linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
* [PATCH 1/2] lockdep: allow instrumenting lockdep.c with KMSAN
@ 2022-11-28  9:45 Alexander Potapenko
  2022-11-28  9:45 ` [PATCH 2/2] kmsan: allow using __msan_instrument_asm_store() inside runtime Alexander Potapenko
  0 siblings, 1 reply; 2+ messages in thread
From: Alexander Potapenko @ 2022-11-28  9:45 UTC (permalink / raw)
  To: glider
  Cc: linux-kernel, akpm, peterz, mingo, will, elver, dvyukov,
	linux-mm, kasan-dev, Eric Biggers

Lockdep and KMSAN used to play badly together, causing deadlocks when
KMSAN instrumentation of lockdep.c called lockdep functions recursively.

Looks like this is no more the case, and a kernel can run (yet slower)
with both KMSAN and lockdep enabled.
This patch should fix false positives on wq_head->lock->dep_map, which
KMSAN used to consider uninitialized because of lockdep.c not being
instrumented.

Link: https://lore.kernel.org/lkml/Y3b9AAEKp2Vr3e6O@sol.localdomain/
Reported-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Alexander Potapenko <glider@google.com>
---
 kernel/locking/Makefile | 1 -
 1 file changed, 1 deletion(-)

diff --git a/kernel/locking/Makefile b/kernel/locking/Makefile
index ea925731fa40f..0db4093d17b8a 100644
--- a/kernel/locking/Makefile
+++ b/kernel/locking/Makefile
@@ -7,7 +7,6 @@ obj-y += mutex.o semaphore.o rwsem.o percpu-rwsem.o
 
 # Avoid recursion lockdep -> sanitizer -> ... -> lockdep.
 KCSAN_SANITIZE_lockdep.o := n
-KMSAN_SANITIZE_lockdep.o := n
 
 ifdef CONFIG_FUNCTION_TRACER
 CFLAGS_REMOVE_lockdep.o = $(CC_FLAGS_FTRACE)
-- 
2.38.1.584.g0f3c55d4c2-goog



^ permalink raw reply	[flat|nested] 2+ messages in thread

* [PATCH 2/2] kmsan: allow using __msan_instrument_asm_store() inside runtime
  2022-11-28  9:45 [PATCH 1/2] lockdep: allow instrumenting lockdep.c with KMSAN Alexander Potapenko
@ 2022-11-28  9:45 ` Alexander Potapenko
  0 siblings, 0 replies; 2+ messages in thread
From: Alexander Potapenko @ 2022-11-28  9:45 UTC (permalink / raw)
  To: glider
  Cc: linux-kernel, akpm, peterz, mingo, will, elver, dvyukov,
	linux-mm, kasan-dev

In certain cases (e.g. when handling a softirq)
__msan_instrument_asm_store(&var, sizeof(var)) may be called with
from within KMSAN runtime, but later the value of @var is used
with !kmsan_in_runtime(), leading to false positives.

Because kmsan_internal_unpoison_memory() doesn't take locks, it should
be fine to call it without kmsan_in_runtime() checks, which fixes the
mentioned false positives.

Signed-off-by: Alexander Potapenko <glider@google.com>
---
 mm/kmsan/instrumentation.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/mm/kmsan/instrumentation.c b/mm/kmsan/instrumentation.c
index 271f135f97a16..770fe02904f36 100644
--- a/mm/kmsan/instrumentation.c
+++ b/mm/kmsan/instrumentation.c
@@ -81,12 +81,16 @@ DECLARE_METADATA_PTR_GETTER(8);
  * Handle a memory store performed by inline assembly. KMSAN conservatively
  * attempts to unpoison the outputs of asm() directives to prevent false
  * positives caused by missed stores.
+ *
+ * __msan_instrument_asm_store() may be called for inline assembly code when
+ * entering or leaving IRQ. We omit the check for kmsan_in_runtime() to ensure
+ * the memory written to in these cases is also marked as initialized.
  */
 void __msan_instrument_asm_store(void *addr, uintptr_t size)
 {
 	unsigned long ua_flags;
 
-	if (!kmsan_enabled || kmsan_in_runtime())
+	if (!kmsan_enabled)
 		return;
 
 	ua_flags = user_access_save();
@@ -103,10 +107,8 @@ void __msan_instrument_asm_store(void *addr, uintptr_t size)
 		user_access_restore(ua_flags);
 		return;
 	}
-	kmsan_enter_runtime();
 	/* Unpoisoning the memory on best effort. */
 	kmsan_internal_unpoison_memory(addr, size, /*checked*/ false);
-	kmsan_leave_runtime();
 	user_access_restore(ua_flags);
 }
 EXPORT_SYMBOL(__msan_instrument_asm_store);
-- 
2.38.1.584.g0f3c55d4c2-goog



^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2022-11-28  9:51 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-28  9:45 [PATCH 1/2] lockdep: allow instrumenting lockdep.c with KMSAN Alexander Potapenko
2022-11-28  9:45 ` [PATCH 2/2] kmsan: allow using __msan_instrument_asm_store() inside runtime Alexander Potapenko

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox