RE: [Xen-ia64-devel] [PATCH] XEN: accelerate guest tr search.

["Magenheimer, Dan (HP Labs Fort Collins)" <dan.magenheimer@xxxxxx>]

I have a couple of concerns with this patch:

1) Because of vcpu_quick_region_check() this patch will
   matter (on Linux) only for code/data in region 7 that
   is not pinned by a guest TR.  Vcpu_quick_region_check() avoids
   checking the guest TRs if accessing a region that has no TRs
   set (e.g. on Linux, the vast majority of misses will never
   check any of the guest TRs).
2) The code assumes that a guest utilizes the TRs in order
   with no holes (e.g. TR0-3 rather than TR0,2,4,6).  This
   is true for Linux but if we are going to add this performance
   shortcut, it might as well be more robust.
3) I think we should be very careful about making changes
   that are intended to improve performance without doing any
   benchmarking.  Many times I have seen code that was intended
   to improve performance actually -- surprise! -- result in
   performance degradation.

> -----Original Message-----
> From: xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx 
> [mailto:xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf 
> Of Xu, Anthony
> Sent: Thursday, March 02, 2006 6:10 AM
> To: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> Subject: [Xen-ia64-devel] [PATCH] XEN: accelerate guest tr search.
> 
> Guest has 8 I/D TR entries, when tlb miss happens, VMM will 
> check these
> 8 TR entries to see whether corresponding tlb entry can be found. In
> fact, guest 
> may not use all of these 8 entry, for example, linux only uses first 2
> ITR and 
> first 4 DTR(maybe not exact). I add two members in vcpu struct,
> itr_maxslot, 
> dtr_maxslot to record max entry number of guest ITR and DTR which are
> used by 
> guest. So when VMM searches guest TRs, it doesn't need to 
> check all TRs,
> just 
> those used by guest. Since searching guest TR is a frequent operation,
> this patch
> should improve performance of dom0 and domU in theory. I 
> don't have data
> :-).
> 
> Signed-off-by: Anthony Xu <anthony.xu@xxxxxxxxx>
> 
> Thanks,
> -Anthony 
> 
> 

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