diff options
author | Thomas Gleixner <tglx@linutronix.de> | 2024-06-10 14:46:36 +0200 |
---|---|---|
committer | Peter Zijlstra <peterz@infradead.org> | 2024-06-11 11:25:23 +0200 |
commit | 83ab38ef0a0b2407d43af9575bb32333fdd74fb2 (patch) | |
tree | 6f704708e2a7335a5f45ac72d542caba237ae338 /kernel/jump_label.c | |
parent | bb9bb45f746b0f9457de9c3fc4da143a6351bdc9 (diff) |
jump_label: Fix concurrency issues in static_key_slow_dec()
The commit which tried to fix the concurrency issues of concurrent
static_key_slow_inc() failed to fix the equivalent issues
vs. static_key_slow_dec():
CPU0 CPU1
static_key_slow_dec()
static_key_slow_try_dec()
key->enabled == 1
val = atomic_fetch_add_unless(&key->enabled, -1, 1);
if (val == 1)
return false;
jump_label_lock();
if (atomic_dec_and_test(&key->enabled)) {
--> key->enabled == 0
__jump_label_update()
static_key_slow_dec()
static_key_slow_try_dec()
key->enabled == 0
val = atomic_fetch_add_unless(&key->enabled, -1, 1);
--> key->enabled == -1 <- FAIL
There is another bug in that code, when there is a concurrent
static_key_slow_inc() which enables the key as that sets key->enabled to -1
so on the other CPU
val = atomic_fetch_add_unless(&key->enabled, -1, 1);
will succeed and decrement to -2, which is invalid.
Cure all of this by replacing the atomic_fetch_add_unless() with a
atomic_try_cmpxchg() loop similar to static_key_fast_inc_not_disabled().
[peterz: add WARN_ON_ONCE for the -1 race]
Fixes: 4c5ea0a9cd02 ("locking/static_key: Fix concurrent static_key_slow_inc()")
Reported-by: Yue Sun <samsun1006219@gmail.com>
Reported-by: Xingwei Lee <xrivendell7@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20240610124406.422897838@linutronix.de
Diffstat (limited to 'kernel/jump_label.c')
-rw-r--r-- | kernel/jump_label.c | 45 |
1 files changed, 29 insertions, 16 deletions
diff --git a/kernel/jump_label.c b/kernel/jump_label.c index 3218fa5688b9..1f05a19918f4 100644 --- a/kernel/jump_label.c +++ b/kernel/jump_label.c @@ -131,7 +131,7 @@ bool static_key_fast_inc_not_disabled(struct static_key *key) STATIC_KEY_CHECK_USE(key); /* * Negative key->enabled has a special meaning: it sends - * static_key_slow_inc() down the slow path, and it is non-zero + * static_key_slow_inc/dec() down the slow path, and it is non-zero * so it counts as "enabled" in jump_label_update(). Note that * atomic_inc_unless_negative() checks >= 0, so roll our own. */ @@ -150,7 +150,7 @@ bool static_key_slow_inc_cpuslocked(struct static_key *key) lockdep_assert_cpus_held(); /* - * Careful if we get concurrent static_key_slow_inc() calls; + * Careful if we get concurrent static_key_slow_inc/dec() calls; * later calls must wait for the first one to _finish_ the * jump_label_update() process. At the same time, however, * the jump_label_update() call below wants to see @@ -247,20 +247,32 @@ EXPORT_SYMBOL_GPL(static_key_disable); static bool static_key_slow_try_dec(struct static_key *key) { - int val; - - val = atomic_fetch_add_unless(&key->enabled, -1, 1); - if (val == 1) - return false; + int v; /* - * The negative count check is valid even when a negative - * key->enabled is in use by static_key_slow_inc(); a - * __static_key_slow_dec() before the first static_key_slow_inc() - * returns is unbalanced, because all other static_key_slow_inc() - * instances block while the update is in progress. + * Go into the slow path if key::enabled is less than or equal than + * one. One is valid to shut down the key, anything less than one + * is an imbalance, which is handled at the call site. + * + * That includes the special case of '-1' which is set in + * static_key_slow_inc_cpuslocked(), but that's harmless as it is + * fully serialized in the slow path below. By the time this task + * acquires the jump label lock the value is back to one and the + * retry under the lock must succeed. */ - WARN(val < 0, "jump label: negative count!\n"); + v = atomic_read(&key->enabled); + do { + /* + * Warn about the '-1' case though; since that means a + * decrement is concurrent with a first (0->1) increment. IOW + * people are trying to disable something that wasn't yet fully + * enabled. This suggests an ordering problem on the user side. + */ + WARN_ON_ONCE(v < 0); + if (v <= 1) + return false; + } while (!likely(atomic_try_cmpxchg(&key->enabled, &v, v - 1))); + return true; } @@ -271,10 +283,11 @@ static void __static_key_slow_dec_cpuslocked(struct static_key *key) if (static_key_slow_try_dec(key)) return; - jump_label_lock(); - if (atomic_dec_and_test(&key->enabled)) + guard(mutex)(&jump_label_mutex); + if (atomic_cmpxchg(&key->enabled, 1, 0)) jump_label_update(key); - jump_label_unlock(); + else + WARN_ON_ONCE(!static_key_slow_try_dec(key)); } static void __static_key_slow_dec(struct static_key *key) |