atomically read page_bitmaps

page_bitmaps does not need to be volatile since it is going to be treated as atomic. In fact, as long as each variable is treated as atomic, we do not need volatile at all, as that means it is modified by outside means, aka hardware, etc.

The atomic macros already ensure optimizations that could cause incorrect values to be used do not happen.

Finally, the first loop is redundant as they are iterated over again in the loop below.

src/shims/atomic_sfb.h

@@ -32,7 +32,7 @@
 // Returns UINT_MAX if all the bits in p were already set.
 DISPATCH_ALWAYS_INLINE
 static inline unsigned int
-os_atomic_set_first_bit(volatile unsigned long *p, unsigned int max)
+os_atomic_set_first_bit(unsigned long *p, unsigned int max)
 {
 	unsigned long val, bit;
 	if (max > (sizeof(val) * 8)) {
@@ -82,7 +82,7 @@ os_atomic_set_first_bit(volatile unsigned long *p, unsigned int max)
 
 DISPATCH_ALWAYS_INLINE
 static inline unsigned int
-os_atomic_set_first_bit(volatile unsigned long *p, unsigned int max_index)
+os_atomic_set_first_bit(unsigned long *p, unsigned int max)
 {
 	unsigned int index;
 	unsigned long b, b_masked;
@@ -94,7 +94,7 @@ os_atomic_set_first_bit(volatile unsigned long *p, unsigned int max_index)
 			os_atomic_rmw_loop_give_up(return UINT_MAX);
 		}
 		index--;
-		if (unlikely(index > max_index)) {
+		if (unlikely(index > max)) {
 			os_atomic_rmw_loop_give_up(return UINT_MAX);
 		}
 		b_masked = b | (1UL << index);