Re: [PATCH 4.9 72/76] arm64: futex: Fix FUTEX_WAKE_OP atomic ops with non-zero result value

[Greg Kroah-Hartman]

On Mon, Apr 15, 2019 at 03:01:51PM -0700, Nathan Chancellor wrote:
> On Mon, Apr 15, 2019 at 08:44:36PM +0200, Greg Kroah-Hartman wrote:
> > From: Will Deacon <will.deacon@xxxxxxx>
> > 
> > commit 045afc24124d80c6998d9c770844c67912083506 upstream.
> > 
> > Rather embarrassingly, our futex() FUTEX_WAKE_OP implementation doesn't
> > explicitly set the return value on the non-faulting path and instead
> > leaves it holding the result of the underlying atomic operation. This
> > means that any FUTEX_WAKE_OP atomic operation which computes a non-zero
> > value will be reported as having failed. Regrettably, I wrote the buggy
> > code back in 2011 and it was upstreamed as part of the initial arm64
> > support in 2012.
> > 
> > The reasons we appear to get away with this are:
> > 
> >   1. FUTEX_WAKE_OP is rarely used and therefore doesn't appear to get
> >      exercised by futex() test applications
> > 
> >   2. If the result of the atomic operation is zero, the system call
> >      behaves correctly
> > 
> >   3. Prior to version 2.25, the only operation used by GLIBC set the
> >      futex to zero, and therefore worked as expected. From 2.25 onwards,
> >      FUTEX_WAKE_OP is not used by GLIBC at all.
> > 
> > Fix the implementation by ensuring that the return value is either 0
> > to indicate that the atomic operation completed successfully, or -EFAULT
> > if we encountered a fault when accessing the user mapping.
> > 
> > Cc: <stable@xxxxxxxxxx>
> > Fixes: 6170a97460db ("arm64: Atomic operations")
> > Signed-off-by: Will Deacon <will.deacon@xxxxxxx>
> > Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> > 
> > ---
> >  arch/arm64/include/asm/futex.h |   16 ++++++++--------
> >  1 file changed, 8 insertions(+), 8 deletions(-)
> > 
> > --- a/arch/arm64/include/asm/futex.h
> > +++ b/arch/arm64/include/asm/futex.h
> > @@ -33,8 +33,8 @@
> >  "	prfm	pstl1strm, %2\n"					\
> >  "1:	ldxr	%w1, %2\n"						\
> >  	insn "\n"							\
> > -"2:	stlxr	%w3, %w0, %2\n"						\
> > -"	cbnz	%w3, 1b\n"						\
> > +"2:	stlxr	%w0, %w3, %2\n"						\
> > +"	cbnz	%w0, 1b\n"						\
> >  "	dmb	ish\n"							\
> >  "3:\n"									\
> >  "	.pushsection .fixup,\"ax\"\n"					\
> > @@ -53,29 +53,29 @@
> >  static inline int
> >  arch_futex_atomic_op_inuser(int op, int oparg, int *oval, u32 __user *uaddr)
> >  {
> > -	int oldval = 0, ret, tmp;
> > +	int oldval, ret, tmp;
> >  
> >  	pagefault_disable();
> >  
> >  	switch (op) {
> >  	case FUTEX_OP_SET:
> > -		__futex_atomic_op("mov	%w0, %w4",
> > +		__futex_atomic_op("mov	%w3, %w4",
> >  				  ret, oldval, uaddr, tmp, oparg);
> >  		break;
> >  	case FUTEX_OP_ADD:
> > -		__futex_atomic_op("add	%w0, %w1, %w4",
> > +		__futex_atomic_op("add	%w3, %w1, %w4",
> >  				  ret, oldval, uaddr, tmp, oparg);
> >  		break;
> >  	case FUTEX_OP_OR:
> > -		__futex_atomic_op("orr	%w0, %w1, %w4",
> > +		__futex_atomic_op("orr	%w3, %w1, %w4",
> >  				  ret, oldval, uaddr, tmp, oparg);
> >  		break;
> >  	case FUTEX_OP_ANDN:
> > -		__futex_atomic_op("and	%w0, %w1, %w4",
> > +		__futex_atomic_op("and	%w3, %w1, %w4",
> >  				  ret, oldval, uaddr, tmp, ~oparg);
> >  		break;
> >  	case FUTEX_OP_XOR:
> > -		__futex_atomic_op("eor	%w0, %w1, %w4",
> > +		__futex_atomic_op("eor	%w3, %w1, %w4",
> >  				  ret, oldval, uaddr, tmp, oparg);
> >  		break;
> >  	default:
> > 
> >
> 
> This causes a (false) build warning with AOSP's GCC 4.9.4 (which is
> used to build nearly all arm64 Android kernels before 4.14):
> 
>   CC      kernel/futex.o
> ../kernel/futex.c: In function 'do_futex':
> ../kernel/futex.c:1492:17: warning: 'oldval' may be used uninitialized in this function [-Wmaybe-uninitialized]
>    return oldval == cmparg;
>                  ^
> In file included from ../kernel/futex.c:69:0:
> ../arch/arm64/include/asm/futex.h:56:6: note: 'oldval' was declared here
>   int oldval, ret, tmp;
>       ^
> 
> The only reason I bring this up is Qualcomm based kernels have a Python
> script that emulates -Werror, meaning this will be fatal for a large
> number of kernels, when this eventually gets merged into them.

Argh, really?  That's a buggy compiler that you have there, as oldval
will be set correctly if all is good, and if not, ret will be and the
code will error out.

Working around broken compilers is not something I really like doing :(

That being said, does this also show up in the 4.19.y and 5.0.y tree
right now?  If not, why not?

thanks,

greg k-h