aboutsummaryrefslogtreecommitdiff
path: root/src
diff options
context:
space:
mode:
authorJens Gustedt <Jens.Gustedt@inria.fr>2018-01-03 14:17:12 +0100
committerRich Felker <dalias@aerifal.cx>2018-01-09 13:10:12 -0500
commit47d0bcd4762f223364e5b58d5a381aaa0cbd7c38 (patch)
treee73dfc04306ff6f03f1e2712aa3c14321866df8a /src
parentb583c5d3b4cc2c54c68eef5eb7855ecfacee8bfc (diff)
downloadmusl-47d0bcd4762f223364e5b58d5a381aaa0cbd7c38.zip
musl-47d0bcd4762f223364e5b58d5a381aaa0cbd7c38.tar.gz
musl-47d0bcd4762f223364e5b58d5a381aaa0cbd7c38.tar.bz2
new lock algorithm with state and congestion count in one atomic int
A variant of this new lock algorithm has been presented at SAC'16, see https://hal.inria.fr/hal-01304108. A full version of that paper is available at https://hal.inria.fr/hal-01236734. The main motivation of this is to improve on the safety of the basic lock implementation in musl. This is achieved by squeezing a lock flag and a congestion count (= threads inside the critical section) into a single int. Thereby an unlock operation does exactly one memory transfer (a_fetch_add) and never touches the value again, but still detects if a waiter has to be woken up. This is a fix of a use-after-free bug in pthread_detach that had temporarily been patched. Therefore this patch also reverts c1e27367a9b26b9baac0f37a12349fc36567c8b6 This is also the only place where internal knowledge of the lock algorithm is used. The main price for the improved safety is a little bit larger code. Under high congestion, the scheduling behavior will be different compared to the previous algorithm. In that case, a successful put-to-sleep may appear out of order compared to the arrival in the critical section.
Diffstat (limited to 'src')
-rw-r--r--src/internal/pthread_impl.h6
-rw-r--r--src/thread/__lock.c55
-rw-r--r--src/thread/pthread_detach.c5
3 files changed, 58 insertions, 8 deletions
diff --git a/src/internal/pthread_impl.h b/src/internal/pthread_impl.h
index 56e1934..602d6f5 100644
--- a/src/internal/pthread_impl.h
+++ b/src/internal/pthread_impl.h
@@ -136,6 +136,12 @@ static inline void __wake(volatile void *addr, int cnt, int priv)
__syscall(SYS_futex, addr, FUTEX_WAKE|priv, cnt) != -ENOSYS ||
__syscall(SYS_futex, addr, FUTEX_WAKE, cnt);
}
+static inline void __futexwait(volatile void *addr, int val, int priv)
+{
+ if (priv) priv = FUTEX_PRIVATE;
+ __syscall(SYS_futex, addr, FUTEX_WAIT|priv, val) != -ENOSYS ||
+ __syscall(SYS_futex, addr, FUTEX_WAIT, val);
+}
void __acquire_ptc(void);
void __release_ptc(void);
diff --git a/src/thread/__lock.c b/src/thread/__lock.c
index 0874c04..45557c8 100644
--- a/src/thread/__lock.c
+++ b/src/thread/__lock.c
@@ -1,15 +1,60 @@
#include "pthread_impl.h"
+/* This lock primitive combines a flag (in the sign bit) and a
+ * congestion count (= threads inside the critical section, CS) in a
+ * single int that is accessed through atomic operations. The states
+ * of the int for value x are:
+ *
+ * x == 0: unlocked and no thread inside the critical section
+ *
+ * x < 0: locked with a congestion of x-INT_MIN, including the thread
+ * that holds the lock
+ *
+ * x > 0: unlocked with a congestion of x
+ *
+ * or in an equivalent formulation x is the congestion count or'ed
+ * with INT_MIN as a lock flag.
+ */
+
void __lock(volatile int *l)
{
- if (libc.threads_minus_1)
- while (a_swap(l, 1)) __wait(l, l+1, 1, 1);
+ if (!libc.threads_minus_1) return;
+ /* fast path: INT_MIN for the lock, +1 for the congestion */
+ int current = a_cas(l, 0, INT_MIN + 1);
+ if (!current) return;
+ /* A first spin loop, for medium congestion. */
+ for (unsigned i = 0; i < 10; ++i) {
+ if (current < 0) current -= INT_MIN + 1;
+ // assertion: current >= 0
+ int val = a_cas(l, current, INT_MIN + (current + 1));
+ if (val == current) return;
+ current = val;
+ }
+ // Spinning failed, so mark ourselves as being inside the CS.
+ current = a_fetch_add(l, 1) + 1;
+ /* The main lock acquisition loop for heavy congestion. The only
+ * change to the value performed inside that loop is a successful
+ * lock via the CAS that acquires the lock. */
+ for (;;) {
+ /* We can only go into wait, if we know that somebody holds the
+ * lock and will eventually wake us up, again. */
+ if (current < 0) {
+ __futexwait(l, current, 1);
+ current -= INT_MIN + 1;
+ }
+ /* assertion: current > 0, the count includes us already. */
+ int val = a_cas(l, current, INT_MIN + current);
+ if (val == current) return;
+ current = val;
+ }
}
void __unlock(volatile int *l)
{
- if (l[0]) {
- a_store(l, 0);
- if (l[1]) __wake(l, 1, 1);
+ /* Check l[0] to see if we are multi-threaded. */
+ if (l[0] < 0) {
+ if (a_fetch_add(l, -(INT_MIN + 1)) != (INT_MIN + 1)) {
+ __wake(l, 1, 1);
+ }
}
}
diff --git a/src/thread/pthread_detach.c b/src/thread/pthread_detach.c
index 1348260..d9c90d1 100644
--- a/src/thread/pthread_detach.c
+++ b/src/thread/pthread_detach.c
@@ -6,11 +6,10 @@ int __pthread_join(pthread_t, void **);
static int __pthread_detach(pthread_t t)
{
/* Cannot detach a thread that's already exiting */
- if (a_swap(t->exitlock, 1))
+ if (a_cas(t->exitlock, 0, INT_MIN + 1))
return __pthread_join(t, 0);
t->detached = 2;
- a_store(t->exitlock, 0);
- __wake(t->exitlock, 1, 1);
+ __unlock(t->exitlock);
return 0;
}