[lldb-dev] Resuming traced process on Linux and SIGSTOP

[Andrew MacPherson]

Just pinging this again in case anyone has a chance to review the small
Linux detach patch here. Thanks!


On Sat, Mar 22, 2014 at 11:05 AM, Andrew MacPherson
<andrew.macp at gmail.com>wrote:

> I'm attaching a patch here that resolves the issue of a signal being
> delivered to the inferior on detach, let me know if I should start a new
> thread for this but since it's related to what we were discussing I'm
> attaching it here. To reproduce the issue just launch a process, then in a
> new terminal:
>
> sudo lldb -p <pid>
> c
> q
>
> When quitting the inferior receives a SIGSTOP because the process is
> already running when the PTRACE_DETACH is sent. I can't find a clear answer
> on what's supposed to happen here (PTRACE_DETACH sent to running thread)
> but this is the behaviour I'm seeing. The docs do indicate that
> PRACE_DETACH is like PTRACE_CONT and should only be sent to a stopped
> thread.
>
> I also tried to add a unit test for this and the previous patch, however I
> hit http://llvm.org/pr16172. The tests these two patches require are
> similar to the one mentioned in that case and I get the same errors in my
> sample test as when running that one:
>
> attach -p <pid> (ok)
> c (ok)
> detach (error: Detach failed: No such process)
>
> And:
>
> attach -p <pid> (ok)
> c (ok)
> process interrupt (ok)
> c (error: Failed to resume process: Resume request failed - process still
> running..)
>
>
> On Fri, Mar 21, 2014 at 8:37 PM, Ed Maste <emaste at freebsd.org> wrote:
>
>> On 21 March 2014 15:14,  <jingham at apple.com> wrote:
>> > I'm pretty sure we have some test cases that test attach.  If we don't,
>> then it would be great to add one.  If we do, then it would be interesting
>> to see why they didn't fail.  For instance, maybe they just didn't bother
>> to try "continue" after the attach.  Just adding that would be good too.
>>
>> We definitely have some, because they used to fail early on in the
>> FreeBSD porting effort.  I suspect that your suggestion is correct and
>> they just don't do anything but attach and then quit.
>>
>> On a related note, I'm pretty sure we don't have tests for detach.
>> I'll see about extending tests to include that at some point.
>>
>> -Ed
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20140325/807de06c/attachment.html>