aboutsummaryrefslogtreecommitdiff
path: root/core/sync
diff options
context:
space:
mode:
authorFeoramund <161657516+Feoramund@users.noreply.github.com>2025-06-03 09:07:38 -0400
committerFeoramund <161657516+Feoramund@users.noreply.github.com>2025-06-03 09:07:38 -0400
commit8cde9dce47712dccc50afe51b0f1b0b16270d54a (patch)
treef57bc1ea52fda6539eff6983fb8b453d1937ceeb /core/sync
parentfcf0d4efa152844a5b0f54b136d3c756e85bcc29 (diff)
Rewrite `Atomic_RW_Mutex`
This patch simplifies the implementation and fixes #5254. Previously, the mutex was set up as if there could be multiple writers, and there seemed to be some confusion as to which `Writer` bits to check, as not all were checked or set at the same time. This could also result in the mutex being left in a non-zero state even after unlocking all locks. All unneeded state has been removed and extra checks have been put in place.
Diffstat (limited to 'core/sync')
-rw-r--r--core/sync/primitives_atomic.odin60
1 files changed, 42 insertions, 18 deletions
diff --git a/core/sync/primitives_atomic.odin b/core/sync/primitives_atomic.odin
index a8a84b2bc..c694e5f96 100644
--- a/core/sync/primitives_atomic.odin
+++ b/core/sync/primitives_atomic.odin
@@ -95,20 +95,16 @@ atomic_mutex_guard :: proc "contextless" (m: ^Atomic_Mutex) -> bool {
Atomic_RW_Mutex_State :: distinct uint
-Atomic_RW_Mutex_State_Half_Width :: size_of(Atomic_RW_Mutex_State)*8/2
-Atomic_RW_Mutex_State_Is_Writing :: Atomic_RW_Mutex_State(1)
-Atomic_RW_Mutex_State_Writer :: Atomic_RW_Mutex_State(1)<<1
-Atomic_RW_Mutex_State_Reader :: Atomic_RW_Mutex_State(1)<<Atomic_RW_Mutex_State_Half_Width
+Atomic_RW_Mutex_State_Is_Writing :: Atomic_RW_Mutex_State(1) << (size_of(Atomic_RW_Mutex_State)*8-1)
+Atomic_RW_Mutex_State_Reader :: Atomic_RW_Mutex_State(1)
+Atomic_RW_Mutex_State_Reader_Mask :: ~Atomic_RW_Mutex_State_Is_Writing
-Atomic_RW_Mutex_State_Writer_Mask :: Atomic_RW_Mutex_State(1<<(Atomic_RW_Mutex_State_Half_Width-1) - 1) << 1
-Atomic_RW_Mutex_State_Reader_Mask :: Atomic_RW_Mutex_State(1<<(Atomic_RW_Mutex_State_Half_Width-1) - 1) << Atomic_RW_Mutex_State_Half_Width
-
-// An Atomic_RW_Mutex is a reader/writer mutual exclusion lock
-// The lock can be held by any arbitrary number of readers or a single writer
-// The zero value for an Atomic_RW_Mutex is an unlocked mutex
+// An Atomic_RW_Mutex is a reader/writer mutual exclusion lock.
+// The lock can be held by any arbitrary number of readers or a single writer.
+// The zero value for an Atomic_RW_Mutex is an unlocked mutex.
//
-// An Atomic_RW_Mutex must not be copied after first use
+// An Atomic_RW_Mutex must not be copied after first use.
Atomic_RW_Mutex :: struct #no_copy {
state: Atomic_RW_Mutex_State,
mutex: Atomic_Mutex,
@@ -118,11 +114,17 @@ Atomic_RW_Mutex :: struct #no_copy {
// atomic_rw_mutex_lock locks rw for writing (with a single writer)
// If the mutex is already locked for reading or writing, the mutex blocks until the mutex is available.
atomic_rw_mutex_lock :: proc "contextless" (rw: ^Atomic_RW_Mutex) {
- _ = atomic_add(&rw.state, Atomic_RW_Mutex_State_Writer)
atomic_mutex_lock(&rw.mutex)
- state := atomic_or(&rw.state, Atomic_RW_Mutex_State_Writer)
+ state := atomic_or(&rw.state, Atomic_RW_Mutex_State_Is_Writing)
if state & Atomic_RW_Mutex_State_Reader_Mask != 0 {
+ // There's at least one reader, so wait for the last one to post the semaphore.
+ //
+ // Because we hold the exclusive lock, no more readers can come in
+ // during this time, which will prevent any situations where the last
+ // reader is pre-empted around the count turning zero, which would
+ // result in the potential for another reader to run amok after the
+ // other posts.
atomic_sema_wait(&rw.sema)
}
}
@@ -138,10 +140,15 @@ atomic_rw_mutex_try_lock :: proc "contextless" (rw: ^Atomic_RW_Mutex) -> bool {
if atomic_mutex_try_lock(&rw.mutex) {
state := atomic_load(&rw.state)
if state & Atomic_RW_Mutex_State_Reader_Mask == 0 {
- _ = atomic_or(&rw.state, Atomic_RW_Mutex_State_Is_Writing)
- return true
+ // Compare-and-exchange for absolute certainty that no one has come in to read.
+ _, ok := atomic_compare_exchange_strong(&rw.state, state, state | Atomic_RW_Mutex_State_Is_Writing)
+ if ok {
+ return true
+ }
}
+ // A reader is active or came in while we have the lock, so we need to
+ // back out.
atomic_mutex_unlock(&rw.mutex)
}
return false
@@ -150,16 +157,22 @@ atomic_rw_mutex_try_lock :: proc "contextless" (rw: ^Atomic_RW_Mutex) -> bool {
// atomic_rw_mutex_shared_lock locks rw for reading (with arbitrary number of readers)
atomic_rw_mutex_shared_lock :: proc "contextless" (rw: ^Atomic_RW_Mutex) {
state := atomic_load(&rw.state)
- for state & (Atomic_RW_Mutex_State_Is_Writing|Atomic_RW_Mutex_State_Writer_Mask) == 0 {
+ for state & Atomic_RW_Mutex_State_Is_Writing == 0 {
ok: bool
state, ok = atomic_compare_exchange_weak(&rw.state, state, state + Atomic_RW_Mutex_State_Reader)
if ok {
+ // We succesfully took the shared reader lock without any writers intervening.
return
}
}
+ // A writer is active or came in while we were trying to get a shared
+ // reader lock, so now we must take the full lock in order to wait for the
+ // writer to give it up.
atomic_mutex_lock(&rw.mutex)
+ // At this point, we have the lock, so we can add to the reader count.
_ = atomic_add(&rw.state, Atomic_RW_Mutex_State_Reader)
+ // Then we give up the lock to let other readers (or writers) come through.
atomic_mutex_unlock(&rw.mutex)
}
@@ -169,6 +182,8 @@ atomic_rw_mutex_shared_unlock :: proc "contextless" (rw: ^Atomic_RW_Mutex) {
if (state & Atomic_RW_Mutex_State_Reader_Mask == Atomic_RW_Mutex_State_Reader) &&
(state & Atomic_RW_Mutex_State_Is_Writing != 0) {
+ // We were the last reader, so post to the writer with the lock who's
+ // waiting to continue.
atomic_sema_post(&rw.sema)
}
}
@@ -176,12 +191,21 @@ atomic_rw_mutex_shared_unlock :: proc "contextless" (rw: ^Atomic_RW_Mutex) {
// atomic_rw_mutex_try_shared_lock tries to lock rw for reading (with arbitrary number of readers)
atomic_rw_mutex_try_shared_lock :: proc "contextless" (rw: ^Atomic_RW_Mutex) -> bool {
state := atomic_load(&rw.state)
- if state & (Atomic_RW_Mutex_State_Is_Writing|Atomic_RW_Mutex_State_Writer_Mask) == 0 {
- _, ok := atomic_compare_exchange_strong(&rw.state, state, state + Atomic_RW_Mutex_State_Reader)
+ // NOTE: We need to check this in a for loop, because it is possible for
+ // another reader to change the underlying state which would cause our
+ // compare-and-exchange to fail.
+ for state & (Atomic_RW_Mutex_State_Is_Writing) == 0 {
+ ok: bool
+ state, ok = atomic_compare_exchange_weak(&rw.state, state, state + Atomic_RW_Mutex_State_Reader)
if ok {
return true
}
}
+ // A writer is active or came in during our lock attempt.
+
+ // We try to take the full lock, and if that succeeds (perhaps because the
+ // writer finished during the time since we failed our CAS), we increment
+ // the reader count and head on.
if atomic_mutex_try_lock(&rw.mutex) {
_ = atomic_add(&rw.state, Atomic_RW_Mutex_State_Reader)
atomic_mutex_unlock(&rw.mutex)