RE: [Xen-devel] [PATCH] [1/3] Xenoprof: Enabling performance profiling in Xen

["Santos, Jose Renato G (Jose Renato Santos)" <joserenato.santos@xxxxxx>]

>> 
>> Is the nmi_sanity_check() needed? We shouldn't lose NMIs: 
>> even if we take one while doing NMI processing, we sholuld 
>> retake the NMI as soon as the first one IRETs.
>> 

  The nmi_sanity_check() should not be needed if everything is
  working fine. However I suggest we should keep it for now
  until we are sure it is safe to remove it. I would also suggest
  that we print a warning message in the console and/or the
  log if it is detected that an NMI was lost. We could built
  our confidence that NMIs are no being lost if we don't
  see those warnings for some time while using oprofile.

>> There used to be a problem whereby the Xen NMI handler would 
>> bail if it detected it had interrupted Xen at an unfortunate 
>> time. I guess the code in xenoprof was probably added to 
>> deal with this situation. If so then it is no longer needed 
>> -- I fix up in a more generic way now (reschedule the NMI 
>> automatically). 
>> 

  That is correct. However I believe Aravind identified a
  problem with the original fix and implemented an alternative fix.
  Since this is a recent fix, we though it would be better
  to keep the nmi_sanity_check() for now.
    

>> Apart from that I guess this is mostly just oprofile source 
>> files pulled out of Linux? Is there much/any other 
>> Xen-specific code in the patch? 
>> 

  It is fair to say that most code came from Linux, but there
  is more Xen specific code besides this. For example, event
  samples must be multiplexed among different domains; samples
  must be delivered through a shared page between Xen and the
  domain, etc. 

  Thanks

  Renato 

  

>>  -- Keir
>> 

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