Hi Yamahata,
Sorry for late response.
This is a great patch,
1. This patch should alleviate pain from migration.
2. VTI-domain and domU now use the same TR number for VHPT mapping, there is
one more TC available to use, and that's good.
A very small comment,
I noticed DomU still use VHPT_ADDR to map VHPT. While in function, for
instance, __vhpt_flush, it uses __va(vhpt_maddr) to access VHPT, because vhpt
is allocated from dom heap, this may cause unnecessary tlb miss.
There are two natural options here.
1. If DomU use __va(vhpt_maddr) to map VHPT, the unnecessary tlb miss can be
eliminated.
2. in function like __vhpt_flush, use VHPT_ADDR to access VHPT.
One more thinking is, seems we plan to expose p2m table to domU, in this case,
if we use VHPT_ADDR to map VHPT, itc can be "emulated" inside domU, when domU
want to insert a mapping, it can look up p2m for machine physical address, then
insert the transferred mapping into VHPT directly.
BTW, this issue is not introduced by this patch; it has been there for long
time.
Thanks,
Anthony
>-----Original Message-----
>From: xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx
>[mailto:xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Isaku Yamahata
>Sent: 2006年10月6日 17:14
>To: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
>Subject: [Xen-ia64-devel] [PATCH] per vcpu vhpt
>
>
>per vcpu vhpt
>implement per vcpu vhpt option. allocate VHPT per vcpu.
>added compile time option, xen_ia64_pervcpu_vhpt=y, to enable it.
>Its default is on.
>added xen boot time option, pervcpu_vhpt=0, to disable it.
>
>This patch focuses on vcpu migration between physical cpus
>becaseu vcpu is heavily migrated with credit scheduler.
>This patch tries to reduce vTLB flush when vcpu is migrated
>
>--
>yamahata
_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel
|
Home