kcsan: Add kcsan_set_access_mask() support

When setting up an access mask with kcsan_set_access_mask(), KCSAN will
only report races if concurrent changes to bits set in access_mask are
observed. Conveying access_mask via a separate call avoids introducing
overhead in the common-case fast-path.

Acked-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

3 files changed, 56 insertions, 5 deletions

diff --git a/kernel/kcsan/core.c b/kernel/kcsan/core.c
index 3f89801161d3..589b1e7f0f25 100644
--- a/kernel/kcsan/core.c
+++ b/kernel/kcsan/core.c
@@ -39,6 +39,7 @@ static DEFINE_PER_CPU(struct kcsan_ctx, kcsan_cpu_ctx) = {
 	.atomic_next		= 0,
 	.atomic_nest_count	= 0,
 	.in_flat_atomic		= false,
+	.access_mask		= 0,
 };
 
 /*
@@ -298,6 +299,15 @@ static noinline void kcsan_found_watchpoint(const volatile void *ptr,
 
 	if (!kcsan_is_enabled())
 		return;
+
+	/*
+	 * The access_mask check relies on value-change comparison. To avoid
+	 * reporting a race where e.g. the writer set up the watchpoint, but the
+	 * reader has access_mask!=0, we have to ignore the found watchpoint.
+	 */
+	if (get_ctx()->access_mask != 0)
+		return;
+
 	/*
 	 * Consume the watchpoint as soon as possible, to minimize the chances
 	 * of !consumed. Consuming the watchpoint must always be guarded by
@@ -341,6 +351,7 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type)
 		u32 _4;
 		u64 _8;
 	} expect_value;
+	unsigned long access_mask;
 	enum kcsan_value_change value_change = KCSAN_VALUE_CHANGE_MAYBE;
 	unsigned long ua_flags = user_access_save();
 	unsigned long irq_flags;
@@ -435,18 +446,27 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type)
 	 * Re-read value, and check if it is as expected; if not, we infer a
 	 * racy access.
 	 */
+	access_mask = get_ctx()->access_mask;
 	switch (size) {
 	case 1:
 		expect_value._1 ^= READ_ONCE(*(const u8 *)ptr);
+		if (access_mask)
+			expect_value._1 &= (u8)access_mask;
 		break;
 	case 2:
 		expect_value._2 ^= READ_ONCE(*(const u16 *)ptr);
+		if (access_mask)
+			expect_value._2 &= (u16)access_mask;
 		break;
 	case 4:
 		expect_value._4 ^= READ_ONCE(*(const u32 *)ptr);
+		if (access_mask)
+			expect_value._4 &= (u32)access_mask;
 		break;
 	case 8:
 		expect_value._8 ^= READ_ONCE(*(const u64 *)ptr);
+		if (access_mask)
+			expect_value._8 &= (u64)access_mask;
 		break;
 	default:
 		break; /* ignore; we do not diff the values */
@@ -460,11 +480,20 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type)
 	if (!remove_watchpoint(watchpoint)) {
 		/*
 		 * Depending on the access type, map a value_change of MAYBE to
-		 * TRUE (require reporting).
+		 * TRUE (always report) or FALSE (never report).
 		 */
-		if (value_change == KCSAN_VALUE_CHANGE_MAYBE && (size > 8 || is_assert)) {
-			/* Always assume a value-change. */
-			value_change = KCSAN_VALUE_CHANGE_TRUE;
+		if (value_change == KCSAN_VALUE_CHANGE_MAYBE) {
+			if (access_mask != 0) {
+				/*
+				 * For access with access_mask, we require a
+				 * value-change, as it is likely that races on
+				 * ~access_mask bits are expected.
+				 */
+				value_change = KCSAN_VALUE_CHANGE_FALSE;
+			} else if (size > 8 || is_assert) {
+				/* Always assume a value-change. */
+				value_change = KCSAN_VALUE_CHANGE_TRUE;
+			}
 		}
 
 		/*
@@ -622,6 +651,12 @@ void kcsan_atomic_next(int n)
 }
 EXPORT_SYMBOL(kcsan_atomic_next);
 
+void kcsan_set_access_mask(unsigned long mask)
+{
+	get_ctx()->access_mask = mask;
+}
+EXPORT_SYMBOL(kcsan_set_access_mask);
+
 void __kcsan_check_access(const volatile void *ptr, size_t size, int type)
 {
 	check_access(ptr, size, type);
diff --git a/kernel/kcsan/kcsan.h b/kernel/kcsan/kcsan.h
index 83a79b08b550..892de5120c1b 100644
--- a/kernel/kcsan/kcsan.h
+++ b/kernel/kcsan/kcsan.h
@@ -99,6 +99,11 @@ enum kcsan_value_change {
 	KCSAN_VALUE_CHANGE_MAYBE,
 
 	/*
+	 * Did not observe a value-change, and it is invalid to report the race.
+	 */
+	KCSAN_VALUE_CHANGE_FALSE,
+
+	/*
 	 * The value was observed to change, and the race should be reported.
 	 */
 	KCSAN_VALUE_CHANGE_TRUE,
diff --git a/kernel/kcsan/report.c b/kernel/kcsan/report.c
index d871476dc134..11c791b886f3 100644
--- a/kernel/kcsan/report.c
+++ b/kernel/kcsan/report.c
@@ -132,6 +132,9 @@ static bool rate_limit_report(unsigned long frame1, unsigned long frame2)
 static bool
 skip_report(enum kcsan_value_change value_change, unsigned long top_frame)
 {
+	/* Should never get here if value_change==FALSE. */
+	WARN_ON_ONCE(value_change == KCSAN_VALUE_CHANGE_FALSE);
+
 	/*
 	 * The first call to skip_report always has value_change==TRUE, since we
 	 * cannot know the value written of an instrumented access. For the 2nd
@@ -493,7 +496,15 @@ void kcsan_report(const volatile void *ptr, size_t size, int access_type,
 
 	kcsan_disable_current();
 	if (prepare_report(&flags, ptr, size, access_type, cpu_id, type)) {
-		if (print_report(ptr, size, access_type, value_change, cpu_id, type) && panic_on_warn)
+		/*
+		 * Never report if value_change is FALSE, only if we it is
+		 * either TRUE or MAYBE. In case of MAYBE, further filtering may
+		 * be done once we know the full stack trace in print_report().
+		 */
+		bool reported = value_change != KCSAN_VALUE_CHANGE_FALSE &&
+				print_report(ptr, size, access_type, value_change, cpu_id, type);
+
+		if (reported && panic_on_warn)
 			panic("panic_on_warn set ...\n");
 
 		release_report(&flags, type);