Merge pull request #22378 from hrydgard/intr-waits

Claude hardware testing: Interrupt waits

Core/HLE/__sceAudio.cpp

@@ -42,6 +42,7 @@
 #include "Core/HLE/sceAudio.h"
 #include "Core/HLE/sceKernel.h"
 #include "Core/HLE/sceKernelThread.h"
+#include "Core/HLE/sceKernelInterrupt.h"
 #include "Core/Util/AudioFormat.h"
 
 // Should be used to lock anything related to the outAudioQueue.
@@ -87,7 +88,12 @@ void __AudioCPUMHzChange() {
 	audioHostIntervalCycles = (int)(usToCycles(1000000ULL) * 512 / hwSampleRate);
 }
 
+// The mixer's DMA is still playing out the last block it was handed, which had samples in it.
+// A buffer arriving then doesn't restart the DMA or get read early: it waits for the next block.
+static bool mixerDMABusy;
+
 void __AudioInit() {
+	mixerDMABusy = false;
 	System_AudioResetStatCounters();
 	mixFrequency = 44100;
 	srcFrequency = 0;
@@ -113,7 +119,7 @@ void __AudioInit() {
 }
 
 void __AudioDoState(PointerWrap &p) {
-	auto s = p.Section("sceAudio", 1, 3);
+	auto s = p.Section("sceAudio", 1, 4);
 	if (!s)
 		return;
 
@@ -180,6 +186,11 @@ void __AudioDoState(PointerWrap &p) {
 		g_audioSRC.DoState(p);
 		__AudioRoutingDoState(p);
 	}
+	if (s >= 4) {
+		Do(p, mixerDMABusy);
+	} else {
+		mixerDMABusy = false;
+	}
 	// For older states the routing modes were read back once per channel, above.
 
 	__AudioCPUMHzChange();
@@ -234,7 +245,7 @@ static void __AudioReScheduleAfterWake() {
 }
 
 // Only channels 0-7. The SRC channel is on its own DMA that the mixer never touches, so the
-// two start independently of each other.
+// two start independently of each other. See mixerDMABusy for the block after the last one.
 static bool __AudioAnyChannelPlaying() {
 	for (const AudioChannel &chan : g_audioChans) {
 		if (chan.sampleAddress != 0) {
@@ -294,7 +305,7 @@ u32 __AudioEnqueue(AudioChannel &chan, u32 samplePtr, int leftVol, int rightVol)
 		chan.rightVolume = rightVol;
 	}
 	// Handing over a buffer while nothing was playing is what starts the DMA.
-	const bool startsDMA = samplePtr != 0 && !__AudioAnyChannelPlaying();
+	const bool startsDMA = samplePtr != 0 && !__AudioAnyChannelPlaying() && !mixerDMABusy;
 	// A null pointer is accepted and leaves the channel idle, but still counts as a buffer's
 	// worth of remaining samples - which is the one case where the two rest-length calls
 	// disagree with each other.
@@ -314,7 +325,7 @@ void __AudioEnqueueOneshot(AudioChannel &chan, u32 samplePtr, u32 sampleCount, u
 	chan.rightVolume = rightVol;
 	chan.remainingSamples = sampleCount;
 
-	const bool startsDMA = samplePtr != 0 && !__AudioAnyChannelPlaying();
+	const bool startsDMA = samplePtr != 0 && !__AudioAnyChannelPlaying() && !mixerDMABusy;
 	chan.sampleAddress = samplePtr;
 	if (startsDMA) {
 		__AudioStartMixerDMA();
@@ -334,8 +345,13 @@ u32 __AudioEnqueueBlocking(AudioChannel &chan, u32 samplePtr, int leftVol, int r
 	if (chan.waitingThread != 0) {
 		return SCE_ERROR_AUDIO_CHANNEL_BUSY;
 	}
-	if (!__KernelIsDispatchEnabled()) {
-		return SCE_KERNEL_ERROR_CAN_NOT_WAIT;
+	// The event flag wait fails at once with interrupts or dispatch disabled, or inside an
+	// interrupt. The driver returns that error without clearing the waiting flag it set, which
+	// leaves the channel busy for good (audio.prx, sceAudioOutputBlocking). That's deliberately not
+	// emulated: whether the channel was busy at that moment is timing, and a small difference in
+	// ours would drop a channel for the rest of the game where hardware wouldn't.
+	if (__IsInInterrupt() || !__KernelIsDispatchEnabled()) {
+		return __IsInInterrupt() ? SCE_KERNEL_ERROR_ILLEGAL_CONTEXT : SCE_KERNEL_ERROR_CAN_NOT_WAIT;
 	}
 
 	chan.waitingThread = __KernelGetCurThread();
@@ -414,13 +430,18 @@ u32 __AudioSRCEnqueueBlocking(AudioSRCChannel &chan, u32 samplePtr, int vol) {
 		return 0;
 	}
 
-	if (chan.completion) {
-		chan.completion = false;
-		return result;
+	// The event flag wait refuses before looking at the flag, so even a completion that's
+	// already there isn't taken - the buffer stays armed and the error comes back.
+	if (__IsInInterrupt()) {
+		return SCE_KERNEL_ERROR_ILLEGAL_CONTEXT;
 	}
 	if (!__KernelIsDispatchEnabled()) {
 		return SCE_KERNEL_ERROR_CAN_NOT_WAIT;
 	}
+	if (chan.completion) {
+		chan.completion = false;
+		return result;
+	}
 
 	chan.waitingThreads.push_back(__KernelGetCurThread());
 	__KernelWaitCurThread(WAITTYPE_AUDIOCHANNEL, PSP_AUDIO_SRC_WAIT_ID, result, 0, false, "blocking audio");
@@ -608,6 +629,7 @@ void __AudioUpdate(bool resetRecording) {
 	memset(mixBuffer, 0, hwBlockSize * 2 * sizeof(s32));
 
 	audioMixing = true;
+	mixerDMABusy = __AudioAnyChannelPlaying();
 	bool woke = false;
 	for (AudioChannel &chan : g_audioChans) {
 		// Deliberately not gated on `reserved`: sceAudioChRelease only clears the

Core/HLE/sceAudio.cpp

@@ -63,6 +63,10 @@ void AudioChannel::DoState(PointerWrap &p) {
 	if (s >= 4) {
 		Do(p, remainingSamples);
 		Do(p, waitingThread);
+		// Briefly, a failed blocking wait left -1 here to keep the channel busy for good.
+		if (waitingThread < 0) {
+			waitingThread = 0;
+		}
 		Do(p, waitingAddress);
 		Do(p, waitingLeftVolume);
 		Do(p, waitingRightVolume);

Core/HLE/sceDisplay.cpp

@@ -135,8 +135,10 @@ std::map<SceUID, int> vblankPausedWaits;
 
 // STATE END
 
-// The vblank period is 731.5 us (0.7315 ms)
-const double vblankMs = 0.7315;
+// tests/display/vblanklen measures 730-770us from sceDisplayWaitVblankStart returning until
+// vblank ends, and an hcount of up to 14 inside it. The wait's own latency is in that, so this is
+// the upper end.
+const double vblankMs = 0.770;
 // These are guesses based on tests.
 const double vsyncStartMs = 0.5925;
 const double vsyncEndMs = 0.7265;

Core/HLE/sceIo.cpp

@@ -474,6 +474,9 @@ static void __IoAsyncNotify(u64 userdata, int cyclesLate) {
 		if (f->closePending) {
 			__IoFreeFd(fd, error);
 		}
+		// Like any interrupt that wakes a thread, this dispatches right away, so a better
+		// priority waiter doesn't sit ready until something else happens to reschedule.
+		__KernelReSchedule("async io completed");
 	}
 }
 
@@ -522,6 +525,7 @@ static void __IoSyncNotify(u64 userdata, int cyclesLate) {
 
 	HLEKernel::ResumeFromWait(threadID, WAITTYPE_IO, fd, result);
 	f->waitingSyncThreads.erase(std::remove(f->waitingSyncThreads.begin(), f->waitingSyncThreads.end(), threadID), f->waitingSyncThreads.end());
+	__KernelReSchedule("io completed");
 }
 
 static void __IoAsyncBeginCallback(SceUID threadID, SceUID prevCallbackId) {

Core/HLE/sceKernelInterrupt.cpp

@@ -512,13 +512,65 @@ int __ReleaseSubIntrHandler(int intrNumber, int subIntrNumber) {
 	return 0;
 }
 
+// What a user mode caller finds on each interrupt, as interruptman.prx checks it: whether a driver
+// has installed a handler for the interrupt at all, whether it has sub-interrupt slots, and whether
+// user handlers are allowed in them. This is system state rather than a rule, read back from
+// intr/registersub and intr/releasesub on a 6.61 PSP (running PSPLink, whose USB drivers may count.)
+// Interrupt 8 had a handler in an older recording and doesn't now, so treat this as approximate.
+enum class IntrUserAccess : u8 {
+	NO_HANDLER,   // SCE_KERNEL_ERROR_NOTFOUND_HANDLER
+	NO_SUBS,      // SCE_KERNEL_ERROR_ILLEGAL_INTRCODE, the sub number is always out of range
+	KERNEL_SUBS,  // SCE_KERNEL_ERROR_ILLEGAL_INTRCODE to register, empty slots to release
+	USER,
+};
+
+static IntrUserAccess GetIntrUserAccess(u32 intrNumber) {
+	switch (intrNumber) {
+	case PSP_GE_INTR:
+	case PSP_VBLANK_INTR:
+		return IntrUserAccess::USER;
+	case 4: case 6: case 21:
+		return IntrUserAccess::KERNEL_SUBS;
+	case 7: case 10: case 12: case 15: case 16: case 17: case 18: case 19: case 20: case 22:
+	case 23: case 24: case 26: case 31: case 36: case 50: case 56: case 57: case 58: case 59:
+	case 60: case 61: case 65:
+		return IntrUserAccess::NO_SUBS;
+	default:
+		return IntrUserAccess::NO_HANDLER;
+	}
+}
+
+// The vblank slots a user handler may take (the rest are the kernel's), and those the display
+// driver already holds.
+static bool IsUserVblankSubIntr(u32 subIntrNumber) {
+	return subIntrNumber < 16;
+}
+static bool IsKernelHeldVblankSubIntr(u32 subIntrNumber) {
+	return (subIntrNumber >= 18 && subIntrNumber <= 20) || (subIntrNumber >= 24 && subIntrNumber <= 26);
+}
+
 u32 sceKernelRegisterSubIntrHandler(u32 intrNumber, u32 subIntrNumber, u32 handler, u32 handlerArg) {
 	if (intrNumber >= PSP_NUMBER_INTERRUPTS) {
 		return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_ILLEGAL_INTRCODE, "invalid interrupt");
 	}
+	switch (GetIntrUserAccess(intrNumber)) {
+	case IntrUserAccess::NO_HANDLER:
+		return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_NOTFOUND_HANDLER, "no handler for this interrupt");
+	case IntrUserAccess::NO_SUBS:
+	case IntrUserAccess::KERNEL_SUBS:
+		return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_ILLEGAL_INTRCODE, "no user subinterrupts");
+	case IntrUserAccess::USER:
+		break;
+	}
 	if (subIntrNumber >= PSP_NUMBER_SUBINTERRUPTS) {
 		return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_ILLEGAL_INTRCODE, "invalid subinterrupt");
 	}
+	if (intrNumber == PSP_VBLANK_INTR) {
+		if (IsKernelHeldVblankSubIntr(subIntrNumber))
+			return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_FOUND_HANDLER, "held by the kernel");
+		if (!IsUserVblankSubIntr(subIntrNumber))
+			return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_ILLEGAL_INTRCODE, "kernel only subinterrupt");
+	}
 
 	u32 error;
 	SubIntrHandler *subIntrHandler = __RegisterSubIntrHandler(intrNumber, subIntrNumber, handler, handlerArg, error);
@@ -539,9 +591,17 @@ u32 sceKernelReleaseSubIntrHandler(u32 intrNumber, u32 subIntrNumber) {
 	if (intrNumber >= PSP_NUMBER_INTERRUPTS) {
 		return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_ILLEGAL_INTRCODE, "invalid interrupt");
 	}
-	if (subIntrNumber >= PSP_NUMBER_SUBINTERRUPTS) {
+	const IntrUserAccess access = GetIntrUserAccess(intrNumber);
+	if (access == IntrUserAccess::NO_HANDLER) {
+		return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_NOTFOUND_HANDLER, "no handler for this interrupt");
+	}
+	if (access == IntrUserAccess::NO_SUBS || subIntrNumber >= PSP_NUMBER_SUBINTERRUPTS) {
 		return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_ILLEGAL_INTRCODE, "invalid subinterrupt");
 	}
+	// User code can't have put anything in the kernel's slots, and can't release what's there.
+	if (access == IntrUserAccess::KERNEL_SUBS || (intrNumber == PSP_VBLANK_INTR && !IsUserVblankSubIntr(subIntrNumber))) {
+		return hleLogError(Log::sceIntc, SCE_KERNEL_ERROR_NOTFOUND_HANDLER, "not a user subinterrupt");
+	}
 
 	u32 error = __ReleaseSubIntrHandler(intrNumber, subIntrNumber);
 	return hleLogDebugOrError(Log::sceIntc, error);

Core/HLE/sceKernelThread.cpp

@@ -1629,8 +1629,17 @@ static PSPThread *__KernelNextThread() {
 		return 0;
 }
 
+// Set by sceKernelStartThread when the new thread outranks the caller, for the reschedule right
+// after it. The firmware hands the CPU straight to the new thread, even when a thread of better
+// priority is ready but hasn't been dispatched (say, one that a sceKernelTerminateThread woke.)
+// See tests/threads/threads/termsuspended.
+static SceUID g_startThreadHandoff = 0;
+
 void __KernelReSchedule(const char *reason)
 {
+	const SceUID handoff = g_startThreadHandoff;
+	g_startThreadHandoff = 0;
+
 	// First, let's check if there are any pending callbacks to trigger.
 	// TODO: Could probably take this out of __KernelReSchedule() which is a bit hot.
 	__KernelCheckCallbacks();
@@ -1643,6 +1652,16 @@ void __KernelReSchedule(const char *reason)
 		return;
 	}
 
+	if (handoff != 0) {
+		u32 error;
+		PSPThread *started = kernelObjects.Get<PSPThread>(handoff, error);
+		if (started && started->nt.status == THREADSTATUS_READY) {
+			threadReadyQueue.remove(started->nt.currentPriority, handoff);
+			__KernelSwitchContext(started, reason);
+			return;
+		}
+	}
+
 	PSPThread *nextThread = __KernelNextThread();
 	if (nextThread) {
 		__KernelSwitchContext(nextThread, reason);
@@ -1900,7 +1919,13 @@ int __KernelCreateThread(const char *threadName, SceUID moduleID, u32 entry, u32
 	// Creating a thread resumes dispatch automatically.  Probably can't create without it.
 	dispatchEnabled = true;
 
-	hleEatCycles(32000);
+	// Measured by tests/threads/scheduling/costs: about 150us, plus filling the stack with 0xFF at
+	// around a cycle per byte - 1.3ms for a 256KB stack.
+	int createCycles = 32000;
+	if ((attr & PSP_THREAD_ATTR_NO_FILLSTACK) == 0 && stacksize > 0) {
+		createCycles += stacksize - stacksize / 64;
+	}
+	hleEatCycles(createCycles);
 	// This won't schedule to the new thread, but it may to one woken from eating cycles.
 	// Technically, this should not eat all at once, and reschedule in the middle, but that's hard.
 	hleReSchedule("thread created");
@@ -1971,8 +1996,10 @@ int __KernelStartThread(SceUID threadToStartID, int argSize, u32 argBlockPtr, bo
 	if (cur && cur->nt.currentPriority > startThread->nt.currentPriority) {
 		KernelValidateThreadTarget(startThread->context.pc);
 		__KernelChangeReadyState(cur, currentThread, true);
-		if (__InterruptsEnabled())
+		if (__InterruptsEnabled()) {
+			g_startThreadHandoff = threadToStartID;
 			hleReSchedule("thread started");
+		}
 	}
 
 	// Starting a thread automatically resumes the dispatch thread if the new thread has worse priority.
@@ -2219,6 +2246,8 @@ int sceKernelDeleteThread(int threadID) {
 			return hleLogError(Log::sceKernel, SCE_KERNEL_ERROR_NOT_DORMANT);
 		}
 
+		// 50-100us whatever the stack size, per tests/threads/scheduling/costs.
+		hleEatCycles(15000);
 		return hleLogDebug(Log::sceKernel, __KernelDeleteThread(threadID, SCE_KERNEL_ERROR_THREAD_TERMINATED, "thread deleted"));
 	}
 }

Core/HLE/sceKernelTime.cpp

@@ -74,7 +74,6 @@ int sceKernelGetSystemTime(u32 sysclockPtr)
 	}
 	VERBOSE_LOG(Log::sceKernel, "sceKernelGetSystemTime(out:%16llx)", t);
 	hleEatCycles(265);
-	hleReSchedule("system time");
 	return hleNoLog(0);
 }
 
@@ -87,7 +86,6 @@ u32 sceKernelGetSystemTimeLow()
 	if (PSP_CoreParameter().compat.flags().KernelGetSystemTimeLowEatMoreCycles) {
 		hleEatCycles(70000);
 	}
-	hleReSchedule("system time");
 	return hleNoLog((u32)t);
 }
 
@@ -96,7 +94,6 @@ u64 sceKernelGetSystemTimeWide()
 	u64 t = CoreTiming::GetGlobalTimeUsScaled();
 	VERBOSE_LOG(Log::sceKernel,"%i=sceKernelGetSystemTimeWide()",(u32)t);
 	hleEatCycles(250);
-	hleReSchedule("system time");
 	return hleNoLog(t);
 }
 
@@ -153,7 +150,6 @@ u32 sceKernelLibcClock()
 {
 	u32 retVal = (u32) CoreTiming::GetGlobalTimeUs();
 	hleEatCycles(330);
-	hleReSchedule("libc clock");
 	return hleLogDebug(Log::sceKernel, retVal);
 }
 
@@ -169,7 +165,6 @@ u32 sceKernelLibcTime(u32 outPtr)
 	else if (outPtr != 0)
 		return 0;
 
-	hleReSchedule("libc time");
 	return hleLogDebug(Log::sceKernel, t);
 }
 
@@ -185,7 +180,6 @@ u32 sceKernelLibcGettimeofday(u32 timeAddr, u32 tzAddr)
 	DEBUG_LOG(Log::sceKernel,"sceKernelLibcGettimeofday(%08x, %08x)", timeAddr, tzAddr);
 	hleEatCycles(1885);
 
-	hleReSchedule("libc timeofday");
 	return hleNoLog(0);
 }
 

Core/HLE/scePower.cpp

@@ -433,16 +433,22 @@ static int sceKernelVolatileMemLock(int type, u32 paddr, u32 psize) {
 	case SCE_KERNEL_ERROR_CAN_NOT_WAIT:
 		{
 			WARN_LOG(Log::HLE, "sceKernelVolatileMemLock(%i, %08x, %08x): dispatch disabled", type, paddr, psize);
-			Memory::WriteOrException_U32(0x08400000, paddr);
-			Memory::WriteOrException_U32(0x00400000, psize);
+			// Only through pointers that are there: intr/waits passes NULL and gets just the error.
+			if (Memory::IsValid4AlignedAddress(paddr))
+				Memory::WriteUnchecked_U32(0x08400000, paddr);
+			if (Memory::IsValid4AlignedAddress(psize))
+				Memory::WriteUnchecked_U32(0x00400000, psize);
 		}
 		break;
 
 	case SCE_KERNEL_ERROR_ILLEGAL_CONTEXT:
 		{
 			WARN_LOG(Log::HLE, "sceKernelVolatileMemLock(%i, %08x, %08x): in interrupt", type, paddr, psize);
-			Memory::WriteOrException_U32(0x08400000, paddr);
-			Memory::WriteOrException_U32(0x00400000, psize);
+			// Only through pointers that are there: intr/waits passes NULL and gets just the error.
+			if (Memory::IsValid4AlignedAddress(paddr))
+				Memory::WriteUnchecked_U32(0x08400000, paddr);
+			if (Memory::IsValid4AlignedAddress(psize))
+				Memory::WriteUnchecked_U32(0x00400000, psize);
 		}
 		break;
 
@@ -505,6 +511,10 @@ static u32 scePowerSetCpuClockFrequency(u32 cpufreq) {
 	if (cpufreq == 0 || cpufreq > 333) {
 		return hleLogWarning(Log::sceMisc, SCE_KERNEL_ERROR_INVALID_VALUE, "invalid frequency");
 	}
+	// The CPU can't run faster than the PLL it's divided from.
+	if ((u64)cpufreq * 1000000 > (u64)pllFreq) {
+		return hleLogWarning(Log::sceMisc, SCE_KERNEL_ERROR_INVALID_VALUE, "above the pll frequency");
+	}
 	if (GetLockedCPUSpeedMhz() > 0) {
 		return hleLogDebug(Log::sceMisc, 0, "locked by user config at %i", GetLockedCPUSpeedMhz());
 	}
@@ -551,7 +561,13 @@ static u32 scePowerGetBusClockFrequencyInt() {
 }
 
 static float scePowerGetCpuClockFrequencyFloat() {
-	float cpuFreq = CoreTiming::GetClockFrequencyHz() / 1000000.0f;
+	// The CPU runs at a multiple of pll/511, and the firmware works the value out in single
+	// precision, as pll * n / 511, rather than from whole Hz - which is off in the last digit
+	// (power/freq).
+	const double step = (double)pllFreq / 511.0;
+	const float steps = (float)std::round(CoreTiming::GetClockFrequencyHz() / step);
+	const float pllMhz = (float)(pllFreq / 1000000.0);
+	float cpuFreq = (pllMhz * steps) / 511.0f;
 	DEBUG_LOG(Log::sceMisc, "%f=scePowerGetCpuClockFrequencyFloat()", (float)cpuFreq);
 	return cpuFreq;
 }

Core/HLE/sceUmd.cpp

@@ -252,7 +252,9 @@ static void __UmdEndCallback(SceUID threadID, SceUID prevCallbackId)
 	else
 	{
 		_dbg_assert_msg_(umdStatTimeoutEvent != -1, "Must have a umd timer");
-		CoreTiming::ScheduleEvent(cyclesLeft, umdStatTimeoutEvent, __KernelGetCurThread());
+		// A deadline of 0 means the wait has no timeout.
+		if (waitDeadline != 0)
+			CoreTiming::ScheduleEvent(cyclesLeft, umdStatTimeoutEvent, __KernelGetCurThread());
 
 		umdWaitingThreads.push_back(threadID);
 
@@ -268,10 +270,15 @@ static int sceUmdCheckMedium() {
 	return hleLogDebug(Log::sceKernel, retVal);
 }
 	
+// A user mode caller can't pass a kernel address. (mediaman.prx checks these with k1.)
+static bool IsUserAddress(u32 addr) {
+	return (addr & 0x80000000) == 0;
+}
+
 static u32 sceUmdGetDiscInfo(u32 infoAddr) {
 	DEBUG_LOG(Log::sceIo, "sceUmdGetDiscInfo(%08x)", infoAddr);
 
-	if (Memory::IsValidAddress(infoAddr)) {
+	if (Memory::IsValidRange(infoAddr, 8) && IsUserAddress(infoAddr) && IsUserAddress(infoAddr + 4)) {
 		auto info = PSPPointer<PspUmdInfo>::Create(infoAddr);
 		if (info->size != 8)
 			return hleLogError(Log::sceIo, SCE_KERNEL_ERROR_ERRNO_INVALID_ARGUMENT);
@@ -283,9 +290,13 @@ static u32 sceUmdGetDiscInfo(u32 infoAddr) {
 	}
 }
 
-static int sceUmdActivate(u32 mode, const char *name) {
+static int sceUmdActivate(u32 mode, u32 namePtr) {
 	if (mode < 1 || mode > 2)
 		return hleLogWarning(Log::sceIo, SCE_KERNEL_ERROR_ERRNO_INVALID_ARGUMENT);
+	// The firmware compares the name before checking the pointer, so a bad one crashes there.
+	const char *name = namePtr != 0 && Memory::IsValidAddress(namePtr) ? Memory::GetCharPointer(namePtr) : nullptr;
+	if (!name || strncmp(name, "disc0:", 7) != 0 || !IsUserAddress(namePtr))
+		return hleLogWarning(Log::sceIo, SCE_KERNEL_ERROR_ERRNO_INVALID_ARGUMENT, "bad name");
 
 	__KernelUmdActivate();
 
@@ -295,11 +306,14 @@ static int sceUmdActivate(u32 mode, const char *name) {
 	return hleLogDebug(Log::sceIo, 0);
 }
 
-static int sceUmdDeactivate(u32 mode, const char *name)
+static int sceUmdDeactivate(u32 mode, u32 namePtr)
 {
 	// Why 18?  No idea.
 	if (mode > 18)
 		return hleLogError(Log::sceIo, SCE_KERNEL_ERROR_ERRNO_INVALID_ARGUMENT);
+	// Unlike sceUmdActivate(), the name isn't compared, and only mode 2 requires one.
+	if ((mode == 2 && namePtr == 0) || !IsUserAddress(namePtr))
+		return hleLogError(Log::sceIo, SCE_KERNEL_ERROR_ERRNO_INVALID_ARGUMENT, "bad name");
 
 	__KernelUmdDeactivate();
 
@@ -314,8 +328,7 @@ static u32 sceUmdRegisterUMDCallBack(u32 cbId)
 {
 	int retVal = 0;
 
-	// TODO: If the callback is invalid, return SCE_KERNEL_ERROR_ERRNO_INVALID_ARGUMENT.
-	if (!kernelObjects.IsValid(cbId)) {
+	if (!kernelObjects.Is<PSPCallback>(cbId)) {
 		retVal = SCE_KERNEL_ERROR_ERRNO_INVALID_ARGUMENT;
 	} else {
 		// There's only ever one.
@@ -363,13 +376,18 @@ static void __UmdStatTimeout(u64 userdata, int cyclesLate)
 	HLEKernel::RemoveWaitingThread(umdWaitingThreads, threadID);
 }
 
-static void __UmdWaitStat(u32 timeout)
+// The firmware waits on an event flag holding the drive state (mediaman.prx), and passes no
+// timeout at all for 0, so that waits forever.
+static void __UmdWaitStat(u32 timeout, bool callbacks)
 {
-	// This happens to be how the hardware seems to time things.
-	if (timeout <= 4)
-		timeout = 15;
-	else if (timeout <= 215)
-		timeout = 250;
+	if (timeout == 0)
+		return;
+
+	// Measured on hardware. Oddly, the CB version doesn't have the shortest step.
+	if (timeout <= 1 && !callbacks)
+		timeout = 25;
+	else if (timeout <= 209)
+		timeout = 240;
 
 	CoreTiming::ScheduleEvent(usToCycles((int) timeout), umdStatTimeoutEvent, __KernelGetCurThread());
 }
@@ -416,7 +434,7 @@ static int sceUmdWaitDriveStatWithTimer(u32 stat, u32 timeout) {
 
 	hleEatCycles(520);
 	if ((stat & __KernelUmdGetState()) == 0) {
-		__UmdWaitStat(timeout);
+		__UmdWaitStat(timeout, false);
 		umdWaitingThreads.push_back(__KernelGetCurThread());
 		__KernelWaitCurThread(WAITTYPE_UMD, 1, stat, 0, false, "umd stat waited with timer");
 		return hleLogDebug(Log::sceIo, 0, "waiting");
@@ -441,11 +459,7 @@ static int sceUmdWaitDriveStatCB(u32 stat, u32 timeout) {
 	hleEatCycles(520);
 	hleCheckCurrentCallbacks();
 	if ((stat & __KernelUmdGetState()) == 0) {
-		if (timeout == 0) {
-			timeout = 8000;
-		}
-
-		__UmdWaitStat(timeout);
+		__UmdWaitStat(timeout, true);
 		umdWaitingThreads.push_back(__KernelGetCurThread());
 		__KernelWaitCurThread(WAITTYPE_UMD, 1, stat, 0, true, "umd stat waited");
 		return hleLogDebug(Log::sceIo, 0, "waiting");
@@ -519,10 +533,10 @@ static u32 sceUmdReplacePermit() {
 
 const HLEFunction sceUmdUser[] = 
 {
-	{0XC6183D47, &WrapI_UC<sceUmdActivate>,               "sceUmdActivate",               'i', "is"},
+	{0XC6183D47, &WrapI_UU<sceUmdActivate>,               "sceUmdActivate",               'i', "is"},
 	{0X6B4A146C, &WrapU_V<sceUmdGetDriveStat>,            "sceUmdGetDriveStat",           'x', ""  },
 	{0X46EBB729, &WrapI_V<sceUmdCheckMedium>,             "sceUmdCheckMedium",            'i', ""  },
-	{0XE83742BA, &WrapI_UC<sceUmdDeactivate>,             "sceUmdDeactivate",             'i', "xs"},
+	{0XE83742BA, &WrapI_UU<sceUmdDeactivate>,             "sceUmdDeactivate",             'i', "xs"},
 	{0X8EF08FCE, &WrapI_U<sceUmdWaitDriveStat>,           "sceUmdWaitDriveStat",          'i', "x" },
 	{0X56202973, &WrapI_UU<sceUmdWaitDriveStatWithTimer>, "sceUmdWaitDriveStatWithTimer", 'i', "xx"},
 	{0X4A9E5E29, &WrapI_UU<sceUmdWaitDriveStatCB>,        "sceUmdWaitDriveStatCB",        'i', "xx"},

docs/pspautotests-hardware.md

@@ -141,15 +141,23 @@ python3 test.py --graphics=software io/shortname/shortname
   builds with pspdev GCC 15 as of the "Make the tests build with a current pspdev toolchain"
   commit; if you hit a broken one anyway, `gentest.py -k` (`--keep`) skips `make` entirely and
   you can build your own target by hand with `make yourtest.prx`.
-- **Rebuilding a `.prx` is not free, so don't regenerate one you didn't change.** The binaries are
-  committed and were built with a much older SDK. Rebuilding with the current toolchain grows them
-  by roughly a third (`testgp.prx`: 117 KB to 191 KB), and it can change what a test *does*:
-  `time_t` is 64-bit now, so `rtc/convert`'s `sceRtcSetTime_t(&pt, 62135596800ULL)` marshals
-  differently than the committed binary and stops matching its own `.expected`. Check
-  `git status` and revert any `.prx` you didn't mean to touch.
-- **A `.prx` you did rebuild deserves a hardware run before you commit it.** Build it, run it, and
-  diff the output against the committed `.expected` - if it differs, decide whether the test
-  genuinely changed or whether the toolchain did. Note the committed `.expected` files have CRLF
+- **When you touch one test, rebuild every `.prx` in its directory and re-record them all.** Many
+  committed binaries were built with a much older SDK, and a rebuild with the current toolchain can
+  change what a test *does*: `time_t` is 64-bit now, so `rtc/convert`'s
+  `sceRtcSetTime_t(&pt, 62135596800ULL)` marshals differently, and in `tests/intr` the SDK's
+  `ModuleMgrForUser` stubs, linked next to the directory's own import of that library, silently did
+  nothing - `waits` printed `sceKernelGetModuleId`'s result for `sceKernelStartModule`. Rebuilding
+  a whole directory at once turns such surprises up while you're there to explain them, rather
+  than one at a time whenever someone happens to touch a neighbour. So run a plain `make` in the
+  directory, record each test on hardware, and commit the binaries and `.expected` files together.
+  (The binaries also grow by roughly a third: `testgp.prx` went from 117 KB to 191 KB.)
+- **Diff every re-recorded `.expected` against the old one before committing.** A difference means
+  the test changed on purpose, the toolchain changed what it does (fix the test), or the firmware
+  did. Many old recordings are likely from a PSP on an earlier firmware than today's 6.61, and a
+  driver change or a bug fix between versions shows up regardless of `sceKernelSetCompiledSdkVersion`
+  (`intr/registersub` depends on which drivers have hooked which interrupts: interrupt 8 had a
+  handler in the old recording and has none on 6.61). Say which in the commit message, and keep
+  the 6.61 result, since that's the firmware PPSSPP models. Note the committed `.expected` files have CRLF
   line endings (they were recorded on Windows) while a fresh run writes LF, so compare with
   `diff <(tr -d '\r' < __testoutput.txt) <(tr -d '\r' < the.expected)`.
 - **Each test PRX stays resident after it runs.** Run a handful back to back and the next

docs/sceAudio.md

@@ -59,11 +59,27 @@ r = enqueue(...);
 if (r != BUSY)         return r;
 if (channel->waiting)  return BUSY;      // somebody else is already parked here
 channel->waiting = 1;
-wait for this channel's bit in the driver's event flag;
+r = wait for this channel's bit in the driver's event flag;
+if (r < 0)             return r;         // waiting is left set! (not emulated)
 retry the enqueue;
 channel->waiting = 0;
 ```
 
+The error path is a real firmware bug. The event flag wait fails at once with interrupts or
+dispatch disabled (`800201a7`) or inside an interrupt (`80020064`), and the driver returns that
+without clearing `waiting`. Nothing ever clears it after that, so the channel stays busy for
+good, and `sceAudioChRelease` refuses to release it.
+
+**The emulator returns the same errors but doesn't leave the flag set.** Whether the channel was
+busy at that moment is a question of timing, and ours is close but not exact: emulating the bug
+means a small scheduling difference could drop a channel for the rest of the game where hardware
+wouldn't. No game can depend on losing a channel, so the risk only runs one way.
+`tests/intr/waits` leaves the case out for this reason.
+
+The SRC channel's wait fails the same way, and since the flag is never looked at, even a
+completion that was already there isn't taken: the buffer stays armed and the error comes back.
+That one is emulated, since nothing is lost - the buffer plays out as usual.
+
 **Only one thread can be parked on a channel.** That single flag is why a game that runs a movie
 thread and a sound-effect thread over one output gets sensible behavior on hardware and did not
 in the emulator: the loser is told the channel is busy and skips its turn, where an emulator that
@@ -121,6 +137,11 @@ reset but no early read, since its DMA feeds the codec directly. Without this th
 `sceAudioGetChannelRestLen` right after an output would depend on where the timer happened to
 be, and would differ from run to run.
 
+That only happens when the DMA is actually idle. After a block that had samples in it, the DMA
+is still playing that block out, so a buffer handed over within it is not read early: it waits
+for the next block like any other. (Also from `tests/intr/waits`, where a second 64-sample buffer
+is still in the channel when the next call looks.)
+
 Two things are still approximate, both below one mix block:
 
 - A descriptor retires when the DMA transfer finishes, which is slightly ahead of the audio

test.py

@@ -180,6 +180,7 @@ tests_good = [
   "ctrl/sampling2/sampling2",
   "ctrl/vblank",
   "display/display",
+  "display/hcount",
   "display/vblankmulti",
   "display/isstate",
   "display/setframebuf",
@@ -266,6 +267,8 @@ tests_good = [
   "intr/intr",
   "intr/mfic",
   "intr/enablesub",
+  "intr/registersub",
+  "intr/releasesub",
   "intr/suspended",
   "intr/vblank/vblank",
   "io/cwd/cwd",
@@ -291,6 +294,7 @@ tests_good = [
   "mstick/mstick",
   "power/cpu",
   "power/power",
+  "power/freq",
   "power/volatile/lock",
   "power/volatile/trylock",
   "power/volatile/unlock",
@@ -403,6 +407,8 @@ tests_good = [
   "threads/threads/stackfree",
   "threads/threads/start",
   "threads/threads/suspend",
+  "threads/threads/terminate",
+  "threads/threads/termsuspended",
   "threads/threads/threadend",
   "threads/threads/threadmanidlist",
   "threads/threads/threadmanidtype",
@@ -450,6 +456,7 @@ tests_good = [
   "utility/savedata/getsize",
   "utility/savedata/makedata",
   "utility/systemparam/systemparam",
+  "umd/api/api",
   "umd/callbacks/umd",
   "umd/wait/wait",
   "umd/register",
@@ -527,7 +534,6 @@ tests_next = [
   "audio/output2/frequency",
   "audio/output2/rest",
   "ccc/convertstring",
-  "display/hcount",
   "font/fonttest",
   "font/charinfo",
   "font/newlib",
@@ -569,8 +575,6 @@ tests_next = [
   "gpu/texmtx/uvs",
   "gpu/textures/size",
   "gpu/triangle/triangle",
-  "intr/registersub",
-  "intr/releasesub",
   "intr/waits",
   "sysmem/kernel/heapgrow",
   "io/file/file",
@@ -590,7 +594,6 @@ tests_next = [
   #"modules/loadexec/loader",
   "net/http/http",
   "net/primary/ether",
-  "power/freq",
   "sysmem/partition",
   # These two mbx tests only appeared to work because they papered over bugs 
 
@@ -598,7 +601,6 @@ tests_next = [
   "threads/scheduling/dispatch",
   "threads/scheduling/scheduling",
   "threads/threads/create",
-  "threads/threads/terminate",
   "threads/tls/memory",
   "threads/vpl/create",
   "umd/io/umd_io",