Re: [PATCH] gdbserver: don't pick a random thread if the current thread dies

[Pedro Alves <palves@redhat.com>]

On 08/24/2015 03:59 PM, Ulrich Weigand wrote:
> Pedro Alves wrote:
> 
>> 	* spu-low.c (spu_resume, spu_request_interrupt): Use the first
>> 	thread's lwp instead of the current thread's.
> 
>>  static void
>>  spu_request_interrupt (void)
>>  {
>> -  syscall (SYS_tkill, ptid_get_lwp (current_ptid), SIGINT);
>> +  struct thread_info *thr = get_first_thread ();
>> +
>> +  syscall (SYS_tkill, ptid_get_lwp (thr), SIGINT);
>>  }
> 
> This doesn't compile due to:
> 
> gdbserver/spu-low.c: In function 'spu_request_interrupt':
> gdbserver/spu-low.c:639: error: incompatible type for argument 1 of 'ptid_get_lwp'
> 
> When adding the obvious fix ("ptid_of (thr)"), it does compile, but now
> gdbserver crashes as soon as GDB attaches to it:

I guess lwpid_of would be a little better even.

> Looks like current_thread is NULL at this point.  Since this is generic
> code, I'm not quite sure if this is a SPU-specific problem or not ...
> 
> Do you think this is related to your change?

Sounds likely.  I can trigger the same on GNU/Linux if I force-disable
the multi-process extensions:

diff --git c/gdb/gdbserver/linux-low.c w/gdb/gdbserver/linux-low.c
index 2bc91c2..44c0feb 100644
--- c/gdb/gdbserver/linux-low.c
+++ w/gdb/gdbserver/linux-low.c
@@ -5785,7 +5785,7 @@ linux_start_non_stop (int nonstop)
 static int
 linux_supports_multi_process (void)
 {
-  return 1;
+  return 0;
 }

 /* Check if fork events are supported.  */


Let me take a closer look.

Thanks,
Pedro Alves