Re: [PATCH] futex: Switch to USER_DS for futex test

From: Andrew Morton
Date: Wed Dec 11 2013 - 17:35:59 EST


On Wed, 11 Dec 2013 11:43:21 +0100 Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:

> Since commit e4f2dfbb5e92be4e46c0625f4f8eb101110f756f ("m68k: implement
> futex.h to support userspace robust futexes and PI mutexes"), the kernel
> crashes during boot up on MC68030:
>
> Data read fault at 0x00000000 in Super Data (pc=0x3afec)
> BAD KERNEL BUSERR
> Oops: 00000000
> Modules linked in:
> PC: [<0003afec>] cmpxchg_futex_value_locked+0x14/0x4a
> SR: 2004 SP: 0082fed4 a2: 0082c000
> d0: 00000000 d1: 00000001 d2: 00000018 d3: 00000000
> d4: 00000061 d5: 00001000 a0: 00000000 a1: 0082e000
> Process swapper (pid: 1, task=0082c000)
> Frame format=B ssw=074d isc=4a80 isb=661c daddr=00000000 dobuf=00000001
> baddr=0003aff2 dibuf=00000000 ver=f
> Stack from 0082ff5c:
> 002b8cb8 0082ff70 00000000 00000000 00000000 00000000 00000000 000020ac
> 00000018 00000007 00000061 00001000 00000000 00000000 002cab50 00002008
> 002b3a56 002b8ca4 0082c3f0 00000000 0082c53c 001e316a 00000000 00000000
> 001e3172 001e316a 000025d4 00000000 00000000 00000000 00000000 00000000
> 00000000 00000000 00000000 00000000 00000000 00000000 00000000 20000000
> 00000000
> Call Trace: [<002b8cb8>] futex_init+0x14/0x54
> [<000020ac>] do_one_initcall+0xa4/0x144
> [<00001000>] kernel_pg_dir+0x0/0x1000
> [<00002008>] do_one_initcall+0x0/0x144
> [<002b3a56>] kernel_init_freeable+0xca/0x152
> [<002b8ca4>] futex_init+0x0/0x54
> [<001e316a>] kernel_init+0x0/0xc8
> [<001e3172>] kernel_init+0x8/0xc8
> [<001e316a>] kernel_init+0x0/0xc8
> [<000025d4>] ret_from_kernel_thread+0xc/0x14
>
> This happens because the futex test in futex_init() lacks a switch to the
> USER_DS address space, while cmpxchg_futex_value_locked() and
> futex_atomic_cmpxchg_inatomic() operate on userspace pointers (albeit NULL
> for this particular test).
>
> Fix this by switching to USER_DS before running the test, and restoring the
> old address space afterwards.
>
> ...
>
> static int __init futex_init(void)
> {
> + mm_segment_t fs;
> u32 curval;
> int i;
>
> @@ -2745,8 +2747,11 @@ static int __init futex_init(void)
> * implementation, the non-functional ones will return
> * -ENOSYS.
> */
> + fs = get_fs();
> + set_fs(USER_DS);
> if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
> futex_cmpxchg_enabled = 1;
> + set_fs(fs);

hm. Are we sure that a reference to NULL will continue to fault on all
architectures when switched to USER_DS?

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/