One perf test for windows VirtualQueryEx calls.

base/memory/i_heart_windows_memory.cc

+#include <cstdio>
+
+#include <vector>
+
+#include "base/command_line.h"
+#include "base/process/launch.h"
+#include "base/process/process.h"
+#include "base/process/process_metrics.h"
+#include "base/threading/platform_thread.h"
+
+namespace {
+using ProcessList = std::vector<base::Process*>;
+
+void SleepForever() {
+  for (;;) {
+    base::PlatformThread::Sleep(base::TimeDelta::FromMinutes(1));
+  }
+}
+
+void LoadDll() {}
+
+void CreateSharedMemoryButDontShare() {}
+
+void ShareMeSomeMemory(int num_regions, ProcessList process_list) {
+  for (base::Process* process : process_list) {
+    CHECK(process);
+  }
+}
+
+void ShareMeSomeCOWsAndTouchThem(int num_regions, ProcessList process_list) {
+  for (base::Process* process : process_list) {
+    CHECK(process);
+  }
+}
+
+void ForceSwap() {
+  char* leak = new char[12345];
+  leak = nullptr;
+}
+
+struct SavedStats {
+  size_t private_bytes;
+  size_t shared_bytes;
+  uint64_t pss_bytes;
+  base::WorkingSetKBytes working_set;
+  base::CommittedKBytes committed;
+};
+
+void SaveStats(SavedStats* saved_stats, base::Process* process) {
+  std::unique_ptr<base::ProcessMetrics> metrics =
+      base::ProcessMetrics::CreateProcessMetrics(process->Handle());
+
+  CHECK(metrics->GetMemoryBytes(&(saved_stats->private_bytes), &(saved_stats->shared_bytes)));
+  CHECK(metrics->GetProportionalSetSizeBytes(&(saved_stats->pss_bytes)));
+  CHECK(metrics->GetWorkingSetKBytes(&saved_stats->working_set));
+  metrics->GetCommittedKBytes(&saved_stats->committed);
+}
+
+void DumpDiff(SavedStats* first, SavedStats* second) {
+  LOG(ERROR) << "  - private kbytes: " << ((int64_t)second->private_bytes - (int64_t)first->private_bytes) / 1024;
+  LOG(ERROR) << "  - shared kbytes: " << ((int64_t)second->shared_bytes - (int64_t)first->shared_bytes) / 1024;
+  LOG(ERROR) << "  - pss kbytes: " << ((int64_t)second->pss_bytes - (int64_t)first->pss_bytes) / 1024;
+  LOG(ERROR) << "  - workingset private kb: " << ((int64_t)second->working_set.priv - (int64_t)first->working_set.priv);
+  LOG(ERROR) << "  - workingset mapped kb: " << ((int64_t)second->working_set.shareable - (int64_t)first->working_set.shareable);
+  LOG(ERROR) << "  - workingset image kb: " << ((int64_t)second->working_set.shared - (int64_t)first->working_set.shared);
+  LOG(ERROR) << "  - committed private kb: " << ((int64_t)second->committed.priv - (int64_t)first->committed.priv);
+  LOG(ERROR) << "  - committed mapped kb: " << ((int64_t)second->committed.mapped - (int64_t)first->committed.mapped);
+  LOG(ERROR) << "  - committed image kb: " << ((int64_t)second->committed.image - (int64_t)first->committed.image);
+  LOG(ERROR) << std::endl << std::endl;
+}
+
+}  // namespace
+
+void malloc1() {
+  LOG(ERROR) << "malloc 16MB, and touch every page";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char*a = (char*)malloc(16 * 1024 * 1024);
+  for (int i = 0; i < 16 * 1024 * 1024; ++i) {
+    a[i] = 'a';
+  }
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void malloc2() {
+  LOG(ERROR) << "malloc 16MB, but only touch every 16th page";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char*a = (char*)malloc(16 * 1024 * 1024);
+  for(int i = 0; i < 16 * 1024 * 1024 / (16 * 4096); ++i) {
+    a[i * 16 * 4096] = 'a';
+  }
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void malloc3() {
+  LOG(ERROR) << "malloc 16MB, touch every page, then free";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char*a = (char*)malloc(16 * 1024 * 1024);
+  for (int i = 0; i < 16 * 1024 * 1024; ++i) {
+    a[i] = 'a';
+  }
+  free(a);
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void malloc4() {
+  LOG(ERROR) << "malloc 16MB in 1024 chunks, touch every page, then free every chunk that isn't a multiple of 16";
+  LOG(ERROR) << "I expect resident memory to hit ~4MB, but instead it stays at 16MB! What's going on?";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  std::vector<char*> chunks;
+  chunks.resize(16 * 1024);
+
+  SaveStats(&first, &process);
+  for (int i = 0; i < 16 * 1024; ++i) {
+    chunks[i] = (char*)malloc(1024);
+    chunks[i][0] = 'a';
+  }
+
+  for (int i = 0; i < 16 * 1024; ++i) {
+    if (i % 16 != 0)
+      free(chunks[i]);
+  }
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void anonymous_file_mapping1() {
+  LOG(ERROR) << "create 16MB anonymous file mapping, and touch every page";
+ base::Process process(base::Process::Current());
+ SavedStats first, second;
+  SaveStats(&first, &process);
+ HANDLE h = CreateFileMapping(INVALID_HANDLE_VALUE, NULL, PAGE_READWRITE, 0,
+                               16 * 1024 * 1024, NULL);
+ char*a = (char*) MapViewOfFile(h, FILE_MAP_READ | FILE_MAP_WRITE, 0, 0, 16 * 1024 *1024);
+ for (int i = 0; i < 16 * 1024 * 1024; ++i)
+   a[i] = 'a';
+ SaveStats(&second, &process);
+ DumpDiff(&first, &second);
+}
+
+void anonymous_file_mapping2() {
+  LOG(ERROR) << "create 16MB anonymous file mapping";
+ base::Process process(base::Process::Current());
+ SavedStats first, second;
+ SaveStats(&first, &process);
+ CreateFileMapping(INVALID_HANDLE_VALUE, NULL, PAGE_READWRITE, 0,
+                               16 * 1024 * 1024, NULL);
+ SaveStats(&second, &process);
+ DumpDiff(&first, &second);
+}
+void anonymous_file_mapping3() {
+  LOG(ERROR) << "create and map 16MB anonymous file mapping";
+ base::Process process(base::Process::Current());
+ SavedStats first, second;
+  SaveStats(&first, &process);
+ HANDLE h = CreateFileMapping(INVALID_HANDLE_VALUE, NULL, PAGE_READWRITE, 0,
+                               16 * 1024 * 1024, NULL);
+ MapViewOfFile(h, FILE_MAP_READ | FILE_MAP_WRITE, 0, 0, 16 * 1024 *1024);
+ SaveStats(&second, &process);
+ DumpDiff(&first, &second);
+}
+
+void virtual_alloc1() {
+  LOG(ERROR) << "virtual_alloc 16MB";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  VirtualAlloc(NULL, 16 * 1024 * 1024, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void virtual_alloc2() {
+  LOG(ERROR) << "virtual_alloc 16MB, then write to 8MB of pages";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char* a = (char*)VirtualAlloc(NULL, 16 * 1024 * 1024, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+  for (int i = 0; i < 8 * 1024 * 1024; ++i) {
+    a[i] = 'a';
+  }
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void virtual_alloc3() {
+  LOG(ERROR) << "virtual_alloc 16MB, then write to 8MB of pages, then decommits 4MB of the written pages";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char* a = (char*)VirtualAlloc(NULL, 16 * 1024 * 1024, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+  for (int i = 0; i < 8 * 1024 * 1024; ++i) {
+    a[i] = 'a';
+  }
+  VirtualFree(a, 4 * 1024 * 1024, MEM_DECOMMIT);
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void virtual_alloc4() {
+  LOG(ERROR) << "virtual_alloc 16MB, then write to 8MB of pages, then reset 4MB of the written pages";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char* a = (char*)VirtualAlloc(NULL, 16 * 1024 * 1024, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+  for (int i = 0; i < 8 * 1024 * 1024; ++i) {
+    a[i] = 'a';
+  }
+  VirtualAlloc(a, 4 * 1024 * 1024, MEM_RESET, PAGE_NOACCESS);
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void virtual_alloc5() {
+  LOG(ERROR) << "virtual_alloc 16MB, then write to 8MB of pages, then discard 4MB of the written pages";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char* a = (char*)VirtualAlloc(NULL, 16 * 1024 * 1024, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+  for (int i = 0; i < 8 * 1024 * 1024; ++i) {
+    a[i] = 'a';
+  }
+  using DiscardVirtualMemoryFunction =
+    DWORD(WINAPI*)(PVOID virtualAddress, SIZE_T size);
+  DiscardVirtualMemoryFunction discard_virtual_memory =
+        reinterpret_cast<DiscardVirtualMemoryFunction>(GetProcAddress(
+            GetModuleHandle(L"Kernel32.dll"), "DiscardVirtualMemory"));
+  discard_virtual_memory(a, 4 * 1024 * 1024);
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+void heap_alloc1() {
+  LOG(ERROR) << "HeapAlloc 16MB";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  HeapAlloc(GetProcessHeap(), 0, 16 * 1024 * 1024);
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+void heap_alloc2() {
+  LOG(ERROR) << "HeapAlloc 16MB, and touch 8MB";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char* a = (char*)HeapAlloc(GetProcessHeap(), 0, 16 * 1024 * 1024);
+  for (int i = 0; i < 8 * 1024 * 1024; ++i) {
+    a[i] = 'a';
+  }
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+void heap_alloc3() {
+  LOG(ERROR) << "HeapAlloc 16MB, and touch 8MB, then decommit 4MB";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char* a = (char*)HeapAlloc(GetProcessHeap(), 0, 16 * 1024 * 1024);
+  for (int i = 0; i < 8 * 1024 * 1024; ++i) {
+    a[i] = 'a';
+  }
+  VirtualFree(a, 4 * 1024 * 1024, MEM_DECOMMIT);
+  SaveStats(&second, &process);
+  
+  DumpDiff(&first, &second);
+}
+
+void leak_1gb() {
+  LOG(ERROR) << "leak 1gb";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+  char* a = (char*)malloc(1024 * 1024 * 1024);
+  for (int i = 0; i < 1024 * 1024 * 1024 / 4096; ++i)
+    a[i * 4096] = 'a';
+  SaveStats(&second, &process);
+  DumpDiff(&first, &second);
+}
+
+void swap1() {
+  LOG(ERROR) << "This will crash your machine. Are you sure you want to continue (y)?";
+  char dummy;
+  scanf("%c", &dummy);
+  if (dummy != 'y')
+    return;
+
+  LOG(ERROR) << "malloc 256MB, touch every page, then spawn children that consume 64GB of memory";
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+
+  char*a = (char*)malloc(256 * 1024 * 1024);
+  for (int i = 0; i < 256 * 1024 * 1024 / 4096; ++i) {
+    a[i * 4096] = 'a';
+  }
+  for (int i = 0; i < 64; ++i) {
+    base::LaunchOptions options;
+    base::CommandLine child_command(*base::CommandLine::ForCurrentProcess());
+    child_command.AppendSwitch("leak_1gb");
+    base::LaunchProcess(child_command, options);
+  }
+ 
+  // Wait 100 seconds for memory to blow up.
+  base::PlatformThread::Sleep(base::TimeDelta::FromSeconds(100));
+  SaveStats(&second, &process);
+  DumpDiff(&first, &second);
+}
+
+void swap2() {
+  LOG(ERROR) << "Spawns 256 process, each of which commits 1GB with VirtualAlloc but only makes 4MB of pages resident.";
+  for (int i = 0; i < 256; ++i) {
+    base::LaunchOptions options;
+    base::CommandLine child_command(*base::CommandLine::ForCurrentProcess());
+    child_command.AppendSwitch("swap2_child");
+    base::LaunchProcess(child_command, options);
+  }
+
+  // Wait 3 seconds for memory to blow up.
+  base::PlatformThread::Sleep(base::TimeDelta::FromSeconds(3));
+
+  LOG(ERROR) << "try to allocate 1GB using malloc";
+  char*b = (char*)malloc(1024 * 1024 * 1024);
+  CHECK(b);
+  for (int i = 0; i < 1024 * 1024 * 1024 / 4096; ++i) {
+    b[i * 4096] = 'a';
+  }
+  LOG(ERROR) << "allocated 1GB using malloc";
+
+  LOG(ERROR) << "try to allocate 1GB using virtual alloc";
+  for (int i = 0; i < 1024; i++) {
+    char* a = (char*)VirtualAlloc(NULL, 1024 * 1024, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+    CHECK(a);
+    a[0] = 'a';
+  }
+  LOG(ERROR) << "allocated 1GB using virtual alloc";
+}
+
+void swap2_child() {
+  base::Process process(base::Process::Current());
+  SavedStats first, second;
+  SaveStats(&first, &process);
+  for (int i = 0; i < 1024; i++) {
+    char* a = (char*)VirtualAlloc(NULL, 1024 * 1024, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+
+    // Occasional errors.
+    if (!a)
+      continue;
+
+    a[0] = 'a';
+  }
+  SaveStats(&second, &process);
+  DumpDiff(&first, &second);
+}
+
+int main(int argc, char* argv[]) {
+  base::CommandLine::Init(argc, argv);
+  const base::CommandLine& current_command =
+      *base::CommandLine::ForCurrentProcess();
+
+  if (current_command.HasSwitch("sleepy")) {
+    new char[base::Process::Current().Pid()];
+    SleepForever();
+  }
+
+  if (current_command.HasSwitch("malloc1")) {
+    malloc1();
+    SleepForever();
+  }
+
+  if (current_command.HasSwitch("malloc2")) {
+    malloc2();
+    SleepForever();
+  }
+
+  if (current_command.HasSwitch("malloc3")) {
+    malloc3();
+    SleepForever();
+  }
+
+  if (current_command.HasSwitch("malloc4")) {
+    malloc4();
+    SleepForever();
+  }
+
+  if (current_command.HasSwitch("anonymous_file_mapping1")) {
+    anonymous_file_mapping1();
+    SleepForever();
+  }
+
+  if (current_command.HasSwitch("anonymous_file_mapping2")) {
+    anonymous_file_mapping2();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("anonymous_file_mapping3")) {
+    anonymous_file_mapping3();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("virtual_alloc1")) {
+    virtual_alloc1();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("virtual_alloc2")) {
+    virtual_alloc2();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("virtual_alloc3")) {
+    virtual_alloc3();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("virtual_alloc4")) {
+    virtual_alloc4();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("virtual_alloc5")) {
+    virtual_alloc4();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("heap_alloc1")) {
+    heap_alloc1();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("heap_alloc2")) {
+    heap_alloc2();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("heap_alloc3")) {
+    heap_alloc3();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("leak_1gb")) {
+    leak_1gb();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("swap1")) {
+    swap1();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("swap2_child")) {
+    swap2_child();
+    SleepForever();
+  }
+  if (current_command.HasSwitch("swap2")) {
+    swap2();
+    SleepForever();
+  }
+  // Launch 2 child processes to mess with.
+  // base::LaunchOptions options;
+  base::Process a(base::Process::Current());
+  // base::CommandLine child_command(current_command);
+  // child_command.AppendSwitch("sleepy");
+  // base::Process b(base::LaunchProcess(child_command, options));
+  // base::Process c(base::LaunchProcess(child_command, options));
+
+  // // Test sharing a DLL.
+  // LoadDll();
+
+  // // Test shared memory that's not shared.
+  // CreateSharedMemoryButDontShare();
+
+  // // Test sharing with just one other process.
+  // ShareMeSomeMemory(2, {&b});
+  // ShareMeSomeMemory(4, {&c});
+
+  // // Test sharing with bunches of processes.
+  // ShareMeSomeMemory(7, {&b, &c});
+
+  // // Test sharing Copy on write pgae.
+  // ShareMeSomeCOWsAndTouchThem(12, {&b, &c});
+
+  // // Force us to swap out memory.
+  // ForceSwap();
+  // DumpStats({&a});
+
+  // for (int i = 0; i < 1000; ++i) {
+  //   char*a = (char*)malloc(4096 * 16);
+  //   *a = 'a';
+  //   // for (int j = 0; j < 16; ++j)
+  //   //   a[j * 4096] = 'a';
+  //   // base::PlatformThread::Sleep(base::TimeDelta::FromSeconds(1));
+  // }
+
+  // // Print out stats.
+  // DumpStats({&a});
+  //DumpStats({&a, &b, &c});
+
+  // Wait until input so it can be inspected.
+  char dummy;
+  scanf("%c", &dummy);
+
+  // Clean up 2 child processes.
+  // b.Terminate(0, true);
+  // c.Terminate(0, true);
+
+  return 0;
+}