From fa228ca637b84e6441879769d54a531ab6aba113 Mon Sep 17 00:00:00 2001 From: Subv Date: Mon, 21 Aug 2017 20:54:29 -0500 Subject: Kernel/Threads: Don't immediately switch to the new main thread when loading a new process. This is necessary for loading multiple processes at the same time. The main thread will be automatically scheduled when necessary once the scheduler runs. --- src/core/hle/kernel/thread.cpp | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) (limited to 'src/core/hle/kernel/thread.cpp') diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index f5f2eb2f7..b957c45dd 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -478,8 +478,6 @@ void Thread::BoostPriority(s32 priority) { } SharedPtr SetupMainThread(u32 entry_point, s32 priority) { - DEBUG_ASSERT(!GetCurrentThread()); - // Initialize new "main" thread auto thread_res = Thread::Create("main", entry_point, priority, 0, THREADPROCESSORID_0, Memory::HEAP_VADDR_END); @@ -489,9 +487,7 @@ SharedPtr SetupMainThread(u32 entry_point, s32 priority) { thread->context.fpscr = FPSCR_DEFAULT_NAN | FPSCR_FLUSH_TO_ZERO | FPSCR_ROUND_TOZERO | FPSCR_IXC; // 0x03C00010 - // Run new "main" thread - SwitchContext(thread.get()); - + // Note: The newly created thread will be run when the scheduler fires. return thread; } -- cgit v1.2.3 From c34ec5e77cd9e83fcf5c929f3951557d5269b7a6 Mon Sep 17 00:00:00 2001 From: Subv Date: Fri, 21 Jul 2017 21:28:03 -0500 Subject: Kernel/Memory: Switch the current page table when a new process is scheduled. --- src/core/hle/kernel/thread.cpp | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'src/core/hle/kernel/thread.cpp') diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index f5f2eb2f7..b7f094f46 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -178,8 +178,18 @@ static void SwitchContext(Thread* new_thread) { Core::CPU().LoadContext(new_thread->context); Core::CPU().SetCP15Register(CP15_THREAD_URO, new_thread->GetTLSAddress()); + + if (!previous_thread || previous_thread->owner_process != current_thread->owner_process) { + Kernel::g_current_process = current_thread->owner_process; + Memory::current_page_table = &Kernel::g_current_process->vm_manager.page_table; + // We have switched processes and thus, page tables, clear the instruction cache so we + // don't keep stale data from the previous process. + Core::CPU().ClearInstructionCache(); + } } else { current_thread = nullptr; + // Note: We do not reset the current process and current page table when idling because + // technically we haven't changed processes, our threads are just paused. } } -- cgit v1.2.3 From b178089251200bd0309afcbcb06b43e7c82dc3bc Mon Sep 17 00:00:00 2001 From: Subv Date: Sat, 22 Jul 2017 19:37:26 -0500 Subject: Kernel/Threads: Don't clear the CPU instruction cache when performing a context switch from an idle thread into a thread in the same process. We were unnecessarily clearing the cache when going from Process A -> Idle -> Process A, this caused extreme performance regressions. --- src/core/hle/kernel/thread.cpp | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'src/core/hle/kernel/thread.cpp') diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index b7f094f46..f77c39d18 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -171,6 +171,8 @@ static void SwitchContext(Thread* new_thread) { // Cancel any outstanding wakeup events for this thread CoreTiming::UnscheduleEvent(ThreadWakeupEventType, new_thread->callback_handle); + auto previous_process = Kernel::g_current_process; + current_thread = new_thread; ready_queue.remove(new_thread->current_priority, new_thread); @@ -179,7 +181,7 @@ static void SwitchContext(Thread* new_thread) { Core::CPU().LoadContext(new_thread->context); Core::CPU().SetCP15Register(CP15_THREAD_URO, new_thread->GetTLSAddress()); - if (!previous_thread || previous_thread->owner_process != current_thread->owner_process) { + if (previous_process != current_thread->owner_process) { Kernel::g_current_process = current_thread->owner_process; Memory::current_page_table = &Kernel::g_current_process->vm_manager.page_table; // We have switched processes and thus, page tables, clear the instruction cache so we -- cgit v1.2.3 From c02bbb7030efd072511bd0051a44d9e503016f74 Mon Sep 17 00:00:00 2001 From: MerryMage Date: Sun, 24 Sep 2017 22:42:42 +0100 Subject: memory: Add GetCurrentPageTable/SetCurrentPageTable Don't expose Memory::current_page_table as a global. --- src/core/hle/kernel/thread.cpp | 11 ++++------- 1 file changed, 4 insertions(+), 7 deletions(-) (limited to 'src/core/hle/kernel/thread.cpp') diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index 324415a36..61378211f 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -178,16 +178,13 @@ static void SwitchContext(Thread* new_thread) { ready_queue.remove(new_thread->current_priority, new_thread); new_thread->status = THREADSTATUS_RUNNING; - Core::CPU().LoadContext(new_thread->context); - Core::CPU().SetCP15Register(CP15_THREAD_URO, new_thread->GetTLSAddress()); - if (previous_process != current_thread->owner_process) { Kernel::g_current_process = current_thread->owner_process; - Memory::current_page_table = &Kernel::g_current_process->vm_manager.page_table; - // We have switched processes and thus, page tables, clear the instruction cache so we - // don't keep stale data from the previous process. - Core::CPU().ClearInstructionCache(); + SetCurrentPageTable(&Kernel::g_current_process->vm_manager.page_table); } + + Core::CPU().LoadContext(new_thread->context); + Core::CPU().SetCP15Register(CP15_THREAD_URO, new_thread->GetTLSAddress()); } else { current_thread = nullptr; // Note: We do not reset the current process and current page table when idling because -- cgit v1.2.3 From 3165466b665185ecbc3e33b02b0b90e25e7248ba Mon Sep 17 00:00:00 2001 From: Subv Date: Tue, 26 Sep 2017 17:40:49 -0500 Subject: Kernel/Thread: Allow specifying which process a thread belongs to when creating it. Don't automatically assume that Thread::Create will only be called when the parent process is currently scheduled. This assumption will be broken when applets or system modules are loaded. --- src/core/hle/kernel/thread.cpp | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-) (limited to 'src/core/hle/kernel/thread.cpp') diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index 61378211f..1033f8552 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -361,7 +361,8 @@ static void ResetThreadContext(ARM_Interface::ThreadContext& context, u32 stack_ } ResultVal> Thread::Create(std::string name, VAddr entry_point, u32 priority, - u32 arg, s32 processor_id, VAddr stack_top) { + u32 arg, s32 processor_id, VAddr stack_top, + SharedPtr owner_process) { // Check if priority is in ranged. Lowest priority -> highest priority id. if (priority > THREADPRIO_LOWEST) { LOG_ERROR(Kernel_SVC, "Invalid thread priority: %d", priority); @@ -375,7 +376,7 @@ ResultVal> Thread::Create(std::string name, VAddr entry_point, // TODO(yuriks): Other checks, returning 0xD9001BEA - if (!Memory::IsValidVirtualAddress(entry_point)) { + if (!Memory::IsValidVirtualAddress(*owner_process, entry_point)) { LOG_ERROR(Kernel_SVC, "(name=%s): invalid entry %08x", name.c_str(), entry_point); // TODO: Verify error return ResultCode(ErrorDescription::InvalidAddress, ErrorModule::Kernel, @@ -399,10 +400,10 @@ ResultVal> Thread::Create(std::string name, VAddr entry_point, thread->wait_address = 0; thread->name = std::move(name); thread->callback_handle = wakeup_callback_handle_table.Create(thread).Unwrap(); - thread->owner_process = g_current_process; + thread->owner_process = owner_process; // Find the next available TLS index, and mark it as used - auto& tls_slots = Kernel::g_current_process->tls_slots; + auto& tls_slots = owner_process->tls_slots; bool needs_allocation = true; u32 available_page; // Which allocated page has free space u32 available_slot; // Which slot within the page is free @@ -426,13 +427,13 @@ ResultVal> Thread::Create(std::string name, VAddr entry_point, // Allocate some memory from the end of the linear heap for this region. linheap_memory->insert(linheap_memory->end(), Memory::PAGE_SIZE, 0); memory_region->used += Memory::PAGE_SIZE; - Kernel::g_current_process->linear_heap_used += Memory::PAGE_SIZE; + owner_process->linear_heap_used += Memory::PAGE_SIZE; tls_slots.emplace_back(0); // The page is completely available at the start available_page = tls_slots.size() - 1; available_slot = 0; // Use the first slot in the new page - auto& vm_manager = Kernel::g_current_process->vm_manager; + auto& vm_manager = owner_process->vm_manager; vm_manager.RefreshMemoryBlockMappings(linheap_memory.get()); // Map the page to the current process' address space. @@ -486,10 +487,10 @@ void Thread::BoostPriority(s32 priority) { current_priority = priority; } -SharedPtr SetupMainThread(u32 entry_point, s32 priority) { +SharedPtr SetupMainThread(u32 entry_point, s32 priority, SharedPtr owner_process) { // Initialize new "main" thread auto thread_res = Thread::Create("main", entry_point, priority, 0, THREADPROCESSORID_0, - Memory::HEAP_VADDR_END); + Memory::HEAP_VADDR_END, owner_process); SharedPtr thread = std::move(thread_res).Unwrap(); -- cgit v1.2.3 From 8432749db7afecc9beea20f993cc036418caaa15 Mon Sep 17 00:00:00 2001 From: Subv Date: Thu, 28 Sep 2017 11:53:32 -0500 Subject: Kernel/Threads: When putting a thread to wait, specify a function to execute when it is awoken. This change makes for a clearer (less confusing) path of execution in the scheduler, now the code to execute when a thread awakes is closer to the code that puts the thread to sleep (WaitSynch1, WaitSynchN). It also allows us to implement the special wake up behavior of ReplyAndReceive without hacking up WaitObject::WakeupAllWaitingThreads. If savestates are desired in the future, we can change this implementation to one similar to the CoreTiming event system, where we first register the callback functions at startup and assign their identifiers to the Thread callback variable instead of directly assigning a lambda to the wake up callback variable. --- src/core/hle/kernel/thread.cpp | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) (limited to 'src/core/hle/kernel/thread.cpp') diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index 61378211f..690cb20b3 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -247,12 +247,15 @@ static void ThreadWakeupCallback(u64 thread_handle, int cycles_late) { if (thread->status == THREADSTATUS_WAIT_SYNCH_ANY || thread->status == THREADSTATUS_WAIT_SYNCH_ALL || thread->status == THREADSTATUS_WAIT_ARB) { - thread->wait_set_output = false; + + // Invoke the wakeup callback before clearing the wait objects + if (thread->wakeup_callback) + thread->wakeup_callback(ThreadWakeupReason::Timeout, thread, nullptr); + // Remove the thread from each of its waiting objects' waitlists for (auto& object : thread->wait_objects) object->RemoveWaitingThread(thread.get()); thread->wait_objects.clear(); - thread->SetWaitSynchronizationResult(RESULT_TIMEOUT); } thread->ResumeFromWait(); @@ -278,6 +281,9 @@ void Thread::ResumeFromWait() { break; case THREADSTATUS_READY: + // The thread's wakeup callback must have already been cleared when the thread was first + // awoken. + ASSERT(wakeup_callback == nullptr); // If the thread is waiting on multiple wait objects, it might be awoken more than once // before actually resuming. We can ignore subsequent wakeups if the thread status has // already been set to THREADSTATUS_READY. @@ -293,6 +299,8 @@ void Thread::ResumeFromWait() { return; } + wakeup_callback = nullptr; + ready_queue.push_back(current_priority, this); status = THREADSTATUS_READY; Core::System::GetInstance().PrepareReschedule(); @@ -394,7 +402,6 @@ ResultVal> Thread::Create(std::string name, VAddr entry_point, thread->nominal_priority = thread->current_priority = priority; thread->last_running_ticks = CoreTiming::GetTicks(); thread->processor_id = processor_id; - thread->wait_set_output = false; thread->wait_objects.clear(); thread->wait_address = 0; thread->name = std::move(name); -- cgit v1.2.3 From a13ab958cbba75bc9abd1ca50f3030a10a75784e Mon Sep 17 00:00:00 2001 From: Huw Pascoe Date: Wed, 27 Sep 2017 00:26:09 +0100 Subject: Fixed type conversion ambiguity --- src/core/hle/kernel/thread.cpp | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) (limited to 'src/core/hle/kernel/thread.cpp') diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index 1033f8552..11f7d2127 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -111,7 +111,7 @@ void Thread::Stop() { Thread* ArbitrateHighestPriorityThread(u32 address) { Thread* highest_priority_thread = nullptr; - s32 priority = THREADPRIO_LOWEST; + u32 priority = THREADPRIO_LOWEST; // Iterate through threads, find highest priority thread that is waiting to be arbitrated... for (auto& thread : thread_list) { @@ -311,7 +311,7 @@ static void DebugThreadQueue() { } for (auto& t : thread_list) { - s32 priority = ready_queue.contains(t.get()); + u32 priority = ready_queue.contains(t.get()); if (priority != -1) { LOG_DEBUG(Kernel, "0x%02X %u", priority, t->GetObjectId()); } @@ -422,7 +422,7 @@ ResultVal> Thread::Create(std::string name, VAddr entry_point, return ERR_OUT_OF_MEMORY; } - u32 offset = linheap_memory->size(); + size_t offset = linheap_memory->size(); // Allocate some memory from the end of the linear heap for this region. linheap_memory->insert(linheap_memory->end(), Memory::PAGE_SIZE, 0); @@ -430,7 +430,7 @@ ResultVal> Thread::Create(std::string name, VAddr entry_point, owner_process->linear_heap_used += Memory::PAGE_SIZE; tls_slots.emplace_back(0); // The page is completely available at the start - available_page = tls_slots.size() - 1; + available_page = static_cast(tls_slots.size() - 1); available_slot = 0; // Use the first slot in the new page auto& vm_manager = owner_process->vm_manager; @@ -457,7 +457,7 @@ ResultVal> Thread::Create(std::string name, VAddr entry_point, return MakeResult>(std::move(thread)); } -void Thread::SetPriority(s32 priority) { +void Thread::SetPriority(u32 priority) { ASSERT_MSG(priority <= THREADPRIO_LOWEST && priority >= THREADPRIO_HIGHEST, "Invalid priority value."); // If thread was ready, adjust queues @@ -470,7 +470,7 @@ void Thread::SetPriority(s32 priority) { } void Thread::UpdatePriority() { - s32 best_priority = nominal_priority; + u32 best_priority = nominal_priority; for (auto& mutex : held_mutexes) { if (mutex->priority < best_priority) best_priority = mutex->priority; @@ -478,7 +478,7 @@ void Thread::UpdatePriority() { BoostPriority(best_priority); } -void Thread::BoostPriority(s32 priority) { +void Thread::BoostPriority(u32 priority) { // If thread was ready, adjust queues if (status == THREADSTATUS_READY) ready_queue.move(this, current_priority, priority); @@ -487,7 +487,7 @@ void Thread::BoostPriority(s32 priority) { current_priority = priority; } -SharedPtr SetupMainThread(u32 entry_point, s32 priority, SharedPtr owner_process) { +SharedPtr SetupMainThread(u32 entry_point, u32 priority, SharedPtr owner_process) { // Initialize new "main" thread auto thread_res = Thread::Create("main", entry_point, priority, 0, THREADPROCESSORID_0, Memory::HEAP_VADDR_END, owner_process); @@ -531,7 +531,7 @@ void Thread::SetWaitSynchronizationOutput(s32 output) { s32 Thread::GetWaitObjectIndex(WaitObject* object) const { ASSERT_MSG(!wait_objects.empty(), "Thread is not waiting for anything"); auto match = std::find(wait_objects.rbegin(), wait_objects.rend(), object); - return std::distance(match, wait_objects.rend()) - 1; + return static_cast(std::distance(match, wait_objects.rend()) - 1); } //////////////////////////////////////////////////////////////////////////////////////////////////// -- cgit v1.2.3 From 8217ed7acb71bfa574e0a29e69b902a0c539c814 Mon Sep 17 00:00:00 2001 From: Subv Date: Fri, 29 Sep 2017 14:47:52 -0500 Subject: Kernel/Thread: Added a helper function to get a thread's command buffer VAddr. --- src/core/hle/kernel/thread.cpp | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'src/core/hle/kernel/thread.cpp') diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index 11f7d2127..6ebc8c151 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -534,6 +534,12 @@ s32 Thread::GetWaitObjectIndex(WaitObject* object) const { return static_cast(std::distance(match, wait_objects.rend()) - 1); } +VAddr Thread::GetCommandBufferAddress() const { + // Offset from the start of TLS at which the IPC command buffer begins. + static constexpr int CommandHeaderOffset = 0x80; + return GetTLSAddress() + CommandHeaderOffset; +} + //////////////////////////////////////////////////////////////////////////////////////////////////// void ThreadingInit() { -- cgit v1.2.3