core: Fixup linux support (#195)

* tls: Implement for linux

* core: Implement memory manager for linux

src/audio_core/sdl_audio.cpp

@@ -10,10 +10,6 @@
 
 namespace Audio {
 
-int SDLAudio::AudioInit() {
-    return SDL_InitSubSystem(SDL_INIT_AUDIO);
-}
-
 int SDLAudio::AudioOutOpen(int type, u32 samples_num, u32 freq,
                            Libraries::AudioOut::OrbisAudioOutParam format) {
     using Libraries::AudioOut::OrbisAudioOutParam;

src/audio_core/sdl_audio.h

@@ -14,7 +14,6 @@ public:
     SDLAudio() = default;
     virtual ~SDLAudio() = default;
 
-    int AudioInit();
     int AudioOutOpen(int type, u32 samples_num, u32 freq,
                      Libraries::AudioOut::OrbisAudioOutParam format);
     s32 AudioOutOutput(s32 handle, const void* ptr);

src/common/io_file.cpp

@@ -183,9 +183,8 @@ void IOFile::Open(const fs::path& path, FileAccessMode mode, FileType type, File
 #endif
 
     if (!IsOpen()) {
-        const auto ec = std::error_code{errno, std::generic_category()};
-        LOG_ERROR(Common_Filesystem, "Failed to open the file at path={}, ec_message={}",
-                  PathToUTF8String(file_path), ec.message());
+        LOG_ERROR(Common_Filesystem, "Failed to open the file at path={}",
+                  PathToUTF8String(file_path));
     }
 }
 

src/common/path_util.cpp

@@ -1,10 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <algorithm>
-#include <sstream>
 #include <unordered_map>
-
 #include "common/logging/log.h"
 #include "common/path_util.h"
 

src/core/address_space.cpp

@@ -10,6 +10,7 @@
 #ifdef _WIN32
 #include <windows.h>
 #else
+#include <fcntl.h>
 #include <sys/mman.h>
 #endif
 
@@ -215,25 +216,96 @@ enum PosixPageProtection {
 
 struct AddressSpace::Impl {
     Impl() {
-        UNREACHABLE();
+        // Allocate virtual address placeholder for our address space.
+        void* hint_address = reinterpret_cast<void*>(SYSTEM_MANAGED_MIN);
+        virtual_size = SystemSize + UserSize;
+        virtual_base = reinterpret_cast<u8*>(
+            mmap(reinterpret_cast<void*>(hint_address), virtual_size, PROT_READ | PROT_WRITE,
+                 MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0));
+        if (virtual_base == MAP_FAILED) {
+            LOG_CRITICAL(Kernel_Vmm, "mmap failed: {}", strerror(errno));
+            throw std::bad_alloc{};
+        }
+        madvise(virtual_base, virtual_size, MADV_HUGEPAGE);
+
+        backing_fd = memfd_create("BackingDmem", 0);
+        if (backing_fd < 0) {
+            LOG_CRITICAL(Kernel_Vmm, "memfd_create failed: {}", strerror(errno));
+            throw std::bad_alloc{};
+        }
+
+        // Defined to extend the file with zeros
+        int ret = ftruncate(backing_fd, BackingSize);
+        if (ret != 0) {
+            LOG_CRITICAL(Kernel_Vmm, "ftruncate failed with {}, are you out-of-memory?",
+                         strerror(errno));
+            throw std::bad_alloc{};
+        }
+
+        // Map backing dmem handle.
+        backing_base = static_cast<u8*>(
+            mmap(nullptr, BackingSize, PROT_READ | PROT_WRITE, MAP_SHARED, backing_fd, 0));
+        if (backing_base == MAP_FAILED) {
+            LOG_CRITICAL(Kernel_Vmm, "mmap failed: {}", strerror(errno));
+            throw std::bad_alloc{};
+        }
+
+        const VAddr start_addr = reinterpret_cast<VAddr>(virtual_base);
+        m_free_regions.insert({start_addr, start_addr + virtual_size});
     }
 
     void* Map(VAddr virtual_addr, PAddr phys_addr, size_t size, PosixPageProtection prot) {
-        UNREACHABLE();
-        return nullptr;
+        m_free_regions.subtract({virtual_addr, virtual_addr + size});
+        const int fd = phys_addr != -1 ? backing_fd : -1;
+        const int host_offset = phys_addr != -1 ? phys_addr : 0;
+        const int flag = phys_addr != -1 ? MAP_SHARED : (MAP_ANONYMOUS | MAP_PRIVATE);
+        void* ret = mmap(reinterpret_cast<void*>(virtual_addr), size, prot, MAP_FIXED | flag, fd,
+                         host_offset);
+        ASSERT_MSG(ret != MAP_FAILED, "mmap failed: {}", strerror(errno));
+        return ret;
     }
 
     void Unmap(VAddr virtual_addr, PAddr phys_addr, size_t size) {
-        UNREACHABLE();
+        // Check to see if we are adjacent to any regions.
+        auto start_address = virtual_addr;
+        auto end_address = start_address + size;
+        auto it = m_free_regions.find({start_address - 1, end_address + 1});
+
+        // If we are, join with them, ensuring we stay in bounds.
+        if (it != m_free_regions.end()) {
+            start_address = std::min(start_address, it->lower());
+            end_address = std::max(end_address, it->upper());
+        }
+
+        // Free the relevant region.
+        m_free_regions.insert({start_address, end_address});
+
+        // Return the adjusted pointers.
+        void* ret = mmap(reinterpret_cast<void*>(start_address), end_address - start_address,
+                         PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
+        ASSERT_MSG(ret != MAP_FAILED, "mmap failed: {}", strerror(errno));
     }
 
     void Protect(VAddr virtual_addr, size_t size, bool read, bool write, bool execute) {
-        UNREACHABLE();
+        int flags = PROT_NONE;
+        if (read) {
+            flags |= PROT_READ;
+        }
+        if (write) {
+            flags |= PROT_WRITE;
+        }
+        if (execute) {
+            flags |= PROT_EXEC;
+        }
+        int ret = mprotect(reinterpret_cast<void*>(virtual_addr), size, flags);
+        ASSERT_MSG(ret == 0, "mprotect failed: {}", strerror(errno));
     }
 
+    int backing_fd;
     u8* backing_base{};
     u8* virtual_base{};
     size_t virtual_size{};
+    boost::icl::interval_set<VAddr> m_free_regions;
 };
 #endif
 

src/core/file_format/pkg.cpp

@@ -1,11 +1,10 @@
 // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <fstream>
 #include <zlib-ng.h>
 #include "common/io_file.h"
-#include "pkg.h"
-#include "pkg_type.h"
+#include "core/file_format/pkg.h"
+#include "core/file_format/pkg_type.h"
 
 static void DecompressPFSC(std::span<const char> compressed_data,
                            std::span<char> decompressed_data) {

src/core/file_format/pkg.h

@@ -4,7 +4,6 @@
 #pragma once
 
 #include <array>
-#include <cstdio>
 #include <filesystem>
 #include <string>
 #include <unordered_map>

src/core/file_format/psf.cpp

@@ -2,11 +2,8 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <cstring>
-#include <fstream>
-#include <iostream>
-
 #include "common/io_file.h"
-#include "psf.h"
+#include "core/file_format/psf.h"
 
 PSF::PSF() = default;
 

src/core/libraries/audio/audioout.cpp

@@ -192,9 +192,8 @@ int PS4_SYSV_ABI sceAudioOutGetSystemState() {
 
 int PS4_SYSV_ABI sceAudioOutInit() {
     audio = std::make_unique<Audio::SDLAudio>();
-    u32 result = audio->AudioInit() == 0 ? ORBIS_OK : ORBIS_AUDIO_OUT_ERROR_NOT_INIT;
-    LOG_INFO(Lib_AudioOut, "AudioInit returned {}", result);
-    return result;
+    LOG_INFO(Lib_AudioOut, "called");
+    return ORBIS_OK;
 }
 
 int PS4_SYSV_ABI sceAudioOutInitIpmiGetSession() {

src/core/libraries/kernel/thread_management.cpp

@@ -393,6 +393,8 @@ void* createMutex(void* addr) {
     if (addr == nullptr || *static_cast<ScePthreadMutex*>(addr) != nullptr) {
         return addr;
     }
+    static std::mutex mutex;
+    std::scoped_lock lk{mutex};
     auto vaddr = reinterpret_cast<u64>(addr);
 
     std::string name = fmt::format("mutex{:#x}", vaddr);
@@ -464,7 +466,7 @@ int PS4_SYSV_ABI scePthreadMutexattrInit(ScePthreadMutexattr* attr) {
 
     int result = pthread_mutexattr_init(&(*attr)->pth_mutex_attr);
 
-    result = (result == 0 ? scePthreadMutexattrSettype(attr, 1) : result);
+    result = (result == 0 ? scePthreadMutexattrSettype(attr, 2) : result);
     result = (result == 0 ? scePthreadMutexattrSetprotocol(attr, 0) : result);
 
     switch (result) {
@@ -1165,6 +1167,10 @@ int PS4_SYSV_ABI posix_pthread_create_name_np(ScePthread* thread, const ScePthre
     return result;
 }
 
+int PS4_SYSV_ABI scePthreadOnce(int* once_control, void (*init_routine)(void)) {
+    return pthread_once(reinterpret_cast<pthread_once_t*>(once_control), init_routine);
+}
+
 void pthreadSymbolsRegister(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("4+h9EzwKF4I", "libkernel", 1, "libkernel", 1, 1, scePthreadAttrSetschedpolicy);
     LIB_FUNCTION("-Wreprtu0Qs", "libkernel", 1, "libkernel", 1, 1, scePthreadAttrSetdetachstate);
@@ -1191,6 +1197,7 @@ void pthreadSymbolsRegister(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("6UgtwV+0zb4", "libkernel", 1, "libkernel", 1, 1, scePthreadCreate);
     LIB_FUNCTION("T72hz6ffq08", "libkernel", 1, "libkernel", 1, 1, scePthreadYield);
     LIB_FUNCTION("-quPa4SEJUw", "libkernel", 1, "libkernel", 1, 1, scePthreadAttrGetstack);
+    LIB_FUNCTION("14bOACANTBo", "libkernel", 1, "libkernel", 1, 1, scePthreadOnce);
 
     // mutex calls
     LIB_FUNCTION("cmo1RIYva9o", "libkernel", 1, "libkernel", 1, 1, scePthreadMutexInit);

src/core/libraries/kernel/thread_management.h

@@ -27,7 +27,7 @@ struct PthreadMutexattrInternal;
 struct PthreadCondInternal;
 struct PthreadCondAttrInternal;
 struct PthreadRwInternal;
-struct PthreadRwLockAttrInernal;
+struct PthreadRwLockAttrInternal;
 
 using SceKernelSchedParam = ::sched_param;
 using ScePthread = PthreadInternal*;
@@ -37,7 +37,7 @@ using ScePthreadMutexattr = PthreadMutexattrInternal*;
 using ScePthreadCond = PthreadCondInternal*;
 using ScePthreadCondattr = PthreadCondAttrInternal*;
 using OrbisPthreadRwlock = PthreadRwInternal*;
-using OrbisPthreadRwlockattr = PthreadRwLockAttrInernal*;
+using OrbisPthreadRwlockattr = PthreadRwLockAttrInternal*;
 
 using pthreadEntryFunc = PS4_SYSV_ABI void* (*)(void*);
 
@@ -86,7 +86,7 @@ struct PthreadCondAttrInternal {
     pthread_condattr_t cond_attr;
 };
 
-struct PthreadRwLockAttrInernal {
+struct PthreadRwLockAttrInternal {
     u8 reserved[64];
     pthread_rwlockattr_t attr_rwlock;
     int type;

src/core/libraries/kernel/threads/kernel_threads_rwlock.cpp

@@ -121,7 +121,7 @@ int PS4_SYSV_ABI posix_pthread_rwlockattr_gettype_np() {
 }
 
 int PS4_SYSV_ABI posix_pthread_rwlockattr_init(OrbisPthreadRwlockattr* attr) {
-    *attr = new PthreadRwLockAttrInernal{};
+    *attr = new PthreadRwLockAttrInternal{};
     int result = pthread_rwlockattr_init(&(*attr)->attr_rwlock);
     if (result != 0) {
         LOG_ERROR(Kernel_Pthread, "posix_pthread_rwlockattr_init: error = {}", result);
@@ -161,7 +161,7 @@ int PS4_SYSV_ABI scePthreadRwlockattrGettype() {
 }
 
 int PS4_SYSV_ABI scePthreadRwlockattrInit(OrbisPthreadRwlockattr* attr) {
-    *attr = new PthreadRwLockAttrInernal{};
+    *attr = new PthreadRwLockAttrInternal{};
     int result = pthread_rwlockattr_init(&(*attr)->attr_rwlock);
     if (result != 0) {
         LOG_ERROR(Kernel_Pthread, "scePthreadRwlockattrInit: error = {}", result);

src/core/tls.cpp

@@ -8,6 +8,8 @@
 
 #ifdef _WIN32
 #include <windows.h>
+#else
+#include <pthread.h>
 #endif
 
 namespace Core {
@@ -53,9 +55,83 @@ Tcb* GetTcbBase() {
     return reinterpret_cast<Tcb*>(TlsGetValue(slot));
 }
 
-void PatchTLS(u64 segment_addr, u64 segment_size, Xbyak::CodeGenerator& c) {
-    using namespace Xbyak::util;
+static void AllocTcbKey() {
+    slot = TlsAlloc();
+}
 
+static void PatchFsAccess(u8* code, const TLSPattern& tls_pattern, Xbyak::CodeGenerator& c) {
+    using namespace Xbyak::util;
+    const auto total_size = tls_pattern.pattern_size + tls_pattern.imm_size;
+
+    // Replace mov instruction with near jump to the trampoline.
+    static constexpr u32 NearJmpSize = 5;
+    auto patch = Xbyak::CodeGenerator(total_size, code);
+    patch.jmp(c.getCurr(), Xbyak::CodeGenerator::LabelType::T_NEAR);
+    patch.nop(total_size - NearJmpSize);
+
+    // Write the trampoline.
+    // The following logic is based on the wine implementation of TlsGetValue
+    // https://github.com/wine-mirror/wine/blob/a27b9551/dlls/kernelbase/thread.c#L719
+    static constexpr u32 TlsSlotsOffset = 0x1480;
+    static constexpr u32 TlsExpansionSlotsOffset = 0x1780;
+    static constexpr u32 TlsMinimumAvailable = 64;
+    const u32 teb_offset = slot < TlsMinimumAvailable ? TlsSlotsOffset : TlsExpansionSlotsOffset;
+    const u32 tls_index = slot < TlsMinimumAvailable ? slot : slot - TlsMinimumAvailable;
+
+    const auto target_reg = Xbyak::Reg64(tls_pattern.target_reg);
+    c.mov(target_reg, teb_offset);
+    c.putSeg(gs);
+    c.mov(target_reg, ptr[target_reg]); // Load the pointer to the table of tls slots.
+    c.mov(target_reg,
+          qword[target_reg + tls_index * sizeof(LPVOID)]); // Load the pointer to our buffer.
+    c.jmp(code + total_size); // Return to the instruction right after the mov.
+}
+
+#else
+
+static pthread_key_t slot = 0;
+
+void SetTcbBase(void* image_address) {
+    ASSERT(pthread_setspecific(slot, image_address) == 0);
+}
+
+Tcb* GetTcbBase() {
+    return reinterpret_cast<Tcb*>(pthread_getspecific(slot));
+}
+
+static void AllocTcbKey() {
+    slot = pthread_key_create(&slot, nullptr);
+}
+
+static void PatchFsAccess(u8* code, const TLSPattern& tls_pattern, Xbyak::CodeGenerator& c) {
+    using namespace Xbyak::util;
+    const auto total_size = tls_pattern.pattern_size + tls_pattern.imm_size;
+
+    // Replace mov instruction with near jump to the trampoline.
+    static constexpr u32 NearJmpSize = 5;
+    auto patch = Xbyak::CodeGenerator(total_size, code);
+    patch.jmp(c.getCurr(), Xbyak::CodeGenerator::LabelType::T_NEAR);
+    patch.nop(total_size - NearJmpSize);
+
+    // Write the trampoline.
+    // The following logic is based on the glibc implementation of pthread_getspecific
+    // https://github.com/bminor/glibc/blob/29807a27/nptl/pthread_getspecific.c#L23
+    static constexpr u32 PthreadKeySecondLevelSize = 32;
+    static constexpr u32 SpecificFirstBlockOffset = 0x308;
+    static constexpr u32 SelfInTcbheadOffset = 16;
+    static constexpr u32 PthreadKeyDataSize = 16;
+    ASSERT(slot < PthreadKeySecondLevelSize);
+
+    const auto target_reg = Xbyak::Reg64(tls_pattern.target_reg);
+    c.putSeg(fs);
+    c.mov(target_reg, qword[SelfInTcbheadOffset]); // Load self member pointer of tcbhead_t.
+    c.add(target_reg, SpecificFirstBlockOffset + sizeof(uintptr_t) + slot * PthreadKeyDataSize);
+    c.jmp(code + total_size); // Return to the instruction right after the mov.
+}
+
+#endif
+
+void PatchTLS(u64 segment_addr, u64 segment_size, Xbyak::CodeGenerator& c) {
     u8* code = reinterpret_cast<u8*>(segment_addr);
     auto remaining_size = segment_size;
 
@@ -89,7 +165,7 @@ void PatchTLS(u64 segment_addr, u64 segment_size, Xbyak::CodeGenerator& c) {
 
             // Allocate slot in the process if not done already.
             if (slot == 0) {
-                slot = TlsAlloc();
+                AllocTcbKey();
             }
 
             // Replace bogus instruction prefix with nops if it exists.
@@ -98,30 +174,8 @@ void PatchTLS(u64 segment_addr, u64 segment_size, Xbyak::CodeGenerator& c) {
                 patch.nop(BadPrefix.size());
             }
 
-            // Replace mov instruction with near jump to the trampoline.
-            static constexpr u32 NearJmpSize = 5;
-            auto patch = Xbyak::CodeGenerator(total_size, code);
-            patch.jmp(c.getCurr(), Xbyak::CodeGenerator::LabelType::T_NEAR);
-            patch.nop(total_size - NearJmpSize);
-
-            // Write the trampoline.
-            // The following logic is based on the wine implementation of TlsGetValue
-            // https://github.com/wine-mirror/wine/blob/a27b9551/dlls/kernelbase/thread.c#L719
-            static constexpr u32 TlsSlotsOffset = 0x1480;
-            static constexpr u32 TlsExpansionSlotsOffset = 0x1780;
-            static constexpr u32 TlsMinimumAvailable = 64;
-            const u32 teb_offset =
-                slot < TlsMinimumAvailable ? TlsSlotsOffset : TlsExpansionSlotsOffset;
-            const u32 tls_index = slot < TlsMinimumAvailable ? slot : slot - TlsMinimumAvailable;
-
-            const auto target_reg = Xbyak::Reg64(tls_pattern.target_reg);
-            c.mov(target_reg, teb_offset);
-            c.putSeg(gs);
-            c.mov(target_reg, ptr[target_reg]); // Load the pointer to the table of tls slots.
-            c.mov(
-                target_reg,
-                qword[target_reg + tls_index * sizeof(LPVOID)]); // Load the pointer to our buffer.
-            c.jmp(code + total_size); // Return to the instruction right after the mov.
+            // Patch access to FS register to a trampoline.
+            PatchFsAccess(code, tls_pattern, c);
 
             // Move ahead in module.
             code += total_size - 1;
@@ -133,20 +187,4 @@ void PatchTLS(u64 segment_addr, u64 segment_size, Xbyak::CodeGenerator& c) {
     }
 }
 
-#else
-
-void SetTcbBase(void* image_address) {
-    UNREACHABLE_MSG("Thread local storage is unimplemented on posix platforms!");
-}
-
-Tcb* GetTcbBase() {
-    UNREACHABLE_MSG("Thread local storage is unimplemented on posix platforms!");
-}
-
-void PatchTLS(u64 segment_addr, u64 segment_size, Xbyak::CodeGenerator& c) {
-    UNREACHABLE_MSG("Thread local storage is unimplemented on posix platforms!");
-}
-
-#endif
-
 } // namespace Core

src/emulator.cpp

@@ -40,10 +40,6 @@ Emulator::Emulator() : window{WindowWidth, WindowHeight, controller} {
     Common::Log::Initialize();
     Common::Log::Start();
 
-    // Start discord integration
-    discord_rpc.init();
-    discord_rpc.update(Discord::RPCStatus::Idling, "");
-
     // Initialize kernel and library facilities.
     Libraries::Kernel::init_pthreads();
     Libraries::InitHLELibs(&linker->GetHLESymbols());
@@ -52,7 +48,6 @@ Emulator::Emulator() : window{WindowWidth, WindowHeight, controller} {
 Emulator::~Emulator() {
     const auto config_dir = Common::FS::GetUserPath(Common::FS::PathType::UserDir);
     Config::save(config_dir / "config.toml");
-    discord_rpc.stop();
 }
 
 void Emulator::Run(const std::filesystem::path& file) {
@@ -116,7 +111,7 @@ void Emulator::Run(const std::filesystem::path& file) {
         std::jthread([this](std::stop_token stop_token) { linker->Execute(); });
 
     // Begin main window loop until the application exits
-    static constexpr std::chrono::microseconds FlipPeriod{100000};
+    static constexpr std::chrono::microseconds FlipPeriod{10};
 
     while (window.isOpen()) {
         window.waitEvent();

src/emulator.h

@@ -7,7 +7,6 @@
 #include <thread>
 
 #include <common/singleton.h>
-#include "common/discord.h"
 #include "core/linker.h"
 #include "input/controller.h"
 #include "sdl_window.h"
@@ -23,7 +22,6 @@ public:
 
 private:
     void LoadSystemModules(const std::filesystem::path& file);
-    Discord::RPC discord_rpc;
     Input::GameController* controller = Common::Singleton<Input::GameController>::Instance();
     Core::Linker* linker = Common::Singleton<Core::Linker>::Instance();
     Frontend::WindowSDL window;

src/sdl_window.cpp

@@ -19,6 +19,7 @@ WindowSDL::WindowSDL(s32 width_, s32 height_, Input::GameController* controller_
     if (SDL_Init(SDL_INIT_VIDEO) < 0) {
         UNREACHABLE_MSG("Failed to initialize SDL video subsystem: {}", SDL_GetError());
     }
+    SDL_InitSubSystem(SDL_INIT_AUDIO);
 
     const std::string title = "shadPS4 v" + std::string(Common::VERSION);
     SDL_PropertiesID props = SDL_CreateProperties();

src/video_core/amdgpu/liverpool.h

@@ -695,7 +695,7 @@ struct Liverpool {
         NumberFormat NumFormat() const {
             // There is a small difference between T# and CB number types, account for it.
             return info.number_type == AmdGpu::NumberFormat::SnormNz ? AmdGpu::NumberFormat::Srgb
-                                                                     : info.number_type;
+                                                                     : info.number_type.Value();
         }
     };
 

src/video_core/amdgpu/resource.h

@@ -32,7 +32,7 @@ struct Buffer {
     };
 
     u32 GetStride() const noexcept {
-        return stride == 0 ? 1U : stride;
+        return stride == 0 ? 1U : stride.Value();
     }
 
     u32 GetStrideElements(u32 element_size) const noexcept {