================
@@ -32,9 +37,42 @@ bool ProcessRunLock::ReadTryLock() {
}
bool ProcessRunLock::ReadUnlock() {
+ Policy policy = PolicyStack::Get().Current();
+ if (policy.capabilities.holds_run_lock)
+ return true;
+
return ::pthread_rwlock_unlock(&m_rwlock) == 0;
}
+bool ProcessRunLock::ProcessRunLocker::TryLock(ProcessRunLock *lock) {
+ if (m_lock) {
+ if (m_lock == lock)
+ return true;
+ Unlock();
+ }
+ if (lock) {
+ if (lock->ReadTryLock()) {
+ m_lock = lock;
+ // Push a policy so re-entrant ReadTryLock calls from the same
+ // thread (e.g. provider Python calling back into SB API) skip
+ // the real lock and avoid deadlocking with a pending writer.
+ Policy policy = PolicyStack::Get().Current();
+ policy.capabilities.holds_run_lock = true;
+ PolicyStack::Get().Push(policy);
----------------
JDevlieghere wrote:
I wonder if we should find a way to make it harder to do the wrong thing. The
policies should be immutable (which they are by virtue of us returning a copy
when you look at the top of the stack). But if you can just modify it
willy-nilly, there's really no safety net.
I'm wondering if ideally, we should have a bunch of factories that do this for
you. For example, you should never set `holds_run_lock` at the same time as say
`can_run_frame_recognizers`.
I'm imagining something like:
```
PolicyStack::PushRunLockPolicy(true/false);
```
which does the clone + change. The `Push` would become private. In that worlds
you can get your hands on a Policy, a modify it if you want, but you can never
push it.
Of course it's possible to bypass that by pushing multiple policies after each
other using the factories, but at that point it's pretty obvious that what
you're doing is wrong. The idea is to make it "easy to do the right thing and
hard to do the wrong thing".
The downside of this approach is a bunch of factory helpers.
https://github.com/llvm/llvm-project/pull/195774
_______________________________________________
lldb-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
