changelog, bump version

commit 3007df0d153ade6d328321dad8f88d63869159c8
Merge: 7557851 cd14d68
Author: Carl Pearson <pearson@illinois.edu>
Date:   Wed Sep 25 09:59:54 2019 -0500

    Merge branch 'master' into feature/aslr

commit 7557851508f0e1c2d75244267c5f78fb3d1ca303
Author: Carl Pearson <pearson@illinois.edu>
Date:   Tue Sep 24 07:50:15 2019 -0500

    wish list

commit 0bbbac31f354c16ae1942dbfe4683a66ec260498
Author: Carl Pearson <pearson@illinois.edu>
Date:   Tue Sep 24 07:49:36 2019 -0500

    ASLR documentation

commit f1ae37e057792696a739e30ecdbd09e071b8a7d4
Author: Carl Pearson <pearson@illinois.edu>
Date:   Tue Sep 24 07:45:54 2019 -0500

    add ASLR interface

CMakeLists.txt

@@ -5,7 +5,7 @@
 # 3.13+ for target_link_directories
 cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
 
-project(perfect LANGUAGES CXX VERSION 0.1.0)
+project(perfect LANGUAGES CXX VERSION 0.5.0)
 message(STATUS "Build type: " ${CMAKE_BUILD_TYPE})
 
 include(CheckLanguage)

README.md

@@ -2,21 +2,24 @@
 
 | Branch | Status |
 |-|-|
-| master |[![Build Status](https://img.shields.io/endpoint.svg?url=https%3A%2F%2Factions-badge.atrox.dev%2Fcwpearson%2Fperfect%2Fbadge%3Fref%3Dmaster&style=flat)](https://actions-badge.atrox.dev/cwpearson/perfect/goto?ref=master)|
+| master |[![Build Status](https://img.shields.io/endpoint.svg?url=https%3A%2F%2Factions-badge.atrox.dev%2Fcwpearson%2Fperfect%2Fbadge%3Fref%3Dmaster&style=flat)](https://actions-badge.atrox.dev/cwpearson/perfect/goto?ref=master) |
 
-CPU/GPU performance control library for benchmarking
-* x86
-* POWER
-* Nvidia
+CPU/GPU Performance control library for benchmarking on Linux, x86, POWER, and Nvidia.
 
 ## Features
 
+- [x] GPU power/utilization/temperature monitoring (nvidia)
 - [x] Disable CPU turbo (linux)
 - [x] Set OS CPU performance mode to maximum (linux)
 - [x] Set GPU clocks (nvidia)
 - [x] Disable GPU turbo (nvidia)
 - [x] Flush addresses from cache (amd64, POWER)
 - [x] CUDA not required (GPU functions will not be compiled)
+- [x] Flush file system caches (linux)
+- [x] Disable ASLR (linux)
+
+## Contributors
+* [Carl Pearson](https://cwpearson.github.io)
 
 ## Installing
 
@@ -67,11 +70,57 @@ perfect::CpuTurboState state;
 PERFECT(perfect::get_cpu_turbo_state(&state));
 ```
 
+## Monitoring
+
+`perfect` can monitor and record GPU activity.
+
+See [examples/gpu_monitor.cu](examples/gpu_monitor.cu)
+
+```c++
+#include "perfect/gpu_monitor.hpp"
+```
+
+* `Monitor(std::ostream *stream)`: create a monitor that will write to `stream`.
+* `void Monitor::start()`: start the monitor
+* `void Monitor::stop()`: terminate the monitor
+* `void Monitor::pause()`: pause the monitor thread
+* `void Monitor::resume()`: resume the monitor thread
+
+### Disable ASLR
+
+`perfect` can disable ASLR
+
+See [tools/no_aslr.cpp](tools/no_aslr.cpp)
+
+```c++
+#include "perfect/aslr.hpp"
+```
+
+* `Result disable_aslr()`: disable ASLR
+* `Result get_aslr(AslrState &state)`: save the current ASLR state
+* `Result set_aslr(const AslrState &state)`: set a previously-saved ASLR state
+
+### Flush file system caches
+
+`perfect` can drop various filesystem caches
+
+See [tools/sync_drop_caches.cpp](tools/sync_drop_caches.cpp)
+
+```c++
+#include "perfect/drop_caches.hpp"
+```
+
+* `Result sync()`: flush filesystem caches to disk
+* `Result drop_caches(DropCaches_t mode)`: remove file system caches
+  * `mode = PAGECACHE`: drop page caches
+  * `mode = ENTRIES`: drop dentries and inodes
+  * `mode = PAGECACHE | ENTRIES`: both
+
 ### CPU Turbo
 
 `perfect` can enable and disable CPU boost through the Intel p-state mechanism or the ACPI cpufreq mechanism.
 
-See [examples/cpu_turbo.cpp].
+See [examples/cpu_turbo.cpp](examples/cpu_turbo.cpp).
 
 
 ```c++
@@ -88,7 +137,7 @@ See [examples/cpu_turbo.cpp].
 
 `perfect` can control the OS governor on linux.
 
-See [examples/os_perf.cpp].
+See [examples/os_perf.cpp](examples/os_perf.cpp).
 
 ```c++
 #include "perfect/os_perf.hpp"
@@ -102,7 +151,7 @@ See [examples/os_perf.cpp].
 
 `perfect` can enable/disable GPU turbo boost.
 
-See [examples/gpu_turbo.cu]
+See [examples/gpu_turbo.cu](examples/gpu_turbo.cu).
 
 ```c++
 #include "perfect/gpu_turbo.hpp"
@@ -118,10 +167,10 @@ See [examples/gpu_turbo.cu]
 
 `perfect` can lock GPU clocks to their maximum values.
 
-See [examples/gpu_clocks.cu]
+See [examples/gpu_clocks.cu](examples/gpu_clocks.cu).
 
 ```c++
-#include "perfect/gpu_clocks.hpp`
+#include "perfect/gpu_clocks.hpp"
 ```
 
 * `Result set_max_gpu_clocks(unsigned int idx)`: Set GPU `idx` clocks to their maximum reported values.
@@ -131,16 +180,53 @@ See [examples/gpu_clocks.cu]
 
 `perfect` can flush data from CPU caches. Unlike the other APIs, these do not return a `Result` because they do not fail.
 
-See [examples/cpu_cache.cpp].
+See [examples/cpu_cache.cpp](examples/cpu_cache.cpp).
 
 ```c++
-#include "perfect/cpu_cache.hpp`
+#include "perfect/cpu_cache.hpp"
 ```
 
 * `void flush_all(void *p, const size_t n)`: Flush all cache lines starting at `p` for `n` bytes.
 
+## Changelog
+
+* v0.5.0
+    * add tools/stress
+    * add tools/max-os-perf
+    * add tools/min-os-perf
+    * add tools/enable-cpu-turbo
+    * add tools/disable-cpu-turbo
+* v0.4.0
+    * Add ASLR interface
+    * Disambiguate some filesystem errors
+    * Fix some powerpc namespace issues
+* v0.3.0
+    * Add filesystem cache interface
+* v0.2.0
+    * add GPU monitoring
+    * Make CUDA optional
+* v0.1.0
+    * cache control
+    * Intel P-State control
+    * linux governor control
+    * POWER cpufreq control
+    * Nvidia GPU boost control
+    * Nvidia GPU clock control
+
 ## Wish List
 
+- [ ] only monitor certain GPUs
+- [ ] hyperthreading interface
+- [ ] process priority interface
+- [ ] A wrapper utility
+    - [ ] disable hyperthreading
+    - [ ] reserve cores 
+    - [ ] set process priority
+    - [ ] disable ASLR
 
-- [ ] Nvidia GPU power monitoring
-- [ ] Nivida GPU utilization monitoring
+## Related
+
+* [LLVM benchmarking instructions](https://llvm.org/docs/Benchmarking.html#linux) covering ASLR, Linux governor, cpuset shielding, SMT, and Intel turbo.
+* [easyperf.net](https://easyperf.net/blog/2019/08/02/Perf-measurement-environment-on-Linux#2-disable-hyper-threading) blog post discussing ACPI/Intel turbo, SMT, Linux governor, CPU affinity, process priority, file system caches, and ASLR. 
+* [temci](https://github.com/parttimenerd/temci) benchmarking tool for cpu sheilding and disabling hyperthreading, among other things.
+* [perflock](https://github.com/aclements/perflock) tool for locking CPU frequency scaling domains

examples/CMakeLists.txt

@@ -51,4 +51,9 @@ endif()
 if(CMAKE_CUDA_COMPILER)
     add_executable(gpu-turbo gpu_turbo.cu)
     target_link_libraries(gpu-turbo perfect)
+endif()
+
+if(CMAKE_CUDA_COMPILER)
+    add_executable(gpu-monitor gpu_monitor.cu)
+    target_link_libraries(gpu-monitor perfect)
 endif()

examples/gpu_monitor.cu

@@ -0,0 +1,25 @@
+#include <chrono>
+#include <iostream>
+#include <thread>
+
+#include "perfect/gpu_monitor.hpp"
+
+int main(void) {
+  using namespace perfect;
+  init();
+
+  // write to stderr
+  Monitor m(&std::cerr);
+
+  // don't record GPU utilization
+  m.config.utilization = false;
+
+  m.start();
+
+  // ctrl-c to exit
+  while (true) {
+    std::this_thread::sleep_for(std::chrono::milliseconds(5000));
+  }
+
+  return 0;
+}

include/perfect/aslr.hpp

@@ -0,0 +1,40 @@
+#pragma once
+
+#include <cerrno>
+#include <iostream>
+
+#ifdef __linux__
+#include "detail/os/linux.hpp"
+#endif
+#include "init.hpp"
+#include "result.hpp"
+
+namespace perfect {
+
+struct AslrState {
+#ifdef __linux__
+  unsigned long persona;
+#else
+#error "unsupported platform"
+#endif
+};
+
+Result get_aslr(AslrState &state) {
+  int persona;
+  PERFECT_SUCCESS_OR_RETURN(detail::get_personality(persona));
+  state.persona = persona;
+  return Result::SUCCESS;
+}
+
+Result set_aslr(const AslrState &state) {
+  return detail::set_personality(state.persona);
+}
+
+Result disable_aslr() {
+  int persona;
+  PERFECT_SUCCESS_OR_RETURN(detail::get_personality(persona));
+  persona |= ADDR_NO_RANDOMIZE;
+  return detail::set_personality(persona);
+}
+
+} // namespace perfect

include/perfect/cpu_set.hpp

@@ -0,0 +1,361 @@
+#pragma once
+
+#include <sys/mount.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+#include <algorithm>
+#include <cstring>
+#include <fstream>
+#include <set>
+#include <sstream>
+#include <string>
+#include <vector>
+
+#include "init.hpp"
+#include "result.hpp"
+
+#define SUCCESS_OR_RETURN(stmt) \
+{\
+  Result _ret; \
+  _ret = (stmt); \
+if (_ret != Result::SUCCESS) {\
+  return _ret;\
+}\
+}
+
+std::set<int> operator-(const std::set<int> &lhs, const std::set<int> &rhs) {
+  std::set<int> result;
+  for (auto e : lhs) {
+    if (0 == rhs.count(e)) {
+      result.insert(e);
+    }
+  }
+  return result;
+}
+
+std::string remove_space(const std::string &s) {
+  std::string result;
+
+  for (auto c : s) {
+    if (!isspace(c)) {
+      result += c;
+    }
+  }
+  return result;
+}
+
+// like "8" or "8-10"
+std::set<int> parse_token(const std::string &token) {
+  // std::cerr << "parse_token: parsing '" << s << "'\n";
+  std::set<int> result;
+
+  std::string s = token;
+  // ignore empty string
+  if (s.empty()) {
+    return result;
+  }
+
+  // remove newline
+  s = remove_space(s);
+
+  size_t pos = 0;
+
+  int first = std::stoi(s, &pos);
+  // std::cerr << "parse_token: found '" << first << "'\n";
+
+  // single int
+  if (pos == s.length()) {
+    result.insert(first);
+    return result;
+  }
+
+  // next char should be a "-"
+  assert(s[pos] == '-');
+
+  std::string rest = s.substr(pos + 1);
+  int second = std::stoi(rest, &pos);
+  // std::cerr << "parse_token: found '" << second << "'\n";
+
+  // insert first-second
+  // std::cerr << "parse_token: range " << first << " to " << second << "\n";
+  for (int i = first; i <= second; ++i) {
+    result.insert(i);
+  }
+  return result;
+}
+
+std::set<int> parse_cpuset(const std::string &s) {
+  // std::cerr << "parse_cpuset: parsing '" << s << "'\n";
+  std::set<int> result;
+
+  std::string token;
+  std::stringstream ss(s);
+  while (std::getline(ss, token, ',')) {
+
+    if ("\n" != token) {
+      auto newCpus = parse_token(token);
+      for (auto cpu : newCpus) {
+        result.insert(cpu);
+      }
+    }
+  }
+
+  return result;
+}
+
+// http://man7.org/linux/man-pages/man7/cpuset.7.html
+namespace perfect {
+class CpuSet {
+public:
+  std::string path_;
+  std::set<int> cpus_;
+  std::set<int> mems_;
+  CpuSet *parent_;
+
+  // make sure cpuset is initialized
+  static Result init() {
+
+    // check for "nodev cpuset" in /proc/filesystems
+
+    // mkdir /dev/cpuset
+    if (mkdir("/dev/cpuset", S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH)) {
+      switch (errno) {
+      case EEXIST: {
+        // okay
+        break;
+      }
+      case EACCES:
+        // std::cerr << "access error in mkdir: " << strerror(errno) << "\n";
+        return Result::NO_PERMISSION;
+      case ENOENT:
+      case EROFS:
+      default:
+        std::cerr << "unhandled error in mkdir: " << strerror(errno) << "\n";
+        return Result::UNKNOWN;
+      }
+
+      // mount -t cpuset none /dev/cpuset
+      if (mount("none", "/dev/cpuset", "cpuset", 0, nullptr)) {
+        switch (errno) {
+        case EEXIST: {
+          // okay
+          break;
+        }
+        case EBUSY: {
+          // FIXME: something is mounted here, assume it is what we want
+          // std::cerr << "EBUSY in mount: " << strerror(errno) << "\n";
+          return Result::SUCCESS;
+        }
+        case EPERM: {
+        // std::cerr << "EPERM in mount: " << strerror(errno) << "\n";
+        return Result::NO_PERMISSION;
+        }
+        case ENOENT:
+        case EROFS:
+        default:
+          std::cerr << "unhandled error in mount: " << strerror(errno) << "\n";
+          return Result::UNKNOWN;
+        }
+      }
+    }
+    return Result::SUCCESS;
+  }
+
+  std::string get_raw_cpus() {
+    std::ifstream is(path_ + "/cpuset.cpus");
+    std::stringstream ss;
+    ss << is.rdbuf();
+    return remove_space(ss.str());
+  }
+
+  std::string get_raw_mems() {
+    std::ifstream is(path_ + "/cpuset.mems");
+    std::stringstream ss;
+    ss << is.rdbuf();
+    return remove_space(ss.str());
+  }
+
+  std::set<int> get_cpus() { return parse_cpuset(get_raw_cpus()); }
+
+  std::set<int> get_mems() { return parse_cpuset(get_raw_mems()); }
+
+  // migrate the caller task from this cpu set to another
+  Result migrate_self_to(CpuSet &other) {
+    // enable memory migration in other
+    other.enable_memory_migration();
+
+    // get my pid
+    pid_t self = this_task();
+
+    // read this tasks and write each line to other.tasks
+    std::ifstream is(path_ + "/tasks");
+    std::string line;
+    while (std::getline(is, line)) {
+      line = remove_space(line);
+      if (std::to_string(self) == line) {
+        // std::cerr << "migrating self task " << line << " to " << other.path
+        //           << "\n";
+        other.write_task(line);
+        return Result::SUCCESS;
+      }
+    }
+    return Result::NO_TASK;
+  }
+
+  // migrate tasks in this cpu set to another
+  Result migrate_tasks_to(CpuSet &other) {
+    // enable memory migration in other
+    SUCCESS_OR_RETURN(other.enable_memory_migration());
+
+    // read this tasks and write each line to other.tasks
+    std::ifstream is(path_ + "/tasks");
+    std::string line;
+    while (std::getline(is, line)) {
+      // std::cerr << "migrating task " << line << " to " << other.path << "\n";
+      other.write_task(line);
+    }
+
+    return Result::SUCCESS;
+  }
+
+  Result enable_memory_migration() {
+    std::ofstream ofs(path_ + "/" + "cpuset.memory_migrate");
+    ofs << "1";
+    ofs.close();
+    if (ofs.fail()) {
+    switch (errno) {
+    case EACCES:
+      return Result::NO_PERMISSION;
+    case ENOENT:
+      return Result::NOT_SUPPORTED;
+    default:
+      return Result::UNKNOWN;
+    }
+  }
+    return Result::SUCCESS;
+  }
+
+  void write_task(const std::string &task) {
+    // write `task` to path/tasks
+    std::ofstream os(path_ + "/tasks");
+    os << task << "\n";
+  }
+
+  // object representing the root CPU set
+  static Result get_root(CpuSet &root) {
+    SUCCESS_OR_RETURN(CpuSet::init());
+    root.path_ = "/dev/cpuset";
+    root.parent_ = nullptr;
+    return Result::SUCCESS;
+  }
+
+  // the ID of this task
+  static pid_t this_task() { return getpid(); }
+
+  Result make_child(CpuSet &child, const std::string &name) {
+
+    if (mkdir((path_ + "/" + name).c_str(),
+              S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH)) {
+      switch (errno) {
+      case EEXIST: {
+        // okay
+        break;
+      }
+      case EACCES:
+        return Result::NO_PERMISSION;
+      case ENOENT:
+      case EROFS:
+      default:
+        return Result::UNKNOWN;
+      }
+    }
+
+    child.path_ = path_ + "/" + name;
+    child.parent_ = this;
+    return Result::SUCCESS;
+  }
+
+  Result enable_cpu(const int cpu) {
+    std::set<int> cpus = get_cpus();
+    cpus.insert(cpu);
+    return write_cpus(cpus);
+  }
+
+  Result enable_cpus(const std::set<int> &cpus) {
+    std::set<int> finalCpus = get_cpus();
+    for (auto cpu : cpus) {
+      finalCpus.insert(cpu);
+    }
+    return write_cpus(finalCpus);
+  }
+
+  // FIXME: check error
+  Result write_cpus(std::set<int> cpus) {
+    std::ofstream os(path_ + "/cpuset.cpus");
+    bool comma = false;
+    for (auto cpu : cpus) {
+      if (comma)
+        os << ",";
+      os << cpu << "-" << cpu;
+      comma = true;
+    }
+    return Result::SUCCESS;
+  }
+
+  // FIXME: check write
+  Result write_mems(std::set<int> mems) {
+    std::ofstream os(path_ + "/cpuset.mems");
+    bool comma = false;
+    for (auto mem : mems) {
+      if (comma)
+        os << ",";
+      os << mem << "-" << mem;
+      comma = true;
+    }
+    return Result::SUCCESS;
+  }
+
+  Result enable_mem(const int mem) {
+    std::set<int> mems = get_mems();
+    mems.insert(mem);
+    return write_mems(mems);
+  }
+
+  Result enable_mems(const std::set<int> &mems) {
+    std::set<int> finalMems = get_mems();
+    for (auto mem : mems) {
+      finalMems.insert(mem);
+    }
+    return write_mems(finalMems);
+  }
+
+  Result destroy() {
+    // remove all child cpu sets
+
+    // move all attached processes back to parent
+    assert(parent_);
+    migrate_tasks_to(*parent_);
+
+    // remove with rmdir
+    if (rmdir(path_.c_str())) {
+      switch (errno) {
+      default:
+        std::cerr << "unhandled error in rmdir: " << strerror(errno) << "\n";
+        return Result::UNKNOWN;
+      }
+    }
+
+    path_ = "";
+    return Result::SUCCESS;
+  }
+
+
+};
+
+  std::ostream &operator<<(std::ostream &s, const CpuSet &c) {
+    s << c.path_;
+    return s;
+  }
+
+} // namespace perfect

include/perfect/detail/cache/power.hpp

@@ -1,5 +1,8 @@
 #pragma once
 
+namespace perfect {
+namespace detail {
+
 inline void flush_line(void *p) {
 
   /*
@@ -32,4 +35,7 @@ inline void barrier_all() {
                : "memory");
 
 
-}
+}
+
+}
+}

include/perfect/detail/fs.hpp

@@ -0,0 +1,51 @@
+#pragma once
+
+#include <fstream>
+#include <string>
+
+#include "../result.hpp"
+
+#ifdef __linux__
+#include "fs/linux.hpp"
+#else
+#error "unsupported platform"
+#endif
+
+namespace perfect {
+namespace detail {
+
+Result write_str(const std::string &path, const std::string &val) {
+
+  if (!path_exists(path)) {
+    std::cerr << "write_str(): does not exist: " << path << "\n";
+    return Result::NOT_SUPPORTED;
+  }
+
+  std::ofstream ofs(path);
+  if (ofs.fail()) {
+    std::cerr << "failed to open " << path << "\n";
+    return Result::NO_PERMISSION;
+  }
+
+  ofs << val;
+  ofs.close();
+  if (ofs.fail()) {
+    switch (errno) {
+    case EACCES:
+    std::cerr << "EACCES when writing to " << path << "\n";
+      return Result::NO_PERMISSION;
+    case EPERM:
+      std::cerr << "EPERM when writing to " << path << "\n";
+      return Result::NO_PERMISSION;
+    case ENOENT:
+    std::cerr << "ENOENT when writing to " << path << "\n";
+      return Result::NOT_SUPPORTED;
+    default:
+      return Result::UNKNOWN;
+    }
+  }
+  return Result::SUCCESS;
+}
+
+} // namespace detail
+} // namespace perfect

include/perfect/detail/fs/linux.hpp

@@ -0,0 +1,31 @@
+#pragma once
+
+#include <string>
+#include <cerrno>
+#include <iostream>
+#include <cassert>
+
+
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+namespace perfect {
+    namespace detail {
+        bool path_exists(const std::string &path) {
+            struct stat sb;
+            if (stat(path.c_str(), &sb)) {
+                switch (errno) {
+                    case ENOENT: return false;
+                    case ENOTDIR: return false;
+                    default: {
+                        std::cerr << "unhandled error in stat() for " << path << "\n";
+                        assert(0);
+                    }
+                }
+            }
+            return true;
+        }
+    }
+}

include/perfect/detail/os/linux.hpp

@@ -12,6 +12,7 @@
 #include <sched.h>
 #include <sys/types.h>
 #include <unistd.h>
+#include <sys/personality.h>
 
 #include "perfect/result.hpp"
 
@@ -88,4 +89,24 @@ size_t cache_linesize() {
 #endif
 }
 
+namespace detail {
+Result get_personality(int &persona) {
+  int ret = personality(0xffffffff);
+  if (-1 == ret) {
+    return Result::UNKNOWN;
+  } else {
+    persona = ret;
+  }
+  return Result::SUCCESS;
+}
+
+Result set_personality(const int persona) {
+  int ret = personality(persona);
+  if (-1 == ret) {
+    return Result::UNKNOWN;
+  }
+  return Result::SUCCESS;
+}
+}
+
 } // namespace perfect

include/perfect/detail/turbo/linux_amd64.hpp

@@ -4,6 +4,7 @@
 #include <fstream>
 
 #include "perfect/result.hpp"
+#include "perfect/detail/fs.hpp"
 
 namespace perfect {
 namespace detail {
@@ -15,19 +16,12 @@ bool has_intel_pstate_no_turbo() {
 Result write_intel_pstate_no_turbo(const std::string &s) {
   assert(has_intel_pstate_no_turbo());
   std::string path("/sys/devices/system/cpu/intel_pstate/no_turbo");
-  std::ofstream ofs(path, std::ofstream::out);
-  ofs << s;
-  ofs.close();
-  if (ofs.fail()) {
-    return Result::NO_PERMISSION;
-  }
-  return Result::SUCCESS;
+  return write_str(path, s);
 }
 
 std::string read_intel_pstate_no_turbo() {
   assert(has_intel_pstate_no_turbo());
   std::string path("/sys/devices/system/cpu/intel_pstate/no_turbo");
-  //   SPDLOG_LOGGER_TRACE(logger::console(), "reading {}", path);
   std::ifstream ifs(path, std::ifstream::in);
   std::string result;
   std::getline(ifs, result);

include/perfect/detail/turbo/linux_power.hpp

@@ -1,31 +1,24 @@
 #pragma once
 
 #include "perfect/result.hpp"
+#include "perfect/detail/fs.hpp"
 
 namespace perfect {
+	namespace detail {
 
 bool has_acpi_cpufreq_boost() {
   return bool(std::ifstream("/sys/devices/system/cpu/cpufreq/boost"));
 }
 
-int write_acpi_cpufreq_boost(const std::string &s) {
+Result write_acpi_cpufreq_boost(const std::string &s) {
   assert(has_acpi_cpufreq_boost());
   std::string path("/sys/devices/system/cpu/cpufreq/boost");
-  SPDLOG_LOGGER_TRACE(logger::console(), "writing to {}", path);
-  std::ofstream ofs(path, std::ofstream::out);
-  ofs << s;
-  ofs.close();
-  if (ofs.fail()) {
-    SPDLOG_LOGGER_TRACE(logger::console(), "error writing to {}", path);
-    return 1;
-  }
-  return 0;
+  return write_str(path, s);
 }
 
 std::string read_acpi_cpufeq_boost() {
   assert(has_acpi_cpufreq_boost());
   std::string path("/sys/devices/system/cpu/cpufreq/boost");
-  SPDLOG_LOGGER_TRACE(logger::console(), "reading {}", path);
   std::ifstream ifs(path, std::ifstream::in);
   std::string result;
   std::getline(ifs, result);
@@ -37,11 +30,12 @@ std::string read_acpi_cpufeq_boost() {
     }
 
     Result disable_cpu_turbo() {
-        write_acpi_cpufeq_boost("0");
+        return write_acpi_cpufreq_boost("0");
     }
 
     Result enable_cpu_turbo() {
-        write_acpi_cpufeq_boost("1");
+        return write_acpi_cpufreq_boost("1");
     }
 
-}
+} // namespace detail
+} // namespace perfect

include/perfect/drop_caches.hpp

@@ -0,0 +1,43 @@
+#pragma once
+
+#include <unistd.h>
+
+#include <fstream>
+#include <iostream>
+
+#include "result.hpp"
+#include "init.hpp"
+#include "detail/fs.hpp"
+
+namespace perfect {
+
+enum DropCaches_t {
+    PAGECACHE = 0x1,
+    ENTRIES = 0x2
+};
+
+
+// commit filesystem caches to disk
+Result sync() {
+    // http://man7.org/linux/man-pages/man2/sync.2.html
+    ::sync(); // always successful
+    return Result::SUCCESS;
+}
+
+Result drop_caches(const DropCaches_t mode) {
+    using detail::write_str;
+    const std::string path = "/proc/sys/vm/drop_caches";
+    if (mode & PAGECACHE & ENTRIES) {
+        PERFECT_SUCCESS_OR_RETURN(write_str(path, "3"));
+    } else if (mode & PAGECACHE) {
+        PERFECT_SUCCESS_OR_RETURN(write_str(path, "1"));
+    } else if (mode & ENTRIES) {
+        PERFECT_SUCCESS_OR_RETURN(write_str(path, "2"));
+    } else {
+        std::cerr << "unexpected mode: " << mode << "\n";
+        return Result::UNKNOWN;
+    }
+    return Result::SUCCESS;
+}
+
+}

include/perfect/gpu_monitor.hpp

@@ -0,0 +1,147 @@
+#pragma once
+
+#ifdef __NVCC__
+#ifndef PERFECT_HAS_CUDA
+#define PERFECT_HAS_CUDA
+#endif
+#endif
+
+#ifdef PERFECT_HAS_CUDA
+#include <nvml.h>
+#endif
+
+#include <atomic>
+#include <chrono>
+#include <string>
+#include <thread>
+
+#include <iostream>
+
+#include "perfect/init.hpp"
+
+namespace perfect {
+
+class Monitor {
+public:
+  struct Config {
+    std::atomic<bool> stop;
+    std::atomic<bool> pause;
+    double samplePeriodMs;
+
+    std::atomic<bool> power;
+    std::atomic<bool> utilization;
+    std::atomic<bool> temperature;
+    std::atomic<bool> pstate;
+    std::ostream *stream_;
+
+    Config(std::ostream *stream)
+        : stop(true), pause(false), power(true), samplePeriodMs(100),
+          utilization(true), temperature(true), pstate(true), stream_(stream) {}
+  };
+
+  std::thread worker;
+  Config config;
+
+  Monitor(std::ostream *stream) : config(stream) {}
+
+  static void worker_func(const Config &cfg) {
+
+    nvmlReturn_t ret;
+    nvmlDevice_t device;
+    unsigned int deviceCount;
+
+    nvmlUtilization_t utilization;
+    unsigned int milliwatts;
+    unsigned int temperature;
+    nvmlPstates_t pState;
+
+    ret = nvmlDeviceGetCount(&deviceCount);
+
+    std::chrono::time_point<std::chrono::system_clock> start;
+
+    while (!cfg.stop.load()) {
+      if (std::chrono::time_point<std::chrono::system_clock>() == start) {
+        start = std::chrono::system_clock::now();
+      }
+      if (!cfg.pause.load()) {
+
+        const double elapsed =
+            (std::chrono::system_clock::now() - start).count() / 1e9 * 1e3;
+
+        for (unsigned int i = 0; i < deviceCount; ++i) {
+
+          (*cfg.stream_) << elapsed << "," << i;
+
+          ret = nvmlDeviceGetHandleByIndex(i, &device);
+
+          if (cfg.power.load()) {
+            ret = nvmlDeviceGetPowerUsage(device, &milliwatts);
+            if (ret == NVML_SUCCESS) {
+              (*cfg.stream_) << "," << milliwatts;
+            } else {
+              (*cfg.stream_) << "," << -1;
+            }
+          } else {
+            (*cfg.stream_) << ","
+                           << "x";
+          }
+          if (cfg.utilization.load()) {
+            // period is between 1 second and 1/6 second depending on product
+            ret = nvmlDeviceGetUtilizationRates(device, &utilization);
+            if (ret == NVML_SUCCESS) {
+              (*cfg.stream_)
+                  << "," << utilization.gpu << "," << utilization.memory;
+            } else {
+              (*cfg.stream_) << "," << -1 << "," << -1;
+            }
+          } else {
+            (*cfg.stream_) << ","
+                           << "x"
+                           << ","
+                           << "x";
+          }
+          if (cfg.temperature.load()) {
+            ret = nvmlDeviceGetTemperature(device, NVML_TEMPERATURE_GPU,
+                                           &temperature);
+            if (ret == NVML_SUCCESS) {
+              (*cfg.stream_) << "," << temperature;
+            } else {
+              (*cfg.stream_) << "," << -1;
+            }
+          } else {
+            (*cfg.stream_) << ","
+                           << "x";
+          }
+          if (cfg.pstate.load()) {
+            ret = nvmlDeviceGetPerformanceState(device, &pState);
+            if (ret == NVML_SUCCESS) {
+              (*cfg.stream_) << "," << pState;
+            } else {
+              (*cfg.stream_) << "," << -1;
+            }
+          } else {
+            (*cfg.stream_) << ","
+                           << "x";
+          }
+
+          (*cfg.stream_) << "\n";
+        }
+      }
+      std::this_thread::sleep_for(std::chrono::milliseconds(100));
+    }
+  }
+
+  void start() {
+    config.stop = false;
+    worker = std::thread(worker_func, std::ref(config));
+  }
+  void stop() {
+    config.stop = true;
+    worker.join();
+  }
+
+  void resume() { config.pause.store(false); }
+  void pause() { config.pause.store(true); }
+};
+
+} // namespace perfect

include/perfect/init.hpp

@@ -1,13 +1,17 @@
 #pragma once
 
 #ifdef __NVCC__
+#ifndef PERFECT_HAS_CUDA
 #define PERFECT_HAS_CUDA
 #endif
+#endif
 
 #ifdef PERFECT_HAS_CUDA
 #include <nvml.h>
 #endif
 
+#include "perfect/result.hpp"
+
 namespace perfect {
 
 /*! initialize the benchmark

include/perfect/os_perf.hpp

@@ -40,6 +40,14 @@ Result os_perf_state_maximum(const int cpu) {
     #endif
 }
 
+Result os_perf_state_minimum(const int cpu) {
+    #ifdef __linux__
+    return set_governor(cpu, "powersave");
+    #else
+    #error "unsupported platform"
+    #endif
+}
+
 Result set_os_perf_state(const int cpu, OsPerfState state) {
         #ifdef __linux__
     return set_governor(cpu, state.governor);

include/perfect/result.hpp

@@ -3,8 +3,10 @@
 #include <cassert>
 
 #ifdef __NVCC__
+#ifndef PERFECT_HAS_CUDA
 #define PERFECT_HAS_CUDA
 #endif
+#endif
 
 #ifdef PERFECT_HAS_CUDA
 #include <nvml.h>
@@ -55,6 +57,8 @@ const char *get_string(const Result &result) {
     return "nvidia-ml returned not supported";
   case Result::NVML_NO_PERMISSION:
     return "nvidia-ml returned no permission";
+  case Result::NVML_UNINITIALIZED:
+    return "nvidia-ml returned uninitialized";
   case Result::NOT_SUPPORTED:
     return "unsupported operation";
   default:
@@ -76,3 +80,12 @@ inline void check(Result result, const char *file, const int line) {
 } // namespace perfect
 
 #define PERFECT(stmt) check(stmt, __FILE__, __LINE__);
+
+#define PERFECT_SUCCESS_OR_RETURN(stmt) \
+{\
+  Result _ret; \
+  _ret = (stmt); \
+if (_ret != Result::SUCCESS) {\
+  return _ret;\
+}\
+}

tools/CMakeLists.txt

@@ -34,6 +34,31 @@ set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} \
  -Wfatal-errors\
  ")
 
-add_executable(enable-turbo enable_turbo.cpp)
-target_link_libraries(enable-turbo perfect)
+add_executable(enable-cpu-turbo enable_cpu_turbo.cpp)
+target_link_libraries(enable-cpu-turbo perfect)
+
+add_executable(disable-cpu-turbo disable_cpu_turbo.cpp)
+target_link_libraries(disable-cpu-turbo perfect)
+
+add_executable(sync-drop-caches sync_drop_caches.cpp)
+target_link_libraries(sync-drop-caches perfect)
+
+add_executable(no-aslr no_aslr.cpp)
+target_link_libraries(no-aslr perfect)
+
+add_executable(max-os-perf max_os_perf.cpp)
+target_link_libraries(max-os-perf perfect)
+
+add_executable(min-os-perf min_os_perf.cpp)
+target_link_libraries(min-os-perf perfect)
+
+## OpenMP
+find_package(OpenMP)
+if (OpenMP_FOUND)
+  add_executable(stress stress.cpp)
+  target_link_libraries(stress perfect)
+  target_link_libraries(stress OpenMP::OpenMP_CXX)
+else(OpenMP_FOUND)
+  message(WARNING "didn't find OpenMP, some benchmarks will be unavailable.")
+endif(OpenMP_FOUND)
 

tools/disable_cpu_turbo.cpp

@@ -0,0 +1,23 @@
+#include <iostream>
+
+#include "perfect/cpu_turbo.hpp"
+
+using namespace perfect;
+
+int main(void) {
+
+  CpuTurboState state;
+
+  perfect::init();
+
+  PERFECT(get_cpu_turbo_state(&state));
+
+  if (!is_turbo_enabled(state)) {
+    std::cerr << "cpu turbo already disabled\n";
+    exit(EXIT_SUCCESS);
+  } else {
+    PERFECT(disable_cpu_turbo());
+    std::cerr << "disabled cpu turbo\n";
+      exit(EXIT_SUCCESS);
+  }
+}

tools/enable_cpu_turbo.cpp

@@ -6,7 +6,6 @@ using namespace perfect;
 
 int main(void) {
 
-  Result ret;
   CpuTurboState state;
 
   perfect::init();

tools/max_os_perf.cpp

@@ -0,0 +1,9 @@
+#include "perfect/os_perf.hpp"
+
+int main(void) {
+  PERFECT(perfect::init());
+
+  for (auto cpu : perfect::cpus()) {
+    PERFECT(perfect::os_perf_state_maximum(cpu));
+  }
+}

tools/min_os_perf.cpp

@@ -0,0 +1,9 @@
+#include "perfect/os_perf.hpp"
+
+int main(void) {
+  PERFECT(perfect::init());
+
+  for (auto cpu : perfect::cpus()) {
+    PERFECT(perfect::os_perf_state_minimum(cpu));
+  }
+}

tools/no_aslr.cpp

@@ -0,0 +1,77 @@
+#include <iostream>
+
+#include <errno.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <cstring>
+#include <vector>
+
+#include "perfect/aslr.hpp"
+
+using namespace perfect;
+
+int main(int argc, char **argv) {
+
+  using namespace perfect;
+
+  PERFECT(init());
+
+  pid_t pid;
+  int status;
+  pid = fork();
+  if (pid == -1) {
+    // pid == -1 means error occured
+    std::cerr << "can't fork, error occured\n";
+    exit(EXIT_FAILURE);
+  } else if (pid == 0) {
+    // in the child process
+
+    // skip the first argument, which is this program
+    std::vector<char*> args;
+    for (int i = 1; i < argc; ++i) {
+      args.push_back(argv[i]);
+    }
+    assert(args.size() > 0);
+    args.push_back(nullptr);
+
+    PERFECT(disable_aslr());
+
+    // the execv() only return if error occured.
+    // The return value is -1
+    return execvp(args[0], args.data());
+  } else {
+    // parent process
+
+    if (waitpid(pid, &status, 0) > 0) {
+
+      if (WIFEXITED(status) && !WEXITSTATUS(status)) {
+        // success
+        exit(status);
+      }
+
+      else if (WIFEXITED(status) && WEXITSTATUS(status)) {
+        if (WEXITSTATUS(status) == 127) {
+
+          // execv failed
+          std::cerr << "execv failed\n";
+          exit(status);
+        } else {
+          std::cerr << "program terminated normally, but returned a non-zero status\n";
+          exit(status);
+        }
+      } else {
+        printf("program didn't terminate normally\n");
+        exit(status);
+      }
+    } else {
+      // waitpid() failed
+      printf("waitpid() failed\n");
+      exit(EXIT_FAILURE);
+    }
+    exit(0);
+  }
+  return 0;
+}

tools/stress.cpp

@@ -0,0 +1,49 @@
+#include <iostream>
+
+#include <cmath>
+
+#include <omp.h>
+#include <cstring>
+#include <chrono>
+#include <vector>
+#include <iostream>
+
+int main(int argc, char **argv) {
+
+size_t numThreads = std::stoi(argv[1]);
+std::vector<size_t> totals(numThreads, 0);
+omp_set_num_threads(numThreads);
+
+auto start = std::chrono::system_clock::now();
+double time = std::stod(argv[2]);
+
+#pragma omp parallel
+{
+  size_t tid = omp_get_thread_num();
+  double a = rand();
+  while (true) {
+    for (size_t i = 0; i < 500; ++i) {
+      double x;
+      asm volatile(""::"r"(a));
+      x = sqrt(a);
+      asm volatile(""::"r"(x));
+
+      asm volatile(""::"r"(a));
+      x = sqrt(a);
+      asm volatile(""::"r"(x));
+    }
+    totals[tid] += 1000;
+    auto elapsed = (std::chrono::system_clock::now() - start).count() / 1e9;
+    if (elapsed > time) {
+      break;
+    }
+  }
+}
+
+size_t sum = 0;
+for (auto t : totals) {
+  sum += t;
+}
+std::cout << (double)sum / time << "\n";
+
+};

tools/sync_drop_caches.cpp

@@ -0,0 +1,14 @@
+#include <iostream>
+
+#include "perfect/drop_caches.hpp"
+
+using namespace perfect;
+
+int main(void) {
+
+  using namespace perfect;
+
+  PERFECT(init());
+  PERFECT(perfect::sync());
+  PERFECT(drop_caches(DropCaches_t(PAGECACHE | ENTRIES)));
+}