Re: [Xen-devel] [PATCH v2] Handle xen_platform_pci=0 case

[Anthony PERARD <anthony.perard@xxxxxxxxxx>]

On Thu, Nov 28, 2013 at 05:07:50PM +0100, Fabio Fantoni wrote:
> Il 28/11/2013 11:03, Fabio Fantoni ha scritto:
> >Il 27/11/2013 19:21, Anthony PERARD ha scritto:
> >>Hi,
> >>
> >>Here is a little patch that attempt to fix the issue regarding
> >>xen_platform_pci=0 not been handled.
> >>
> >>There is one patch left from the previous version. The patch that was
> >>adding
> >>qemu_machine_override have been removed as it is unnecessary. If someone
> >>wants
> >>to change the -machine, it can always add it to device_model_args_hvm,
> >>as QEMU
> >>appear to use the last one.
> >>
> >>Regards,
> >>
> >>Anthony PERARD (1):
> >>   libxl: Handle xen_platform_pci=0 case with qemu-xen.
> >>
> >>  tools/libxl/libxl_dm.c | 12 ++++++++++--
> >>  1 file changed, 10 insertions(+), 2 deletions(-)
> >>
> >
> >The patch seems good and is working, the problem of kernel panic on hvm
> >linux domU with xen_platform_pci=0 seems problem of xen modules kernel
> >side.
> >I think that xen modules of linux kernel need to be fixed and/or improved
> >(with backports to lts kernel versions).
> >Probably xen modules can cause an issue also on other case, for example
> >this kernel problem with spice vdagent where I not found solution for now:
> >http://lists.freedesktop.org/archives/spice-devel/2013-October/015185.html
> >
> >
> >I also tried xen_platform_pci=0 on windows 7 with gplpv installed and gave
> >blu screen, this should be windows problem, is know that is not adactive
> >and give blu screen also on phisical hardware when mainly components
> >change.
> >Without gplpv windows 7 boots also with xen_platform_pci=0.
> >Probably I also found that with xen_platform_pci=0 seems that solve qxl
> >"refresh problem" on windows.
> >I'll try to further understand what xen cause conflicts or unforeseen
> >cases.
> 
> I did another tests and I can confirm that the main big problem with qxl on
> windows domUs without gplpv and with xen_platform_pci=0 seems missed!
> Time ago I tested only removing gplpv but the problem remain.
> 
> The only component missed in this case with qxl working is -device
> xen-platform, right?

Sorry, I don't know. Never tryed PV driver on Windows and never tryed QXL
on Windows. But I can tell you that you should not need to add "-device
xen-platform" anywhere. If you xen_platform_pci=1 (or nothing), you will
get the xen-platform. If you have xen_platform_pci=0, then you say you
don't want the xen-platform.

> I tried to found possible cause of the qxl problem and I found this doc
> patch:
> http://lists.xen.org/archives/html/xen-devel/2013-11/msg01897.html
> I saw this part:
> >+    Also the size parameter (default 0x400000) can be used to specify the
> >+    size of the single MMIO BAR that the device exposes. This area may be
> >+    used by drivers for mapping grant tables, etc.
> can be the cause of possible conflict with qxl?
> If not what other things can be different in the case of xen_platform_pci=0?

The only difference between xen_platform_pci=0 and =1 is the presence of
the "xen-platform" device which is seen as a PCI device by a guest, and
which is used by PV drivers.

-- 
Anthony PERARD

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