From 006875ed4f223283e50c70cd1cd03791e6a3ded6 Mon Sep 17 00:00:00 2001
From: nmzik <Nmzik@mail.ru>
Date: Wed, 7 Oct 2026 22:27:36 +0400
Subject: [PATCH] graphics: pause memory-tracking spin loops

---
 src/graphics/host_gpu/pageManager.cpp | 14 +++++++++-
 src/graphics/host_gpu/regionManager.h | 14 +++++++++-
 tests/PageManagerTests.cpp            | 40 +++++++++++++++++++++++++--
 3 files changed, 64 insertions(+), 4 deletions(-)

diff --git a/src/graphics/host_gpu/pageManager.cpp b/src/graphics/host_gpu/pageManager.cpp
index 00debbbd..75829739 100644
--- a/src/graphics/host_gpu/pageManager.cpp
+++ b/src/graphics/host_gpu/pageManager.cpp
@@ -15,6 +15,12 @@
 #include <mutex>
 #include <vector>
 
+#if defined(_MSC_VER)
+#include <intrin.h>
+#elif defined(__x86_64__)
+#include <xmmintrin.h>
+#endif
+
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #ifndef NOMINMAX
 #define NOMINMAX
@@ -62,7 +68,13 @@ class SpinGuard final {
 public:
 	explicit SpinGuard(std::atomic_flag& lock): m_lock(lock) {
 		while (m_lock.test_and_set(std::memory_order_acquire)) {
-			std::atomic_signal_fence(std::memory_order_seq_cst);
+#if defined(__x86_64__) || defined(_M_X64)
+			_mm_pause();
+#elif defined(_M_ARM64)
+			__yield();
+#elif defined(__aarch64__)
+			asm volatile("yield");
+#endif
 		}
 	}
 	~SpinGuard() { m_lock.clear(std::memory_order_release); }
diff --git a/src/graphics/host_gpu/regionManager.h b/src/graphics/host_gpu/regionManager.h
index 4c95cc2c..9fae4ef7 100644
--- a/src/graphics/host_gpu/regionManager.h
+++ b/src/graphics/host_gpu/regionManager.h
@@ -9,6 +9,12 @@
 #include <mutex>
 #include <utility>
 
+#if defined(_MSC_VER)
+#include <intrin.h>
+#elif defined(__x86_64__)
+#include <xmmintrin.h>
+#endif
+
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #ifndef NOMINMAX
 #define NOMINMAX
@@ -32,7 +38,13 @@ public:
 		const auto thread = CurrentThread();
 		while (m_lock.test_and_set(std::memory_order_acquire)) {
 			EXIT_NOT_IMPLEMENTED(m_owner.load(std::memory_order_relaxed) == thread);
-			std::atomic_signal_fence(std::memory_order_seq_cst);
+#if defined(__x86_64__) || defined(_M_X64)
+			_mm_pause();
+#elif defined(_M_ARM64)
+			__yield();
+#elif defined(__aarch64__)
+			asm volatile("yield");
+#endif
 		}
 		m_owner.store(thread, std::memory_order_relaxed);
 	}
diff --git a/tests/PageManagerTests.cpp b/tests/PageManagerTests.cpp
index effb8c95..fe13707f 100644
--- a/tests/PageManagerTests.cpp
+++ b/tests/PageManagerTests.cpp
@@ -1,12 +1,15 @@
 #include "common/virtualMemory.h"
 #include "graphics/host_gpu/pageManager.h"
 
+#include <barrier>
 #include <cstdint>
 #include <cstdio>
 #include <cstdlib>
 #include <cstring>
 #include <memory>
+#include <mutex>
 #include <string>
+#include <thread>
 #include <vector>
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
@@ -167,6 +170,7 @@ struct ProtectionCall {
   uint64_t size;
 };
 std::vector<ProtectionCall> g_protection_ranges;
+std::mutex g_protection_log_mutex;
 
 bool ProtectAddressSpace(uint64_t vaddr, uint64_t size,
                          Common::VirtualMemory::Mode mode) {
@@ -177,8 +181,11 @@ bool ProtectAddressSpace(uint64_t vaddr, uint64_t size,
     protection = PAGE_READWRITE;
   }
   DWORD old_protection = 0;
-  g_protection_calls++;
-  g_protection_ranges.push_back({vaddr, size});
+  {
+    std::lock_guard lock(g_protection_log_mutex);
+    g_protection_calls++;
+    g_protection_ranges.push_back({vaddr, size});
+  }
   return VirtualProtect(reinterpret_cast<void *>(vaddr), size, protection,
                         &old_protection) != 0;
 }
@@ -547,6 +554,34 @@ void TestReadWriteWatcherInteractions() {
   Check(VirtualFree(memory, 0, MEM_RELEASE) != 0, "VirtualFree failed");
 }
 
+void TestConcurrentSharedWatchers() {
+  constexpr size_t worker_count = 8;
+  PageManager manager;
+  auto *memory = Allocate(TRACKER_PAGE_SIZE);
+  const auto address = reinterpret_cast<uint64_t>(memory);
+  bool watched = false;
+  std::barrier phase(worker_count, [&]() noexcept {
+    watched = !watched;
+    Check(Protection(memory) == (watched ? PAGE_READONLY : PAGE_READWRITE),
+          "concurrent watchers lost protection or failed to restore writes");
+  });
+  std::vector<std::jthread> workers;
+  for (size_t i = 0; i < worker_count; i++) {
+    workers.emplace_back([&] {
+      for (size_t round = 0; round < 100; round++) {
+        manager.UpdatePageWatchers<true>(address, TRACKER_PAGE_SIZE);
+        phase.arrive_and_wait();
+        manager.UpdatePageWatchers<false>(address, TRACKER_PAGE_SIZE);
+        phase.arrive_and_wait();
+      }
+    });
+  }
+  for (auto &worker : workers) {
+    worker.join();
+  }
+  Check(VirtualFree(memory, 0, MEM_RELEASE) != 0, "VirtualFree failed");
+}
+
 [[noreturn]] void RunDeathCase(const char *name) {
   PageManager manager;
   const auto page_size = manager.GetPageSize();
@@ -669,6 +704,7 @@ int main(int argc, char **argv) {
   TestRegionMaskWatcherRanges();
   TestRegionEndpointBatching();
   TestReadWriteWatcherInteractions();
+  TestConcurrentSharedWatchers();
   TestFatalPaths();
   std::puts("PageManagerTests: all cases passed");
   return 0;