Merge yuzu-emu#12749

src/common/CMakeLists.txt

@@ -106,6 +106,7 @@ add_library(common STATIC
     precompiled_headers.h
     quaternion.h
     range_map.h
+    range_mutex.h
     reader_writer_queue.h
     ring_buffer.h
     ${CMAKE_CURRENT_BINARY_DIR}/scm_rev.cpp

src/common/range_mutex.h

@@ -0,0 +1,93 @@
+// SPDX-FileCopyrightText: 2024 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <condition_variable>
+#include <mutex>
+
+#include "common/intrusive_list.h"
+
+namespace Common {
+
+class ScopedRangeLock;
+
+class RangeMutex {
+public:
+    explicit RangeMutex() = default;
+    ~RangeMutex() = default;
+
+private:
+    friend class ScopedRangeLock;
+
+    void Lock(ScopedRangeLock& l);
+    void Unlock(ScopedRangeLock& l);
+    bool HasIntersectionLocked(ScopedRangeLock& l);
+
+private:
+    std::mutex m_mutex;
+    std::condition_variable m_cv;
+
+    using LockList = Common::IntrusiveListBaseTraits<ScopedRangeLock>::ListType;
+    LockList m_list;
+};
+
+class ScopedRangeLock : public Common::IntrusiveListBaseNode<ScopedRangeLock> {
+public:
+    explicit ScopedRangeLock(RangeMutex& mutex, u64 address, u64 size)
+        : m_mutex(mutex), m_address(address), m_size(size) {
+        if (m_size > 0) {
+            m_mutex.Lock(*this);
+        }
+    }
+    ~ScopedRangeLock() {
+        if (m_size > 0) {
+            m_mutex.Unlock(*this);
+        }
+    }
+
+    u64 GetAddress() const {
+        return m_address;
+    }
+
+    u64 GetSize() const {
+        return m_size;
+    }
+
+private:
+    RangeMutex& m_mutex;
+    const u64 m_address{};
+    const u64 m_size{};
+};
+
+inline void RangeMutex::Lock(ScopedRangeLock& l) {
+    std::unique_lock lk{m_mutex};
+    m_cv.wait(lk, [&] { return !HasIntersectionLocked(l); });
+    m_list.push_back(l);
+}
+
+inline void RangeMutex::Unlock(ScopedRangeLock& l) {
+    {
+        std::scoped_lock lk{m_mutex};
+        m_list.erase(m_list.iterator_to(l));
+    }
+    m_cv.notify_all();
+}
+
+inline bool RangeMutex::HasIntersectionLocked(ScopedRangeLock& l) {
+    const auto cur_begin = l.GetAddress();
+    const auto cur_last = l.GetAddress() + l.GetSize() - 1;
+
+    for (const auto& other : m_list) {
+        const auto other_begin = other.GetAddress();
+        const auto other_last = other.GetAddress() + other.GetSize() - 1;
+
+        if (cur_begin <= other_last && other_begin <= cur_last) {
+            return true;
+        }
+    }
+
+    return false;
+}
+
+} // namespace Common

src/core/device_memory_manager.h

@@ -10,6 +10,7 @@
 #include <mutex>
 
 #include "common/common_types.h"
+#include "common/range_mutex.h"
 #include "common/scratch_buffer.h"
 #include "common/virtual_buffer.h"
 
@@ -204,7 +205,7 @@ private:
         (1ULL << (device_virtual_bits - page_bits)) / subentries;
     using CachedPages = std::array<CounterEntry, num_counter_entries>;
     std::unique_ptr<CachedPages> cached_pages;
-    std::mutex counter_guard;
+    Common::RangeMutex counter_guard;
     std::mutex mapping_guard;
 };
 

src/core/device_memory_manager.inc

@@ -508,12 +508,7 @@ void DeviceMemoryManager<Traits>::UnregisterProcess(Asid asid) {
 
 template <typename Traits>
 void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size, s32 delta) {
-    std::unique_lock<std::mutex> lk(counter_guard, std::defer_lock);
+    Common::ScopedRangeLock lk(counter_guard, addr, size);
-    const auto Lock = [&] {
-        if (!lk) {
-            lk.lock();
-        }
-    };
     u64 uncache_begin = 0;
     u64 cache_begin = 0;
     u64 uncache_bytes = 0;
@@ -548,7 +543,6 @@ void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size
             }
             uncache_bytes += Memory::YUZU_PAGESIZE;
         } else if (uncache_bytes > 0) {
-            Lock();
             MarkRegionCaching(memory_device_inter, uncache_begin << Memory::YUZU_PAGEBITS,
                               uncache_bytes, false);
             uncache_bytes = 0;
@@ -559,7 +553,6 @@ void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size
             }
             cache_bytes += Memory::YUZU_PAGESIZE;
         } else if (cache_bytes > 0) {
-            Lock();
             MarkRegionCaching(memory_device_inter, cache_begin << Memory::YUZU_PAGEBITS, cache_bytes,
                               true);
             cache_bytes = 0;
@@ -567,12 +560,10 @@ void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size
         vpage++;
     }
     if (uncache_bytes > 0) {
-        Lock();
         MarkRegionCaching(memory_device_inter, uncache_begin << Memory::YUZU_PAGEBITS, uncache_bytes,
                           false);
     }
     if (cache_bytes > 0) {
-        Lock();
         MarkRegionCaching(memory_device_inter, cache_begin << Memory::YUZU_PAGEBITS, cache_bytes,
                           true);
     }

src/core/memory.cpp

@@ -1093,6 +1093,20 @@ bool Memory::InvalidateNCE(Common::ProcessAddress vaddr, size_t size) {
         [&] { rasterizer = true; });
     if (rasterizer) {
         impl->InvalidateGPUMemory(ptr, size);
+
+        const auto type = impl->current_page_table->pointers[vaddr >> YUZU_PAGEBITS].Type();
+        if (type == Common::PageType::RasterizerCachedMemory) {
+            // Check if device mapped. If not, this bugged and we can unmark.
+            DAddr addr{};
+            Common::ScratchBuffer<u32> buffer;
+            impl->gpu_device_memory->ApplyOpOnPointer(ptr, buffer,
+                                                      [&](DAddr address) { addr = address; });
+
+            if (addr == 0) {
+                LOG_ERROR(HW_Memory, "Fixing unmapped cached region {:#x}", GetInteger(vaddr));
+                impl->RasterizerMarkRegionCached(GetInteger(vaddr), size, false);
+            }
+        }
     }
 
 #ifdef __linux__

src/video_core/buffer_cache/buffer_cache.h

@@ -1546,7 +1546,10 @@ void BufferCache<P>::ImmediateUploadMemory([[maybe_unused]] Buffer& buffer,
             std::span<const u8> upload_span;
             const DAddr device_addr = buffer.CpuAddr() + copy.dst_offset;
             if (IsRangeGranular(device_addr, copy.size)) {
-                upload_span = std::span(device_memory.GetPointer<u8>(device_addr), copy.size);
+                auto* const ptr = device_memory.GetPointer<u8>(device_addr);
+                if (ptr != nullptr) {
+                    upload_span = std::span(ptr, copy.size);
+                }
             } else {
                 if (immediate_buffer.empty()) {
                     immediate_buffer = ImmediateBuffer(largest_copy);

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -243,10 +243,12 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf
     const u64 size_in_bytes{Tegra::Texture::CalculateSize(
         true, bytes_per_pixel, framebuffer.stride, framebuffer.height, 1, block_height_log2, 0)};
     const u8* const host_ptr{device_memory.GetPointer<u8>(framebuffer_addr)};
+    if (host_ptr != nullptr) {
         const std::span<const u8> input_data(host_ptr, size_in_bytes);
         Tegra::Texture::UnswizzleTexture(gl_framebuffer_data, input_data, bytes_per_pixel,
-                                     framebuffer.width, framebuffer.height, 1, block_height_log2,
+                                         framebuffer.width, framebuffer.height, 1,
-                                     0);
+                                         block_height_log2, 0);
+    }
 
     glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
     glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(framebuffer.stride));

src/video_core/renderer_vulkan/vk_blit_screen.cpp

@@ -230,9 +230,11 @@ void BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
         const u64 tiled_size{Tegra::Texture::CalculateSize(true, bytes_per_pixel,
                                                            framebuffer.stride, framebuffer.height,
                                                            1, block_height_log2, 0)};
+        if (host_ptr != nullptr) {
             Tegra::Texture::UnswizzleTexture(
                 mapped_span.subspan(image_offset, linear_size), std::span(host_ptr, tiled_size),
                 bytes_per_pixel, framebuffer.width, framebuffer.height, 1, block_height_log2, 0);
+        }
 
         const VkBufferImageCopy copy{
             .bufferOffset = image_offset,

src/video_core/renderer_vulkan/vk_query_cache.cpp

@@ -1064,8 +1064,6 @@ public:
                 }
             });
         }
-        auto* ptr = device_memory.GetPointer<u8>(new_query->dependant_address);
-        ASSERT(ptr != nullptr);
 
         new_query->dependant_manage = must_manage_dependance;
         pending_flush_queries.push_back(index);
@@ -1104,10 +1102,12 @@ public:
                 tfb_streamer.Free(query->dependant_index);
             } else {
                 u8* pointer = device_memory.GetPointer<u8>(query->dependant_address);
+                if (pointer != nullptr) {
                     u32 result;
                     std::memcpy(&result, pointer, sizeof(u32));
                     num_vertices = static_cast<u64>(result) / query->stride;
                 }
+            }
             query->value = [&]() -> u64 {
                 switch (query->topology) {
                 case Maxwell3D::Regs::PrimitiveTopology::Points:
@@ -1360,7 +1360,9 @@ bool QueryCacheRuntime::HostConditionalRenderingCompareValues(VideoCommon::Looku
     const auto check_value = [&](DAddr address) {
         u8* ptr = impl->device_memory.GetPointer<u8>(address);
         u64 value{};
+        if (ptr != nullptr) {
             std::memcpy(&value, ptr, sizeof(value));
+        }
         return value == 0;
     };
     std::array<VideoCommon::LookupData*, 2> objects{&object_1, &object_2};