Re: Wrong data type in function unpack_varlen_hex()

["zhigang gong" <zhigang dot gong at gmail dot com>]

On 7/13/06, Daniel Jacobowitz <drow@false.org> wrote:
> On Fri, Jun 16, 2006 at 08:58:33PM +0800, zhigang gong wrote:
> > Hi,
> >
> >        I am working on an mips4k platform. When I added hw watchpoint
> > support for my target board and debug it with gdb. The write
> > watchpoint works fine, but rwatch and awatch doesn't work. After trace
> > the source code of the gdb, I found there is a bug in
> > unpack_varlen_hex. The local variable retval is a signed integer. For
> > my case, the ULONGEST is a 64bit integer type. So when the
> > watchpoint's address is 0x8XXXXXXX, the "retval" will be 0x8XXXXXXX,
> > and pass its value to variable "result", the "result"'s value will be
> > sign extended to 0xFFFFFFFF8XXXXXXX. Then when i set a rwatch point,
> > the address matching will fail when the read watchpoint ocurred.
> >     The patch is as belows. And I test it,
>
> Thank you for the patch.  This is definitely an improvement, so I have
> committed it.
>
> There may be more problems here: addresses on MIPS are generally
> considered to be signed, so it might be a bug somewhere else in GDB
> that you're getting 0x000000008xxxxxxx.  But, we'll worry about that
> only if it causes a problem for someone else.
I have worried about that too. As I foud that in gdb the signed extention should
be the default way. And I also found the watchpoint address
0x000000008xxxxxxx was calculated from the address expression, which
is inputted in the command line as belows:
gdb ) rwatch  *0x8xxxxxxx
Maybe  change the watchpoint address expression calculating, make it
to do an signed extension, is better.

> --
> Daniel Jacobowitz
> CodeSourcery