Re: utrace regressions (was: -mm merge plans for 2.6.21)

[Roland McGrath <roland@redhat.com>]

> On Sat, Feb 17, 2007 at 06:35:31PM -0800, Roland McGrath wrote:
> > > Looking at mainline x86_64 ptrace code I think hole for u_debugreg[4]
> > > and [5] is also needed.
> >
> > It's not.  The utrace_regset for the debugregs already has that behavior
> > for those two words, so mapping all 8 uarea words to the regset is fine.
> 
> Sorry, I don't get it. Choosing segment from x86_64_uarea is done before
> calling regset->set and regset->get as well as before zero-filling. No
> segment for u_debugreg[4] and [5] means -EIO before segment handlers
> will have a chance to be called.
> 
> Do you want to consolidate these two?
> 
>     {offsetof(struct user, u_debugreg[0]), offsetof(struct user, u_debugreg[4]), 3, 0},
>     {offsetof(struct user, u_debugreg[6]), offsetof(struct user, u_debugreg[8]), 3, 6 * sizeof(long)},

Oops!  I was misremembering what was in x86_64_uarea when I wrote that.
I've indeed fixed it to match what I thought it was:

	{offsetof(struct user, u_debugreg[0]),
	 offsetof(struct user, u_debugreg[8]), 3, 0},


Thanks,
Roland