[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] RE: latest xen-unstable fails to boot on Dell D630 (likely hpet/Cstate problem)



FYI, 20073+20093+20149 boots properly and xend starts
WITH max_cstate=2, but dom0 FAILs to boot unless
max_cstate=2 is added as a Xen boot parameter.

So I still think something changed at 20073 that
causes Merom+RHEL5dom0 to fail to boot due to not
recovering from deep C-state (after dom0 runs
/sbin/hwclock ... Ke Yu knows how to reproduce
the problem).

Thanks,
Dan

> -----Original Message-----
> From: Zhang, Xiantao [mailto:xiantao.zhang@xxxxxxxxx]
> Sent: Tuesday, December 08, 2009 6:44 PM
> To: Dan Magenheimer; Yu, Ke; Xen-Devel (E-mail)
> Subject: RE: latest xen-unstable fails to boot on Dell D630 (likely
> hpet/Cstate problem)
> 
> 
> Dan, 
> Don't use Cset20073 for testing separately, since it needs 
> two minor fixes check-ined by the Cset #20093 and #20149.  
> Except this, Keir also has a typo in Cset #20076 fixed by 
> Cset #20092. In addition, one serious issue is also 
> introduced in #Cset20084 which is fixed in Cset #20140.  I 
> remembered Pod also has issues which can crash hypervisor 
> before Cset #20100. Thus, it is too hard to identify this 
> issue through bisect before #Cset20149, since these issues 
> are introduced and fixed crossedly.   Certainly, if you want 
> to test Cset #20073, you at least have to apply the 
> Cset#20093 and #20149 on top of it.  :)   
> Xiantao
> 
> 
> Dan Magenheimer wrote:
> >> But I'll give bisecting a try.
> > 
> > Looks like the problem has been around for awhile.  It appears
> > the problem starts at c/s 20073.  Xiantao cc'ed since 20073
> > was his patch.
> > 
> > 20070 boots OK without max_cstate=2
> > 
> > 20072 boots most of the way without max_cstate=2 but crashes
> >       before a login prompt (when xend is starting I think)
> > 
> > 20073 FAILS to boot without max_cstate=2 but crashes
> >       before a login prompt
> > 
> > 20082 FAILS to boot without max_cstate=2 but crashes
> >       before a login prompt with max_cstate=2
> > 
> > 20143 FAILS to boot without max_cstate=2 but boots OK
> >       with max_cstate=2
> > 
> > Note that I have NOT bisected tools, just the hypervisor
> > so the crashes are likely due to a newer xend failing on
> > an older hypervisor (which is irrelevant to this problem).
> > 
> >> -----Original Message-----
> >> From: Dan Magenheimer
> >> Sent: Tuesday, December 08, 2009 10:42 AM
> >> To: Yu, Ke; Xen-Devel (E-mail)
> >> Subject: RE: latest xen-unstable fails to boot on Dell D630 (likely
> >> hpet/Cstate problem) 
> >> 
> >> 
> >>> case, if convenient, could you help to do some bisect to see
> >>> which cset cause this bug?
> >> 
> >> I can do this, but because it is often no longer easy to
> >> bisect Xen because of interdependencies with other
> >> components, I was hoping that Keir or you or someone might
> >> have some idea of what changeset might have caused the regression.
> >> But I'll give bisecting a try.
> >> 
> >>> max_cstate=2), when dom0 hangs, is xen still alive, E.g. can
> >>> Xen response to three Ctrl-'A' in serial?
> >> 
> >> Unfortunately, I can't seem to get a Xen console working on
> >> the Merom machine, and the problem can't be reproduced on
> >> my other machine where the Xen console is working (because
> >> Conroe doesn't support deep C).
> >> 
> >>> -----Original Message-----
> >>> From: Yu, Ke [mailto:ke.yu@xxxxxxxxx]
> >>> Sent: Tuesday, December 08, 2009 12:08 AM
> >>> To: Dan Magenheimer; Xen-Devel (E-mail)
> >>> Subject: RE: latest xen-unstable fails to boot on Dell 
> D630 (likely
> >>> hpet/Cstate problem) 
> >>> 
> >>> 
> >>>> -----Original Message-----
> >>>> In this thread, I observed that I was unable to
> >>>> provoke deep C state (C3) on my Dell D630, which has
> >>>> a Intel Merom (dual-core laptop) processor.  At that
> >>>> time, when I tried enabling hpetbroadcast, dom0 boot failed.
> >>>> 
> >>>> http://lists.xensource.com/archives/html/xen-devel/2009-10/ms
> >>>> g01027.html 
> >>>> 
> >>>> As it turned out, all RHEL5-based (maybe RHEL4- also) dom0
> >>>> default installation run /sbin/hwclock, which IIRC takes
> >>>> the RTC away from Xen and gives it to dom0.  Since the
> >>>> Xen hpet emulation does not do RTC emulation, bad things
> >>>> then happen when a deep Cstate is entered (dom0 apparently
> >>>> never wakes up).  I think Ke Yu has also reproduced this problem.
> >>>> 
> >>>> Sometime in the last few weeks, some patch in xen-unstable
> >>>> apparently changed some defaults and xen-unstable will
> >>>> no longer boot with this processor/dom0, with or without
> >>>> hpetbroadcast on the Xen command line.  However, specifying
> >>>> max_cstate=2 on the Xen command line allows a successful
> >>>> dom0 boot, so I suspect the problem is the same (or at
> >>>> least very similar).
> >>>> 
> >>>> I did a quick scan for hpet changes and found c/s 20497,
> >>>> but backing it out made no difference.
> >>>> 
> >>>> I have a workaround for now, but since it is likely that
> >>>> many customers (including all of Oracle's OVS customers)
> >>>> use a RHEL5-based dom0 boot sequence, and Merom processors
> >>>> work fine otherwise, it would be nice to get this identified
> >>>> and fixed before 4.0.
> >>> 
> >>> Let's firstly figure out which component the issue resides.
> >>> 
> >>> Firstly, in the default boot (i.e. without specifying
> >>> max_cstate=2), when dom0 hangs, is xen still alive, E.g. can
> >>> Xen response to three Ctrl-'A' in serial?
> >>> 
> >>> If only dom0 hangs, it is probably that RTC malfunction make
> >>> incorrect dom0 time and lead dom0 fail to boot. Then RTC
> >>> emulation in hypervisor should fix this issue.
> >>> 
> >>> If Xen also hangs, it should be another bug, i.e. hpet
> >>> broadcast does not wake up CPU in deep C states. in this
> >>> case, if convenient, could you help to do some bisect to see
> >>> which cset cause this bug?
> >>> 
> >>> Best Regards
> >>> Ke
> 
>

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


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.