libbpf, xsk: Add libbpf_smp_store_release libbpf_smp_load_acquire

Now that the AF_XDP rings have load-acquire/store-release semantics,
move libbpf to that as well.

The library-internal libbpf_smp_{load_acquire,store_release} are only
valid for 32-bit words on ARM64.

Also, remove the barriers that are no longer in use.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/20210305094113.413544-3-bjorn.topel@gmail.com

tools/lib/bpf/libbpf_util.h

@@ -5,6 +5,7 @@
 #define __LIBBPF_LIBBPF_UTIL_H
 
 #include <stdbool.h>
+#include <linux/compiler.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -15,29 +16,56 @@ extern "C" {
  * application that uses libbpf.
  */
 #if defined(__i386__) || defined(__x86_64__)
-# define libbpf_smp_rmb() asm volatile("" : : : "memory")
-# define libbpf_smp_wmb() asm volatile("" : : : "memory")
-# define libbpf_smp_mb() \
-	asm volatile("lock; addl $0,-4(%%rsp)" : : : "memory", "cc")
-/* Hinders stores to be observed before older loads. */
-# define libbpf_smp_rwmb() asm volatile("" : : : "memory")
+# define libbpf_smp_store_release(p, v)					\
+	do {								\
+		asm volatile("" : : : "memory");			\
+		WRITE_ONCE(*p, v);					\
+	} while (0)
+# define libbpf_smp_load_acquire(p)					\
+	({								\
+		typeof(*p) ___p1 = READ_ONCE(*p);			\
+		asm volatile("" : : : "memory");			\
+		___p1;							\
+	})
 #elif defined(__aarch64__)
-# define libbpf_smp_rmb() asm volatile("dmb ishld" : : : "memory")
-# define libbpf_smp_wmb() asm volatile("dmb ishst" : : : "memory")
-# define libbpf_smp_mb() asm volatile("dmb ish" : : : "memory")
-# define libbpf_smp_rwmb() libbpf_smp_mb()
-#elif defined(__arm__)
-/* These are only valid for armv7 and above */
-# define libbpf_smp_rmb() asm volatile("dmb ish" : : : "memory")
-# define libbpf_smp_wmb() asm volatile("dmb ishst" : : : "memory")
-# define libbpf_smp_mb() asm volatile("dmb ish" : : : "memory")
-# define libbpf_smp_rwmb() libbpf_smp_mb()
-#else
-/* Architecture missing native barrier functions. */
-# define libbpf_smp_rmb() __sync_synchronize()
-# define libbpf_smp_wmb() __sync_synchronize()
-# define libbpf_smp_mb() __sync_synchronize()
-# define libbpf_smp_rwmb() __sync_synchronize()
+# define libbpf_smp_store_release(p, v)					\
+		asm volatile ("stlr %w1, %0" : "=Q" (*p) : "r" (v) : "memory")
+# define libbpf_smp_load_acquire(p)					\
+	({								\
+		typeof(*p) ___p1;					\
+		asm volatile ("ldar %w0, %1"				\
+			      : "=r" (___p1) : "Q" (*p) : "memory");	\
+		__p1;							\
+	})
+#elif defined(__riscv)
+# define libbpf_smp_store_release(p, v)					\
+	do {								\
+		asm volatile ("fence rw,w" : : : "memory");		\
+		WRITE_ONCE(*p, v);					\
+	} while (0)
+# define libbpf_smp_load_acquire(p)					\
+	({								\
+		typeof(*p) ___p1 = READ_ONCE(*p);			\
+		asm volatile ("fence r,rw" : : : "memory");		\
+		___p1;							\
+	})
+#endif
+
+#ifndef libbpf_smp_store_release
+#define libbpf_smp_store_release(p, v)					\
+	do {								\
+		__sync_synchronize();					\
+		WRITE_ONCE(*p, v);					\
+	} while (0)
+#endif
+
+#ifndef libbpf_smp_load_acquire
+#define libbpf_smp_load_acquire(p)					\
+	({								\
+		typeof(*p) ___p1 = READ_ONCE(*p);			\
+		__sync_synchronize();					\
+		___p1;							\
+	})
 #endif
 
 #ifdef __cplusplus

tools/lib/bpf/xsk.h

@@ -96,7 +96,8 @@ static inline __u32 xsk_prod_nb_free(struct xsk_ring_prod *r, __u32 nb)
 	 * this function. Without this optimization it whould have been
 	 * free_entries = r->cached_prod - r->cached_cons + r->size.
 	 */
-	r->cached_cons = *r->consumer + r->size;
+	r->cached_cons = libbpf_smp_load_acquire(r->consumer);
+	r->cached_cons += r->size;
 
 	return r->cached_cons - r->cached_prod;
 }
@@ -106,7 +107,7 @@ static inline __u32 xsk_cons_nb_avail(struct xsk_ring_cons *r, __u32 nb)
 	__u32 entries = r->cached_prod - r->cached_cons;
 
 	if (entries == 0) {
-		r->cached_prod = *r->producer;
+		r->cached_prod = libbpf_smp_load_acquire(r->producer);
 		entries = r->cached_prod - r->cached_cons;
 	}
 
@@ -129,9 +130,7 @@ static inline void xsk_ring_prod__submit(struct xsk_ring_prod *prod, __u32 nb)
 	/* Make sure everything has been written to the ring before indicating
 	 * this to the kernel by writing the producer pointer.
 	 */
-	libbpf_smp_wmb();
-
-	*prod->producer += nb;
+	libbpf_smp_store_release(prod->producer, *prod->producer + nb);
 }
 
 static inline __u32 xsk_ring_cons__peek(struct xsk_ring_cons *cons, __u32 nb, __u32 *idx)
@@ -139,11 +138,6 @@ static inline __u32 xsk_ring_cons__peek(struct xsk_ring_cons *cons, __u32 nb, __
 	__u32 entries = xsk_cons_nb_avail(cons, nb);
 
 	if (entries > 0) {
-		/* Make sure we do not speculatively read the data before
-		 * we have received the packet buffers from the ring.
-		 */
-		libbpf_smp_rmb();
-
 		*idx = cons->cached_cons;
 		cons->cached_cons += entries;
 	}
@@ -161,9 +155,8 @@ static inline void xsk_ring_cons__release(struct xsk_ring_cons *cons, __u32 nb)
 	/* Make sure data has been read before indicating we are done
 	 * with the entries by updating the consumer pointer.
 	 */
-	libbpf_smp_rwmb();
+	libbpf_smp_store_release(cons->consumer, *cons->consumer + nb);
 
-	*cons->consumer += nb;
 }
 
 static inline void *xsk_umem__get_data(void *umem_area, __u64 addr)