/src/gos/gfx_* -> /src/gos/gos_*

1 files changed, 248 insertions, 0 deletions

diff --git a/src/gos/gos_linux.c b/src/gos/gos_linux.c
new file mode 100644
index 00000000..e6fbe26c
--- /dev/null
+++ b/src/gos/gos_linux.c
@@ -0,0 +1,248 @@
+/*
+ * This file is subject to the terms of the GFX License. If a copy of
+ * the license was not distributed with this file, you can obtain one at:
+ *
+ *              http://ugfx.org/license.html
+ */
+
+#include "gfx.h"
+
+#if GFX_USE_OS_LINUX
+
+// Linux seems to have deprecated pthread_yield() and now says to use sched_yield()
+#define		USE_SCHED_NOT_PTHREAD_YIELD		TRUE
+
+#include <stdio.h>
+#include <unistd.h>
+#include <time.h>
+#if USE_SCHED_NOT_PTHREAD_YIELD
+	#include <sched.h>
+	#define linuxyield()	sched_yield()
+#else
+	#define linuxyield()	pthread_yield()
+#endif
+
+static gfxMutex		SystemMutex;
+
+void _gosInit(void)
+{
+	/* No initialization of the operating system itself is needed */
+	gfxMutexInit(&SystemMutex);
+}
+
+void _gosDeinit(void)
+{
+	/* ToDo */
+}
+
+void gfxSystemLock(void) {
+	gfxMutexEnter(&SystemMutex);
+}
+
+void gfxSystemUnlock(void) {
+	gfxMutexExit(&SystemMutex);
+}
+
+void gfxYield(void) {
+	linuxyield();
+}
+
+void gfxHalt(const char *msg) {
+	if (msg)
+		fprintf(stderr, "%s\n", msg);
+	exit(1);
+}
+
+void gfxSleepMilliseconds(delaytime_t ms) {
+	struct timespec	ts;
+
+	switch(ms) {
+		case TIME_IMMEDIATE:
+			linuxyield();
+			return;
+
+		case TIME_INFINITE:
+			while(1)
+				sleep(60);
+			return;
+
+		default:
+			ts.tv_sec = ms / 1000;
+			ts.tv_nsec = (ms % 1000) * 1000000;
+			nanosleep(&ts, 0);
+			return;
+	}
+}
+
+void gfxSleepMicroseconds(delaytime_t us) {
+	struct timespec	ts;
+
+	switch(us) {
+		case TIME_IMMEDIATE:
+			linuxyield();
+			return;
+
+		case TIME_INFINITE:
+			while(1)
+				sleep(60);
+			return;
+
+		default:
+			ts.tv_sec = us / 1000000;
+			ts.tv_nsec = (us % 1000000) * 1000;
+			nanosleep(&ts, 0);
+			return;
+	}
+}
+
+systemticks_t gfxSystemTicks(void) {
+	struct timespec	ts;
+
+	clock_gettime(CLOCK_MONOTONIC, &ts);
+	return ts.tv_sec * 1000 + ts.tv_nsec / 1000000;
+}
+
+gfxThreadHandle gfxThreadCreate(void *stackarea, size_t stacksz, threadpriority_t prio, DECLARE_THREAD_FUNCTION((*fn),p), void *param) {
+	gfxThreadHandle		th;
+	(void)				stackarea;
+	(void)				stacksz;
+	(void)				prio;
+
+	// Implementing priority with pthreads is a rats nest that is also pthreads implementation dependent.
+	// Only some pthreads schedulers support it, some implementations use the operating system process priority mechanisms.
+	// Even those that do support it can have different ranges of priority and "normal" priority is an undefined concept.
+	// Across different UNIX style operating systems things can be very different (let alone OS's such as Windows).
+	// Even just Linux changes the way priority works with different kernel schedulers and across kernel versions.
+	// For these reasons we ignore the priority.
+
+	if (pthread_create(&th, 0, fn, param))
+		return 0;
+
+	return th;
+}
+
+threadreturn_t gfxThreadWait(gfxThreadHandle thread) {
+	threadreturn_t	retval;
+
+	if (pthread_join(thread, &retval))
+		return 0;
+
+	return retval;
+}
+
+#if GFX_USE_POSIX_SEMAPHORES
+	void gfxSemInit(gfxSem *pSem, semcount_t val, semcount_t limit) {
+		pSem->max = limit;
+		sem_init(&pSem->sem, 0, val);
+	}
+	void gfxSemDestroy(gfxSem *pSem) {
+		sem_destroy(&pSem->sem);
+	}
+	bool_t gfxSemWait(gfxSem *pSem, delaytime_t ms) {
+		switch (ms) {
+		case TIME_INFINITE:
+			return sem_wait(&pSem->sem) ? FALSE : TRUE;
+
+		case TIME_IMMEDIATE:
+			return sem_trywait(&pSem->sem) ? FALSE : TRUE;
+
+		default:
+			{
+				struct timespec	tm;
+
+				clock_gettime(CLOCK_REALTIME, &tm);
+				tm.tv_sec += ms / 1000;
+				tm.tv_nsec += (ms % 1000) * 1000000;
+				return sem_timedwait(&pSem->sem, &tm) ? FALSE : TRUE;
+			}
+		}
+	}
+	void gfxSemSignal(gfxSem *pSem) {
+		if (gfxSemCounter(pSem) < pSem->max)
+			sem_post(&pSem->sem);
+	}
+	semcount_t gfxSemCounter(gfxSem *pSem) {
+		int	res;
+
+		res = 0;
+		sem_getvalue(&pSem->sem, &res);
+		return res;
+	}
+#else
+	void gfxSemInit(gfxSem *pSem, semcount_t val, semcount_t limit) {
+		pthread_mutex_init(&pSem->mtx, 0);
+		pthread_cond_init(&pSem->cond, 0);
+		pthread_mutex_lock(&pSem->mtx);
+		pSem->cnt = val;
+		pSem->max = limit;
+		pthread_mutex_unlock(&pSem->mtx);
+	}
+	void gfxSemDestroy(gfxSem *pSem) {
+		pthread_mutex_destroy(&pSem->mtx);
+		pthread_cond_destroy(&pSem->cond);
+	}
+	bool_t gfxSemWait(gfxSem *pSem, delaytime_t ms) {
+		pthread_mutex_lock(&pSem->mtx);
+
+		switch (ms) {
+			case TIME_INFINITE:
+				while (!pSem->cnt)
+					pthread_cond_wait(&pSem->cond, &pSem->mtx);
+				break;
+
+			case TIME_IMMEDIATE:
+				if (!pSem->cnt) {
+					pthread_mutex_unlock(&pSem->mtx);
+					return FALSE;
+				}
+				break;
+
+			default:
+				{
+					struct timespec	tm;
+
+					clock_gettime(CLOCK_REALTIME, &tm);
+					tm.tv_sec += ms / 1000;
+					tm.tv_nsec += (ms % 1000) * 1000000;
+					while (!pSem->cnt) {
+						// We used to test the return value for ETIMEDOUT. This doesn't
+						//	work in some current pthread libraries which return -1 instead
+						//	and set errno to ETIMEDOUT. So, we will return FALSE on any error
+						//	including a ETIMEDOUT.
+						if (pthread_cond_timedwait(&pSem->cond, &pSem->mtx, &tm)) {
+							pthread_mutex_unlock(&pSem->mtx);
+							return FALSE;
+						}
+					}
+				}
+				break;
+		}
+
+		pSem->cnt--;
+		pthread_mutex_unlock(&pSem->mtx);
+		return TRUE;
+	}
+	void gfxSemSignal(gfxSem *pSem) {
+		pthread_mutex_lock(&pSem->mtx);
+
+		if (pSem->cnt < pSem->max) {
+			pSem->cnt++;
+			pthread_cond_signal(&pSem->cond);
+		}
+
+		pthread_mutex_unlock(&pSem->mtx);
+	}
+	semcount_t gfxSemCounter(gfxSem *pSem) {
+		semcount_t	res;
+
+		// The locking is really only required if obtaining the count is a divisible operation
+		//	which it might be on a 8/16 bit processor with a 32 bit semaphore count.
+		pthread_mutex_lock(&pSem->mtx);
+		res = pSem->cnt;
+		pthread_mutex_unlock(&pSem->mtx);
+
+		return res;
+	}
+#endif // GFX_USE_POSIX_SEMAPHORES
+
+#endif /* GFX_USE_OS_LINUX */