[nce] Adjusted precise access fault window handling + decay mechanism

src/core/arm/nce/arm_nce.cpp

@@ -45,6 +45,7 @@ using namespace Common::Literals;
 constexpr u32 StackSize = 128_KiB;
 constexpr u64 SplitPageAccessWindow = 64;
 constexpr size_t MaxPreciseAccessPages = 256;
+constexpr u8 MaxPreciseAccessPageWeight = 4;
 
 [[nodiscard]] constexpr u64 AlignDownPage(u64 addr) {
     return addr & ~u64{Memory::YUZU_PAGEMASK};
@@ -203,7 +204,7 @@ bool ArmNce::HandleGuestAccessFault(GuestContext* guest_ctx, void* raw_info, voi
 
     if (prefer_precise_channel) {
         if (auto next_pc = MatchAndExecuteOneInstruction(memory, &host_ctx, fpctx); next_pc) {
-            parent->MarkPreciseAccessPage(fault_addr);
+            parent->MarkPreciseAccessFaultWindow(fault_addr);
             host_ctx.pc = *next_pc;
             return true;
         }
@@ -224,7 +225,7 @@ bool ArmNce::HandleGuestAccessFault(GuestContext* guest_ctx, void* raw_info, voi
     }
 
     if (auto next_pc = MatchAndExecuteOneInstruction(memory, &host_ctx, fpctx); next_pc) {
-        parent->MarkPreciseAccessPage(fault_addr);
+        parent->MarkPreciseAccessFaultWindow(fault_addr);
         host_ctx.pc = *next_pc;
         return true;
     }
@@ -248,10 +249,44 @@ bool ArmNce::IsPreciseAccessPage(u64 addr) const {
 
 void ArmNce::MarkPreciseAccessPage(u64 addr) {
     const std::scoped_lock lk{m_precise_pages_guard};
-    if (m_precise_pages.size() >= MaxPreciseAccessPages) {
-        m_precise_pages.clear();
+    const u64 page = AlignDownPage(addr);
+    if (auto it = m_precise_pages.find(page); it != m_precise_pages.end()) {
+        it->second = std::min<u8>(MaxPreciseAccessPageWeight, static_cast<u8>(it->second + 1));
+        return;
+    }
+
+    while (m_precise_pages.size() >= MaxPreciseAccessPages) {
+        DecayPreciseAccessPagesLocked();
+    }
+
+    m_precise_pages.emplace(page, 1);
+}
+
+void ArmNce::MarkPreciseAccessFaultWindow(u64 addr) {
+    MarkPreciseAccessPage(addr);
+
+    if (!IsNearPageBoundary(addr)) {
+        return;
+    }
+
+    const u64 page_offset = addr & Memory::YUZU_PAGEMASK;
+    if (page_offset < SplitPageAccessWindow && addr >= Memory::YUZU_PAGESIZE) {
+        MarkPreciseAccessPage(addr - Memory::YUZU_PAGESIZE);
+    }
+    if (page_offset + SplitPageAccessWindow > Memory::YUZU_PAGESIZE) {
+        MarkPreciseAccessPage(addr + Memory::YUZU_PAGESIZE);
+    }
+}
+
+void ArmNce::DecayPreciseAccessPagesLocked() {
+    for (auto it = m_precise_pages.begin(); it != m_precise_pages.end();) {
+        if (it->second > 1) {
+            --it->second;
+            ++it;
+        } else {
+            it = m_precise_pages.erase(it);
+        }
     }
-    m_precise_pages.insert(AlignDownPage(addr));
 }
 
 void ArmNce::LockThread(Kernel::KThread* thread) {

src/core/arm/nce/arm_nce.h

@@ -7,7 +7,7 @@
 #pragma once
 
 #include <mutex>
-#include <unordered_set>
+#include <unordered_map>
 
 #include "core/arm/arm_interface.h"
 #include "core/arm/nce/guest_context.h"
@@ -83,6 +83,8 @@ private:
 
     bool IsPreciseAccessPage(u64 addr) const;
     void MarkPreciseAccessPage(u64 addr);
+    void MarkPreciseAccessFaultWindow(u64 addr);
+    void DecayPreciseAccessPagesLocked();
 
 public:
     Core::System& m_system;
@@ -96,7 +98,7 @@ public:
     Kernel::KThread* m_running_thread{};
 
     mutable std::mutex m_precise_pages_guard{};
-    std::unordered_set<u64> m_precise_pages{};
+    std::unordered_map<u64, u8> m_precise_pages{};
 
     // Stack for signal processing.
     std::unique_ptr<u8[]> m_stack{};