locking/atomic: arm: fix sync ops

The sync_*() ops on arch/arm are defined in terms of the regular bitops
with no special handling. This is not correct, as UP kernels elide
barriers for the fully-ordered operations, and so the required ordering
is lost when such UP kernels are run under a hypervsior on an SMP
system.

Fix this by defining sync ops with the required barriers.

Note: On 32-bit arm, the sync_*() ops are currently only used by Xen,
which requires ARMv7, but the semantics can be implemented for ARMv6+.

Fixes: e54d2f61528165bb ("xen/arm: sync_bitops")
Signed-off-by: Mark Rutland <[e-mail]>
Signed-off-by: Peter Zijlstra (Intel) <[e-mail]>
Reviewed-by: Kees Cook <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]

arch/arm/include/asm/assembler.h

···
 #endif
 	.endm
 
+/*
+ * Raw SMP data memory barrier
+ */
+	.macro	__smp_dmb mode
+#if __LINUX_ARM_ARCH__ >= 7
+	.ifeqs "\mode","arm"
+	dmb	ish
+	.else
+	W(dmb)	ish
+	.endif
+#elif __LINUX_ARM_ARCH__ == 6
+	mcr	p15, 0, r0, c7, c10, 5	@ dmb
+#else
+	.error "Incompatible SMP platform"
+#endif
+	.endm
+
 #if defined(CONFIG_CPU_V7M)
 	/*
 	 * setmode is used to assert to be in svc mode during boot. For v7-M

arch/arm/include/asm/sync_bitops.h

···
  * ops which are SMP safe even on a UP kernel.
  */
 
+/*
+ * Unordered
+ */
+
 #define sync_set_bit(nr, p)		_set_bit(nr, p)
 #define sync_clear_bit(nr, p)		_clear_bit(nr, p)
 #define sync_change_bit(nr, p)		_change_bit(nr, p)
-#define sync_test_and_set_bit(nr, p)	_test_and_set_bit(nr, p)
-#define sync_test_and_clear_bit(nr, p)	_test_and_clear_bit(nr, p)
-#define sync_test_and_change_bit(nr, p)	_test_and_change_bit(nr, p)
 #define sync_test_bit(nr, addr)		test_bit(nr, addr)
-#define arch_sync_cmpxchg		arch_cmpxchg
 
+/*
+ * Fully ordered
+ */
+
+int _sync_test_and_set_bit(int nr, volatile unsigned long * p);
+#define sync_test_and_set_bit(nr, p)	_sync_test_and_set_bit(nr, p)
+
+int _sync_test_and_clear_bit(int nr, volatile unsigned long * p);
+#define sync_test_and_clear_bit(nr, p)	_sync_test_and_clear_bit(nr, p)
+
+int _sync_test_and_change_bit(int nr, volatile unsigned long * p);
+#define sync_test_and_change_bit(nr, p)	_sync_test_and_change_bit(nr, p)
+
+#define arch_sync_cmpxchg(ptr, old, new)				\
+({									\
+	__typeof__(*(ptr)) __ret;					\
+	__smp_mb__before_atomic();					\
+	__ret = arch_cmpxchg_relaxed((ptr), (old), (new));		\
+	__smp_mb__after_atomic();					\
+	__ret;								\
+})
 
 #endif

arch/arm/lib/bitops.h

···
 ENDPROC(\name		)
 	.endm
 
-	.macro	testop, name, instr, store
+	.macro	__testop, name, instr, store, barrier
 ENTRY(	\name		)
 UNWIND(	.fnstart	)
 	ands	ip, r1, #3
···
 	mov	r0, r0, lsr #5
 	add	r1, r1, r0, lsl #2	@ Get word offset
 	mov	r3, r2, lsl r3		@ create mask
-	smp_dmb
+	\barrier
 #if __LINUX_ARM_ARCH__ >= 7 && defined(CONFIG_SMP)
 	.arch_extension	mp
 	ALT_SMP(W(pldw)	[r1])
···
 	strex	ip, r2, [r1]
 	cmp	ip, #0
 	bne	1b
-	smp_dmb
+	\barrier
 	cmp	r0, #0
 	movne	r0, #1
 2:	bx	lr
 UNWIND(	.fnend		)
 ENDPROC(\name		)
+	.endm
+
+	.macro	testop, name, instr, store
+	__testop \name, \instr, \store, smp_dmb
+	.endm
+
+	.macro	sync_testop, name, instr, store
+	__testop \name, \instr, \store, __smp_dmb
 	.endm
 #else
 	.macro	bitop, name, instr

arch/arm/lib/testchangebit.S

···
                 .text
 
 testop	_test_and_change_bit, eor, str
+
+#if __LINUX_ARM_ARCH__ >= 6
+sync_testop	_sync_test_and_change_bit, eor, str
+#endif

arch/arm/lib/testclearbit.S

···
                 .text
 
 testop	_test_and_clear_bit, bicne, strne
+
+#if __LINUX_ARM_ARCH__ >= 6
+sync_testop	_sync_test_and_clear_bit, bicne, strne
+#endif

arch/arm/lib/testsetbit.S

···
                 .text
 
 testop	_test_and_set_bit, orreq, streq
+
+#if __LINUX_ARM_ARCH__ >= 6
+sync_testop	_sync_test_and_set_bit, orreq, streq
+#endif