Regression with Xen 4.1.5 (cs 23459 compared to cs 23431)

Regression with Xen 4.1.5 (cs 23459 compared to cs 23431)

[Konrad Rzeszutek Wilk]

Hey,

So I tried to start an PV guest with Fri Feb 15 15:31:55 2013 +0100 23459:9f12bdd6b7f0
and I kept on getting:
xc: error: panic: xc_dom_core.c:315: xc_dom_do_gunzip: inflate failed (rc=-3): Internal error

Then rebuilt it with cs 23431 and it booted fine. Hadn't tried any specific
bisection but any thoughts off which one I ought to try out first?

Re: Regression with Xen 4.1.5 (cs 23459 compared to cs 23431)

[Ian Campbell]

On Thu, 2013-02-21 at 13:00 +0000, Konrad Rzeszutek Wilk wrote:
> Hey,
> 
> So I tried to start an PV guest with Fri Feb 15 15:31:55 2013 +0100 23459:9f12bdd6b7f0
> and I kept on getting:
> xc: error: panic: xc_dom_core.c:315: xc_dom_do_gunzip: inflate failed (rc=-3): Internal error
> 
> Then rebuilt it with cs 23431 and it booted fine. Hadn't tried any specific
> bisection but any thoughts off which one I ought to try out first?

Tools changesets in that range are:
        $ hg log -r  23431:23459 tools/
        changeset:   23452:47c7b8531923
        user:        Ian Jackson <ian.jackson@eu.citrix.com>
        date:        Thu Feb 07 14:26:29 2013 +0000
        summary:     tools/ocaml: oxenstored: Be more paranoid about ring reading
        
        changeset:   23453:130446135528
        user:        Ian Jackson <Ian.Jackson@eu.citrix.com>
        date:        Thu Feb 07 14:26:37 2013 +0000
        summary:     oxenstored: Enforce a maximum message size of 4096 bytes
        
        changeset:   23457:8792a805cc9a
        user:        Ian Campbell <ian.campbell@citrix.com>
        date:        Fri Feb 15 11:50:45 2013 +0000
        summary:     tools/ocaml: oxenstored: correctly handle a full ring.

None of which look even vaguely plausible to me. There's some XZ fixes
on the hypervisor side, but this is a guest and GZIP so that seem
unlikely too.

A gzip internal error seems like it should come from zlib and not
something on the Xen side. Was the guest image completely unchanged in
your two tests? Any chance the rebuild picked up a different zlib?

Ian.

Re: Regression with Xen 4.1.5 (cs 23459 compared to cs 23431)

[Jan Beulich]

>>> On 21.02.13 at 14:00, Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> wrote:
> So I tried to start an PV guest with Fri Feb 15 15:31:55 2013 +0100 
> 23459:9f12bdd6b7f0
> and I kept on getting:
> xc: error: panic: xc_dom_core.c:315: xc_dom_do_gunzip: inflate failed 
> (rc=-3): Internal error
> 
> Then rebuilt it with cs 23431 and it booted fine. Hadn't tried any specific
> bisection but any thoughts off which one I ought to try out first?

Apart from a couple of QEMU_TAG updates and the oxenstored
security fixes I don't see any tools changes since 4.1.4 at all, so
a tools side regression seems rather unlikely. And the hypervisor
side changes also don't look like they could cause this sort of a
failure.

You didn't happen to have some stale (from another version's
build) shared libraries left installed?

Jan

Re: Regression with Xen 4.1.5 (cs 23459 compared to cs 23431)

[Konrad Rzeszutek Wilk]

On Thu, Feb 21, 2013 at 01:12:34PM +0000, Jan Beulich wrote:
> >>> On 21.02.13 at 14:00, Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> wrote:
> > So I tried to start an PV guest with Fri Feb 15 15:31:55 2013 +0100 
> > 23459:9f12bdd6b7f0
> > and I kept on getting:
> > xc: error: panic: xc_dom_core.c:315: xc_dom_do_gunzip: inflate failed 
> > (rc=-3): Internal error
> > 
> > Then rebuilt it with cs 23431 and it booted fine. Hadn't tried any specific
> > bisection but any thoughts off which one I ought to try out first?
> 
> Apart from a couple of QEMU_TAG updates and the oxenstored
> security fixes I don't see any tools changes since 4.1.4 at all, so
> a tools side regression seems rather unlikely. And the hypervisor
> side changes also don't look like they could cause this sort of a
> failure.
> 
> You didn't happen to have some stale (from another version's
> build) shared libraries left installed?

No. The build thing does a nice clean slate of this.
> 
> Jan
> 

Re: Regression with Xen 4.1.5 (cs 23459 compared to cs 23431)

[Konrad Rzeszutek Wilk]

On Thu, Feb 21, 2013 at 08:00:13AM -0500, Konrad Rzeszutek Wilk wrote:
> Hey,
> 
> So I tried to start an PV guest with Fri Feb 15 15:31:55 2013 +0100 23459:9f12bdd6b7f0
> and I kept on getting:
> xc: error: panic: xc_dom_core.c:315: xc_dom_do_gunzip: inflate failed (rc=-3): Internal error
> 
> Then rebuilt it with cs 23431 and it booted fine. Hadn't tried any specific
> bisection but any thoughts off which one I ought to try out first?

So I tried also 23430 (RELEASE_4_1_4) and it hit the same issue. Then I tried on a
different box - an Intel one - and I did not see any issues.

I am wondering if that particular AMD box is suspect (F1A75-M, BIOS 0406 06/11/2011,
AMD A8-3850, Fam 18).
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>