diff --git a/Core/HLE/sceKernelSemaphore.cpp b/Core/HLE/sceKernelSemaphore.cpp
index d23861dd6..200343e6e 100644
--- a/Core/HLE/sceKernelSemaphore.cpp
+++ b/Core/HLE/sceKernelSemaphore.cpp
@@ -15,6 +15,7 @@
 // Official git repository and contact information can be found at
 // https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
 
+#include <algorithm>
 #include "HLE.h"
 #include "../MIPS/MIPS.h"
 #include "../../Core/CoreTiming.h"
@@ -279,6 +280,15 @@ void __KernelSemaTimeout(u64 userdata, int cycleslate)
 	if (timeoutPtr != 0)
 		Memory::Write_U32(0, timeoutPtr);
 
+	SceUID semaID = __KernelGetWaitID(threadID, error);
+	Semaphore *s = kernelObjects.Get<Semaphore>(semaID, error);
+	if (s)
+	{
+		// This thread isn't waiting anymore.
+		s->waitingThreads.erase(std::remove(s->waitingThreads.begin(), s->waitingThreads.end(), threadID), s->waitingThreads.end());
+		s->ns.numWaitThreads--;
+	}
+
 	__KernelResumeThreadFromWait(threadID, SCE_KERNEL_ERROR_WAIT_TIMEOUT);
 }
 
@@ -301,6 +311,12 @@ void __KernelWaitSema(SceUID id, int wantedCount, u32 timeoutPtr, const char *ba
 	Semaphore *s = kernelObjects.Get<Semaphore>(id, error);
 	if (s)
 	{
+		if (wantedCount > s->ns.maxCount || wantedCount <= 0)
+		{
+			RETURN(SCE_KERNEL_ERROR_ILLEGAL_COUNT);
+			return;
+		}
+
 		// We need to set the return value BEFORE processing callbacks / etc.
 		RETURN(0);
 
diff --git a/Core/HLE/sceKernelThread.cpp b/Core/HLE/sceKernelThread.cpp
index 49d645e25..5b35f8c05 100644
--- a/Core/HLE/sceKernelThread.cpp
+++ b/Core/HLE/sceKernelThread.cpp
@@ -402,7 +402,21 @@ u32 __KernelGetWaitTimeoutPtr(SceUID threadID, u32 &error)
 	}
 	else
 	{
-		ERROR_LOG(HLE, "__KernelGetWaitValue ERROR: thread %i", threadID);
+		ERROR_LOG(HLE, "__KernelGetWaitTimeoutPtr ERROR: thread %i", threadID);
+		return 0;
+	}
+}
+
+SceUID __KernelGetWaitID(SceUID threadID, u32 &error)
+{
+	Thread *t = kernelObjects.Get<Thread>(threadID, error);
+	if (t)
+	{
+		return t->nt.waitID;
+	}
+	else
+	{
+		ERROR_LOG(HLE, "__KernelGetWaitID ERROR: thread %i", threadID);
 		return 0;
 	}
 }
diff --git a/Core/HLE/sceKernelThread.h b/Core/HLE/sceKernelThread.h
index 57f07b8e1..208ec838b 100644
--- a/Core/HLE/sceKernelThread.h
+++ b/Core/HLE/sceKernelThread.h
@@ -107,6 +107,7 @@ u32 __KernelResumeThreadFromWait(SceUID threadID, int retval);
 
 u32 __KernelGetWaitValue(SceUID threadID, u32 &error);
 u32 __KernelGetWaitTimeoutPtr(SceUID threadID, u32 &error);
+SceUID __KernelGetWaitID(SceUID threadID, u32 &error);
 void __KernelWaitCurThread(WaitType type, SceUID waitId, u32 waitValue, u32 timeoutPtr, bool processCallbacks);
 void __KernelReSchedule(const char *reason = "no reason");
 void __KernelReSchedule(bool doCallbacks, const char *reason);
diff --git a/test.py b/test.py
index b6bbb1967..b91fa7957 100644
--- a/test.py
+++ b/test.py
@@ -34,7 +34,6 @@ tests_good = [
   "threads/mutex/mutex",
   "threads/mutex/delete/delete",
   "threads/semaphores/semaphores",
-  "threads/semaphores/cancel/cancel",
   "threads/semaphores/delete/delete",
   "threads/semaphores/poll/poll",
   "threads/semaphores/refer/refer",
@@ -56,6 +55,7 @@ tests_next = [
   "threads/mutex/try/try",
   "threads/mutex/unlock/unlock",
   "threads/scheduling/scheduling",
+  "threads/semaphores/cancel/cancel",
   "threads/semaphores/create/create",
   "threads/semaphores/priority/priority",
   "threads/semaphores/wait/wait",