From fb32f0284f86d5a613095a2965e6f99fb45c01b3 Mon Sep 17 00:00:00 2001
From: Andrea Terzolo <andreaterzolo3@gmail.com>
Date: Fri, 13 Sep 2024 17:34:52 +0200
Subject: [PATCH] fix(libscap): use the correct memory barrier for ARM64

Signed-off-by: Andrea Terzolo <andreaterzolo3@gmail.com>
---
 .../libpman/src/ringbuffer_definitions.h      | 21 +------
 userspace/libscap/engine/bpf/scap_bpf.h       |  9 +--
 userspace/libscap/linux/barrier.h             | 59 ++++++++++++++++++-
 3 files changed, 62 insertions(+), 27 deletions(-)

diff --git a/userspace/libpman/src/ringbuffer_definitions.h b/userspace/libpman/src/ringbuffer_definitions.h
index 8c134b7a79..1f224926bc 100644
--- a/userspace/libpman/src/ringbuffer_definitions.h
+++ b/userspace/libpman/src/ringbuffer_definitions.h
@@ -20,6 +20,7 @@ limitations under the License.
 
 #include "state.h"
 #include <linux/bpf.h>
+#include <libscap/scap_barrier.h>
 
 /* Taken from libbpf: /src/ringbuf.c */
 struct ring {
@@ -50,23 +51,3 @@ static inline int roundup_len(uint32_t len) {
 	/* round up to 8 byte alignment */
 	return (len + 7) / 8 * 8;
 }
-
-/* Taken from libbpf: `include/linux/compiler.h` */
-
-#define READ_ONCE(x) (*(volatile typeof(x) *)&x)
-#define WRITE_ONCE(x, v) (*(volatile typeof(x) *)&x) = (v)
-
-#define barrier() asm volatile("" ::: "memory")
-
-#define smp_store_release(p, v) \
-	do {                        \
-		barrier();              \
-		WRITE_ONCE(*p, v);      \
-	} while(0)
-
-#define smp_load_acquire(p)              \
-	({                                   \
-		typeof(*p) ___p = READ_ONCE(*p); \
-		barrier();                       \
-		___p;                            \
-	})
diff --git a/userspace/libscap/engine/bpf/scap_bpf.h b/userspace/libscap/engine/bpf/scap_bpf.h
index 6c454646fc..a61dd3d5c2 100644
--- a/userspace/libscap/engine/bpf/scap_bpf.h
+++ b/userspace/libscap/engine/bpf/scap_bpf.h
@@ -20,6 +20,7 @@ limitations under the License.
 
 #include <libscap/compat/bpf.h>
 #include <libscap/compat/perf_event.h>
+#include <libscap/scap_barrier.h>
 
 struct perf_event_sample {
 	struct perf_event_header header;
@@ -49,9 +50,7 @@ static inline void scap_bpf_get_buf_pointers(scap_device *dev,
 	*phead = header->data_head;
 	*ptail = header->data_tail;
 
-	// clang-format off
-	asm volatile("" ::: "memory");
-	// clang-format on
+	mem_barrier();
 
 	uint64_t cons = *ptail % header->data_size;  // consumer position
 	uint64_t prod = *phead % header->data_size;  // producer position
@@ -154,9 +153,7 @@ static inline void scap_bpf_advance_tail(struct scap_device *dev) {
 
 	header = (struct perf_event_mmap_page *)dev->m_buffer;
 
-	// clang-format off
-	asm volatile("" ::: "memory");
-	// clang-format on
+	mem_barrier();
 
 	ASSERT(dev->m_lastreadsize > 0);
 	/* `header->data_tail` is the consumer position. */
diff --git a/userspace/libscap/linux/barrier.h b/userspace/libscap/linux/barrier.h
index 5cfc22f600..4bfaa856d2 100644
--- a/userspace/libscap/linux/barrier.h
+++ b/userspace/libscap/linux/barrier.h
@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: Apache-2.0
 /*
-Copyright (C) 2023 The Falco Authors.
+Copyright (C) 2024 The Falco Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -17,4 +17,61 @@ limitations under the License.
 */
 #pragma once
 
+// This is taken from kernel headers `/include/linux/compiler.h`
+// Used by libpman and scap_bpf engine
+
+#define READ_ONCE(x) (*(volatile typeof(x) *)&x)
+#define WRITE_ONCE(x, v) (*(volatile typeof(x) *)&x) = (v)
+
+#define barrier() asm volatile("" ::: "memory")
+
+#if defined(__x86_64__)
+
+#define smp_mb() asm volatile("lock; addl $0,-132(%%rsp)" ::: "memory", "cc")
+
+#define smp_store_release(p, v) \
+	do {                        \
+		barrier();              \
+		WRITE_ONCE(*p, v);      \
+	} while(0)
+
+#define smp_load_acquire(p)              \
+	({                                   \
+		typeof(*p) ___p = READ_ONCE(*p); \
+		barrier();                       \
+		___p;                            \
+	})
+
+#elif defined(__aarch64__)
+
+#define smp_mb() asm volatile("dmb ish" ::: "memory")
+
+#endif
+
+#ifndef smp_mb
+#define smp_mb() __sync_synchronize()
+#endif
+
+#ifndef smp_store_release
+#define smp_store_release(p, v) \
+	do {                        \
+		smp_mb();               \
+		WRITE_ONCE(*p, v);      \
+	} while(0)
+#endif
+
+#ifndef smp_load_acquire
+#define smp_load_acquire(p)              \
+	({                                   \
+		typeof(*p) ___p = READ_ONCE(*p); \
+		smp_mb();                        \
+		___p;                            \
+	})
+#endif
+
+// This is defined by us
+#if defined(__x86_64__)
+#define mem_barrier() asm volatile("" ::: "memory")
+#else
 #define mem_barrier() __sync_synchronize()
+#endif