Re: [Xen-devel] [PATCH v2] xen, kdump: handle pv domain in paddr_vmcoreinfo_note()

From: Daniel Kiper
Date: Fri Apr 14 2017 - 18:27:02 EST


On Fri, Apr 14, 2017 at 06:53:36PM +0200, Petr Tesarik wrote:
> On Tue, 11 Apr 2017 19:20:08 +0200
> Daniel Kiper <daniel.kiper@xxxxxxxxxx> wrote:
> > On Tue, Apr 11, 2017 at 04:59:16PM +0200, Petr Tesarik wrote:
> >[...]
> > > Tested-by: Petr Tesarik <ptesarik@xxxxxxxx>
> > >
> > > I copied the complete /proc/vmcore to a directory on disk. Exactly
> > > as expected, crash works both without the patch and with the patch, as
> > > it does not use VMCOREINFO at all (instead, crash obtains the
> > > information from kernel debuginfo directly).
> >
> > Thanks for doing the tests. I suppose that you have tested HVM guests.
>
> Not really. I crashed Dom0, which is in turn sent to the hypervisor, so
> the result is a complete host dump, including Xen hypervisor data and
> all domains.

OK.

> > IIRC, PV guests are not supported by crash right now due to p2m VMA
> > mapping. At least it was an issue some time ago. Is it still valid?
>
> Yes, this is correct. I tested this behaviour a few weeks ago.

Thanks for update.

> > Anyway, one guy in Oracle works on fix for that issue and I do review.
> > We are going to post it in 2-3 weeks.
>
> All right. FYI I do not plan to put much effort into it, as my focus has

OK.

> shifted towards libkdumpfile (https://github.com/ptesarik/libkdumpfile),
> and this library can open PV guest dump files without any issues.

Great! AIUI, it reminds my idea to make such think. However, I have not
have time to make it happen. Is it based on makedumpfile or written from
scratch? Do you plan support for Linux kernel dumps and/or Xen ones?

Daniel