Hi Yamahata,
>From: Isaku Yamahata [mailto:yamahata@xxxxxxxxxxxxx]
>Sent: 2006年9月29日 18:27
>To: Xu, Anthony
>Cc: Magnus Damm; Tristan Gingold; xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
>Subject: Re: [Xen-ia64-devel][PATCH][RFC] Task: support huge page RE:
>[Xen-ia64-devel] Xen/IA64 Healthiness Report -Cset#11460
>This is an issue.
>Assuming that fall back path should be implemented eventually
>and we want to support more than two page size (64MB, 1GB, 4GB).
>we have to distinguish page-size of given gpfn page somehow.
>
>
>> For HVM side, there is no choice; XEN must allocate 256M contiguous memory
>> chunks
>for HVM domain.
>
>Here I don't agree with you.
>If we can detect 265M page use by heuristic, xen/ia64 reallocates
>256MB contiguous memory, pause the domain, exchange memory, then
>unpause the domain.
>This event happens very rare so that it is acceptable to pause/unpase
>domain.
>
>Then, how can it be detected?
>Currently I have very vague idea though.
>For example, tlb insert with 256MB page size on rr4 can be detected.
>Does this sound too complex for the first prottype?
>
Good thinking, you are talking about dynamically setup 256M contiguous memory
chunks when Guest requires it.
As I know huge tlb is used frequently in real IA64 server. If every time XEN
need to pause the domain, exchange memory, then unpause the domain, this will
make domain very slow. And exchanging memory is time-consuming; it may cause
Guest softlockup or some timeout.
Yes, I think it's too complex for the first prototype.
In the first prototype, the simplest way is allocate 256M contiguous memory
chunks for domain, which will get domain better performance. Any TLB entries
whose page size is less than 256M can be inserted into machine TLB or VHPT (if
the page size is equal to preferred page size of that region).
After this, we can look at other possibility.
Thanks,
Anthony
>
>
>
>--
>yamahata
_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel
|