Xen devel - Dec 2005 - RE: RE: event delay issue on SMP machine when xen0 is SMPenabled

>From: Li, Xin B
>Sent: 2005年12月9日 16:50
>
>>> BTW, why vcpu other than vcpu0 won''t handle event by
default?
>>
>>We could allow any vcpu to process any pending notifications. It would
>>mean vcpus outside an irq''s affinity set could end up
processing an
>>interrupt and I''m not sure if that''s a good thing. It
would actually
>>slightly simplify evtchn.c though (no need to apply a per-cpu mask to
>>the event-channel port array).
>>
>
>I have a question on the following function, why l1 is updated only one
>time, while l2 is updated in each loop? I think they should handle in
>the same way. Anyway, in current code, l2 &= ~(1UL << l2i); is not
>needed.
Does that mean in the middle of processing l2, some bits previously active may
become inactive due to unmasked by other vcpus? So conservative check in each
loop may be necessary. ;-)

Thanks,
Kevin
>Thanks
>-Xin
>
>/* NB. Interrupts are disabled on entry. */
>asmlinkage void evtchn_do_upcall(struct pt_regs *regs)
>{
>	unsigned long  l1, l2;
>	unsigned int   l1i, l2i, port;
>	int            irq, cpu = smp_processor_id();
>	shared_info_t *s = HYPERVISOR_shared_info;
>	vcpu_info_t   *vcpu_info = &s->vcpu_info[cpu];
>
>	vcpu_info->evtchn_upcall_pending = 0;
>
>	/* NB. No need for a barrier here -- XCHG is a barrier on x86.
>*/
>	l1 = xchg(&vcpu_info->evtchn_pending_sel, 0);
>	while (l1 != 0) {
>		l1i = __ffs(l1);
>		l1 &= ~(1UL << l1i);
>
>		while ((l2 = active_evtchns(cpu, s, l1i)) != 0) {
>			l2i = __ffs(l2);
>			l2 &= ~(1UL << l2i);
>
>			port = (l1i * BITS_PER_LONG) + l2i;
>			if ((irq = evtchn_to_irq[port]) != -1)
>				do_IRQ(irq, regs);
>			else
>				evtchn_device_upcall(port);
>		}
>	}
>}
>
>_______________________________________________
>Xen-devel mailing list
>Xen-devel@lists.xensource.com
>http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 9 Dec 2005, at 08:59, Tian, Kevin wrote:
>> I have a question on the following function, why l1 is updated only 
>> one
>> time, while l2 is updated in each loop? I think they should handle in
>> the same way. Anyway, in current code, l2 &= ~(1UL << l2i);
is not
>> needed.
>
> Does that mean in the middle of processing l2, some bits previously 
> active may become inactive due to unmasked by other vcpus? So 
> conservative check in each loop may be necessary. ;-)
You can reenter evtchn_do_upcall() on the same CPU from within an IRQ 
handler that does not specify SA_INTERRUPT. It will then handle all the 
pending l2''s and the saved l2 in your outer stack activation is now 
stale. You then end up double-executing IRQs occasionally which I think 
we found to be a bad thing in a few cases.

  -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel