- removed HLE mem "hack" and replaced with kernel mem region

- added a helper function for getting command buffer for services
- fixed bug where GSP DMA was incorrectly being done in DataSynchronizationBarrier (instead of gsp_TriggerCmdReqQueue)

src/core/hle/coprocessor.cpp

@@ -9,39 +9,15 @@
 
 namespace HLE {
 
-enum {
-    CMD_GX_REQUEST_DMA  = 0x00000000,
-};
-
 /// Data synchronization barrier
-u32 DataSynchronizationBarrier(u32* command_buffer) {
-    u32 command = command_buffer[0];
-
-    switch (command) {
-
-    case CMD_GX_REQUEST_DMA:
-        {
-            u32* src = (u32*)Memory::GetPointer(command_buffer[1]);
-            u32* dst = (u32*)Memory::GetPointer(command_buffer[2]);
-            u32 size = command_buffer[3];
-            memcpy(dst, src, size);
-        }
-        break;
-
-    default:
-        ERROR_LOG(OSHLE, "MRC::DataSynchronizationBarrier unknown command 0x%08X", command);
-        return -1;
-    }
-
+u32 DataSynchronizationBarrier() {
     return 0;
 }
 
 /// Returns the coprocessor (in this case, syscore) command buffer pointer
 Addr GetThreadCommandBuffer() {
     // Called on insruction: mrc p15, 0, r0, c13, c0, 3
-    // Returns an address in OSHLE memory for the CPU to read/write to
-    RETURN(CMD_BUFFER_ADDR);
-    return CMD_BUFFER_ADDR;
+    return Memory::KERNEL_MEMORY_VADDR;
 }
 
 /// Call an MCR (move to coprocessor from ARM register) instruction in HLE
@@ -49,7 +25,7 @@ s32 CallMCR(u32 instruction, u32 value) {
     CoprocessorOperation operation = (CoprocessorOperation)((instruction >> 20) & 0xFF);
     ERROR_LOG(OSHLE, "unimplemented MCR instruction=0x%08X, operation=%02X, value=%08X", 
         instruction, operation, value);
-    return -1;
+    return 0;
 }
 
 /// Call an MRC (move to ARM register from coprocessor) instruction in HLE
@@ -59,7 +35,7 @@ s32 CallMRC(u32 instruction) {
     switch (operation) {
 
     case DATA_SYNCHRONIZATION_BARRIER:
-        return DataSynchronizationBarrier((u32*)Memory::GetPointer(PARAM(0)));
+        return DataSynchronizationBarrier();
 
     case CALL_GET_THREAD_COMMAND_BUFFER:
         return GetThreadCommandBuffer();
@@ -68,7 +44,7 @@ s32 CallMRC(u32 instruction) {
         ERROR_LOG(OSHLE, "unimplemented MRC instruction 0x%08X", instruction);
         break;
     }
-    return -1;
+    return 0;
 }
 
 } // namespace

src/core/hle/hle.cpp

@@ -15,49 +15,6 @@ namespace HLE {
 
 static std::vector<ModuleDef> g_module_db;
 
-u8* g_command_buffer = NULL;    ///< Command buffer used for sharing between appcore and syscore
-
-// Read from memory used by CTROS HLE functions
-template <typename T>
-inline void Read(T &var, const u32 addr) {
-    if (addr >= HLE::CMD_BUFFER_ADDR && addr < HLE::CMD_BUFFER_ADDR_END) {
-        var = *((const T*)&g_command_buffer[addr & CMD_BUFFER_MASK]);
-    } else {
-        ERROR_LOG(HLE, "unknown read from address %08X", addr);
-    }
-}
-
-// Write to memory used by CTROS HLE functions
-template <typename T>
-inline void Write(u32 addr, const T data) {
-    if (addr >= HLE::CMD_BUFFER_ADDR && addr < HLE::CMD_BUFFER_ADDR_END) {
-        *(T*)&g_command_buffer[addr & CMD_BUFFER_MASK] = data;
-    } else {
-        ERROR_LOG(HLE, "unknown write to address %08X", addr);
-    }
-}
-
-u8 *GetPointer(const u32 addr) {
-    if (addr >= HLE::CMD_BUFFER_ADDR && addr < HLE::CMD_BUFFER_ADDR_END) {
-        return g_command_buffer + (addr & CMD_BUFFER_MASK);
-    } else {
-        ERROR_LOG(HLE, "unknown pointer from address %08X", addr);
-        return 0;
-    }
-}
-
-// Explicitly instantiate template functions because we aren't defining this in the header:
-
-template void Read<u64>(u64 &var, const u32 addr);
-template void Read<u32>(u32 &var, const u32 addr);
-template void Read<u16>(u16 &var, const u32 addr);
-template void Read<u8>(u8 &var, const u32 addr);
-
-template void Write<u64>(u32 addr, const u64 data);
-template void Write<u32>(u32 addr, const u32 data);
-template void Write<u16>(u32 addr, const u16 data);
-template void Write<u8>(u32 addr, const u8 data);
-
 const FunctionDef* GetSyscallInfo(u32 opcode) {
     u32 func_num = opcode & 0xFFFFFF; // 8 bits
     if (func_num > 0xFF) {
@@ -91,8 +48,6 @@ void RegisterAllModules() {
 
 void Init() {
     Service::Init();
-
-    g_command_buffer = new u8[CMD_BUFFER_SIZE];
     
     RegisterAllModules();
 
@@ -102,8 +57,6 @@ void Init() {
 void Shutdown() {
     Service::Shutdown();
 
-    delete g_command_buffer;
-
     g_module_db.clear();
 
     NOTICE_LOG(HLE, "shutdown OK");

src/core/hle/hle.h

@@ -17,13 +17,6 @@
 
 namespace HLE {
 
-enum {
-    CMD_BUFFER_ADDR     = 0xA0010000,    ///< Totally arbitrary unused address space
-    CMD_BUFFER_SIZE     = 0x10000,
-    CMD_BUFFER_MASK     = (CMD_BUFFER_SIZE - 1),
-    CMD_BUFFER_ADDR_END = (CMD_BUFFER_ADDR + CMD_BUFFER_SIZE),
-};
-
 typedef u32 Addr;
 typedef void (*Func)();
 
@@ -39,20 +32,6 @@ struct ModuleDef {
     const FunctionDef*  func_table;
 };
 
-// Read from memory used by CTROS HLE functions
-template <typename T>
-inline void Read(T &var, const u32 addr);
-
-// Write to memory used by CTROS HLE functions
-template <typename T>
-inline void Write(u32 addr, const T data);
-
-u8* GetPointer(const u32 Address);
-
-inline const char* GetCharPointer(const u32 address) {
-    return (const char *)GetPointer(address);
-}
-
 void RegisterModule(std::string name, int num_functions, const FunctionDef *func_table);
 
 void CallSyscall(u32 opcode);

src/core/hle/service/apt.cpp

@@ -18,7 +18,7 @@ void Initialize(Service::Interface* self) {
 }
 
 void GetLockHandle(Service::Interface* self) {
-    u32* cmd_buff = (u32*)HLE::GetPointer(HLE::CMD_BUFFER_ADDR + Service::kCommandHeaderOffset);
+    u32* cmd_buff = Service::GetCommandBuffer();
     cmd_buff[5] = 0x00000000; // TODO: This should be an actual mutex handle
 }
 

src/core/hle/service/gsp.cpp

@@ -4,6 +4,7 @@
 
 
 #include "common/log.h"
+#include "common/bit_field.h"
 
 #include "core/mem_map.h"
 #include "core/hle/hle.h"
@@ -11,11 +12,57 @@
 
 #include "core/hw/lcd.h"
 
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+/// GSP shared memory GX command buffer header
+union GX_CmdBufferHeader {
+    u32 hex;
+
+    // Current command index. This index is updated by GSP module after loading the command data, 
+    // right before the command is processed. When this index is updated by GSP module, the total 
+    // commands field is decreased by one as well.
+    BitField<0,8,u32>   index;
+    
+    // Total commands to process, must not be value 0 when GSP module handles commands. This must be
+    // <=15 when writing a command to shared memory. This is incremented by the application when 
+    // writing a command to shared memory, after increasing this value TriggerCmdReqQueue is only 
+    // used if this field is value 1.
+    BitField<8,8,u32>  number_commands;
+    
+    // Must not be value 1. When the error-code u32 is set, this u8 is set to value 0x80.
+    BitField<16,8,u32> unk_0;
+
+    // Bit 0 must not be set
+    BitField<24,8,u32> unk_1;
+};
+
+/// Gets the address of the start (header) of a command buffer in GSP shared memory
+static inline u32 GX_GetCmdBufferAddress(u32 thread_id) {
+    return (0x10002000 + 0x800 + (thread_id * 0x200));
+}
+
+/// Gets a pointer to the start (header) of a command buffer in GSP shared memory
+static inline u8* GX_GetCmdBufferPointer(u32 thread_id, u32 offset=0) {
+    return Memory::GetPointer(GX_GetCmdBufferAddress(thread_id) + offset);
+}
+
+/// Finishes execution of a GSP command
+void GX_FinishCommand(u32 thread_id) {
+    GX_CmdBufferHeader* header = (GX_CmdBufferHeader*)GX_GetCmdBufferPointer(thread_id);
+    header->number_commands = header->number_commands - 1;
+}
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Namespace GSP_GPU
 
 namespace GSP_GPU {
 
+u32 g_thread_id = 0;
+
+enum {
+    CMD_GX_REQUEST_DMA  = 0x00000000,
+};
+
 enum {
     REG_FRAMEBUFFER_1   = 0x00400468,
     REG_FRAMEBUFFER_2   = 0x00400494,
@@ -26,7 +73,7 @@ void ReadHWRegs(Service::Interface* self) {
     static const u32 framebuffer_1[] = {LCD::PADDR_VRAM_TOP_LEFT_FRAME1, LCD::PADDR_VRAM_TOP_RIGHT_FRAME1};
     static const u32 framebuffer_2[] = {LCD::PADDR_VRAM_TOP_LEFT_FRAME2, LCD::PADDR_VRAM_TOP_RIGHT_FRAME2};
 
-    u32* cmd_buff = (u32*)HLE::GetPointer(HLE::CMD_BUFFER_ADDR + Service::kCommandHeaderOffset);
+    u32* cmd_buff = Service::GetCommandBuffer();
     u32 reg_addr = cmd_buff[1];
     u32 size = cmd_buff[2];
     u32* dst = (u32*)Memory::GetPointer(cmd_buff[0x41]);
@@ -50,18 +97,37 @@ void ReadHWRegs(Service::Interface* self) {
         break;
 
     default:
-        ERROR_LOG(OSHLE, "GSP_GPU::ReadHWRegs unknown register read at address %08X", reg_addr);
+        ERROR_LOG(GSP, "ReadHWRegs unknown register read at address %08X", reg_addr);
     }
 
 }
 
 void RegisterInterruptRelayQueue(Service::Interface* self) {
-    u32* cmd_buff = (u32*)HLE::GetPointer(HLE::CMD_BUFFER_ADDR + Service::kCommandHeaderOffset);
+    u32* cmd_buff = Service::GetCommandBuffer();
     u32 flags = cmd_buff[1];
     u32 event_handle = cmd_buff[3]; // TODO(bunnei): Implement event handling
+    
+    cmd_buff[2] = g_thread_id;          // ThreadID
     cmd_buff[4] = self->NewHandle();
+}
+
+/// This triggers handling of the GX command written to the command buffer in shared memory.
+void TriggerCmdReqQueue(Service::Interface* self) {
+    GX_CmdBufferHeader* header = (GX_CmdBufferHeader*)GX_GetCmdBufferPointer(g_thread_id);
+    u32* cmd_buff = (u32*)GX_GetCmdBufferPointer(g_thread_id, 0x20 + (header->index * 0x20));
 
-    return;
+    switch (cmd_buff[0]) {
+
+    // GX request DMA - typically used for copying memory from GSP heap to VRAM
+    case CMD_GX_REQUEST_DMA:
+        memcpy(Memory::GetPointer(cmd_buff[2]), Memory::GetPointer(cmd_buff[1]), cmd_buff[3]);
+        break;
+
+    default:
+        ERROR_LOG(GSP, "TriggerCmdReqQueue unknown command 0x%08X", cmd_buff[0]);
+    }
+    
+    GX_FinishCommand(g_thread_id);
 }
 
 const Interface::FunctionInfo FunctionTable[] = {
@@ -76,7 +142,7 @@ const Interface::FunctionInfo FunctionTable[] = {
     {0x00090082, NULL,                          "InvalidateDataCache"},
     {0x000A0044, NULL,                          "RegisterInterruptEvents"},
     {0x000B0040, NULL,                          "SetLcdForceBlack"},
-    {0x000C0000, NULL,                          "TriggerCmdReqQueue"},
+    {0x000C0000, TriggerCmdReqQueue,            "TriggerCmdReqQueue"},
     {0x000D0140, NULL,                          "SetDisplayTransfer"},
     {0x000E0180, NULL,                          "SetTextureCopy"},
     {0x000F0200, NULL,                          "SetMemoryFill"},

src/core/hle/service/service.h

@@ -10,6 +10,7 @@
 
 #include "common/common.h"
 #include "common/common_types.h"
+#include "core/mem_map.h"
 #include "core/hle/syscall.h"
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -22,6 +23,15 @@ typedef s32 NativeUID;                          ///< Native handle for a service
 static const int kMaxPortSize           = 0x08; ///< Maximum size of a port name (8 characters)
 static const int kCommandHeaderOffset   = 0x80; ///< Offset into command buffer of header
 
+/**
+ * Returns a pointer to the command buffer in kernel memory
+ * @param offset Optional offset into command buffer
+ * @return Pointer to command buffer
+ */
+inline static u32* GetCommandBuffer(const int offset=0) {
+    return (u32*)Memory::GetPointer(Memory::KERNEL_MEMORY_VADDR + kCommandHeaderOffset + offset);
+}
+
 class Manager;
 
 /// Interface to a CTROS service
@@ -81,7 +91,7 @@ public:
      * @return Return result of svcSendSyncRequest passed back to user app
      */
     Syscall::Result Sync() {
-        u32* cmd_buff = (u32*)HLE::GetPointer(HLE::CMD_BUFFER_ADDR + kCommandHeaderOffset);
+        u32* cmd_buff = GetCommandBuffer();
         auto itr = m_functions.find(cmd_buff[0]);
 
         if (itr == m_functions.end()) {

src/core/hle/service/srv.cpp

@@ -18,7 +18,7 @@ void Initialize(Service::Interface* self) {
 
 void GetServiceHandle(Service::Interface* self) {
     Syscall::Result res = 0;
-    u32* cmd_buff = (u32*)HLE::GetPointer(HLE::CMD_BUFFER_ADDR + Service::kCommandHeaderOffset);
+    u32* cmd_buff = Service::GetCommandBuffer();
 
     std::string port_name = std::string((const char*)&cmd_buff[1], 0, Service::kMaxPortSize);
     Service::Interface* service = Service::g_manager->FetchFromPortName(port_name);

src/core/mem_map_funcs.cpp

@@ -48,11 +48,9 @@ inline void _Read(T &var, const u32 addr) {
 
     const u32 vaddr = _VirtualAddress(addr);
 
-    // Memory allocated for HLE use that can be addressed from the emulated application
-    // The primary use of this is sharing a commandbuffer between the HLE OS (syscore) and the LLE
-    // core running the user application (appcore)
-    if (vaddr >= HLE::CMD_BUFFER_ADDR && vaddr < HLE::CMD_BUFFER_ADDR_END) {
-        HLE::Read<T>(var, vaddr);
+    // Kernel memory command buffer
+    if (vaddr >= KERNEL_MEMORY_VADDR && vaddr < KERNEL_MEMORY_VADDR_END) {
+        var = *((const T*)&g_kernel_mem[vaddr & KERNEL_MEMORY_MASK]);
 
     // Hardware I/O register reads
     // 0x10XXXXXX- is physical address space, 0x1EXXXXXX is virtual address space
@@ -92,11 +90,9 @@ template <typename T>
 inline void _Write(u32 addr, const T data) {
     u32 vaddr = _VirtualAddress(addr);
     
-    // Memory allocated for HLE use that can be addressed from the emulated application
-    // The primary use of this is sharing a commandbuffer between the HLE OS (syscore) and the LLE
-    // core running the user application (appcore)
-    if (vaddr >= HLE::CMD_BUFFER_ADDR && vaddr < HLE::CMD_BUFFER_ADDR_END) {
-        HLE::Write<T>(vaddr, data);
+    // Kernel memory command buffer
+    if (vaddr >= KERNEL_MEMORY_VADDR && vaddr < KERNEL_MEMORY_VADDR_END) {
+        *(T*)&g_kernel_mem[vaddr & KERNEL_MEMORY_MASK] = data;
 
     // Hardware I/O register writes
     // 0x10XXXXXX- is physical address space, 0x1EXXXXXX is virtual address space
@@ -140,8 +136,12 @@ inline void _Write(u32 addr, const T data) {
 u8 *GetPointer(const u32 addr) {
     const u32 vaddr = _VirtualAddress(addr);
 
+    // Kernel memory command buffer
+    if (vaddr >= KERNEL_MEMORY_VADDR && vaddr < KERNEL_MEMORY_VADDR_END) {
+        return g_kernel_mem + (vaddr & KERNEL_MEMORY_MASK);
+
     // ExeFS:/.code is loaded here
-    if ((vaddr >= EXEFS_CODE_VADDR)  && (vaddr < EXEFS_CODE_VADDR_END)) {
+    } else if ((vaddr >= EXEFS_CODE_VADDR)  && (vaddr < EXEFS_CODE_VADDR_END)) {
         return g_exefs_code + (vaddr & EXEFS_CODE_MASK);
 
     // FCRAM - GSP heap