Re: [Qemu-devel] [PATCH v3] e1000: add interrupt mitigation support

[Vincenzo Maffione]

Actually it's not always zero because

- presave() calls e1000_mit_timer(), which sets mit_timer_on to 0 and
calls set_interrupt_cause(ICR);

- if there are pending events, set_interrupt_cause(ICR) detects a
raising edge, and doesn't filter out the interrupt since mit_timer_on
== 0. Therefore, the timer is reloaded and mit_timer_on is set to 1.

However, I've just realized that this is a mistake, since it would
need the post_load() to call qemu_mod_timer() when
mit_timer_on is migrated as "1", otherwise the interrupt delivery
would deadlock on the migrated system.

I think it's way better not to migrate mit_timer_on, so that the
migrated system sees mit_timer_on==0  and doesn't need to call
qemu_mod_timer() in post_load(). The first interrupt request
(set_interrupt_cause()) will cause an immediate interrupt.


Thanks,
  Vincenzo

2013/7/29 Stefan Hajnoczi <address@hidden>:
> On Fri, Jul 26, 2013 at 05:02:50PM +0200, Vincenzo Maffione wrote:
>> +static const VMStateDescription vmstate_e1000_mit_state = {
>> +    .name = "e1000/mit_state",
>> +    .version_id = 1,
>> +    .minimum_version_id = 1,
>> +    .minimum_version_id_old = 1,
>> +    .fields    = (VMStateField[]) {
>> +        VMSTATE_UINT32(mac_reg[RDTR], E1000State),
>> +        VMSTATE_UINT32(mac_reg[RADV], E1000State),
>> +        VMSTATE_UINT32(mac_reg[TADV], E1000State),
>> +        VMSTATE_UINT32(mac_reg[ITR], E1000State),
>> +        VMSTATE_BOOL(mit_timer_on, E1000State),
>
> Why is this migrated?  I it is always 0 since pre_save calls
> set_interrupt_cause(s, 0, s->mac_reg[ICR]).



-- 
Vincenzo Maffione