Re: [PATCH] Fix "PC register is not available" issue

[Joel Brobecker <brobecker at adacore dot com>]

Hi Eli,

On Mon, Mar 17, 2014 at 09:43:23PM +0200, Eli Zaretskii wrote:
> This problem was raised and mentioned several times over the last few
> years.  It was discussed here in the thread which started with this
> message:
> 
>   https://sourceware.org/ml/gdb-patches/2013-06/msg00237.html

Thanks for looking into this!

> So I came up with the patch below.  The idea is that setting
> th->suspended to -1 just signals to GDB that the thread isn't
> suspended, and shouldn't be resumed; otherwise, we simply ignore the
> failure to suspend the thread, although the warning is still printed.
> 
> I am running with this patch for almost a month, and the dreaded "PC
> register is not available" message didn't appear even once.  No more
> botched debugging sessions!  The test program in PR/14018 also runs
> indefinitely until interrupted, instead of stopping.
> 
> So I suggest to install this.  OK?  Should I also close the Bugzilla
> PR?
> 
> --- gdb/windows-nat.c~0	2014-02-06 04:21:29.000000000 +0200
> +++ gdb/windows-nat.c	2014-02-26 22:27:10.225625000 +0200
> @@ -313,9 +313,10 @@ thread_rec (DWORD id, int get_context)
>  		    warning (_("SuspendThread (tid=0x%x) failed."
>  			       " (winerr %u)"),
>  			     (unsigned) id, (unsigned) err);
> -		    return NULL;
> +		    th->suspended = -1;
>  		  }
> -		th->suspended = 1;
> +		else
> +		  th->suspended = 1;
>  	      }
>  	    else if (get_context < 0)
>  	      th->suspended = -1;

Generally speaking, it seems to make sense to me that we would
mark as unsuspended threads that we cannot suspend. But one question
that rises from doing that is: how does the rest of GDB handle
this thread? In particular, does the thread show up in "info thread"
and is the user able to switch to that thread? etc? Certainly, if
the current situation is to leave the user stranded, the suggested
approach cannot only be an improvement...

Another thought I had on your patch is that we might want to limit
the warning to situation where the return code is not a permission
denied.

It would also have been nice to double-check that the thread in
question, when the error shows up, is indeed on a system thread
unrelated to our program (Eg: the thread created when we try to
interrupt the program with ctrl-c or ctrl-break). Not sure if
there is a way to determine that, though.

I would have looked into that, but I don't have much time this week, and
might be equally busy the following 2 weeks, so I didn't want to delay
my answer any further. Overall, your patch looks very promising.

Sorry I can't be anymore support for the moment.
-- 
Joel