Xen project Mailing List

On Fri, Aug 30, 2013 at 12:45 AM, Andrew Cooper <andrew.cooper3@xxxxxxxxxx> wrote:

On 29/08/13 20:10, Rushikesh Jadhav wrote:

The problematic spinlock is at address 0xffff831960e72e90, which is sadly a dynamically allocated one so cant be traced back to a symbol using the symbol table.

Thanks. How easy or hard it is to trace such thing ? I tried google for xen crash analyze and came up with https://github.com/xenserver/xen-crashdump-analyser & http://xenbits.xen.org/people/andrewcoop/.

The xen crashdump analyser is my new replacement to an old tool we used, although my xenbits page is a tad out of date. For now, github should be considered the canonical source, although there has been interest in getting it into the main Xen tree (perhaps when I gain enough tuits).

Yes, I saw that from code and switched to github now. Thanks.

Having said that, you are using Xen 3.4 which is ages out of date, and in fact, probably using XenServer 5.6SP2 (so shouldn't be using xen-devel anyway).

It is quite an old yet stable host with trusted guests. I wanted more information about Xen stack traces and whats the best way to read them hence sent on the list.

Current crash did not generate a core dump, hence I tried crashdump analyzer from xenbits.xen.org on other core dumps but that fails with

INFO Elf CORE crash file: /tmp/core.kdump.1405

ERROR Unexpected class 1

ERROR Failed to parse the crash file

in xen-crashdump-analyser.log.

There is a reason I threw kdump away and wrote the crashdump analyser; It would crash all over the place. That crash file is from the kdump utility, not from Xen. The crashdump analyser will not be able to parse it.

I have only one binary dump file in each crash logs /var/crash/ as core.kdump.1405. Rest all are .log files which does not seem to be parsed by the github tool. Is there any other file I should be looking at ? /proc/vmcore looks to be default file for tool but its not present.

~Andrew

Re: [Xen-devel] Xen host crash