add proper fuxed-based locking for stdio

previously, stdio used spinlocks, which would be unacceptable if we
ever add support for thread priorities, and which yielded
pathologically bad performance if an application attempted to use
flockfile on a key file as a major/primary locking mechanism.

i had held off on making this change for fear that it would hurt
performance in the non-threaded case, but actually support for
recursive locking had already inflicted that cost. by having the
internal locking functions store a flag indicating whether they need
to perform unlocking, rather than using the actual recursive lock
counter, i was able to combine the conditionals at unlock time,
eliminating any additional cost, and also avoid a nasty corner case
where a huge number of calls to ftrylockfile could cause deadlock
later at the point of internal locking.

this commit also fixes some issues with usage of pthread_self
conflicting with __attribute__((const)) which resulted in crashes with
some compiler versions/optimizations, mainly in flockfile prior to
pthread_create.

src/stdio/__fdopen.c

@@ -39,11 +39,13 @@ FILE *__fdopen(int fd, const char *mode)
 	f->seek = __stdio_seek;
 	f->close = __stdio_close;
 
+	if (!libc.threaded) f->lock = -1;
+
 	/* Add new FILE to open file list */
 	OFLLOCK();
-	f->next = ofl_head;
-	if (ofl_head) ofl_head->prev = f;
-	ofl_head = f;
+	f->next = libc.ofl_head;
+	if (libc.ofl_head) libc.ofl_head->prev = f;
+	libc.ofl_head = f;
 	OFLUNLOCK();
 
 	return f;

src/stdio/__lockfile.c

@@ -1,20 +1,18 @@
 #include "stdio_impl.h"
 #include "pthread_impl.h"
 
-void __lockfile(FILE *f)
+int __lockfile(FILE *f)
 {
-	int spins=10000;
-	int tid;
-
-	if (f->lock < 0) return;
-	tid = __pthread_self()->tid;
-	if (f->lock == tid) {
-		while (f->lockcount == INT_MAX);
-		f->lockcount++;
-		return;
-	}
-	while (a_cas(&f->lock, 0, tid))
-		if (spins) spins--, a_spin();
-		else __syscall(SYS_sched_yield);
-	f->lockcount = 1;
+	int owner, tid = __pthread_self()->tid;
+	if (f->lock == tid)
+		return 0;
+	while ((owner = a_cas(&f->lock, 0, tid)))
+		__wait(&f->lock, &f->waiters, owner, 1);
+	return f->lockcount = 1;
+}
+
+void __unlockfile(FILE *f)
+{
+	a_store(&f->lock, 0);
+	if (f->waiters) __wake(&f->lock, 1, 1);
 }

src/stdio/fclose.c

@@ -9,7 +9,7 @@ int fclose(FILE *f)
 		OFLLOCK();
 		if (f->prev) f->prev->next = f->next;
 		if (f->next) f->next->prev = f->prev;
-		if (ofl_head == f) ofl_head = f->next;
+		if (libc.ofl_head == f) libc.ofl_head = f->next;
 		OFLUNLOCK();
 	}
 

src/stdio/fflush.c

@@ -22,8 +22,8 @@ static int __fflush_unlocked(FILE *f)
 }
 
 /* stdout.c will override this if linked */
-static FILE *const __dummy = 0;
-weak_alias(__dummy, __stdout_to_flush);
+static FILE *const dummy = 0;
+weak_alias(dummy, __stdout_used);
 
 int fflush(FILE *f)
 {
@@ -37,13 +37,13 @@ int fflush(FILE *f)
 		return r;
 	}
 
-	r = __stdout_to_flush ? fflush(__stdout_to_flush) : 0;
+	r = __stdout_used ? fflush(__stdout_used) : 0;
 
 	OFLLOCK();
-	for (f=ofl_head; f; f=next) {
+	for (f=libc.ofl_head; f; f=next) {
 		FLOCK(f);
 		//OFLUNLOCK();
-		r |= __fflush_unlocked(f);
+		if (f->wpos > f->wbase) r |= __fflush_unlocked(f);
 		//OFLLOCK();
 		next = f->next;
 		FUNLOCK(f);

src/stdio/fgetc.c

@@ -3,9 +3,10 @@
 int fgetc(FILE *f)
 {
 	int c;
-	FLOCK(f);
+	if (f->lock < 0 || !__lockfile(f))
+		return getc_unlocked(f);
 	c = getc_unlocked(f);
-	FUNLOCK(f);
+	__unlockfile(f);
 	return c;
 }
 

src/stdio/fgetwc.c

@@ -34,7 +34,6 @@ wint_t __fgetwc_unlocked(FILE *f)
 		if (l == -1) return WEOF;
 	}
 
-	FUNLOCK(f);
 	return wc;
 }
 

src/stdio/flockfile.c

@@ -3,6 +3,8 @@
 
 void flockfile(FILE *f)
 {
-	if (!libc.threaded) pthread_self();
-	__lockfile(f);
+	while (ftrylockfile(f)) {
+		int owner = f->lock;
+		if (owner) __wait(&f->lock, &f->waiters, owner, 1);
+	}
 }

src/stdio/fputc.c

@@ -2,9 +2,10 @@
 
 int fputc(int c, FILE *f)
 {
-	FLOCK(f);
+	if (f->lock < 0 || !__lockfile(f))
+		return putc_unlocked(c, f);
 	c = putc_unlocked(c, f);
-	FUNLOCK(f);
+	__unlockfile(f);
 	return c;
 }
 

src/stdio/ftell.c

@@ -3,10 +3,8 @@
 off_t __ftello_unlocked(FILE *f)
 {
 	off_t pos = f->seek(f, 0, SEEK_CUR);
-	if (pos < 0) {
-		FUNLOCK(f);
-		return pos;
-	}
+	if (pos < 0) return pos;
+
 	/* Adjust for data in buffer. */
 	return pos - (f->rend - f->rpos) + (f->wpos - f->wbase);
 }

src/stdio/ftrylockfile.c

@@ -5,11 +5,12 @@ int ftrylockfile(FILE *f)
 {
 	int tid = pthread_self()->tid;
 	if (f->lock == tid) {
-		if (f->lockcount == INT_MAX)
+		if (f->lockcount == LONG_MAX)
 			return -1;
 		f->lockcount++;
 		return 0;
 	}
+	if (f->lock < 0) f->lock = 0;
 	if (f->lock || a_cas(&f->lock, 0, tid))
 		return -1;
 	f->lockcount = 1;

src/stdio/funlockfile.c

@@ -3,5 +3,5 @@
 
 void funlockfile(FILE *f)
 {
-	FUNLOCK(f);
+	if (!--f->lockcount) __unlockfile(f);
 }

src/stdio/stderr.c

@@ -10,5 +10,7 @@ static FILE f = {
 	.write = __stdio_write,
 	.seek = __stdio_seek,
 	.close = __stdio_close,
+	.lock = -1,
 };
 FILE *const stderr = &f;
+FILE *const __stderr_used = &f;

src/stdio/stdin.c

@@ -9,5 +9,7 @@ static FILE f = {
 	.read = __stdio_read,
 	.seek = __stdio_seek,
 	.close = __stdio_close,
+	.lock = -1,
 };
 FILE *const stdin = &f;
+FILE *const __stdin_used = &f;

src/stdio/stdout.c

@@ -10,8 +10,7 @@ static FILE f = {
 	.write = __stdout_write,
 	.seek = __stdio_seek,
 	.close = __stdio_close,
+	.lock = -1,
 };
 FILE *const stdout = &f;
-
-/* overrides symbol in fflush.c, used for flushing NULL */
-FILE *const __stdout_to_flush = &f;
+FILE *const __stdout_used = &f;