kgdb over ethernet?

kgdb over ethernet?

[Jan Engelhardt]

Hi,


the DocBook files for the recently-merged kgdb mention kgdboe, as does 
the doc on kgdb.sf.net, but I do not see kgdboe anywhere in the source. 
Is oe actually merged? I'm kinda short on rs232...

thanks,
Jan

Re: kgdb over ethernet?

[Jason Wessel]

Jan Engelhardt wrote:
> Hi,
>
>
> the DocBook files for the recently-merged kgdb mention kgdboe, as does 
> the doc on kgdb.sf.net, but I do not see kgdboe anywhere in the source. 
> Is oe actually merged? I'm kinda short on rs232...
>   
Well if you have at least one you have enough because you can share kgdb
with the console.

> thanks,
> Jan
>   

kgdboe is not in the mainline kernel. 

It is a feature that might end up being re-implemented at some point or
perhaps merged as is in the future.  The current kgdboe suffers from
some reliability problems related to the need to sync interrupts prior
to using the network device hardware.  This is due to the way the
"polled mode" support is implemented in the NET POLL API.  There is
presently no good means to avoid certain deadlock scenarios.

That being said if you want to work with the latest kgdb + the latest
kernel work it is available.  First you clone Linus's tree and then do
something like:

git remote add kgdb-next
git://git.kernel.org/pub/scm/linux/kernel/git/jwessel/linux-2.6-kgdb.git
git remote update
git checkout kgdb-next

And later when if you update it, you would want make sure and use the
"+" type pull because the branch gets shuffled quite often with patches
pulled in and out of the tree.  IE:

git pull kgdb-next +kgdb-next:kgdb-next

Jason.

Re: kgdb over ethernet?

[Stefan Richter]

> Jan Engelhardt wrote:
>> the DocBook files for the recently-merged kgdb mention kgdboe, as does 
>> the doc on kgdb.sf.net, but I do not see kgdboe anywhere in the source. 
>> Is oe actually merged? I'm kinda short on rs232...

Not what you asked for, but... perhaps you have a spare FireWire port?
>From Documentation/debugging-via-ohci1394.txt:

| There is also a gdb proxy for firewire which allows to use gdb to
| access data which can be referenced from symbols found by gdb in
| vmlinux:
| - ftp://ftp.suse.de/private/bk/firewire/tools/fireproxy-0.33.tar.bz2
|
| The latest version of this gdb proxy (fireproxy-0.34) can communicate
| (not yet stable) with kgdb over an memory-based communication module
| (kgdbom).
-- 
Stefan Richter
-=====-==--- -=-- ==-=-
http://arcgraph.de/sr/

Re: kgdb over ethernet?

[Jan Engelhardt]

On Saturday 2008-04-26 16:30, Stefan Richter wrote:

>> Jan Engelhardt wrote:
>> > the DocBook files for the recently-merged kgdb mention kgdboe, as does the
>> > doc on kgdb.sf.net, but I do not see kgdboe anywhere in the source. Is oe
>> > actually merged? I'm kinda short on rs232...
>
> Not what you asked for, but... perhaps you have a spare FireWire port?

Not either. But see 
http://jengelh.medozas.de/2008/0425-vmware-serial.php for the prime 
solution that I have working now.

Re: kgdb over ethernet?

[Jan Engelhardt]

On Friday 2008-04-25 19:32, Jason Wessel wrote:
>Jan Engelhardt wrote:
>>
>> the DocBook files for the recently-merged kgdb mention kgdboe, as does 
>> the doc on kgdb.sf.net, but I do not see kgdboe anywhere in the source. 
>> Is oe actually merged? I'm kinda short on rs232...
>
>Well if you have at least one you have enough because you can share kgdb
>with the console.

kgdb is running, though I wonder why breakpoints are not seemingly hit.

(gdb) b vfs_readdir
cBreakpoint 2 at 0xc0270c11: file fs/readdir.c, line 23.
(gdb) c
Continuing.

Then I issued ls -l on tty1 (kgdb is on ttyS0), but it did not
happen to stop at vfs_readdir, though I would assume vfs_readdir
was entered as part of /bin/ls.
Ideas?

Re: kgdb over ethernet?

[Jason Wessel]

Jan Engelhardt wrote:
> On Friday 2008-04-25 19:32, Jason Wessel wrote:
>> Jan Engelhardt wrote:
>>> the DocBook files for the recently-merged kgdb mention kgdboe, as does
>>> the doc on kgdb.sf.net, but I do not see kgdboe anywhere in the source.
>>> Is oe actually merged? I'm kinda short on rs232...
>> Well if you have at least one you have enough because you can share kgdb
>> with the console.
>
> kgdb is running, though I wonder why breakpoints are not seemingly hit.
>
> (gdb) b vfs_readdir
> cBreakpoint 2 at 0xc0270c11: file fs/readdir.c, line 23.
> (gdb) c
> Continuing.
>
> Then I issued ls -l on tty1 (kgdb is on ttyS0), but it did not
> happen to stop at vfs_readdir, though I would assume vfs_readdir
> was entered as part of /bin/ls.
> Ideas?

I know very little about what vmware may or may not be doing here.

I tried your test with the tip of the kernel tree this morning with a
arch=i386 kernel and it seems to work fine on real HW and qemu.  You
might consider checking if the kgdb regression tests pass.  You can
change your .config to have:

CONFIG_KGDB_TESTS=y
CONFIG_KGDB_TESTS_ON_BOOT=y
CONFIG_KGDB_TESTS_BOOT_STRING="V1F100"

This will verify if the kgdb core is work correctly on your platform,
and if some part fails we can go from there.  Depending on if vmware
implements HW breakpoints the HW breakpoint tests could certainly
fail.  As an example, qemu does not implement HW breakpoints so those
particular tests always fail, but the failure does not kill the kernel
and the test suite simply moves on to the next test.

Jason.