Provide versions of mutex, condvar, and aio init that never fail.

If the underlying platform fails (FreeBSD is the only one I'm aware
of that does this!), we use a global lock or condition variable instead.
This means that our lock initializers never ever fail.

Probably we could eliminate most of this for Linux and Darwin, since
on those platforms, mutex and condvar initialization reasonably never
fails.  Initial benchmarks show little difference either way -- so we
can revisit (optimize) later.

This removes a lot of otherwise untested code in error cases and so forth,
improving coverage and resilience in the face of allocation failures.

Platforms other than POSIX should follow a similar pattern if they need
this.  (VxWorks, I'm thinking of you.)  Most sane platforms won't have
an issue here, since normally these initializations do not need to allocate
memory. (Reportedly, even FreeBSD has plans to "fix" this in libthr2.)

While here, some bugs were fixed in initialization & teardown.

The fallback code is properly tested with dedicated test cases.

1 files changed, 11 insertions, 6 deletions

diff --git a/src/platform/posix/posix_impl.h b/src/platform/posix/posix_impl.h
index 46ebbc1d..81fbd48b 100644
--- a/src/platform/posix/posix_impl.h
+++ b/src/platform/posix/posix_impl.h
@@ -54,9 +54,19 @@ extern int nni_plat_errno(int);
 // elsewhere.
 
 struct nni_plat_mtx {
-	int             init;
 	pthread_t       owner;
 	pthread_mutex_t mtx;
+	int             fallback;
+	int             flags;
+};
+
+struct nni_plat_cv {
+	pthread_cond_t cv;
+	nni_plat_mtx * mtx;
+	int            fallback;
+	int            flags;
+	int            gen;
+	int            wake;
 };
 
 struct nni_plat_thr {
@@ -65,11 +75,6 @@ struct nni_plat_thr {
 	void *arg;
 };
 
-struct nni_plat_cv {
-	pthread_cond_t cv;
-	nni_plat_mtx * mtx;
-};
-
 #endif
 
 extern int  nni_posix_pollq_sysinit(void);