Re: [Xen-devel] PVH Dom0 with latest Linux kernels

[Mukesh Rathor <mukesh.rathor@xxxxxxxxxx>]

On Fri, 14 Mar 2014 13:48:22 -0400
Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx> wrote:

> On Thu, Mar 13, 2014 at 07:09:54PM -0700, Mukesh Rathor wrote:
> > On Fri, 28 Feb 2014 17:15:08 -0800
> > Mukesh Rathor <mukesh.rathor@xxxxxxxxxx> wrote:
> > 
> > > On Tue, 18 Feb 2014 17:47:04 -0800
> > > Mukesh Rathor <mukesh.rathor@xxxxxxxxxx> wrote:
> > > 
> > > > On Tue, 18 Feb 2014 17:48:02 +0100
> > > > Roger Pau Monnà <roger.pau@xxxxxxxxxx> wrote:
> > > > 
> > > > > Hello,
> > > > > 
> > > > > I've tried to boot a PVH Dom0 using Linux 3.14.0-rc3 and
> > > > > Mukesh's PVH Dom0 Xen tree
> > > > > (https://oss.oracle.com/git/?p=mrathor/xen.git;a=shortlog;
> > > > > h=dom0pvh-v7), and got the following crash. Do you have any
> > > > > new Xen Dom0 series that I could use to test PVH Dom0?
> > > > > 
> > > > 
> > > > It won't work. The final linux patches were changed a bit, and
> > > > it makes an hcall that will cause mem corruption in xen. So,
> > > > just hang in there a bit, new patches coming up in few days.
> > > > 
> > > > thanks
> > > > Mukesh
> > > 
> > > Hey Roger,
> > > 
> > > Sorry, still don't have the new version out. Just running into one
> > > issue after another... I had to refresh the tree, that broke my
> > > tools as they needed newer glibc, which caused me to reinstall,
> > > which caused dom0 to break causing me to rebuild that entire
> > > thing... anyways, I ve got stuff up, but guests are not starting
> > > right now. Debugging. Will keep you posted.
> > 
> > JFYI... I am able to boot dom0 in PVH mode, then start PV and PVH
> > guest. But for HVM guest, during boot I see PF fautls in different
> > places during load of acpi-cpufreq module. Very hard to debug since
> > faults come from different place everytime. I'm now trying to
> > figure what load_module does thruout.
> > 
> > [    3.804141] BUG: unable to handle kernel paging request at
> > 00000014bf0d13c8 [    3.805127] IP: [<ffffffff8104c87c>]
> > apply_relocate_add+0x7c/0x150 [    3.805127] PGD 0
> > 
> > ------
> > [    3.779263] BUG: unable to handle kernel NULL pointer
> > dereference at           (null) [    3.780250] IP:
> > [<ffffffff810eadf3>] add_unformed_module+0x23/0x150 [    3.780250]
> > PGD 0 [    3.780250] Oops: 0002 [#1] SMP
> > --------
> > [    3.975346] BUG: unable to handle kernel paging request at
> > 0000000000003100 [    3.976333] IP: [<ffffffff8130276d>]
> > memcpy+0xd/0x110 [    3.976333] PGD 0
> > [    3.976333]  [<ffffffff810ede36>] ?
> > layout_and_allocate+0x756/0xc00 [    3.976333]
> > [<ffffffff810ee3f1>] load_module+0x111/0x18b0
> > 
> > 
> > Strange, not sure why the problem is on PVH dom0 only... anyways,
> > debugging. The problem is intermittent and sometimes it boots just
> > fine...
> 
> The other thing that might be worth thinking - the virtual address
> that modules are loaded on are way way at the end of the pagetables.
> Could it be that the EPT is limited and said L4 entry (or L3) is
> shared with Xen or hadn't been properly setup?
> 
> Oh wait, this is for HVM guests! And of course this problem you
> see only occurs _after_ you have started PV and PVH guest right? If
> you start the HVM guests first there are no problems?

no, after refresh to latest fc19 base, it's not dependent on PV/PVH.

> Could you describe the steps and perhaps post a link to your updated
> git tree for both Linux and Xen? More folks reproducing it could help
> with this. (Like it helped with the other WP flag).

Yeah, that's what I was thinking about, but just now I was able
to come up with reproducing it with running insmod in a loop. So, I now
need to understand load module code and debug it. Unfortunately, any
thing with timing, like printk, stepping with debugger, .. makes it go
away. I think I can solve it soon now after I get back to it.

thanks
mukesh


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