RE: [PATCH] powerpc: Select HAVE_FUTEX_CMPXCHG

[David Laight]

From: Samuel Holland
> Sent: 19 September 2020 04:20
> 
> On powerpc, access_ok() succeeds for the NULL pointer. This breaks the
> dynamic check in futex_detect_cmpxchg(), which expects -EFAULT. As a
> result, robust futex operations are not functional on powerpc.

access_ok(NULL, sane_count) will succeed on all (maybe most) architectures.
All access_ok() does is check that kernel addresses aren't referenced.
(access_ok(kernel_adress, 0) is also likely to succeed.)

It is the access to user-address 0 that is expected to fault.
If this isn't faulting something else is wrong.

Historically (at least pre-elf, if not before) user programs
were linked to address zero - so the page was mapped.
(Linux may be too new to actually require it.)
Not sure what 'wine' requires for win-32 execuatbles.

ISTR there are also some 'crazy' ARM? cpu that read the interrupt
vectors from address 0 in user-space.

So assuming:

static void __init futex_detect_cmpxchg(void)
{
#ifndef CONFIG_HAVE_FUTEX_CMPXCHG
        u32 curval;

        /*
         * This will fail and we want it. Some arch implementations do
         * runtime detection of the futex_atomic_cmpxchg_inatomic()
         * functionality. We want to know that before we call in any
         * of the complex code paths. Also we want to prevent
         * registration of robust lists in that case. NULL is
         * guaranteed to fault and we get -EFAULT on functional
         * implementation, the non-functional ones will return
         * -ENOSYS.
         */
        if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
                futex_cmpxchg_enabled = 1;
#endif
}

will fail -EFAULT because user address 0 is invalid seems hopeful.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)