Re: [PATCH] hw/timer/etraxfs_timer: Add vmstate for ETRAX timers

[Laurent Vivier]

Le 18/12/2021 à 03:28, Richard Henderson a écrit :
> On 12/17/21 3:37 PM, Philippe Mathieu-Daudé wrote:
> > ping?
> >
> > On 11/6/21 11:56, Philippe Mathieu-Daudé wrote:
> > > Add the vmstate for the ETRAX timers.
> > > This is in theory a migration compatibility break
> > > for the 'AXIS devboard 88' CRIS machine.
> > >
> > > Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
> > > ---
> > >   hw/timer/etraxfs_timer.c | 34 +++++++++++++++++++++++++++++++++-
> > >   1 file changed, 33 insertions(+), 1 deletion(-)
>
>
> In that it matches another similar timer device:
> Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
>
>
> > > +static const VMStateDescription vmstate_etraxfs = {
> > > +    .name = "etraxfs",
> > > +    .version_id = 0,
> > > +    .minimum_version_id = 0,
> > > +    .fields = (VMStateField[]) {
> > > +        VMSTATE_PTIMER(ptimer_t0, ETRAXTimerState),
> > > +        VMSTATE_PTIMER(ptimer_t1, ETRAXTimerState),
> > > +        VMSTATE_PTIMER(ptimer_wd, ETRAXTimerState),
> > > +
> > > +        VMSTATE_UINT32(wd_hits, ETRAXTimerState),
> > > +
> > > +        VMSTATE_UINT32(rw_tmr0_div, ETRAXTimerState),
> > > +        VMSTATE_UINT32(r_tmr0_data, ETRAXTimerState),
> > > +        VMSTATE_UINT32(rw_tmr0_ctrl, ETRAXTimerState),
> > > +
> > > +        VMSTATE_UINT32(rw_tmr1_div, ETRAXTimerState),
> > > +        VMSTATE_UINT32(r_tmr1_data, ETRAXTimerState),
> > > +        VMSTATE_UINT32(rw_tmr1_ctrl, ETRAXTimerState),
> > > +
> > > +        VMSTATE_UINT32(rw_wd_ctrl, ETRAXTimerState),
> > > +
> > > +        VMSTATE_UINT32(rw_intr_mask, ETRAXTimerState),
> > > +        VMSTATE_UINT32(rw_ack_intr, ETRAXTimerState),
> > > +        VMSTATE_UINT32(r_intr, ETRAXTimerState),
> > > +        VMSTATE_UINT32(r_masked_intr, ETRAXTimerState),
> > > +
> > > +        VMSTATE_END_OF_LIST()
> > > +    }
> > > +};
>
> What I don't understand is how these controls get applied to qemu_irq after vmload, here or in any 
> other device.  It seems like we should have some post_load hook that calls timer_update_irq, etc.

FWIW, in VMSTATE_PTIMER(), we use a vmstate_ptimer struct that registers a vmstate_info_timer with 
VMSTATE_TIMER_PTR(). vmstate_info_timer uses timer_get() to update or delete the timer when it is 
loaded.

Applied to my trivial-patches branch.

Thanks,
Laurent