[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PATCH 3/8] qemu-thread-win32: cleanup, fix, document QemuEvent
From: |
Paolo Bonzini |
Subject: |
[PATCH 3/8] qemu-thread-win32: cleanup, fix, document QemuEvent |
Date: |
Fri, 3 Mar 2023 18:19:34 +0100 |
QemuEvent is currently broken on ARM due to missing memory barriers
after qatomic_*(). Apart from adding the memory barrier, a closer look
reveals some unpaired memory barriers that are not really needed and
complicated the functions unnecessarily, as well as some optimizations
that I couldn't quite prove to be correct.
Finally, the code is relying on a memory barrier in ResetEvent(); the
barrier _ought_ to be there but there is really no documentation about
it; it only affects the slow path, so make it explicit.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
util/qemu-thread-win32.c | 78 +++++++++++++++++++++++++++-------------
1 file changed, 53 insertions(+), 25 deletions(-)
diff --git a/util/qemu-thread-win32.c b/util/qemu-thread-win32.c
index 69db254ac7c1..eff664ae6b31 100644
--- a/util/qemu-thread-win32.c
+++ b/util/qemu-thread-win32.c
@@ -272,12 +272,20 @@ void qemu_event_destroy(QemuEvent *ev)
void qemu_event_set(QemuEvent *ev)
{
assert(ev->initialized);
- /* qemu_event_set has release semantics, but because it *loads*
+
+ /*
+ * Pairs with memory barrier in qemu_event_reset.
+ *
+ * qemu_event_set has release semantics, but because it *loads*
* ev->value we need a full memory barrier here.
*/
smp_mb();
if (qatomic_read(&ev->value) != EV_SET) {
- if (qatomic_xchg(&ev->value, EV_SET) == EV_BUSY) {
+ int old = qatomic_xchg(&ev->value, EV_SET);
+
+ /* Pairs with memory barrier after ResetEvent. */
+ smp_mb__after_rmw();
+ if (old == EV_BUSY) {
/* There were waiters, wake them up. */
SetEvent(ev->event);
}
@@ -286,17 +294,19 @@ void qemu_event_set(QemuEvent *ev)
void qemu_event_reset(QemuEvent *ev)
{
- unsigned value;
-
assert(ev->initialized);
- value = qatomic_read(&ev->value);
- smp_mb_acquire();
- if (value == EV_SET) {
- /* If there was a concurrent reset (or even reset+wait),
- * do nothing. Otherwise change EV_SET->EV_FREE.
- */
- qatomic_or(&ev->value, EV_FREE);
- }
+
+ /*
+ * If there was a concurrent reset (or even reset+wait),
+ * do nothing. Otherwise change EV_SET->EV_FREE.
+ */
+ qatomic_or(&ev->value, EV_FREE);
+
+ /*
+ * Order reset before checking the condition in the caller.
+ * Pairs with the first memory barrier in qemu_event_set().
+ */
+ smp_mb__after_rmw();
}
void qemu_event_wait(QemuEvent *ev)
@@ -304,29 +314,47 @@ void qemu_event_wait(QemuEvent *ev)
unsigned value;
assert(ev->initialized);
+
+ /*
+ * This read does not have any particular ordering requirements;
+ * if it moves earlier, we might miss qemu_event_set() and go down the
+ * slow path unnecessarily, but ultimately the memory barrier below,
+ * plus the internal synchronization of the kernel event, will ensure
+ * the check is done correctly.
+ */
value = qatomic_read(&ev->value);
- smp_mb_acquire();
if (value != EV_SET) {
if (value == EV_FREE) {
- /* qemu_event_set is not yet going to call SetEvent, but we are
- * going to do another check for EV_SET below when setting EV_BUSY.
- * At that point it is safe to call WaitForSingleObject.
+ /*
+ * Here the underlying kernel event is reset, but qemu_event_set is
+ * not yet going to call SetEvent. However, there will be another
+ * check for EV_SET below when setting EV_BUSY. At that point it
+ * is safe to call WaitForSingleObject.
*/
ResetEvent(ev->event);
- /* Tell qemu_event_set that there are waiters. No need to retry
- * because there cannot be a concurrent busy->free transition.
- * After the CAS, the event will be either set or busy.
+ /*
+ * It is not clear whether ResetEvent provides this barrier; kernel
+ * APIs (KeResetEvent/KeClearEvent) do not. Better safe than
sorry!
+ */
+ smp_mb();
+
+ /*
+ * Leave the event reset and tell qemu_event_set that there are
+ * waiters. No need to retry, because there cannot be a concurrent
+ * busy->free transition. After the CAS, the event will be either
+ * set or busy.
*/
if (qatomic_cmpxchg(&ev->value, EV_FREE, EV_BUSY) == EV_SET) {
- value = EV_SET;
- } else {
- value = EV_BUSY;
+ return;
}
}
- if (value == EV_BUSY) {
- WaitForSingleObject(ev->event, INFINITE);
- }
+
+ /*
+ * ev->value is now EV_BUSY. Since we didn't observe EV_SET,
+ * qemu_event_set() must observe EV_BUSY and call SetEvent().
+ */
+ WaitForSingleObject(ev->event, INFINITE);
}
}
--
2.39.1
- [PATCH 0/8] Fix missing memory barriers on ARM, Paolo Bonzini, 2023/03/03
- [PATCH 1/8] qatomic: add smp_mb__before/after_rmw(), Paolo Bonzini, 2023/03/03
- [PATCH 2/8] qemu-thread-posix: cleanup, fix, document QemuEvent, Paolo Bonzini, 2023/03/03
- [PATCH 3/8] qemu-thread-win32: cleanup, fix, document QemuEvent,
Paolo Bonzini <=
- Re: [PATCH 3/8] qemu-thread-win32: cleanup, fix, document QemuEvent, Richard Henderson, 2023/03/05
- Re: [PATCH 3/8] qemu-thread-win32: cleanup, fix, document QemuEvent, David Hildenbrand, 2023/03/06
- Re: [PATCH 3/8] qemu-thread-win32: cleanup, fix, document QemuEvent, Paolo Bonzini, 2023/03/06
- Re: [PATCH 3/8] qemu-thread-win32: cleanup, fix, document QemuEvent, David Hildenbrand, 2023/03/06
- Re: [PATCH 3/8] qemu-thread-win32: cleanup, fix, document QemuEvent, Paolo Bonzini, 2023/03/06
[PATCH 4/8] edu: add smp_mb__after_rmw(), Paolo Bonzini, 2023/03/03