[ob] Don't clobber inferior_ptid in read_pc_pid

[ob] Don't clobber inferior_ptid in read_pc_pid

[Daniel Jacobowitz]

Another issue found in testing on arm-linux.  A return was added to this
function back in June; if we return from the middle of it, we leave
inferior_ptid set to the wrong thread.  This causes a "!ptid_equal
(ecs->ptid, inferior_ptid)" test to fail, since we called read_pc_pid
with ecs->ptid.  That leads to not calling context_switch; which clobbers
the stepping range for the previous thread; which causes stepping to stop
unexpectedly.

I'll commit this patch as obvious in a day or two.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

2004-02-28  Daniel Jacobowitz  <drow@mvista.com>

	* regcache.c (read_pc_pid): Restore inferior_ptid after reading
	PC_REGNUM.

Index: gdb/regcache.c
===================================================================
RCS file: /big/fsf/rsync/src-cvs/src/gdb/regcache.c,v
retrieving revision 1.108
diff -u -p -r1.108 regcache.c
--- gdb/regcache.c	15 Feb 2004 15:22:06 -0000	1.108
+++ gdb/regcache.c	27 Feb 2004 19:31:13 -0000
@@ -1332,8 +1332,7 @@ read_pc_pid (ptid_t ptid)
   else if (PC_REGNUM >= 0)
     {
       CORE_ADDR raw_val = read_register_pid (PC_REGNUM, ptid);
-      CORE_ADDR pc_val = ADDR_BITS_REMOVE (raw_val);
-      return pc_val;
+      pc_val = ADDR_BITS_REMOVE (raw_val);
     }
   else
     internal_error (__FILE__, __LINE__, "read_pc_pid: Unable to find PC");

Re: [ob] Don't clobber inferior_ptid in read_pc_pid

[Andrew Cagney]

> Another issue found in testing on arm-linux.  A return was added to this
> function back in June; if we return from the middle of it, we leave
> inferior_ptid set to the wrong thread.  This causes a "!ptid_equal
> (ecs->ptid, inferior_ptid)" test to fail, since we called read_pc_pid
> with ecs->ptid.  That leads to not calling context_switch; which clobbers
> the stepping range for the previous thread; which causes stepping to stop
> unexpectedly.
> 
> I'll commit this patch as obvious in a day or two.

Can you please commit it now?

Andrew

Re: [ob] Don't clobber inferior_ptid in read_pc_pid

[Daniel Jacobowitz]

On Sat, Feb 28, 2004 at 02:32:55PM -0500, Andrew Cagney wrote:
> >Another issue found in testing on arm-linux.  A return was added to this
> >function back in June; if we return from the middle of it, we leave
> >inferior_ptid set to the wrong thread.  This causes a "!ptid_equal
> >(ecs->ptid, inferior_ptid)" test to fail, since we called read_pc_pid
> >with ecs->ptid.  That leads to not calling context_switch; which clobbers
> >the stepping range for the previous thread; which causes stepping to stop
> >unexpectedly.
> >
> >I'll commit this patch as obvious in a day or two.
> 
> Can you please commit it now?

Sorry, I had left before this message arrived.

My goal in waiting was to retest on another target, which I did not
have time for this morning, and to wait for the release branch to be
confusion on my own part about the timing, since as of your
next-to-last announcement you were planning on back-dating the release
branch.  These are patches I consider suitable for the release branch
and it's not much extra work for me to retest and commit them on two
branches.

When I check in patches immediately people complain that I am acting
impetuously.  When I wait you ask me to commit the patch now.  When I
get back and see your message I get:

> Daniel, what the f- are you doing?  I'm half way through cutting a
> branch, the last thing I need is you being a fly-by nighter checking in
> random stuff.

OK, I'm wrong no matter what I do.  I'll come back to fixing ARM in a
few days when the dust has settled.

FYI, I stopped committing to the release branch as soon as I saw your
message asking me to.  You don't need to curse at me about it.  In
fact, I'd appreciate it if you wouldn't.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: [ob] Don't clobber inferior_ptid in read_pc_pid

[Andrew Cagney]

> On Sat, Feb 28, 2004 at 02:32:55PM -0500, Andrew Cagney wrote:
> 
>>>> >Another issue found in testing on arm-linux.  A return was added to this
>>>> >function back in June; if we return from the middle of it, we leave
>>>> >inferior_ptid set to the wrong thread.  This causes a "!ptid_equal
>>>> >(ecs->ptid, inferior_ptid)" test to fail, since we called read_pc_pid
>>>> >with ecs->ptid.  That leads to not calling context_switch; which clobbers
>>>> >the stepping range for the previous thread; which causes stepping to stop
>>>> >unexpectedly.
>>>> >
>>>> >I'll commit this patch as obvious in a day or two.

It isn't obvious, but it does appear to be correct.

>>> Can you please commit it now?
> 
> 
> Sorry, I had left before this message arrived.
> 
> My goal in waiting was to retest on another target, which I did not
> have time for this morning, and to wait for the release branch to be
> confusion on my own part about the timing, since as of your
> next-to-last announcement you were planning on back-dating the release
> branch.  These are patches I consider suitable for the release branch
> and it's not much extra work for me to retest and commit them on two
> branches.
> 
> When I check in patches immediately people complain that I am acting
> impetuously.  When I wait you ask me to commit the patch now.  When I
> get back and see your message I get:

My post clearly has context - if that change were committed before the 
branch was cut, life will be easy for all concerned.

It doesn't say commit to an un-announced branch the moment you see the 
first hint that it is being created.  That will make the release 
engineer's life hell.

I'm going to alter the releng notes to point out that the first mention 
of the branch should be its announcement.  That hopefully will stop 
future occurances of this problem.

Andrew

Re: [ob] Don't clobber inferior_ptid in read_pc_pid

[Daniel Jacobowitz]

On Sat, Feb 28, 2004 at 12:35:41PM -0500, Daniel Jacobowitz wrote:
> Another issue found in testing on arm-linux.  A return was added to this
> function back in June; if we return from the middle of it, we leave
> inferior_ptid set to the wrong thread.  This causes a "!ptid_equal
> (ecs->ptid, inferior_ptid)" test to fail, since we called read_pc_pid
> with ecs->ptid.  That leads to not calling context_switch; which clobbers
> the stepping range for the previous thread; which causes stepping to stop
> unexpectedly.

> 2004-02-28  Daniel Jacobowitz  <drow@mvista.com>
> 
> 	* regcache.c (read_pc_pid): Restore inferior_ptid after reading
> 	PC_REGNUM.

I have committed this to HEAD.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer