Sorry for the late reply. I was off attending the Xen Developer
Conference, and then testing out the latest kexec code, and am just
getting back to this,
On 10/23/13 12:37, Daniel Kiper wrote:
On Tue, Oct 22, 2013 at 07:03:19PM +0200, Daniel Kiper wrote:
> Hi,
>
> Sorry for late reply but I am very busy now.
>
> On Thu, Oct 17, 2013 at 03:25:25PM -0400, Don Slutz wrote:
>> On 10/16/13 09:49, Petr Tesarik wrote:
>>> On Tue, 15 Oct 2013 12:02:51 -0400
>>> Don Slutz <dslutz(a)verizon.com> wrote:
>>>
>>>> I have some code that allows this. See the following mail thread:
>>>>
>>>>
http://permalink.gmane.org/gmane.comp.emulators.xen.devel/174807
>>>>
>>>> The questions are:
>>>>
>>>> 1. Does remote access have a specification?
>>>> 2. Is it supported?
>>> I have never even made it to work. But since there's now some new use
>>> for it, I wouldn't object reviving the code.
>>>
>> Good.
>>>> 3. Should the code be part of xen or crash?
>>> I think it should be part of xen, because it needs the xen development
>>> files, which would be a new dependency for the crash utility. Also, the
>>> development cycle is more aligned with xen than crash.
>>>
>>> Just my two cents,
>>> Petr T
>> That is why I started with it in xen. Not sure where it will end up.
> I do not know protocol details but I think that it is worth checking
> gdbsx tool from Xen (xen/tools/debugger/gdbsx) and gdb protocol spec
> (you could expose GDB interface from QEMU so it could be useful in HVM
> case). Maybe crash protocol is very similar or even identical with GDB
> protocol because a large part of crash is GDB itself.
I can see using gdbsx and
gdb. However I have found crash to be easier
in lots of cases. I could have used xg from here if I did not already
have code based on xenctx.
Ugh... I was in a hurry and forgot about final statement. Sorry for
that.
No problem.
So we could just use existing solutions if they are compatible with
protocol used by crash.
Daniel
The only existing solution that I know of is:
xl dump-core <Domain> <filename>
crash <filename>
But this way does require disk space, and the delay while writing the
file. For me it took ~2 minutes to write out a 3G core file. I also do
not see the current registers being saved in the core file.
-Don Slutz