Re: [Xen-devel] VMX status report. Xen:25256 & Dom0:d93dc5c...

["Ren, Yongjie" <yongjie.ren@xxxxxxxxx>]

> -----Original Message-----
> From: Konrad Rzeszutek Wilk [mailto:konrad.wilk@xxxxxxxxxx]
> Sent: Thursday, May 10, 2012 10:39 PM
> To: Ren, Yongjie
> Cc: Wu, GabrielX; xen-devel@xxxxxxxxxxxxx
> Subject: Re: [Xen-devel] VMX status report. Xen:25256 & Dom0:d93dc5c...
> 
> > > > xen-changeset:   25256
> > > > Dom0:          linux.git  3.1.0-rc7 (commit: d93dc5c...)
> > >
> > > Please update your dom0.
> >
> > I think we can use 3.4-RCx kernel as dom0 when sending the report next
> time.
> 
> Excellent!
> >
> > > >
> > >
> ============================================================
> > > =====
> > > >
> > > > New issue(1)
> > > > ==============
> > > > 1. cpu weight out of range error when create hvm domU
> > > >   http://bugzilla.xen.org/bugzilla/show_bug.cgi?id=1818
> > >
> > > And what is the guest config? Does it have any cpu weights?
> >
> > Added the config in the BZ.  It doesn't have any cpu weight in config.
> > Some other person also reported this issue in the mailing list several days
> ago.
> 
> I think some of the Citrix folks were going to take a look at this.
> 
> > > >   http://bugzilla.xen.org/bugzilla/show_bug.cgi?id=1730
> > > > 3. [VT-D]fail to detach NIC from guest
> > >
> > > Hmm, the last update is from a year ago. Are you sure this
> > > is an issue?
> > >
> > Yeah, it still exist.  It may be a similar issue with BZ# 1812.
> > According to our recent testing, it something about VNC console.
> > Also added some latest info in the BZ.
> >
> > > >   http://bugzilla.xen.org/bugzilla/show_bug.cgi?id=1736
> > > > 4. Sometimes Xen panic on ia32pae Sandybridge when restore guest
> > > >   http://bugzilla.xen.org/bugzilla/show_bug.cgi?id=1747
> > >
> > > the bug talks about 2.6.32. Can you update it to include the
> > > 3.4 (or 3.3) dmesg output?
> > >
> > This is about 32bit Xen.
> > As we don't test 32bit Xen for a long time, we may update it when we're
> free.
> >
> > > > 5. [VT-D] device reset fail when create/destroy guest
> > > >   http://bugzilla.xen.org/bugzilla/show_bug.cgi?id=1752
> > >
> > > Does this happen if you manually echo 1 > ../reset? And is this
> > > an issue with the kernel if you upgrade it to 3.4-rc5?
> > >
> > When doing 'echo 1 >../reset', I get the same error as list in the BZ.
> > "/sys/bus/pci/devices/0000:09:00.0/reset returned -1: Invalid
> argument".
> 
> OK, so then the device can't do reset's properly. Is this the
> physical adaptor or the virtual one?
> 
It's a physical adaptor.

> > This bug only exists when we use 'pcistub' to hide a device.
> 
> Ok, so it looks like upstream Linux can't do this properly on that device.
> 
Yes, I think so. To do some fix on 'pcistub' module might fix this issue.

> >
> > > > 8. after detaching a VF from a guest, shutdown the guest is very slow
> > > >   http://bugzilla.xen.org/bugzilla/show_bug.cgi?id=1812
> > >
> > > Where does it slow down? When bringing the NIC down or in specific
> > > cases?
> > > Is this an issue with if you are using a different guest (say Fedora Core
> 16?)
> > >
> > We found it is related to 'stdgva' option in guest config file.
> > If 'stdvga=1' and 'vnc=1', the guest is not slow when shutdown.
> > If 'stdvga=0' and 'vnc=1', guest shutdown will be very slow.
> 
> What does turning standard VGA and VNC on mean to the guest? Doesn't
> that mean
> that the VGA adapter is gone from the guest? And is this issue only
> present if you have PCI devices in the guest? Meaning do you get this
> if you boot a normal HVM guest and do 'stdvga=0' and 'vnc='1?

'stdvga=1' will make the guest use a emulated VGA named "Technical Corp. Device 
1111".
'stdvga=0' will make the guest use a emulated VGA "Cirrus Logic GD5446".
We found this issue only exist when we shut down the guest in its VNC after we 
hot remove a PCI device.
As the manual page says, "If your guest supports VBE 2.0 or later (e.g. Windows 
XP onwards) then you should enable this (stdvga=1)."
I think most of the OS we use today support VBE 2.0 or later. 
And if 'stdvga=1', this issue will not exist.
So, can we set stdvga to TRUE by default in libxl code?

--Jay

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel