1 /* 2 * Copyright (c) 2021-2022 Huawei Device Co., Ltd. 3 * Licensed under the Apache License, Version 2.0 (the "License"); 4 * you may not use this file except in compliance with the License. 5 * You may obtain a copy of the License at 6 * 7 * http://www.apache.org/licenses/LICENSE-2.0 8 * 9 * Unless required by applicable law or agreed to in writing, software 10 * distributed under the License is distributed on an "AS IS" BASIS, 11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 * See the License for the specific language governing permissions and 13 * limitations under the License. 14 */ 15 #ifndef SUBCOMMAND_RECORD_H 16 #define SUBCOMMAND_RECORD_H 17 18 // some debug feaure 19 #define HIDEBUG_RECORD_NOT_PROCESS 0 20 #define HIDEBUG_RECORD_NOT_PROCESS_VM 0 21 #define HIDEBUG_RECORD_NOT_SAVE 0 22 #define HIDEBUG_SKIP_PROCESS_SYMBOLS 0 23 #define HIDEBUG_SKIP_MATCH_SYMBOLS 0 24 #define HIDEBUG_SKIP_LOAD_KERNEL_SYMBOLS 0 25 #define HIDEBUG_SKIP_SAVE_SYMBOLS 0 26 #define USE_COLLECT_SYMBOLIC 1 27 28 #include <functional> 29 #include <thread> 30 #include <unordered_map> 31 #include <unordered_set> 32 #include <chrono> 33 #include "perf_event_record.h" 34 #include "perf_events.h" 35 #include "perf_file_writer.h" 36 #include "subcommand.h" 37 #include "virtual_runtime.h" 38 39 namespace OHOS { 40 namespace Developtools { 41 namespace HiPerf { 42 class SubCommandRecord : public SubCommand { 43 public: 44 static constexpr int DEFAULT_CPU_PERCENT = 25; 45 static constexpr int MIN_CPU_PERCENT = 1; 46 static constexpr int MAX_CPU_PERCENT = 100; 47 static constexpr int MIN_SAMPLE_FREQUENCY = 1; 48 static constexpr int MAX_SAMPLE_FREQUENCY = 100000; 49 static constexpr int DEFAULT_MMAP_PAGES = 256; 50 static constexpr int MIN_PERF_MMAP_PAGE = 2; 51 static constexpr int MAX_PERF_MMAP_PAGE = 1024; 52 static constexpr int DEFAULT_CHECK_APP_MS = 10; 53 static constexpr int MIN_CHECK_APP_MS = 1; 54 static constexpr int MAX_CHECK_APP_MS = 200; 55 static constexpr float MIN_STOP_SECONDS = 0.100; 56 static constexpr float MAX_STOP_SECONDS = 10000.0; 57 SubCommandRecord()58 SubCommandRecord() 59 // clang-format off 60 : SubCommand("record", "Collect performance sample information", 61 "Usage: hiperf record [options] [command [command-args]]\n" 62 " Collect performance sampling information of running [command].\n" 63 " The default options are: -c <all cpu> --cpu-limit 25 -d 10000.0 -e hw-cpu-cycles\n" 64 " -f 4000 -m 1024 -o /data/local/tmp/perf.data.\n" 65 " -a\n" 66 " Collect system-wide information.\n" 67 " for measures all processes/threads\n" 68 " This requires CAP_PERFMON (since Linux 5.8) or CAP_SYS_ADMIN capability or a\n" 69 " /proc/sys/kernel/perf_event_paranoid value of less than 1.\n" 70 " --exclude-hiperf\n" 71 " Don't record events issued by hiperf itself.\n" 72 " -c <cpuid>[<,cpuid>]...\n" 73 " cpuid should be 0,1,2...\n" 74 " Limit the CPU that collects data.\n" 75 " 0 means cpu0, 1 means cpu1 ...\n" 76 " --cpu-limit <percent>\n" 77 " Set the max percent of cpu time used for recording.\n" 78 " percent is in range [1-100], default is 25.\n" 79 " -d <sec>\n" 80 " stop in <sec> seconds. floating point number. seconds is in range [0.100-10000.0]\n" 81 " default is 10000.0\n" 82 " -f <freq>\n" 83 " Set event sampling frequency. default is 4000 samples every second.\n" 84 " check /proc/sys/kernel/perf_event_max_sample_rate for maximum allowed frequency\n" 85 " --period <num>\n" 86 " Set event sampling period for tracepoint events. recording one sample when <num> events happen.\n" 87 " The default <num> is 1\n" 88 " -e <event1[:<u|k>]>[,event1[:<u|k>]]...\n" 89 " Customize the name of the event that needs to be sampled.\n" 90 " The name can use the names listed in the list parameter.\n" 91 " It can also be represented by the value of 0x<hex>.\n" 92 " u - monitor user space events only\n" 93 " k - monitor kernel space events only\n" 94 " -g <event1[:<u|k>]>[,event1[:<u|k>]]...\n" 95 " Put the events into a group, can set multiple groups by multiple -g\n" 96 " PMU is required to report data in designated groups\n" 97 " limited by HW capability, too many events cannot be reported in the same sampling)\n" 98 " --no-inherit\n" 99 " Don't trace child processes.\n" 100 " -p <pid1>[,pid2]...\n" 101 " Limit the process id of the collection target. Conflicts with the -a option.\n" 102 " -t <tid1>[,tid2]...\n" 103 " Limit the thread id of the collection target. Conflicts with the -a option.\n" 104 " --exclude-tid <tid1>[,tid2]...\n" 105 " Exclude threads of the collection target by thread ids. Conflicts with the -a option.\n" 106 " --exclude-thread <tname1>[,tname2]...\n" 107 " Exclude threads of the collection target by thread names. Conflicts with the -a option.\n" 108 " --offcpu\n" 109 " Trace when threads are scheduled off cpu.\n" 110 " -j <branch_filter1>[,branch_filter2]...\n" 111 " taken branch stack sampling, filter can be:\n" 112 " any: any type of branch\n" 113 " any_call: any function call or system call\n" 114 " any_ret: any function return or system call return\n" 115 " ind_call: any indirect branch\n" 116 " call: direct calls, including far (to/from kernel) calls\n" 117 " u: only when the branch target is at the user level\n" 118 " k: only when the branch target is in the kernel\n" 119 " requires at least one of any, any_call, any_ret, ind_call\n" 120 " -s / --call-stack <fp|dwarf[,size]>\n" 121 " Setup and enable call stack (stack chain/backtrace) recording, Default is 'fp'.\n" 122 " the value can be:\n" 123 " fp: frame pointer\n" 124 " dwarf: DWARF's CFI - Call Frame Information\n" 125 " 'dwarf,size' set sample stack size, size should be in 8~65528 and 8 byte aligned. \n" 126 " as the method to collect the information used to show the call stacks.\n" 127 " --delay-unwind\n" 128 " If '-s dwarf' used, stack will be unwind while recording, use this option to switch\n" 129 " to unwind after recording.\n" 130 " --disable-unwind\n" 131 " If '-s dwarf' is used, stack will be unwind while recording by default\n" 132 " use this option to disable unwinding.\n" 133 " --disable-callstack-expand\n" 134 " If '-s dwarf' is used, to break the 64k stack limit, callstack is merged by default\n" 135 " to build more complete call stack. that may not be correct sometimes.\n" 136 " --clockid <clock_id>\n" 137 " Set the clock id to use for the various time fields in the perf_event_type records.\n" 138 " monotonic and monotonic_raw are supported,\n" 139 " some events might also allow boottime, realtime and clock_tai.\n" 140 " --symbol-dir <dir>\n" 141 " Set directory to look for symbol files, used for unwinding. \n" 142 " -m <mmap_pages>\n" 143 " Number of the mmap pages, used to receiving record data from kernel,\n" 144 " must be a power of two, rang[2,1024], default is 1024.\n" 145 " --app <package_name>\n" 146 " Collect profile info for an OHOS app, the app must be debuggable.\n" 147 " Record will exit if the process is not started within 10 seconds.\n" 148 " --chkms <millisec>\n" 149 " Set the interval of querying the <package_name>.\n" 150 " <millisec> is in range [1-200], default is 10.\n" 151 " --data-limit <SIZE[K|M|G]>\n" 152 " Stop recording after SIZE bytes of records. Default is unlimited.\n" 153 " -o <output_file_name>\n" 154 " Set output file name, default is /data/local/tmp/perf.data.\n" 155 " -z\n" 156 " Compress record data.\n" 157 " --restart\n" 158 " Collect performance counter information of application startup.\n" 159 " Record will exit if the process is not started within 30 seconds.\n" 160 " --verbose\n" 161 " Show more detailed reports.\n" 162 " --control <command>\n" 163 " Control sampling by <command>, the <command> can be:\n" 164 " prepare: set arguments and prepare sampling\n" 165 " start: start sampling\n" 166 " pause: pause sampling\n" 167 " resume: resume sampling\n" 168 " stop: stop sampling\n" 169 ) 170 // clang-format on 171 { 172 } 173 174 ~SubCommandRecord(); 175 bool OnSubCommand(std::vector<std::string> &args) override; 176 bool ParseOption(std::vector<std::string> &args) override; 177 void DumpOptions(void) const override; 178 179 static bool RegisterSubCommandRecord(void); 180 181 private: 182 PerfEvents perfEvents_; 183 184 bool targetSystemWide_ = false; 185 bool compressData_ = false; 186 bool noInherit_ = false; 187 bool excludeHiperf_ = false; 188 bool offCPU_ = false; 189 bool delayUnwind_ = false; 190 bool disableUnwind_ = false; 191 bool disableCallstackExpend_ = false; 192 bool verboseReport_ = false; 193 float timeStopSec_ = PerfEvents::DEFAULT_TIMEOUT; 194 int frequency_ = 0; 195 int period_ = 0; 196 int cpuPercent_ = DEFAULT_CPU_PERCENT; 197 int mmapPages_ = MAX_PERF_MMAP_PAGE; 198 std::vector<std::string> symbolDir_ = {}; 199 std::string outputFilename_ = "/data/local/tmp/perf.data"; 200 std::string appPackage_ = {}; 201 int checkAppMs_ = DEFAULT_CHECK_APP_MS; 202 std::string clockId_ = {}; 203 std::string strLimit_ = {}; 204 std::vector<pid_t> selectCpus_ = {}; 205 std::vector<pid_t> selectPids_ = {}; 206 std::vector<pid_t> selectTids_ = {}; 207 std::vector<pid_t> excludeTids_ = {}; 208 bool restart_ = false; 209 std::vector<std::string> selectEvents_ = {}; 210 std::vector<std::vector<std::string>> selectGroups_ = {}; 211 std::vector<std::string> callStackType_ = {}; 212 std::vector<std::string> vecBranchFilters_ = {}; 213 std::vector<std::string> trackedCommand_ = {}; 214 std::vector<std::string> excludeThreadNames_ = {}; 215 216 bool GetOptions(std::vector<std::string> &args); 217 bool CheckOptions(); 218 bool CheckDataLimitOption(); 219 bool CheckSelectCpuPidOption(); 220 bool GetOptionFrequencyAndPeriod(std::vector<std::string> &args); 221 222 bool isCallStackDwarf_ = false; 223 bool isCallStackFp_ = false; 224 uint32_t callStackDwarfSize_ = MAX_SAMPLE_STACK_SIZE; 225 uint64_t branchSampleType_ = 0; 226 uint64_t dataSizeLimit_ = 0; 227 bool isDataSizeLimitStop_ = false; 228 229 std::unique_ptr<PerfFileWriter> fileWriter_ = nullptr; 230 231 // for client 232 int clientPipeInput_ = -1; 233 int clientPipeOutput_ = -1; 234 int nullFd_ = -1; 235 std::thread clientCommandHanle_; 236 bool clientExit_ = false; 237 void ClientCommandHandle(); 238 bool ClientCommandResponse(bool OK); 239 bool IsSamplingRunning(); 240 // for cmdline client 241 std::string controlCmd_ = {}; 242 bool isFifoServer_ = false; 243 bool isFifoClient_ = false; 244 bool ProcessControl(); 245 bool CreateFifoServer(); 246 bool SendFifoAndWaitReply(const std::string &cmd, const std::chrono::milliseconds &timeOut); 247 bool WaitFifoReply(int fd, const std::chrono::milliseconds &timeOut); 248 void CloseClientThread(); 249 250 bool PreparePerfEvent(); 251 bool PrepareSysKernel(); 252 bool PrepareVirtualRuntime(); 253 254 size_t recordSamples_ = 0; 255 size_t recordNoSamples_ = 0; 256 257 bool isNeedSetPerfHarden_ = false; 258 259 // callback to process record 260 bool ProcessRecord(std::unique_ptr<PerfEventRecord>); 261 bool SaveRecord(std::unique_ptr<PerfEventRecord>); 262 263 // file format like as 0,1-3,4-6,7,8 264 uint32_t GetCountFromFile(const std::string &fileName); 265 std::string GetCpuDescFromFile(); 266 bool AddCpuFeature(); 267 void AddMemTotalFeature(); 268 void AddEventDescFeature(); 269 void AddRecordTimeFeature(); 270 void AddWorkloadCmdFeature(); 271 void AddCommandLineFeature(); 272 void AddCpuOffFeature(); 273 bool AddFeatureRecordFile(); 274 275 bool CreateInitRecordFile(bool compressData = false); 276 bool FinishWriteRecordFile(); 277 bool PostProcessRecordFile(); 278 bool RecordCompleted(); 279 #ifdef HIPERF_DEBUG_TIME 280 void ReportTime(); 281 #endif 282 283 bool CollectionSymbol(std::unique_ptr<PerfEventRecord> record); 284 285 bool SetPerfLimit(const std::string& file, int value, std::function<bool (int, int)> const& cmd, 286 const std::string& param); 287 bool SetPerfCpuMaxPercent(); 288 bool SetPerfMaxSampleRate(); 289 bool SetPerfEventMlock(); 290 bool SetPerfHarden(); 291 292 bool TraceOffCpu(); 293 bool ParseCallStackOption(const std::vector<std::string> &callStackType); 294 bool ParseDataLimitOption(const std::string &str); 295 bool ParseBranchSampleType(const std::vector<std::string> &vecBranchSampleTypes); 296 bool ParseControlCmd(const std::string cmd); 297 bool CheckTargetProcessOptions(); 298 bool CheckTargetPids(); 299 300 VirtualRuntime virtualRuntime_; 301 #if USE_COLLECT_SYMBOLIC 302 std::unordered_set<uint64_t> kernelSymbolsHits_; 303 std::unordered_map<pid_t, std::unordered_set<uint64_t>> userSymbolsHits_; 304 void SymbolicHits(); 305 #endif 306 307 #ifdef HIPERF_DEBUG_TIME 308 std::chrono::microseconds prcessRecordTimes_ = std::chrono::microseconds::zero(); 309 std::chrono::microseconds saveRecordTimes_ = std::chrono::microseconds::zero(); 310 std::chrono::microseconds saveFeatureTimes_ = std::chrono::microseconds::zero(); 311 #endif 312 std::chrono::time_point<std::chrono::steady_clock> startSaveFileTimes_; 313 }; 314 } // namespace HiPerf 315 } // namespace Developtools 316 } // namespace OHOS 317 #endif // SUBCOMMAND_RECORD_H 318