This is the mail archive of the
gdb@sourceware.org
mailing list for the GDB project.
Re: Inadvertently run inferior threads
- From: Pedro Alves <palves at redhat dot com>
- To: Eli Zaretskii <eliz at gnu dot org>
- Cc: gdb at sourceware dot org
- Date: Mon, 15 Jun 2015 18:56:30 +0100
- Subject: Re: Inadvertently run inferior threads
- Authentication-results: sourceware.org; auth=none
- References: <83h9tq3zu3 dot fsf at gnu dot org> <55043A63 dot 6020103 at redhat dot com> <8361a339xd dot fsf at gnu dot org> <5504555C dot 804 at redhat dot com> <550458E0 dot 10206 at redhat dot com> <83y4jrsgui dot fsf at gnu dot org> <557ECCA5 dot 7050506 at redhat dot com> <83vbepngxm dot fsf at gnu dot org> <557EEF0E dot 1040400 at redhat dot com> <83ioaoopkh dot fsf at gnu dot org>
On 06/15/2015 06:21 PM, Eli Zaretskii wrote:
>> Date: Mon, 15 Jun 2015 16:28:14 +0100
>> From: Pedro Alves <palves@redhat.com>
>> CC: gdb@sourceware.org
>>
>>>> So that should mean that even for GNU/Linux, it should be possible
>>>> to end in the exact same, when any thread other than the one that we
>>>> had started the infcall in reports an event that doesn't cause a stop.
>>>> E.g., a thread specific breakpoint, a "handle nostop" signal, etc.
>>>
>>> Does that involve minus_one_ptid on GNU/Linux as well? If not, that
>>> call will not mark all threads as running, will it?
>>
>> It does. user_visible_resume_ptid returns the same irrespective
>> of target_can_async.
>
> Then how come your test program ended up showing just that additional
> thread running? You even said you couldn't explain why I saw all my
> threads running on Windows.
Because linux-nat.c does not report any event to infrun when a
thread is created, it just immediately goes back to waiting, unlike
windows-nat.c, which reports TARGET_WAITKIND_SPURIOUS. By previous
test program did not have the new thread trip on any event after
its creation.
That's why I said:
~~~
So that should mean that even for GNU/Linux, it should be possible
to end in the exact same, when any thread other than the one that we
had started the infcall in reports an event that doesn't cause a stop.
E.g., a thread specific breakpoint, a "handle nostop" signal, etc.
~~~
And I just confirmed it, with a thread-specific breakpoint that
trips on the wrong thread, thus, reporting a trap to the core,
which does not cause a user-visible stop, and then ends up
marking all threads running when we gdb internally re-resumes
the program. Like so:
~~~
#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
#include <pthread.h>
volatile int count;
static int
foo (void)
{
usleep (1);
}
static void *
thread_function (void *arg)
{
pthread_t thread;
printf ("created thread %d\n", count++);
while (1)
{
foo ();
}
}
void
new_thread (void)
{
pthread_t thread;
pthread_create (&thread, NULL, thread_function, NULL);
}
int
main (int argc, char **argv)
{
while (1)
{
usleep (1);
}
}
~~~
~~~
(gdb) start
...
Temporary breakpoint 1, main (argc=1, argv=0x7fffffffd858) at threads_infcall.c:39
39 usleep (1);
(gdb) b foo thread 1
Breakpoint 2 at 0x4006a4: file threads_infcall.c, line 11.
(gdb) p new_thread ()
[New Thread 0x7ffff7fc1700 (LWP 4928)]
created thread 1
$1 = void
(gdb) info threads
Id Target Id Frame
2 Thread 0x7ffff7fc1700 (LWP 4928) "threads_infcall" (running)
* 1 Thread 0x7ffff7fc2740 (LWP 4924) "threads_infcall" main (argc=1, argv=0x7fffffffd858) at threads_infcall.c:39
(gdb) p new_thread ()
[New Thread 0x7ffff77c0700 (LWP 4929)]
created thread 2
$2 = void
(gdb) info threads
Id Target Id Frame
3 Thread 0x7ffff77c0700 (LWP 4929) "threads_infcall" (running)
2 Thread 0x7ffff7fc1700 (LWP 4928) "threads_infcall" (running)
* 1 Thread 0x7ffff7fc2740 (LWP 4924) "threads_infcall" (running)
(gdb)
~~~
Thanks,
Pedro Alves