gdb + remote qemu, Ctrl-C does not work

gdb + remote qemu, Ctrl-C does not work

[dpc]

Hi,

I have a little problem with arm-none-eabi-gdb that I'm debugging
remote qemu-system-arm targets.

The problem is: Ctrl-C does not break qemu target, but disconnects
qemu completely.

(gdb) c
Continuing.
^CRemote communication error.  Target disconnected.: Connection reset by peer.


GDB is not offering me the Ctrl-C sequence to break execution (like my
native gdb does) so I guess I should use some different options
because what I know is (I've tried), that sending SIGINT to
arm-none-eabi-gdb process does break the execution. I see no reason
Ctrl-C could not work if sending SIGINT does.

I'm using default options and here are some relevant version strings:

$ arm-none-eabi-gdb --version
GNU gdb (Sourcery G++ Lite 2010.09-51) 7.2.50.20100908-cvs
$ uname -a
Linux mutex 2.6.35-27-generic #48-Ubuntu SMP Tue Feb 22 20:25:46 UTC
2011 x86_64 GNU/Linux

Much appreciated.

Regards,
Dawid

--
http://dpc.ucore.info

Re: gdb + remote qemu, Ctrl-C does not work

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 444 bytes --]

On Tuesday, March 01, 2011 09:47:57 dpc@ucore.info wrote:
> I have a little problem with arm-none-eabi-gdb that I'm debugging
> remote qemu-system-arm targets.
> 
> The problem is: Ctrl-C does not break qemu target, but disconnects
> qemu completely.
> 
> (gdb) c
> Continuing.
> ^CRemote communication error.  Target disconnected.: Connection reset by
> peer.

last i looked at this, it seemed like a qemu problem, not gdb
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: gdb + remote qemu, Ctrl-C does not work

[dpc]

On Tue, Mar 1, 2011 at 7:35 PM, Mike Frysinger <vapier@gentoo.org> wrote:
> On Tuesday, March 01, 2011 09:47:57 dpc@ucore.info wrote:
>> I have a little problem with arm-none-eabi-gdb that I'm debugging
>> remote qemu-system-arm targets.

> last i looked at this, it seemed like a qemu problem, not gdb
> -mike

Clearly the gdb has some way to break gdb if sending SIGINT to gdb
works. Any way to emulate this on Ctrl-C?

-- 
http://dpc.ucore.info

Re: gdb + remote qemu, Ctrl-C does not work

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 604 bytes --]

On Tuesday, March 01, 2011 17:52:29 dpc@ucore.info wrote:
> On Tue, Mar 1, 2011 at 7:35 PM, Mike Frysinger <vapier@gentoo.org> wrote:
> > On Tuesday, March 01, 2011 09:47:57 dpc@ucore.info wrote:
> >> I have a little problem with arm-none-eabi-gdb that I'm debugging
> >> remote qemu-system-arm targets.
> > 
> > last i looked at this, it seemed like a qemu problem, not gdb
> 
> Clearly the gdb has some way to break gdb if sending SIGINT to gdb
> works.

if the remote stub does things wrong, there's nothing the gdb client should be 
doing to work around it.  fix the remote stub.
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: gdb + remote qemu, Ctrl-C does not work

[dpc]

On Wed, Mar 2, 2011 at 12:50 AM, Mike Frysinger <vapier@gentoo.org> wrote:
>> Clearly the gdb has some way to break gdb if sending SIGINT to gdb
>> works.
>
> if the remote stub does things wrong, there's nothing the gdb client should be
> doing to work around it.  fix the remote stub.

That's the possibility as well. Someone / I could try fix qemu sources.

However I don't understand why gdb is doing something different on
SIGINT and something different on Ctrl-C.


-- 
http://dpc.ucore.info

Re: gdb + remote qemu, Ctrl-C does not work

[Pedro Alves]

On Wednesday 02 March 2011 07:29:38, dpc@ucore.info wrote:
> That's the possibility as well. Someone / I could try fix qemu sources.
> 
> However I don't understand why gdb is doing something different on
> SIGINT and something different on Ctrl-C.

Do you have qemu running on the same terminal as gdb?

-- 
Pedro Alves

Re: gdb + remote qemu, Ctrl-C does not work

[Andreas Schwab]

"dpc@ucore.info" <dpc@ucore.info> writes:

> However I don't understand why gdb is doing something different on
> SIGINT and something different on Ctrl-C.

It's a matter of who receives the SIGINT.  The action of Ctrl-C is
dependent on the terminal setting.

Andreas.

-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."

Re: gdb + remote qemu, Ctrl-C does not work

[dpc]

On Wed, Mar 2, 2011 at 10:45 AM, Pedro Alves <pedro@codesourcery.com> wrote:
> Do you have qemu running on the same terminal as gdb?

Well yes, I guess. I'm running it in the same script like this:

qemu someoptions &
sleep 1
gdb -x somefile


-- 
http://dpc.ucore.info

Re: gdb + remote qemu, Ctrl-C does not work

[Pedro Alves]

On Wednesday 02 March 2011 12:29:40, dpc@ucore.info wrote:
> On Wed, Mar 2, 2011 at 10:45 AM, Pedro Alves <pedro@codesourcery.com> wrote:
> > Do you have qemu running on the same terminal as gdb?
> 
> Well yes, I guess. I'm running it in the same script like this:
> 
> qemu someoptions &
> sleep 1
> gdb -x somefile

Your ctrl-c is reaching qemu as well, which is making it
commit suicide.  Notice on your OP:

 Continuing.
 ^CRemote communication error.  Target disconnected.: Connection reset by peer.
                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

This not a GDB problem.

I'd run qemu through a pipe ("target remote | qemu") instead
of using tcp.

-- 
Pedro Alves

Re: gdb + remote qemu, Ctrl-C does not work

[Tristan Gingold]

On Mar 2, 2011, at 1:49 PM, Pedro Alves wrote:

> On Wednesday 02 March 2011 12:29:40, dpc@ucore.info wrote:
>> On Wed, Mar 2, 2011 at 10:45 AM, Pedro Alves <pedro@codesourcery.com> wrote:
>>> Do you have qemu running on the same terminal as gdb?
>> 
>> Well yes, I guess. I'm running it in the same script like this:
>> 
>> qemu someoptions &
>> sleep 1
>> gdb -x somefile
> 
> Your ctrl-c is reaching qemu as well, which is making it
> commit suicide.

Usually background processes don't get C-c.

>  Notice on your OP:
> 
> Continuing.
> ^CRemote communication error.  Target disconnected.: Connection reset by peer.
>                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

IIRC we also had this issue.  This was due to the fact that qemu exits before acking the packet.

Tristan.

Re: gdb + remote qemu, Ctrl-C does not work

[Pedro Alves]

On Wednesday 02 March 2011 13:07:19, Tristan Gingold wrote:
> >  Notice on your OP:
> > 
> > Continuing.
> > ^CRemote communication error.  Target disconnected.: Connection reset by peer.
> >                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> 
> IIRC we also had this issue.  This was due to the fact that qemu exits before acking the packet.
> 

I've seen that happen on a "quit/kill", due to the fact that
the 'k' packet (kill) does not require a reply.

But in this case, why would qemu exit at all before acking
the packet?  This is about interrupting the target with ctrl-c, not
killing/quiting.

-- 
Pedro Alves

Re: gdb + remote qemu, Ctrl-C does not work

[Tristan Gingold]

On Mar 2, 2011, at 3:16 PM, Pedro Alves wrote:

> On Wednesday 02 March 2011 13:07:19, Tristan Gingold wrote:
>>> Notice on your OP:
>>> 
>>> Continuing.
>>> ^CRemote communication error.  Target disconnected.: Connection reset by peer.
>>>                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>> 
>> IIRC we also had this issue.  This was due to the fact that qemu exits before acking the packet.
>> 
> 
> I've seen that happen on a "quit/kill", due to the fact that
> the 'k' packet (kill) does not require a reply.
> 
> But in this case, why would qemu exit at all before acking
> the packet?  This is about interrupting the target with ctrl-c, not
> killing/quiting.

Right.  Why I was so confused ? :-)

Tristan.

Re: gdb + remote qemu, Ctrl-C does not work

[dpc]

On Wed, Mar 2, 2011 at 2:07 PM, Tristan Gingold <gingold@adacore.com> wrote:
>> Your ctrl-c is reaching qemu as well, which is making it
>> commit suicide.
>
> Usually background processes don't get C-c.

I am not sure, but I'd guess so. But what Pedro Alves says would
explained everything perfectly.

I'm planing to use internal bash "trap" command to handle ctrl-c myself.

But I'll try the trick with a pipe too.
-- 
http://dpc.ucore.info

Re: gdb + remote qemu, Ctrl-C does not work

[Joel Brobecker]

> > Usually background processes don't get C-c.
> 
> I am not sure, but I'd guess so. But what Pedro Alves says would
> explained everything perfectly.

That's only the case with interactive shells! In non-interactive
shell sessions (typically a script), background process are still
launched with the same process group as the parent script.  Since
the SIGINT is propagated to all processes sharing the foreground
process group ID, this should explain why qemu is receiving the
SIGINT as well.

-- 
Joel

Re: gdb + remote qemu, Ctrl-C does not work

[Andreas Schwab]

Joel Brobecker <brobecker@adacore.com> writes:

> That's only the case with interactive shells! In non-interactive
> shell sessions (typically a script), background process are still
> launched with the same process group as the parent script.

This is not about interactive vs. non-interactive, but a matter of job
control.  You can enable/disable job control independent of whether the
shell is interactive.

Andreas.

-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."