Port page_manager from readbacks-poc

src/video_core/buffer_cache/buffer_cache.cpp

@@ -36,7 +36,7 @@ BufferCache::BufferCache(const Vulkan::Instance& instance_, Vulkan::Scheduler& s
       bda_pagetable_buffer{instance, scheduler, MemoryUsage::DeviceLocal,
                            0,        AllFlags,  BDA_PAGETABLE_SIZE},
       fault_buffer(instance, scheduler, MemoryUsage::DeviceLocal, 0, AllFlags, FAULT_BUFFER_SIZE),
-      memory_tracker{&tracker} {
+      memory_tracker{tracker} {
     Vulkan::SetObjectName(instance.GetDevice(), gds_buffer.Handle(), "GDS Buffer");
     Vulkan::SetObjectName(instance.GetDevice(), bda_pagetable_buffer.Handle(),
                           "BDA Page Table Buffer");

src/video_core/buffer_cache/memory_tracker_base.h

@@ -19,11 +19,11 @@ public:
     static constexpr size_t MANAGER_POOL_SIZE = 32;
 
 public:
-    explicit MemoryTracker(PageManager* tracker_) : tracker{tracker_} {}
+    explicit MemoryTracker(PageManager& tracker_) : tracker{&tracker_} {}
     ~MemoryTracker() = default;
 
     /// Returns true if a region has been modified from the CPU
-    [[nodiscard]] bool IsRegionCpuModified(VAddr query_cpu_addr, u64 query_size) noexcept {
+    bool IsRegionCpuModified(VAddr query_cpu_addr, u64 query_size) noexcept {
         return IteratePages<true>(
             query_cpu_addr, query_size, [](RegionManager* manager, u64 offset, size_t size) {
                 return manager->template IsRegionModified<Type::CPU>(offset, size);
@@ -31,7 +31,7 @@ public:
     }
 
     /// Returns true if a region has been modified from the GPU
-    [[nodiscard]] bool IsRegionGpuModified(VAddr query_cpu_addr, u64 query_size) noexcept {
+    bool IsRegionGpuModified(VAddr query_cpu_addr, u64 query_size) noexcept {
         return IteratePages<false>(
             query_cpu_addr, query_size, [](RegionManager* manager, u64 offset, size_t size) {
                 return manager->template IsRegionModified<Type::GPU>(offset, size);
@@ -57,8 +57,7 @@ public:
     }
 
     /// Call 'func' for each CPU modified range and unmark those pages as CPU modified
-    template <typename Func>
+    void ForEachUploadRange(VAddr query_cpu_range, u64 query_size, auto&& func) {
-    void ForEachUploadRange(VAddr query_cpu_range, u64 query_size, Func&& func) {
         IteratePages<true>(query_cpu_range, query_size,
                            [&func](RegionManager* manager, u64 offset, size_t size) {
                                manager->template ForEachModifiedRange<Type::CPU, true>(
@@ -67,17 +66,12 @@ public:
     }
 
     /// Call 'func' for each GPU modified range and unmark those pages as GPU modified
-    template <bool clear, typename Func>
+    template <bool clear>
-    void ForEachDownloadRange(VAddr query_cpu_range, u64 query_size, Func&& func) {
+    void ForEachDownloadRange(VAddr query_cpu_range, u64 query_size, auto&& func) {
         IteratePages<false>(query_cpu_range, query_size,
                             [&func](RegionManager* manager, u64 offset, size_t size) {
-                                if constexpr (clear) {
+                                manager->template ForEachModifiedRange<Type::GPU, clear>(
-                                    manager->template ForEachModifiedRange<Type::GPU, true>(
+                                    manager->GetCpuAddr() + offset, size, func);
-                                        manager->GetCpuAddr() + offset, size, func);
-                                } else {
-                                    manager->template ForEachModifiedRange<Type::GPU, false>(
-                                        manager->GetCpuAddr() + offset, size, func);
-                                }
                             });
     }
 

src/video_core/buffer_cache/word_manager.h

@@ -10,8 +10,9 @@
 
 #ifdef __linux__
 #include "common/adaptive_mutex.h"
-#endif
+#else
 #include "common/spin_lock.h"
+#endif
 #include "common/types.h"
 #include "video_core/page_manager.h"
 
@@ -56,7 +57,7 @@ public:
         return cpu_addr;
     }
 
-    static u64 ExtractBits(u64 word, size_t page_start, size_t page_end) {
+    static constexpr u64 ExtractBits(u64 word, size_t page_start, size_t page_end) {
         constexpr size_t number_bits = sizeof(u64) * 8;
         const size_t limit_page_end = number_bits - std::min(page_end, number_bits);
         u64 bits = (word >> page_start) << page_start;
@@ -64,7 +65,7 @@ public:
         return bits;
     }
 
-    static std::pair<size_t, size_t> GetWordPage(VAddr address) {
+    static constexpr std::pair<size_t, size_t> GetWordPage(VAddr address) {
         const size_t converted_address = static_cast<size_t>(address);
         const size_t word_number = converted_address / BYTES_PER_WORD;
         const size_t amount_pages = converted_address % BYTES_PER_WORD;
@@ -104,13 +105,12 @@ public:
         }
     }
 
-    template <typename Func>
+    void IteratePages(u64 mask, auto&& func) const {
-    void IteratePages(u64 mask, Func&& func) const {
         size_t offset = 0;
         while (mask != 0) {
             const size_t empty_bits = std::countr_zero(mask);
             offset += empty_bits;
-            mask = mask >> empty_bits;
+            mask >>= empty_bits;
 
             const size_t continuous_bits = std::countr_one(mask);
             func(offset, continuous_bits);
@@ -155,8 +155,8 @@ public:
      * @param size            Size in bytes of the CPU range to loop over
      * @param func            Function to call for each turned off region
      */
-    template <Type type, bool clear, typename Func>
+    template <Type type, bool clear>
-    void ForEachModifiedRange(VAddr query_cpu_range, s64 size, Func&& func) {
+    void ForEachModifiedRange(VAddr query_cpu_range, s64 size, auto&& func) {
         std::scoped_lock lk{lock};
         static_assert(type != Type::Untracked);
 
@@ -177,11 +177,9 @@ public:
             if constexpr (clear) {
                 if constexpr (type == Type::CPU) {
                     UpdateProtection<true>(index, untracked[index], mask);
-                }
-                state_words[index] &= ~mask;
-                if constexpr (type == Type::CPU) {
                     untracked[index] &= ~mask;
                 }
+                state_words[index] &= ~mask;
             }
             const size_t base_offset = index * PAGES_PER_WORD;
             IteratePages(word, [&](size_t pages_offset, size_t pages_size) {
@@ -245,11 +243,12 @@ private:
      */
     template <bool add_to_tracker>
     void UpdateProtection(u64 word_index, u64 current_bits, u64 new_bits) const {
+        constexpr s32 delta = add_to_tracker ? 1 : -1;
         u64 changed_bits = (add_to_tracker ? current_bits : ~current_bits) & new_bits;
         VAddr addr = cpu_addr + word_index * BYTES_PER_WORD;
         IteratePages(changed_bits, [&](size_t offset, size_t size) {
-            tracker->UpdatePagesCachedCount(addr + offset * BYTES_PER_PAGE, size * BYTES_PER_PAGE,
+            tracker->UpdatePageWatchers<delta>(addr + offset * BYTES_PER_PAGE,
-                                            add_to_tracker ? 1 : -1);
+                                               size * BYTES_PER_PAGE);
         });
     }
 

src/video_core/page_manager.cpp

@@ -1,11 +1,8 @@
 // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#include <thread>
 #include <boost/icl/interval_set.hpp>
-#include "common/alignment.h"
 #include "common/assert.h"
-#include "common/error.h"
 #include "common/signal_context.h"
 #include "core/memory.h"
 #include "core/signals.h"
@@ -15,8 +12,10 @@
 #ifndef _WIN64
 #include <sys/mman.h>
 #ifdef ENABLE_USERFAULTFD
+#include <thread>
 #include <fcntl.h>
 #include <linux/userfaultfd.h>
+#include "common/error.h"
 #include <poll.h>
 #include <sys/ioctl.h>
 #endif
@@ -24,14 +23,43 @@
 #include <windows.h>
 #endif
 
+#ifdef __linux__
+#include "common/adaptive_mutex.h"
+#else
+#include "common/spin_lock.h"
+#endif
+
 namespace VideoCore {
 
-constexpr size_t PAGESIZE = 4_KB;
+constexpr size_t PAGE_SIZE = 4_KB;
-constexpr size_t PAGEBITS = 12;
+constexpr size_t PAGE_BITS = 12;
 
-#ifdef ENABLE_USERFAULTFD
 struct PageManager::Impl {
-    Impl(Vulkan::Rasterizer* rasterizer_) : rasterizer{rasterizer_} {
+    struct PageState {
+        u8 num_watchers{};
+
+        Core::MemoryPermission Perm() const noexcept {
+            return num_watchers == 0 ? Core::MemoryPermission::ReadWrite
+                                     : Core::MemoryPermission::Read;
+        }
+
+        template <s32 delta>
+        u8 AddDelta() {
+            if constexpr (delta == 1) {
+                return ++num_watchers;
+            } else {
+                ASSERT_MSG(num_watchers > 0, "Not enough watchers");
+                return --num_watchers;
+            }
+        }
+    };
+
+    static constexpr size_t ADDRESS_BITS = 40;
+    static constexpr size_t NUM_ADDRESS_PAGES = 1ULL << (40 - PAGE_BITS);
+    inline static Vulkan::Rasterizer* rasterizer;
+#ifdef ENABLE_USERFAULTFD
+    Impl(Vulkan::Rasterizer* rasterizer_) {
+        rasterizer = rasterizer_;
         uffd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK | UFFD_USER_MODE_ONLY);
         ASSERT_MSG(uffd != -1, "{}", Common::GetLastErrorMsg());
 
@@ -63,7 +91,8 @@ struct PageManager::Impl {
         ASSERT_MSG(ret != -1, "Uffdio unregister failed");
     }
 
-    void Protect(VAddr address, size_t size, bool allow_write) {
+    void Protect(VAddr address, size_t size, Core::MemoryPermission perms) {
+        bool allow_write = True(perms & Core::MemoryPermission::Write);
         uffdio_writeprotect wp;
         wp.range.start = address;
         wp.range.len = size;
@@ -118,12 +147,9 @@ struct PageManager::Impl {
         }
     }
 
-    Vulkan::Rasterizer* rasterizer;
     std::jthread ufd_thread;
     int uffd;
-};
 #else
-struct PageManager::Impl {
     Impl(Vulkan::Rasterizer* rasterizer_) {
         rasterizer = rasterizer_;
 
@@ -141,12 +167,10 @@ struct PageManager::Impl {
         // No-op
     }
 
-    void Protect(VAddr address, size_t size, bool allow_write) {
+    void Protect(VAddr address, size_t size, Core::MemoryPermission perms) {
         auto* memory = Core::Memory::Instance();
         auto& impl = memory->GetAddressSpace();
-        impl.Protect(address, size,
+        impl.Protect(address, size, perms);
-                     allow_write ? Core::MemoryPermission::ReadWrite
-                                 : Core::MemoryPermission::Read);
     }
 
     static bool GuestFaultSignalHandler(void* context, void* fault_address) {
@@ -157,23 +181,65 @@ struct PageManager::Impl {
         return false;
     }
 
-    inline static Vulkan::Rasterizer* rasterizer;
-};
 #endif
+    template <s32 delta>
+    void UpdatePageWatchers(VAddr addr, u64 size) {
+        std::scoped_lock lk(lock);
+
+        size_t page = addr >> PAGE_BITS;
+        auto perms = cached_pages[page].Perm();
+        u64 range_begin = 0;
+        u64 range_bytes = 0;
+
+        const auto release_pending = [&] {
+            if (range_bytes > 0) {
+                Protect(range_begin << PAGE_BITS, range_bytes, perms);
+                range_bytes = 0;
+            }
+        };
+
+        // Iterate requested pages
+        const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE);
+        for (; page != page_end; ++page) {
+            PageState& state = cached_pages[page];
+
+            // Apply the change to the page state
+            const u8 new_count = state.AddDelta<delta>();
+
+            // If the protection changed flush pending (un)protect action
+            if (auto new_perms = state.Perm(); new_perms != perms) [[unlikely]] {
+                release_pending();
+                perms = new_perms;
+            }
+
+            // If the page must be (un)protected, add it to the pending range
+            if ((new_count == 0 && delta < 0) || (new_count == 1 && delta > 0)) {
+                if (range_bytes == 0) {
+                    range_begin = page;
+                }
+                range_bytes += PAGE_SIZE;
+            } else {
+                release_pending();
+            }
+        }
+        
+        // Flush pending (un)protect action
+        release_pending();
+    }
+
+    std::array<PageState, NUM_ADDRESS_PAGES> cached_pages{};
+#ifdef __linux__
+    Common::AdaptiveMutex lock;
+#else
+    Common::SpinLock lock;
+#endif
+};
 
 PageManager::PageManager(Vulkan::Rasterizer* rasterizer_)
-    : impl{std::make_unique<Impl>(rasterizer_)}, rasterizer{rasterizer_} {}
+    : impl{std::make_unique<Impl>(rasterizer_)} {}
 
 PageManager::~PageManager() = default;
 
-VAddr PageManager::GetPageAddr(VAddr addr) {
-    return Common::AlignDown(addr, PAGESIZE);
-}
-
-VAddr PageManager::GetNextPageAddr(VAddr addr) {
-    return Common::AlignUp(addr + 1, PAGESIZE);
-}
-
 void PageManager::OnGpuMap(VAddr address, size_t size) {
     impl->OnMap(address, size);
 }
@@ -182,41 +248,12 @@ void PageManager::OnGpuUnmap(VAddr address, size_t size) {
     impl->OnUnmap(address, size);
 }
 
-void PageManager::UpdatePagesCachedCount(VAddr addr, u64 size, s32 delta) {
+template <s32 delta>
-    static constexpr u64 PageShift = 12;
+void PageManager::UpdatePageWatchers(VAddr addr, u64 size) const {
-
+    impl->UpdatePageWatchers<delta>(addr, size);
-    std::scoped_lock lk{lock};
-    const u64 num_pages = ((addr + size - 1) >> PageShift) - (addr >> PageShift) + 1;
-    const u64 page_start = addr >> PageShift;
-    const u64 page_end = page_start + num_pages;
-
-    const auto pages_interval =
-        decltype(cached_pages)::interval_type::right_open(page_start, page_end);
-    if (delta > 0) {
-        cached_pages.add({pages_interval, delta});
-    }
-
-    const auto& range = cached_pages.equal_range(pages_interval);
-    for (const auto& [range, count] : boost::make_iterator_range(range)) {
-        const auto interval = range & pages_interval;
-        const VAddr interval_start_addr = boost::icl::first(interval) << PageShift;
-        const VAddr interval_end_addr = boost::icl::last_next(interval) << PageShift;
-        const u32 interval_size = interval_end_addr - interval_start_addr;
-        ASSERT_MSG(rasterizer->IsMapped(interval_start_addr, interval_size),
-                   "Attempted to track non-GPU memory at address {:#x}, size {:#x}.",
-                   interval_start_addr, interval_size);
-        if (delta > 0 && count == delta) {
-            impl->Protect(interval_start_addr, interval_size, false);
-        } else if (delta < 0 && count == -delta) {
-            impl->Protect(interval_start_addr, interval_size, true);
-        } else {
-            ASSERT(count >= 0);
-        }
-    }
-
-    if (delta < 0) {
-        cached_pages.add({pages_interval, delta});
-    }
 }
 
+template void PageManager::UpdatePageWatchers<1>(VAddr addr, u64 size) const;
+template void PageManager::UpdatePageWatchers<-1>(VAddr addr, u64 size) const;
+
 } // namespace VideoCore

src/video_core/page_manager.h

@@ -4,11 +4,7 @@
 #pragma once
 
 #include <memory>
-#include <boost/icl/interval_map.hpp>
+#include "common/alignment.h"
-#ifdef __linux__
-#include "common/adaptive_mutex.h"
-#endif
-#include "common/spin_lock.h"
 #include "common/types.h"
 
 namespace Vulkan {
@@ -18,6 +14,8 @@ class Rasterizer;
 namespace VideoCore {
 
 class PageManager {
+    static constexpr size_t PAGE_BITS = 12;
+    static constexpr size_t PAGE_SIZE = 1ULL << PAGE_BITS;
 public:
     explicit PageManager(Vulkan::Rasterizer* rasterizer);
     ~PageManager();
@@ -28,22 +26,23 @@ public:
     /// Unregister a range of gpu memory that was unmapped.
     void OnGpuUnmap(VAddr address, size_t size);
 
-    /// Increase/decrease the number of surface in pages touching the specified region
+    /// Updates watches in the pages touching the specified region.
-    void UpdatePagesCachedCount(VAddr addr, u64 size, s32 delta);
+    template <s32 delta>
+    void UpdatePageWatchers(VAddr addr, u64 size) const;
 
-    static VAddr GetPageAddr(VAddr addr);
+    /// Returns page aligned address.
-    static VAddr GetNextPageAddr(VAddr addr);
+    static constexpr VAddr GetPageAddr(VAddr addr) {
+        return Common::AlignDown(addr, PAGE_SIZE);
+    }
+
+    /// Returns address of the next page.
+    static constexpr VAddr GetNextPageAddr(VAddr addr) {
+        return Common::AlignUp(addr + 1, PAGE_SIZE);
+    }
 
 private:
     struct Impl;
     std::unique_ptr<Impl> impl;
-    Vulkan::Rasterizer* rasterizer;
-    boost::icl::interval_map<VAddr, s32> cached_pages;
-#ifdef PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP
-    Common::AdaptiveMutex lock;
-#else
-    Common::SpinLock lock;
-#endif
 };
 
 } // namespace VideoCore

src/video_core/texture_cache/texture_cache.cpp

@@ -672,7 +672,7 @@ void TextureCache::TrackImage(ImageId image_id) {
         // Re-track the whole image
         image.track_addr = image_begin;
         image.track_addr_end = image_end;
-        tracker.UpdatePagesCachedCount(image_begin, image.info.guest_size, 1);
+        tracker.UpdatePageWatchers<1>(image_begin, image.info.guest_size);
     } else {
         if (image_begin < image.track_addr) {
             TrackImageHead(image_id);
@@ -695,7 +695,7 @@ void TextureCache::TrackImageHead(ImageId image_id) {
     ASSERT(image.track_addr != 0 && image_begin < image.track_addr);
     const auto size = image.track_addr - image_begin;
     image.track_addr = image_begin;
-    tracker.UpdatePagesCachedCount(image_begin, size, 1);
+    tracker.UpdatePageWatchers<1>(image_begin, size);
 }
 
 void TextureCache::TrackImageTail(ImageId image_id) {
@@ -711,7 +711,7 @@ void TextureCache::TrackImageTail(ImageId image_id) {
     const auto addr = image.track_addr_end;
     const auto size = image_end - image.track_addr_end;
     image.track_addr_end = image_end;
-    tracker.UpdatePagesCachedCount(addr, size, 1);
+    tracker.UpdatePageWatchers<1>(addr, size);
 }
 
 void TextureCache::UntrackImage(ImageId image_id) {
@@ -724,7 +724,7 @@ void TextureCache::UntrackImage(ImageId image_id) {
     image.track_addr = 0;
     image.track_addr_end = 0;
     if (size != 0) {
-        tracker.UpdatePagesCachedCount(addr, size, -1);
+        tracker.UpdatePageWatchers<-1>(addr, size);
     }
 }
 
@@ -743,7 +743,7 @@ void TextureCache::UntrackImageHead(ImageId image_id) {
         // Cehck its hash later.
         MarkAsMaybeDirty(image_id, image);
     }
-    tracker.UpdatePagesCachedCount(image_begin, size, -1);
+    tracker.UpdatePageWatchers<-1>(image_begin, size);
 }
 
 void TextureCache::UntrackImageTail(ImageId image_id) {
@@ -762,7 +762,7 @@ void TextureCache::UntrackImageTail(ImageId image_id) {
         // Cehck its hash later.
         MarkAsMaybeDirty(image_id, image);
     }
-    tracker.UpdatePagesCachedCount(addr, size, -1);
+    tracker.UpdatePageWatchers<-1>(addr, size);
 }
 
 void TextureCache::DeleteImage(ImageId image_id) {