Re: Timer

On May 7, 2013, at 4:48 PM, Pedro Alves wrote:

> On 05/07/2013 07:42 AM, Marc Brünink wrote:
>> 
>> On May 7, 2013, at 12:43 PM, Doug Evans wrote:
>> 
>>> On Mon, May 6, 2013 at 2:52 AM, Marc Brünink <[email protected]> wrote:
>>>> I want to execute a piece of code at regular intervals. Actually I'm sampling $pc.
>>>> (let's not go into detail why I use gdb)
>>>> 
>>>> My current solution just starts another process that sends a SIGTRAP to the debugged application. Using a simple script I can print the $pc.
>>>> 
>>>> However, I just realised that this approach does not work too well. If gdb is stopped due to a breakpoint it will interpret the received SIGTRAP as another hit of the very same breakpoint.
>>>> 
>>>> Reproduce:
>>>> 1. Attach to any program
>>>> 2. Create any breakpoint
>>>> 3. Wait until breakpoint is hit
>>>> 4. Send SIGTRAP to debugged application
>>>> 5. continue
>>>> 
>>>> Actually, now that I think about it, I should have anticipated this behaviour.
>>>> Is there a better way to execute a piece of code at regular intervals?
> 
> You could also use another signal instead of SIGTRAP.

Yes, this is probably the way to go. However, I remember having some issues with different signals. Esp if an applications depends on the delivery of a signal and I use it to implement the timer interrupt. But I suppose using SIGPROF or something similar should be fine.

> 
>>> bash$ man setitimer
>> 
>> I suppose you are suggesting to modify either GDB or the application. This is exactly what I don't want. Any other way to accomplish this (using gdb)?
> 
> You could use LD_PRELOAD to inject a library that uses setitimer into your program.

Possible, but contradicts the gdb-only approach. 


> I guess you could do it with gdb python scripting too.

This would be nice but does not work. As far as I remember there is a sigsupend in linux-nat.c which will thwart using a simple threading.Timer. But I might be wrong here. Whatever the reason, it does not work.


Marc