single software TLB vs. multiple software TLBs (was RE: [Xen-ia64-devel] [PATCH] [Resend]Enable hash vtlb)

["Dong, Eddie" <eddie.dong@xxxxxxxxx>]

Tristan:
        You are probably avocating for the flexibility of hash TLB too from the 
issue :-)
        Yes, when multiple TLBs are supported, i.e. hash TLB here, those big 
guest page translations can be put into software TLB using gmfn (leave to 
future tuning). When it is inserted into machine TLB or VHPT, it can be 
breakdown to many small page size translations (up to the balance of 
performance and functionality). I.e. the machine rr4.ps could be 16KB for guest 
vrr4.ps=256M. Another point is that VHPT hash uses VPN and rid as hash index, 
here the VPN means 256M/16K=16K contiguous index. In this way the hash result 
will be not duplicated as if the hash table can accomodate 16K+ entries. So 
forward progress can be guaranteed by the software TLB + VHPT.

        Hope this helpful! 
  Thx,eddie

-----Original Message-----
From: xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx 
[mailto:xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Tristan Gingold
Sent: 2006年4月12日 16:38
To: Xu, Anthony; xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-ia64-devel] [PATCH] [Resend]Enable hash vtlb

Le Mercredi 12 Avril 2006 10:01, Xu, Anthony a écrit :
> From: Tristan Gingold [mailto:Tristan.Gingold@xxxxxxxx]
>
> >Sent: 2006年4月12日 15:53
> >To: Xu, Anthony; xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> >Subject: Re: [Xen-ia64-devel] [PATCH] [Resend]Enable hash vtlb
> >
> >Le Vendredi 07 Avril 2006 21:02, Xu, Anthony a écrit :
> >> Hash vTLB is intended to address SMP scalability for large system.
> >
> >If I understand correctly, the Hash vTLB patch doesn't handle itc whose ps
> > > rr.ps (there is a panic here).
>
> Correct, Hash vTLB doesn't handle multiple page sizes on the same region.
> Linux doesn't use multiple page sizes on the same region.
> Linux uses region 4 to handle hugetlb like 256M, rr4.ps is set to 256M.
>
> >After a few minutes of thinking, I don't see how this could be done.
> >Any hint ?
>
> Break this huge tlb into several small tlbs whose page size is equal to
> rr.ps.
I am not sure this works.  During insertion an entry can erase another entry 
from the same itc.  And if a fault occurs within the erased entry, there is 
no forward progression!

Tristan.

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

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