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 0.3.0)
+project(perfect LANGUAGES CXX VERSION 0.5.0)
 message(STATUS "Build type: " ${CMAKE_BUILD_TYPE})
 
 include(CheckLanguage)

README.md

@@ -2,12 +2,9 @@
 
 | 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
 
@@ -19,6 +16,10 @@ CPU/GPU performance control library for benchmarking
 - [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
 
@@ -85,6 +86,20 @@ See [examples/gpu_monitor.cu](examples/gpu_monitor.cu)
 * `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
@@ -175,6 +190,16 @@ See [examples/cpu_cache.cpp](examples/cpu_cache.cpp).
 
 ## 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
@@ -191,8 +216,17 @@ See [examples/cpu_cache.cpp](examples/cpu_cache.cpp).
 ## Wish List
 
 - [ ] only monitor certain GPUs
+- [ ] hyperthreading interface
+- [ ] process priority interface
 - [ ] A wrapper utility
     - [ ] disable hyperthreading
     - [ ] reserve cores 
     - [ ] set process priority
     - [ ] disable ASLR
+
+## 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

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/detail/cache/power.hpp

@@ -1,5 +1,8 @@
 #pragma once
 
+namespace perfect {
+namespace detail {
+
 inline void flush_line(void *p) {
 
   /*
@@ -33,3 +36,6 @@ inline void barrier_all() {
 
 
 }
+
+}
+}

include/perfect/detail/fs.hpp

@@ -3,14 +3,28 @@
 #include <fstream>
 #include <string>
 
-#include "perfect/result.hpp"
+#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()) {
-    return Result::NOT_SUPPORTED;
+    std::cerr << "failed to open " << path << "\n";
+    return Result::NO_PERMISSION;
   }
 
   ofs << val;
@@ -24,6 +38,7 @@ Result write_str(const std::string &path, const std::string &val) {
       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;

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/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

@@ -57,6 +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:

tools/CMakeLists.txt

@@ -34,8 +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";
+
+};