Re: [Patch] x86_64: TASK_SIZE fixes for compatibility mode processes

[Andrew Morton <akpm@osdl.org>]

"Siddha, Suresh B" <suresh.b.siddha@intel.com> wrote:
>
> On Thu, Jun 02, 2005 at 01:50:13PM -0700, Andrew Morton wrote:
> > "Siddha, Suresh B" <suresh.b.siddha@intel.com> wrote:
> > >
> > > +#define TASK_SIZE_OF(child) 	((test_tsk_thread_flag(child, TIF_IA32)) ? IA32_PAGE_OFFSET : TASK_SIZE64)
> > 
> > The task size is an attribute of the task's mm_struct, not of the task.
> 
> ia64, ppc64 and s390 seems be getting this info from thread_info or 
> thread_struct in the task struct.

I know.  I'm claiming that this is conceptually wrong.

> > The place where this tends to come unstuck is when a 32-bit task holds a
> > reference on a 64-bit tasks's task_struct via a read of a /proc file.  If
> > the 64-bit task exits then it is the 32-bit task who does the final freeing
> > of the 64-bit tasks's task_struct and mm_struct.  (and all vice-versa, of
> > course).  Will your patch handle this race scenario correctly?
> 
> In recent kernels, instead of TASK_SIZE, "-1" is getting passed to unmap_vmas()
> from exit_mmap. Same case with ceiling (set to "0") for free_pgtables().
> It shouldn't be a problem with this, right?

Yeah, I recall that hack being added.  Might work.