Misc fixes and implement osStopThread (partial), osEepromWrite, and osEepromRead (#67)

* Fix rdram allocation to only allow kseg0 accesses

* Implement osEepromWrite and osEepromRead, fix implicit casting error in save_write

* Partially implement osStopThread

* Fix audio allowing games to queue 0 samples

* Remove commented out line

librecomp/include/librecomp/addresses.hpp

@@ -6,8 +6,10 @@
 #include "librecomp/recomp.h"
 
 namespace recomp {
-    // 2GB (Addressable upper half of rdram)
-    constexpr size_t mem_size = 2U * 1024U * 1024U * 1024U;
+    // 512GB (kseg0 size)
+    constexpr size_t mem_size = 512U * 1024U * 1024U;
+    // 2GB (Addressable upper half of the address space)
+    constexpr size_t allocation_size = 2048U * 1024U * 1024U;
     // We need a place in rdram to hold the PI handles, so pick an address in extended rdram
     constexpr int32_t cart_handle = 0x80800000;
     constexpr int32_t drive_handle = (int32_t)(cart_handle + sizeof(OSPiHandle));

librecomp/src/eep.cpp

@@ -16,7 +16,16 @@ extern "C" void osEepromProbe_recomp(uint8_t* rdram, recomp_context* ctx) {
 }
 
 extern "C" void osEepromWrite_recomp(uint8_t* rdram, recomp_context* ctx) {
-    assert(false);// ctx->r2 = 8; // CONT_NO_RESPONSE_ERROR
+    uint8_t eep_address = ctx->r5;
+    gpr buffer = ctx->r6;
+    int32_t nbytes = eeprom_block_size;
+
+    assert(!(nbytes & 7));
+    assert(eep_address * eeprom_block_size + nbytes <= eep16_size);
+
+    save_write(rdram, buffer, eep_address * eeprom_block_size, nbytes);
+
+    ctx->r2 = 0;
 }
 
 extern "C" void osEepromLongWrite_recomp(uint8_t* rdram, recomp_context* ctx) {
@@ -33,7 +42,16 @@ extern "C" void osEepromLongWrite_recomp(uint8_t* rdram, recomp_context* ctx) {
 }
 
 extern "C" void osEepromRead_recomp(uint8_t* rdram, recomp_context* ctx) {
-    assert(false);// ctx->r2 = 8; // CONT_NO_RESPONSE_ERROR
+    uint8_t eep_address = ctx->r5;
+    gpr buffer = ctx->r6;
+    int32_t nbytes = eeprom_block_size;
+
+    assert(!(nbytes & 7));
+    assert(eep_address * eeprom_block_size + nbytes <= eep16_size);
+
+    save_read(rdram, buffer, eep_address * eeprom_block_size, nbytes);
+
+    ctx->r2 = 0;
 }
 
 extern "C" void osEepromLongRead_recomp(uint8_t* rdram, recomp_context* ctx) {

librecomp/src/pi.cpp

@@ -154,7 +154,7 @@ void save_write_ptr(const void* in, uint32_t offset, uint32_t count) {
 void save_write(RDRAM_ARG PTR(void) rdram_address, uint32_t offset, uint32_t count) {
     {
         std::lock_guard lock { save_context.save_buffer_mutex };
-        for (uint32_t i = 0; i < count; i++) {
+        for (gpr i = 0; i < count; i++) {
             save_context.save_buffer[offset + i] = MEM_B(i, rdram_address);
         }
     }
@@ -164,7 +164,7 @@ void save_write(RDRAM_ARG PTR(void) rdram_address, uint32_t offset, uint32_t cou
 
 void save_read(RDRAM_ARG PTR(void) rdram_address, uint32_t offset, uint32_t count) {
     std::lock_guard lock { save_context.save_buffer_mutex };
-    for (size_t i = 0; i < count; i++) {
+    for (gpr i = 0; i < count; i++) {
         MEM_B(i, rdram_address) = save_context.save_buffer[offset + i];
     }
 }

librecomp/src/recomp.cpp

@@ -609,15 +609,30 @@ void recomp::start(
     }
 
     // Allocate rdram without comitting it. Use a platform-specific virtual allocation function
-    // that initializes to zero.
+    // that initializes to zero. Protect the region above the memory size to catch accesses to invalid addresses.
     uint8_t* rdram;
     bool alloc_failed;
 #ifdef _WIN32
-    rdram = reinterpret_cast<uint8_t*>(VirtualAlloc(nullptr, mem_size, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE));
+    rdram = reinterpret_cast<uint8_t*>(VirtualAlloc(nullptr, allocation_size, MEM_COMMIT | MEM_RESERVE, PAGE_NOACCESS));
+    DWORD old_protect = 0;
     alloc_failed = (rdram == nullptr);
+    if (!alloc_failed) {
+        // VirtualProtect returns 0 on failure.
+        alloc_failed = (VirtualProtect(rdram, mem_size, PAGE_READWRITE, &old_protect) == 0);
+        if (alloc_failed) {
+            VirtualFree(rdram, 0, MEM_RELEASE);
+        }
+    }
 #else
-    rdram = (uint8_t*)mmap(NULL, mem_size, PROT_READ | PROT_WRITE, MAP_ANON | MAP_PRIVATE, -1, 0);
+    rdram = (uint8_t*)mmap(NULL, allocation_size, PROT_READ | PROT_WRITE, MAP_ANON | MAP_PRIVATE, -1, 0);
     alloc_failed = rdram == reinterpret_cast<uint8_t*>(MAP_FAILED);
+    if (!alloc_failed) {
+        // mprotect returns -1 on failure.
+        alloc_failed = (mprotect(rdram, mem_size, PROT_READ | PROT_WRITE) == -1);
+        if (alloc_failed) {
+            munmap(rdram, allocation_size);
+        }
+    }
 #endif
 
     if (alloc_failed) {
@@ -659,7 +674,7 @@ void recomp::start(
     free_failed = (VirtualFree(rdram, 0, MEM_RELEASE) == 0);
 #else
     // munmap returns -1 on failure.
-    free_failed = (munmap(rdram, mem_size) == -1);
+    free_failed = (munmap(rdram, allocation_size) == -1);
 #endif
 
     if (free_failed) {

ultramodern/src/audio.cpp

@@ -30,7 +30,7 @@ void ultramodern::queue_audio_buffer(RDRAM_ARG PTR(int16_t) audio_data_, uint32_
     uint32_t sample_count = byte_count / sizeof(int16_t);
 
     // Queue the swapped audio data.
-    if (audio_callbacks.queue_samples) {
+    if (sample_count > 0 && audio_callbacks.queue_samples) {
         audio_callbacks.queue_samples(TO_PTR(int16_t, audio_data_), sample_count);
     }
 }
@@ -52,16 +52,16 @@ uint32_t ultramodern::get_remaining_audio_bytes() {
     else {
         buffered_byte_count = 100;
     }
-     // Adjust the reported count to be some number of refreshes in the future, which helps ensure that
-     // there are enough samples even if the audio thread experiences a small amount of lag. This prevents
-     // audio popping on games that use the buffered audio byte count to determine how many samples
-     // to generate.
-     uint32_t samples_per_vi = (sample_rate / 60);
-     if (buffered_byte_count > static_cast<uint32_t>(buffer_offset_frames * sizeof(int16_t) * samples_per_vi)) {
-         buffered_byte_count -= static_cast<uint32_t>(buffer_offset_frames * sizeof(int16_t) * samples_per_vi);
-     }
-     else {
-         buffered_byte_count = 0;
-     }
-     return buffered_byte_count;
+    // Adjust the reported count to be some number of refreshes in the future, which helps ensure that
+    // there are enough samples even if the audio thread experiences a small amount of lag. This prevents
+    // audio popping on games that use the buffered audio byte count to determine how many samples
+    // to generate.
+    uint32_t samples_per_vi = (sample_rate / 60);
+    if (buffered_byte_count > static_cast<uint32_t>(buffer_offset_frames * sizeof(int16_t) * samples_per_vi)) {
+        buffered_byte_count -= static_cast<uint32_t>(buffer_offset_frames * sizeof(int16_t) * samples_per_vi);
+    }
+    else {
+        buffered_byte_count = 0;
+    }
+    return buffered_byte_count;
 }

ultramodern/src/threads.cpp

@@ -147,7 +147,7 @@ void ultramodern::set_native_thread_priority(ThreadPriority pri) {}
 #endif
 
 void wait_for_resumed(RDRAM_ARG UltraThreadContext* thread_context) {
-    TO_PTR(OSThread, ultramodern::this_thread())->context->running.wait();
+    thread_context->running.wait();
     // If this thread's context was replaced by another thread or deleted, destroy it again from its own context.
     // This will trigger thread cleanup instead.
     if (TO_PTR(OSThread, ultramodern::this_thread())->context != thread_context) {
@@ -198,7 +198,10 @@ static void _thread_func(RDRAM_ARG PTR(OSThread) self_, PTR(thread_func_t) entry
     debug_printf("[Thread] Thread waiting to be started: %d\n", self->id);
 
     // Wait until the thread is marked as running.
-    wait_for_resumed(PASS_RDRAM thread_context);
+    try {
+        wait_for_resumed(PASS_RDRAM thread_context);
+    } catch (ultramodern::thread_terminated& terminated) {
+    }
 
     // Make sure the thread wasn't replaced or destroyed before it was started.
     if (self->context == thread_context) {
@@ -228,11 +231,6 @@ extern "C" void osStartThread(RDRAM_ARG PTR(OSThread) t_) {
     OSThread* t = TO_PTR(OSThread, t_);
     debug_printf("[os] Start Thread %d\n", t->id);
 
-    // Wait until the thread is initialized to indicate that it's ready to be started.
-    t->context->initialized.wait();
-
-    debug_printf("[os] Thread %d is ready to be started\n", t->id);
-
     // If this is a game thread, insert the new thread into the running queue and then check the running queue.
     if (thread_self) {
         ultramodern::schedule_running_thread(PASS_RDRAM t_);
@@ -259,12 +257,26 @@ extern "C" void osCreateThread(RDRAM_ARG PTR(OSThread) t_, OSId id, PTR(thread_f
 
     // Spawn a new thread, which will immediately pause itself and wait until it's been started.
     // Pass the context as an argument to the thread function to ensure that it can't get cleared before the thread captures its value.
-    t->context = new UltraThreadContext{};
-    t->context->host_thread = std::thread{_thread_func, PASS_RDRAM t_, entrypoint, arg, t->context};
+    UltraThreadContext* context = new UltraThreadContext{};
+    t->context = context;
+    context->host_thread = std::thread{_thread_func, PASS_RDRAM t_, entrypoint, arg, t->context};
+
+    // Wait until the thread is initialized to indicate that it's ready to be started.
+    context->initialized.wait();
+    debug_printf("[os] Thread %d is ready to be started\n", t->id);
 }
 
 extern "C" void osStopThread(RDRAM_ARG PTR(OSThread) t_) {
-    assert(false);
+    if (t_ == NULLPTR) {
+        t_ = thread_self;
+    }
+    // Check if the thread is stopping itself (arg is null or thread_self).
+    if (t_ == thread_self) {
+        ultramodern::run_next_thread_and_wait(PASS_RDRAM1);
+    }
+    else {
+        assert(false);
+    }
 }
 
 extern "C" void osDestroyThread(RDRAM_ARG PTR(OSThread) t_) {