wait_on_bit: add an acquire memory barrier

There are several places in the kernel where wait_on_bit is not followed
by a memory barrier (for example, in drivers/md/dm-bufio.c:new_read).

On architectures with weak memory ordering, it may happen that memory
accesses that follow wait_on_bit are reordered before wait_on_bit and
they may return invalid data.

Fix this class of bugs by introducing a new function "test_bit_acquire"
that works like test_bit, but has acquire memory ordering semantics.

Signed-off-by: Mikulas Patocka <[e-mail]>
Acked-by: Will Deacon <[e-mail]>
Cc: [e-mail]
Signed-off-by: Linus Torvalds <[e-mail]>

Documentation/atomic_bitops.txt

···
 
  - RMW operations that have a return value are fully ordered.
 
- - RMW operations that are conditional are unordered on FAILURE,
-   otherwise the above rules apply. In the case of test_and_set_bit_lock(),
-   if the bit in memory is unchanged by the operation then it is deemed to have
-   failed.
+ - RMW operations that are conditional are fully ordered.
 
-Except for a successful test_and_set_bit_lock() which has ACQUIRE semantics and
-clear_bit_unlock() which has RELEASE semantics.
+Except for a successful test_and_set_bit_lock() which has ACQUIRE semantics,
+clear_bit_unlock() which has RELEASE semantics and test_bit_acquire which has
+ACQUIRE semantics.
 
 Since a platform only has a single means of achieving atomic operations
 the same barriers as for atomic_t are used, see atomic_t.txt.

arch/x86/include/asm/bitops.h

···
 		(addr[nr >> _BITOPS_LONG_SHIFT])) != 0;
 }
 
+static __always_inline bool constant_test_bit_acquire(long nr, const volatile unsigned long *addr)
+{
+	bool oldbit;
+
+	asm volatile("testb %2,%1"
+		     CC_SET(nz)
+		     : CC_OUT(nz) (oldbit)
+		     : "m" (((unsigned char *)addr)[nr >> 3]),
+		       "i" (1 << (nr & 7))
+		     :"memory");
+
+	return oldbit;
+}
+
 static __always_inline bool variable_test_bit(long nr, volatile const unsigned long *addr)
 {
 	bool oldbit;
···
 arch_test_bit(unsigned long nr, const volatile unsigned long *addr)
 {
 	return __builtin_constant_p(nr) ? constant_test_bit(nr, addr) :
+					  variable_test_bit(nr, addr);
+}
+
+static __always_inline bool
+arch_test_bit_acquire(unsigned long nr, const volatile unsigned long *addr)
+{
+	return __builtin_constant_p(nr) ? constant_test_bit_acquire(nr, addr) :
 					  variable_test_bit(nr, addr);
 }
 

include/asm-generic/bitops/generic-non-atomic.h

···
 #define __ASM_GENERIC_BITOPS_GENERIC_NON_ATOMIC_H
 
 #include <linux/bits.h>
+#include <asm/barrier.h>
 
 #ifndef _LINUX_BITOPS_H
 #error only <linux/bitops.h> can be included directly
···
 	return 1UL & (addr[BIT_WORD(nr)] >> (nr & (BITS_PER_LONG-1)));
 }
 
+/**
+ * generic_test_bit_acquire - Determine, with acquire semantics, whether a bit is set
+ * @nr: bit number to test
+ * @addr: Address to start counting from
+ */
+static __always_inline bool
+generic_test_bit_acquire(unsigned long nr, const volatile unsigned long *addr)
+{
+	unsigned long *p = ((unsigned long *)addr) + BIT_WORD(nr);
+	return 1UL & (smp_load_acquire(p) >> (nr & (BITS_PER_LONG-1)));
+}
+
 /*
  * const_*() definitions provide good compile-time optimizations when
  * the passed arguments can be resolved at compile time.
···
 #define const___test_and_set_bit	generic___test_and_set_bit
 #define const___test_and_clear_bit	generic___test_and_clear_bit
 #define const___test_and_change_bit	generic___test_and_change_bit
+#define const_test_bit_acquire		generic_test_bit_acquire
 
 /**
  * const_test_bit - Determine whether a bit is set

include/asm-generic/bitops/instrumented-non-atomic.h

···
 	return arch_test_bit(nr, addr);
 }
 
+/**
+ * _test_bit_acquire - Determine, with acquire semantics, whether a bit is set
+ * @nr: bit number to test
+ * @addr: Address to start counting from
+ */
+static __always_inline bool
+_test_bit_acquire(unsigned long nr, const volatile unsigned long *addr)
+{
+	instrument_atomic_read(addr + BIT_WORD(nr), sizeof(long));
+	return arch_test_bit_acquire(nr, addr);
+}
+
 #endif /* _ASM_GENERIC_BITOPS_INSTRUMENTED_NON_ATOMIC_H */

include/asm-generic/bitops/non-atomic.h

···
 #define arch___test_and_change_bit generic___test_and_change_bit
 
 #define arch_test_bit generic_test_bit
+#define arch_test_bit_acquire generic_test_bit_acquire
 
 #include <asm-generic/bitops/non-instrumented-non-atomic.h>
 

include/asm-generic/bitops/non-instrumented-non-atomic.h

···
 #define ___test_and_change_bit	arch___test_and_change_bit
 
 #define _test_bit		arch_test_bit
+#define _test_bit_acquire	arch_test_bit_acquire
 
 #endif /* __ASM_GENERIC_BITOPS_NON_INSTRUMENTED_NON_ATOMIC_H */

include/linux/bitops.h

···
 #define __test_and_clear_bit(nr, addr)	bitop(___test_and_clear_bit, nr, addr)
 #define __test_and_change_bit(nr, addr)	bitop(___test_and_change_bit, nr, addr)
 #define test_bit(nr, addr)		bitop(_test_bit, nr, addr)
+#define test_bit_acquire(nr, addr)	bitop(_test_bit_acquire, nr, addr)
 
 /*
  * Include this here because some architectures need generic_ffs/fls in

include/linux/buffer_head.h

···
 	 * make it consistent with folio_test_uptodate
 	 * pairs with smp_mb__before_atomic in set_buffer_uptodate
 	 */
-	return (smp_load_acquire(&bh->b_state) & (1UL << BH_Uptodate)) != 0;
+	return test_bit_acquire(BH_Uptodate, &bh->b_state);
 }
 
 #define bh_offset(bh)		((unsigned long)(bh)->b_data & ~PAGE_MASK)

include/linux/wait_bit.h

···
 wait_on_bit(unsigned long *word, int bit, unsigned mode)
 {
 	might_sleep();
-	if (!test_bit(bit, word))
+	if (!test_bit_acquire(bit, word))
 		return 0;
 	return out_of_line_wait_on_bit(word, bit,
 				       bit_wait,
···
 wait_on_bit_io(unsigned long *word, int bit, unsigned mode)
 {
 	might_sleep();
-	if (!test_bit(bit, word))
+	if (!test_bit_acquire(bit, word))
 		return 0;
 	return out_of_line_wait_on_bit(word, bit,
 				       bit_wait_io,
···
 		    unsigned long timeout)
 {
 	might_sleep();
-	if (!test_bit(bit, word))
+	if (!test_bit_acquire(bit, word))
 		return 0;
 	return out_of_line_wait_on_bit_timeout(word, bit,
 					       bit_wait_timeout,
···
 		   unsigned mode)
 {
 	might_sleep();
-	if (!test_bit(bit, word))
+	if (!test_bit_acquire(bit, word))
 		return 0;
 	return out_of_line_wait_on_bit(word, bit, action, mode);
 }

kernel/sched/wait_bit.c

···
 		prepare_to_wait(wq_head, &wbq_entry->wq_entry, mode);
 		if (test_bit(wbq_entry->key.bit_nr, wbq_entry->key.flags))
 			ret = (*action)(&wbq_entry->key, mode);
-	} while (test_bit(wbq_entry->key.bit_nr, wbq_entry->key.flags) && !ret);
+	} while (test_bit_acquire(wbq_entry->key.bit_nr, wbq_entry->key.flags) && !ret);
 
 	finish_wait(wq_head, &wbq_entry->wq_entry);