Re: [Patch] Fix memory corruption when using multiple hardware watchpoints.

[Daniel Jacobowitz <drow at false dot org>]

On Wed, Sep 10, 2008 at 01:14:34PM -0700, David Daney wrote:
> In bpstat_stop_status there is special code to clear the breakpoint_at
> element for a hardware watchpoint if it could be invalidated by a
> remove_breakpoints(), insert_breakpoints() pair.  The comment explains
> that we do this to "Prevent further code from trying to use it."
> 
> The problem with the current code is that it assumes that there is only
> a single hardware breakpoint and breaks out of the loop when it is
> found.  If there are multiple hardware breakpoints, any but the first
> are left with breakpoint_at pointing at now freed memory leading to Bad
>  Things if the memory happens to get clobbered.
> 
> My fix is to continue through the entire list clearing breakpoint_at for
>  *all* hardware watchpoints and then outside of the loop calling
> remove_breakpoints(), insert_breakpoints() if any were found.
> 
> Testing on x86_64-pc-linux-gnu.  OK to commit (if no regressions)?
> 
> 2008-09-10  David Daney  <ddaney@avtrex.com>
> 
> 	* breakpoint.c (bpstat_stop_status): Clear breakpoint_at for
> 	all hardware bpstats.

OK.

On Tue, Sep 16, 2008 at 11:33:39AM -0700, David Daney wrote:
> This patch:
> http://sourceware.org/ml/gdb-patches/2008-09/msg00222.html
> 
> Is awaiting review.
> 
> It fixes 10 FAILs in the testsuite on x86_64-pc-linux:

All of these look like threading tests with no watchpoints; I suspect
this is just noise.

-- 
Daniel Jacobowitz
CodeSourcery