OSDN Git Service

ARM: futex: Address build warning
authorThomas Gleixner <tglx@linutronix.de>
Tue, 14 Apr 2020 09:07:22 +0000 (11:07 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Wed, 6 May 2020 22:41:47 +0000 (00:41 +0200)
commit8101b5a1531f3390b3a69fa7934c70a8fd6566ad
tree4332a54862007c203fa57b7afca7295e3d0b0c69
parent3c40cdb0e93ec166f1fa4fee1eb62d45b5451515
ARM: futex: Address build warning

Stephen reported the following build warning on a ARM multi_v7_defconfig
build with GCC 9.2.1:

kernel/futex.c: In function 'do_futex':
kernel/futex.c:1676:17: warning: 'oldval' may be used uninitialized in this function [-Wmaybe-uninitialized]
 1676 |   return oldval == cmparg;
      |          ~~~~~~~^~~~~~~~~
kernel/futex.c:1652:6: note: 'oldval' was declared here
 1652 |  int oldval, ret;
      |      ^~~~~~

introduced by commit a08971e9488d ("futex: arch_futex_atomic_op_inuser()
calling conventions change").

While that change should not make any difference it confuses GCC which
fails to work out that oldval is not referenced when the return value is
not zero.

GCC fails to properly analyze arch_futex_atomic_op_inuser(). It's not the
early return, the issue is with the assembly macros. GCC fails to detect
that those either set 'ret' to 0 and set oldval or set 'ret' to -EFAULT
which makes oldval uninteresting. The store to the callsite supplied oldval
pointer is conditional on ret == 0.

The straight forward way to solve this is to make the store unconditional.

Aside of addressing the build warning this makes sense anyway because it
removes the conditional from the fastpath. In the error case the stored
value is uninteresting and the extra store does not matter at all.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/87pncao2ph.fsf@nanos.tec.linutronix.de
arch/arm/include/asm/futex.h