kc3-lang/SDL

diff --git a/docs/README-n3ds.md b/docs/README-n3ds.md
index 66e194d..e9e7c7d 100644
--- a/docs/README-n3ds.md
+++ b/docs/README-n3ds.md
@@ -25,3 +25,4 @@ cmake --install build
 -   SDL2main should be used to ensure ROMFS is enabled.
 -   By default, the extra L2 cache and higher clock speeds of the New 2/3DS lineup are enabled. If you wish to turn it off, use `osSetSpeedupEnable(false)` in your main function.
 -   `SDL_GetBasePath` returns the romfs root instead of the executable's directory.
+-   The Nintendo 3DS uses a cooperative threading model on a single core, meaning a thread will never yield unless done manually through the `SDL_Delay` functions, or blocking waits (`SDL_LockMutex`, `SDL_SemWait`, `SDL_CondWait`, `SDL_WaitThread`). To avoid starving other threads, `SDL_SemTryWait` and `SDL_SemWaitTimeout` will yield if they fail to acquire the semaphore, see https://github.com/libsdl-org/SDL/pull/6776 for more information.
diff --git a/src/thread/n3ds/SDL_syssem.c b/src/thread/n3ds/SDL_syssem.c
index 1a4109b..e7cbacb 100644
--- a/src/thread/n3ds/SDL_syssem.c
+++ b/src/thread/n3ds/SDL_syssem.c
@@ -27,14 +27,16 @@
 #include <3ds.h>
 
 #include "SDL_thread.h"
+#include "SDL_timer.h"
+
+int WaitOnSemaphoreFor(SDL_sem *sem, Uint32 timeout);
 
 struct SDL_semaphore
 {
     LightSemaphore semaphore;
 };
 
-SDL_sem *
-SDL_CreateSemaphore(Uint32 initial_value)
+SDL_sem *SDL_CreateSemaphore(Uint32 initial_value)
 {
     SDL_sem *sem;
 
@@ -59,9 +61,7 @@ SDL_CreateSemaphore(Uint32 initial_value)
 */
 void SDL_DestroySemaphore(SDL_sem *sem)
 {
-    if (sem) {
-        SDL_free(sem);
-    }
+    SDL_free(sem);
 }
 
 int SDL_SemTryWait(SDL_sem *sem)
@@ -70,35 +70,49 @@ int SDL_SemTryWait(SDL_sem *sem)
         return SDL_InvalidParamError("sem");
     }
 
-    return SDL_SemWaitTimeout(sem, 0);
+    if (LightSemaphore_TryAcquire(&sem->semaphore, 1) != 0) {
+        /* If we failed, yield to avoid starvation on busy waits */
+        svcSleepThread(1);
+        return SDL_MUTEX_TIMEDOUT;
+    }
+
+    return 0;
 }
 
 int SDL_SemWaitTimeout(SDL_sem *sem, Uint32 timeout)
 {
-    int retval;
-
     if (sem == NULL) {
         return SDL_InvalidParamError("sem");
     }
 
     if (timeout == SDL_MUTEX_MAXWAIT) {
         LightSemaphore_Acquire(&sem->semaphore, 1);
-        retval = 0;
-    } else {
-        int return_code = LightSemaphore_TryAcquire(&sem->semaphore, 1);
-        if (return_code != 0) {
-            for (u32 i = 0; i < timeout; i++) {
-                svcSleepThread(1000000LL);
-                return_code = LightSemaphore_TryAcquire(&sem->semaphore, 1);
-                if (return_code == 0) {
-                    break;
-                }
-            }
+        return 0;
+    }
+
+    if (LightSemaphore_TryAcquire(&sem->semaphore, 1) != 0) {
+        return WaitOnSemaphoreFor(sem, timeout);
+    }
+
+    return 0;
+}
+
+int WaitOnSemaphoreFor(SDL_sem *sem, Uint32 timeout)
+{
+    Uint64 stop_time = SDL_GetTicks64() + timeout;
+    Uint64 current_time = SDL_GetTicks64();
+    while (current_time < stop_time) {
+        if (LightSemaphore_TryAcquire(&sem->semaphore, 1) == 0) {
+            return 0;
         }
-        retval = return_code != 0 ? SDL_MUTEX_TIMEDOUT : 0;
+        /* 100 microseconds seems to be the sweet spot */
+        svcSleepThread(100000LL);
+        current_time = SDL_GetTicks64();
     }
 
-    return retval;
+    /* If we failed, yield to avoid starvation on busy waits */
+    svcSleepThread(1);
+    return SDL_MUTEX_TIMEDOUT;
 }
 
 int SDL_SemWait(SDL_sem *sem)
@@ -106,8 +120,7 @@ int SDL_SemWait(SDL_sem *sem)
     return SDL_SemWaitTimeout(sem, SDL_MUTEX_MAXWAIT);
 }
 
-Uint32
-SDL_SemValue(SDL_sem *sem)
+Uint32 SDL_SemValue(SDL_sem *sem)
 {
     if (sem == NULL) {
         SDL_InvalidParamError("sem");
diff --git a/src/thread/n3ds/SDL_systhread.c b/src/thread/n3ds/SDL_systhread.c
index caf2df7..926dd34 100644
--- a/src/thread/n3ds/SDL_systhread.c
+++ b/src/thread/n3ds/SDL_systhread.c
@@ -49,8 +49,9 @@ static void ThreadEntry(void *arg)
 
 int SDL_SYS_CreateThread(SDL_Thread *thread)
 {
-    s32 priority = N3DS_THREAD_PRIORITY_MEDIUM;
+    s32 priority;
     size_t stack_size = GetStackSize(thread->stacksize);
+    svcGetThreadPriority(&priority, CUR_THREAD_HANDLE);
 
     thread->handle = threadCreate(ThreadEntry,
                                   thread,
diff --git a/test/testsem.c b/test/testsem.c
index 2b36a3a..8d8a249 100644
--- a/test/testsem.c
+++ b/test/testsem.c
@@ -210,6 +210,8 @@ TestOverheadContended(SDL_bool try_wait)
         }
         /* Make sure threads consumed everything */
         while (SDL_SemValue(sem)) {
+            /* Friendlier with cooperative threading models */
+            SDL_Delay(1);
         }
     }
     end_ticks = SDL_GetTicks();