RE: [Xen-ia64-devel][PATCH] Fix a bug related to vcpu migration

["Xu, Anthony" <anthony.xu@xxxxxxxxx>]

>From: Isaku Yamahata [mailto:yamahata@xxxxxxxxxxxxx]
>Sent: 2006年9月21日 10:14
>To: Xu, Anthony
>Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
>Subject: Re: [Xen-ia64-devel][PATCH] Fix a bug related to vcpu migration
>
>
>On Thu, Sep 21, 2006 at 09:49:01AM +0800, Xu, Anthony wrote:
>> When destroying domain, only LPs, this domain is running on, do local
>> TLB purge,
>> Other LPs may still have TLBs belonging to this domain,
>> When another domain is created, this may have the same region ID range
>> with old domain,
>> and may conflict with old TLB belonging to old domain.
>> I did observe this scenario, after destroying&creating domain several
>> times,
>> Domain crashing happens.
>>
>> This patch is intended to resolve this issue.
>> Just before launching a VCPU on a LP, Xen purges local TLB on this LP.
>
>Hi Anthony.
>domain_flush_tlb_vhpt() called by arch_domain_destroy()
>addresses this scneario.
>I don't think the patch resolves the domain crash issue because it
>flushes TLB before region ID range is reused.
Yes, you are right, domain_flush_tlb_vhpt() can addresses this scenario.
One comment about flush_tlb_vhpt_all,

static void flush_tlb_vhpt_all (struct domain *d)
{
        /* First VHPT.  */
>>>> if d is not VMX domain
        vhpt_flush ();

        /* Then mTLB.  */
        local_flush_tlb_all ();
}

>
>Have you observed domain crash with the patch?
>If the domain crash disappeared with your patch,
>we might have different issues, I guess.
Before this patch, both dom0 and VTI domain hang after several create&destroy.
After this patch, both dom0 and VTI domain are alive after more than one 
hundred create&destroy.
Seems this issue is related with TLB, I'll continue to debug this.


Thanks,
Anthony



>
>--
>yamahata

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel