Is there a reason HPET is not by default turned on for HVM guests? While Xen itself sets it to on, the tools - afaics - default it back to off. Thanks, Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Due to xen-unstable:17017. Now that HPET model is integrated with vpt.c better, I think we could enable vhpet by default. -- Keir On 18/12/2009 16:18, "Jan Beulich" <JBeulich@novell.com> wrote:> Is there a reason HPET is not by default turned on for HVM guests? While > Xen itself sets it to on, the tools - afaics - default it back to off. > > Thanks, Jan > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
IIRC HPET required some timer_mode work because certain kernel versions had problems when ticks were delivered too closely together. Dave Winchell did this at Virtual Iron, but left early this year, and his work was plumbed all the way through to the hardware HPET (because it preceded work that made xen system time more reliable) so resurrecting his VI patch and munging it to apply to xen-unstable may take significant work and testing (that I may have to eventually get around to, but not soon). Are you suggesting that an HVM use of HPET should be plumbed through to the hardware HPET? Hardware HPET access is verrrrryyyyyy slow and save/restore/migration would likely be a problem. And in any case, what do you expect to gain by using HPET instead of another clocksource? Turning on vhpet by default (on top of Xen system time) may turn out to be OK, but unless there''s a really good reason and a great deal of testing done soon, I wouldn''t suggest flipping the default prior to 4.0. Thanks, Dan> -----Original Message----- > From: Keir Fraser [mailto:keir.fraser@eu.citrix.com] > Sent: Friday, December 18, 2009 9:38 AM > To: Jan Beulich; xen-devel@lists.xensource.com > Cc: Dan Magenheimer > Subject: Re: [Xen-devel] HPET in HVM guests > > > Due to xen-unstable:17017. Now that HPET model is integrated > with vpt.c > better, I think we could enable vhpet by default. > > -- Keir > > On 18/12/2009 16:18, "Jan Beulich" <JBeulich@novell.com> wrote: > > > Is there a reason HPET is not by default turned on for HVM > guests? While > > Xen itself sets it to on, the tools - afaics - default it > back to off. > > > > Thanks, Jan > > > > > > _______________________________________________ > > Xen-devel mailing list > > Xen-devel@lists.xensource.com > > http://lists.xensource.com/xen-devel > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
>>> Dan Magenheimer <dan.magenheimer@oracle.com> 18.12.09 18:55 >>> >Are you suggesting that an HVM use of HPET should be >plumbed through to the hardware HPET? Hardware HPET >access is verrrrryyyyyy slow and save/restore/migration >would likely be a problem.No, I''m just talking about what you appear to call vHPET.>And in any case, what do you expect to gain by using >HPET instead of another clocksource?Certain Windows certification seem the require the presence (and visibility through ACPI tables) of HPET. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> From: Jan Beulich [mailto:JBeulich@novell.com] > Sent: Monday, December 21, 2009 2:43 AM > To: Dan Magenheimer > Cc: Keir Fraser; xen-devel@lists.xensource.com > Subject: RE: [Xen-devel] HPET in HVM guests > > >>> Dan Magenheimer <dan.magenheimer@oracle.com> 18.12.09 18:55 >>> > >Are you suggesting that an HVM use of HPET should be > >plumbed through to the hardware HPET? Hardware HPET > >access is verrrrryyyyyy slow and save/restore/migration > >would likely be a problem. > > No, I''m just talking about what you appear to call vHPET. > > >And in any case, what do you expect to gain by using > >HPET instead of another clocksource? > > Certain Windows certification seem the require the presence (and > visibility through ACPI tables) of HPET.Do the WHQL rules allow for turning on a parameter in the VM config file? If so, setting hpet=1 should work (though I haven''t tested it in quite awhile so bitrot might have set in). Dan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
>>> Dan Magenheimer <dan.magenheimer@oracle.com> 21.12.09 16:51 >>> >Do the WHQL rules allow for turning on a parameter >in the VM config file?I have no idea.>If so, setting hpet=1 should >work (though I haven''t tested it in quite awhile so >bitrot might have set in).Setting hpet=1 in the VM config addresses this - that''s why the question really just was about the default the tools use in the absence of a hpetsetting. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Keir, would you consider backing out 20709? I don''t think Jan has confirmed a need to change the default, just a question as to why the default was set as it was. Given Dave Winchell''s error measurements here: http://lists.xensource.com/archives/html/xen-devel/2008-06/msg00181.html and the long discussion that followed (which I don''t think was ever resolved or had patches applied) and given that we don''t have bug reports for the current default, I think it would be a bad idea to change the default, certainly without specific testing on Windows and 2.6.9-based Linux kernels to see if the problems Dave reported recur.> -----Original Message----- > From: Jan Beulich [mailto:JBeulich@novell.com] > Sent: Monday, December 21, 2009 9:07 AM > To: Dan Magenheimer > Cc: Keir Fraser; xen-devel@lists.xensource.com > Subject: RE: [Xen-devel] HPET in HVM guests > > > >>> Dan Magenheimer <dan.magenheimer@oracle.com> 21.12.09 16:51 >>> > >Do the WHQL rules allow for turning on a parameter > >in the VM config file? > > I have no idea. > > >If so, setting hpet=1 should > >work (though I haven''t tested it in quite awhile so > >bitrot might have set in). > > Setting hpet=1 in the VM config addresses this - that''s why > the question > really just was about the default the tools use in the > absence of a hpet> setting. > > Jan > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Okay. -- Keir On 21/12/2009 16:34, "Dan Magenheimer" <dan.magenheimer@oracle.com> wrote:> Keir, would you consider backing out 20709? I don''t think > Jan has confirmed a need to change the default, just a > question as to why the default was set as it was. > > Given Dave Winchell''s error measurements here: > http://lists.xensource.com/archives/html/xen-devel/2008-06/msg00181.html > and the long discussion that followed (which I don''t > think was ever resolved or had patches applied) and > given that we don''t have bug reports for the current > default, I think it would be a bad idea to change the > default, certainly without specific testing on Windows > and 2.6.9-based Linux kernels to see if the problems > Dave reported recur. > >> -----Original Message----- >> From: Jan Beulich [mailto:JBeulich@novell.com] >> Sent: Monday, December 21, 2009 9:07 AM >> To: Dan Magenheimer >> Cc: Keir Fraser; xen-devel@lists.xensource.com >> Subject: RE: [Xen-devel] HPET in HVM guests >> >> >>>>> Dan Magenheimer <dan.magenheimer@oracle.com> 21.12.09 16:51 >>> >>> Do the WHQL rules allow for turning on a parameter >>> in the VM config file? >> >> I have no idea. >> >>> If so, setting hpet=1 should >>> work (though I haven''t tested it in quite awhile so >>> bitrot might have set in). >> >> Setting hpet=1 in the VM config addresses this - that''s why >> the question >> really just was about the default the tools use in the >> absence of a hpet>> setting. >> >> Jan >> >>_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel