>>> +
>>> +    /*
>>> +     * See:
>>> +     * https://www.kernel.org/doc/Documentation/virtual/kvm/timekeeping.txt
>>> +     */
>>> +    QLIST_FOREACH(s, &rtc_devices, link) {
>>> +        s->cmos_data[RTC_REG_B] |= REG_B_UIE;
>>> +        s->cmos_data[RTC_REG_C] |= REG_C_IRQF | REG_C_UF;
>>> +        qemu_irq_raise(s->irq);
>>> +    }
>>> +}
>>> +
>> Note for later: qmp_rtc_notify() works on all realized mc146818rtc
>> devices.  Other kinds of RTC devices are silently ignored.  Just like
>> qmp_rtc_reset_reinjection().
>
> IMO to avoid any future ambiguity (in heterogeneous machines), this
> command must take a QOM device path (or a list of) and only notify
> those.

Let's compare:

• With QOM path:

  · You need to know the machine's RTC device(s).

    Unfortunately, this is bothersome, as the QOM path is not stable.

    For Q35, it's generally "/machine/unattached/device[N]/rtc", but N
    varies with configuration (TCG N=2, KVM N=3 for me), and it might
    vary with machine type version.  That's because the machine code
    creates ICH9-LPC without a proper name.  We do that a lot.  I hate
    it.

    Likewise for i440FX with PIIX3 instead of ICH9-LPC.

    For isapc, it's /machine/unattached/device[3].  I suspect the 3
    isn't reliable there, either.

    microvm doesn't seem to have an RTC by default.

  · If the device so named doesn't support IRQ inject, the command
    should fail.

  · Could be generalized to non-RTC devices when that's useful.

• Broadcast:

  · You don't need to know the machine's RTC device(s).

  · If there are multiple RTC devices that support IRQ inject, we inject
    for each of them.  There is no way to select specific RTCs.

  · If there is no RTC device that supports IRQ inject, the command does
    nothing silently.

    I don't like silent failures.  It could be made to fail instead.

If it wasn't for the unstable QOM path problem, I'd advise against
the broadcast interface.

Thoughts?