RE: [Xen-ia64-devel] RE: [PATCH] Patch to make latest hg multi-domainback to work

["Tian, Kevin" <kevin.tian@xxxxxxxxx>]

>From: Magenheimer, Dan (HP Labs Fort Collins)
[mailto:dan.magenheimer@xxxxxx]
>
>>      Could you elaborate more how your latest patch works
>> differently and fix the potential issue?
>>
>> -               *pteval = vcpu->arch.dtlb_pte;
>> +               if (vcpu->domain==dom0 && !in_tpa) *pteval =
>> trp->page_flags;
>> +               else *pteval = vcpu->arch.dtlb_pte;
>> +               printf("DTLB MATCH... NEW, DOM%s, %s\n",
>> vcpu->domain==dom0?
>> +                       "0":"U",
>> in_tpa?"vcpu_tpa":"ia64_do_page_fault");
>>
>>      The new limitation seems only for dom0, while dom0 has
>> exactly same guest physical address as machine address. Based
>> upon this assumption, trp->page_flags actually equals to
>> guest pte (vcpu->arch.dtlb_pte)? So I'm not sure about the
>> trick here behind.
>
>Using more printfs, it appears that the problem is that
>one of the "page flags" has PL2 and the other has PL0.
>
>Dan

Since vcpu_itc_no_srlz is always invoked after translate_domain_pte
which will force _PAGE_PL_2 on, it seems unlikely to insert a machine TC
entry with incorrect privilege level. After some search for reference to
dtlb/dtlb_pte, I found hyper_ptc_ga though not the full reason, but
possible a problematic point? In that section of code, only
vcpu->arch.dtlb/itlb is marked as non-present by:

adds r25=IA64_VCPU_DTLB_OFFSET,r27
adds r26=IA64_VCPU_ITLB_OFFSET,r27;;
...
st8 [r25]=r24                   // set 1-entry i/dtlb as not present

Instead vcpu->arch.dtlb_pte is not touched. So after my patch changed
match_dtlb to return guest pte, above problematic code may leave them
out of sync there. Comments?

Thanks,
Kevin

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