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 static constexpr int MIN_SAVED_CMDLINES_SIZE = 512; 58 static constexpr int DEFAULT_SAVED_CMDLINES_SIZE = 2048; 59 static constexpr int MAX_SAVED_CMDLINES_SIZE = 4096; 60 static constexpr uint64_t MIN_BACKTRACK_TIME_SEC = 5; 61 static constexpr uint64_t DEFAULT_BACKTRACK_TIME_SEC = 10; 62 static constexpr uint64_t MAX_BACKTRACK_TIME_SEC = 30; 63 SubCommandRecord()64 SubCommandRecord() 65 // clang-format off 66 : SubCommand("record", "Collect performance sample information", 67 "Usage: hiperf record [options] [command [command-args]]\n" 68 " Collect performance sampling information of running [command].\n" 69 " The default options are: -c <all cpu> --cpu-limit 25 -d 10000.0 -e hw-cpu-cycles\n" 70 " -f 4000 -m 1024 -o /data/local/tmp/perf.data.\n" 71 " -a\n" 72 " Collect system-wide information.\n" 73 " for measures all processes/threads\n" 74 " This requires CAP_PERFMON (since Linux 5.8) or CAP_SYS_ADMIN capability or a\n" 75 " /proc/sys/kernel/perf_event_paranoid value of less than 1.\n" 76 " --exclude-hiperf\n" 77 " Don't record events issued by hiperf itself.\n" 78 " -c <cpuid>[<,cpuid>]...\n" 79 " cpuid should be 0,1,2...\n" 80 " Limit the CPU that collects data.\n" 81 " 0 means cpu0, 1 means cpu1 ...\n" 82 " --cpu-limit <percent>\n" 83 " Set the max percent of cpu time used for recording.\n" 84 " percent is in range [1-100], default is 25.\n" 85 " -d <sec>\n" 86 " stop in <sec> seconds. floating point number. seconds is in range [0.100-10000.0]\n" 87 " default is 10000.0\n" 88 " -f <freq>\n" 89 " Set event sampling frequency. default is 4000 samples every second.\n" 90 " check /proc/sys/kernel/perf_event_max_sample_rate for maximum allowed frequency\n" 91 " --period <num>\n" 92 " Set event sampling period for tracepoint events. recording one sample when <num> events happen.\n" 93 " The default <num> is 1\n" 94 " -e <event1[:<u|k>]>[,event1[:<u|k>]]...\n" 95 " Customize the name of the event that needs to be sampled.\n" 96 " The name can use the names listed in the list parameter.\n" 97 " It can also be represented by the value of 0x<hex>.\n" 98 " u - monitor user space events only\n" 99 " k - monitor kernel space events only\n" 100 " -g <event1[:<u|k>]>[,event1[:<u|k>]]...\n" 101 " Put the events into a group, can set multiple groups by multiple -g\n" 102 " PMU is required to report data in designated groups\n" 103 " limited by HW capability, too many events cannot be reported in the same sampling)\n" 104 " --no-inherit\n" 105 " Don't trace child processes.\n" 106 " -p <pid1>[,pid2]...\n" 107 " Limit the process id of the collection target. Conflicts with the -a option.\n" 108 " -t <tid1>[,tid2]...\n" 109 " Limit the thread id of the collection target. Conflicts with the -a option.\n" 110 " --exclude-tid <tid1>[,tid2]...\n" 111 " Exclude threads of the collection target by thread ids. Conflicts with the -a option.\n" 112 " --exclude-thread <tname1>[,tname2]...\n" 113 " Exclude threads of the collection target by thread names. Conflicts with the -a option.\n" 114 " --exclude-process <pname1>[,pname2]...\n" 115 " Exclude processes by process names. Must be used with -a.\n" 116 " --offcpu\n" 117 " Trace when threads are scheduled off cpu.\n" 118 " -j <branch_filter1>[,branch_filter2]...\n" 119 " taken branch stack sampling, filter can be:\n" 120 " any: any type of branch\n" 121 " any_call: any function call or system call\n" 122 " any_ret: any function return or system call return\n" 123 " ind_call: any indirect branch\n" 124 " ind_jmp: any indirect jump\n" 125 " cond: conditional branches\n" 126 " call: direct calls, including far (to/from kernel) calls\n" 127 " u: only when the branch target is at the user level\n" 128 " k: only when the branch target is in the kernel\n" 129 " requires at least one of any, any_call, any_ret, ind_call, ind_jmp, cond, call\n" 130 " -s / --call-stack <fp|dwarf[,size]>\n" 131 " Setup and enable call stack (stack chain/backtrace) recording, Default is 'fp'.\n" 132 " the value can be:\n" 133 " fp: frame pointer\n" 134 " dwarf: DWARF's CFI - Call Frame Information\n" 135 " 'dwarf,size' set sample stack size, size should be in 8~65528 and 8 byte aligned. \n" 136 " as the method to collect the information used to show the call stacks.\n" 137 " --kernel-callchain\n" 138 " collect kernel callchain, must used with -s fp/dwarf simultaneously.\n" 139 " --callchain-useronly\n" 140 " collect only user callchain.\n" 141 " --delay-unwind\n" 142 " If '-s dwarf' used, stack will be unwind while recording, use this option to switch\n" 143 " to unwind after recording.\n" 144 " --disable-unwind\n" 145 " If '-s dwarf' is used, stack will be unwind while recording by default\n" 146 " use this option to disable unwinding.\n" 147 " --disable-callstack-expand\n" 148 " If '-s dwarf' is used, to break the 64k stack limit, callstack is merged by default\n" 149 " to build more complete call stack. that may not be correct sometimes.\n" 150 " --enable-debuginfo-symbolic\n" 151 " If '-s fp/dwarf' is used, symbols in .gnu_debugdata section of an elf, also called minidebuginfo\n" 152 " will be parsed, if not use this option, we will not parse minidebuginfo by default.\n" 153 " --clockid <clock_id>\n" 154 " Set the clock id to use for the various time fields in the perf_event_type records.\n" 155 " monotonic and monotonic_raw are supported,\n" 156 " some events might also allow boottime, realtime and clock_tai.\n" 157 " --symbol-dir <dir>\n" 158 " Set directory to look for symbol files, used for unwinding. \n" 159 " -m <mmap_pages>\n" 160 " Number of the mmap pages, used to receiving record data from kernel,\n" 161 " must be a power of two, rang[2,1024], default is 1024.\n" 162 " --app <package_name>\n" 163 " Collect profile info for an OHOS app, the app must be debuggable.\n" 164 " Record will exit if the process is not started within 20 seconds.\n" 165 " --chkms <millisec>\n" 166 " Set the interval of querying the <package_name>.\n" 167 " <millisec> is in range [1-200], default is 10.\n" 168 " --data-limit <SIZE[K|M|G]>\n" 169 " Stop recording after SIZE bytes of records. Default is unlimited.\n" 170 " -o <output_file_name>\n" 171 " Set output file name, default is /data/local/tmp/perf.data.\n" 172 " -z\n" 173 " Compress record data.\n" 174 " --restart\n" 175 " Collect performance counter information of application startup.\n" 176 " Record will exit if the process is not started within 30 seconds.\n" 177 " --verbose\n" 178 " Show more detailed reports.\n" 179 " --control <command>\n" 180 " Control sampling by <command>, the <command> can be:\n" 181 " prepare: set arguments and prepare sampling\n" 182 " start: start sampling\n" 183 " pause: pause sampling\n" 184 " resume: resume sampling\n" 185 " output: output sampling data\n" 186 " stop: stop sampling\n" 187 " --dedup_stack\n" 188 " Remove duplicated stacks in perf record, conflicts with -a, only restrain using with -p\n" 189 " --cmdline-size <size>\n" 190 " set value to /sys/kernel/tracing/saved_cmdlines_size\n" 191 " the value should be between 512 and 4096\n" 192 " --report\n" 193 " Report with callstack after record. Conflicts with the -a option.\n" 194 " --backtrack\n" 195 " Collect data of the previous period. only restrain using with --control.\n" 196 " --backtrack-sec\n" 197 " If '--backtrack' is used, stop in <sec> seconds. seconds is in range [5-30]\n" 198 " default is 10\n" 199 " --dumpoptions\n" 200 " Dump command options.\n" 201 ) 202 // clang-format on 203 { 204 } 205 206 ~SubCommandRecord(); 207 HiperfError OnSubCommand(std::vector<std::string>& args) override; 208 bool ParseOption(std::vector<std::string> &args) override; 209 void DumpOptions(void) const override; 210 211 // add args for hisysevent 212 void AddReportArgs(CommandReporter& reporter) override; 213 214 static bool RegisterSubCommandRecord(void); 215 std::map<const std::string, uint64_t> speOptMap_ = { 216 {"branch_filter", 0}, {"load_filter", 0}, 217 {"store_filter", 0}, {"ts_enable", 0}, 218 {"pa_enable", 0}, {"jitter", 0}, 219 {"min_latency", 0}, {"event_filter", 0}, 220 {"pct_enable", 0}, 221 }; 222 223 static SubCommand& GetInstance(); 224 225 private: 226 PerfEvents perfEvents_; 227 228 bool targetSystemWide_ = false; 229 bool compressData_ = false; 230 bool noInherit_ = false; 231 bool excludeHiperf_ = false; 232 bool offCPU_ = false; 233 bool delayUnwind_ = false; 234 bool disableUnwind_ = false; 235 bool disableCallstackExpend_ = false; 236 bool enableDebugInfoSymbolic_ = false; 237 bool verboseReport_ = false; 238 bool kernelCallChain_ = true; 239 bool callChainUserOnly_ = false; 240 bool report_ = false; 241 float timeStopSec_ = PerfEvents::DEFAULT_TIMEOUT; 242 int frequency_ = 0; 243 int period_ = 0; 244 int cpuPercent_ = DEFAULT_CPU_PERCENT; 245 int mmapPages_ = MAX_PERF_MMAP_PAGE; 246 int cmdlinesSize_ = DEFAULT_SAVED_CMDLINES_SIZE; 247 int oldCmdlinesSize_ = 0; 248 std::vector<std::string> symbolDir_ = {}; 249 std::string outputFilename_ = "/data/local/tmp/perf.data"; 250 std::string appPackage_ = {}; 251 int checkAppMs_ = DEFAULT_CHECK_APP_MS; 252 std::string clockId_ = {}; 253 std::string strLimit_ = {}; 254 std::vector<pid_t> selectCpus_ = {}; 255 std::vector<pid_t> selectPids_ = {}; 256 std::vector<pid_t> selectTids_ = {}; 257 std::vector<pid_t> inputPidTidArgs_ = {}; 258 bool restart_ = false; 259 std::vector<std::string> selectEvents_ = {}; 260 std::vector<std::vector<std::string>> selectGroups_ = {}; 261 std::vector<std::string> callStackType_ = {}; 262 std::vector<std::string> vecBranchFilters_ = {}; 263 std::vector<std::string> trackedCommand_ = {}; 264 265 // for exclude process and thread 266 std::vector<pid_t> excludeTidArgs_ = {}; 267 std::vector<std::string> excludeThreadNameArgs_ = {}; 268 std::vector<std::string> excludeProcessNameArgs_ = {}; 269 std::set<pid_t> excludePids_ = {}; 270 std::set<pid_t> excludeTids_ = {}; 271 void CollectExcludeThread(); 272 void SetExcludeHiperf(); 273 bool IsThreadExcluded(pid_t pid, pid_t tid); 274 275 // for background track 276 bool backtrack_ = false; 277 uint64_t backtrackTime_ = DEFAULT_BACKTRACK_TIME_SEC; // 10 seconds 278 bool outputEnd_ = false; 279 bool PreOutputRecordFile(); 280 void OutputRecordFile(); 281 bool PostOutputRecordFile(bool output); 282 283 bool GetOptions(std::vector<std::string> &args); 284 bool CheckArgsRange(); 285 bool CheckExcludeArgs(); 286 bool CheckOptions(); 287 bool GetSpeOptions(); 288 bool CheckDataLimitOption(); 289 bool CheckSelectCpuPidOption(); 290 bool GetOptionFrequencyAndPeriod(std::vector<std::string> &args); 291 292 bool isCallStackDwarf_ = false; 293 bool isCallStackFp_ = false; 294 uint32_t callStackDwarfSize_ = MAX_SAMPLE_STACK_SIZE; 295 uint64_t branchSampleType_ = 0; 296 uint64_t dataSizeLimit_ = 0; 297 bool isDataSizeLimitStop_ = false; 298 299 std::unique_ptr<PerfFileWriter> fileWriter_ = nullptr; 300 301 // for client 302 int clientPipeInput_ = -1; 303 int clientPipeOutput_ = -1; 304 int nullFd_ = -1; 305 std::thread clientCommandHanle_; 306 bool clientRunning_ = true; 307 struct ControlCommandHandler { 308 std::function<bool()> preProcess = []() -> bool { 309 return false; 310 }; 311 std::function<void(bool)> postProcess = [](bool) {}; 312 }; 313 std::unordered_map<std::string, ControlCommandHandler> controlCommandHandlerMap_ = {}; 314 inline void CreateClientThread(); 315 void ClientCommandHandle(); 316 void InitControlCommandHandlerMap(); 317 void DispatchControlCommand(const std::string& command); 318 bool ClientCommandResponse(bool response); 319 bool ClientCommandResponse(const std::string& str); 320 bool IsSamplingRunning(); 321 322 // for cmdline client 323 bool allowIpc_ = true; 324 std::string controlCmd_ = {}; 325 bool isFifoServer_ = false; 326 bool isFifoClient_ = false; 327 bool dedupStack_ = false; 328 std::map<pid_t, std::vector<pid_t>> mapPids_; 329 bool ProcessControl(); 330 void ProcessStopCommand(bool ret); 331 void ProcessOutputCommand(bool ret); 332 bool CreateFifoServer(); 333 bool SendFifoAndWaitReply(const std::string &cmd, const std::chrono::milliseconds &timeOut); 334 bool WaitFifoReply(int fd, const std::chrono::milliseconds &timeOut); 335 void WaitFifoReply(int fd, const std::chrono::milliseconds &timeOut, std::string& reply); 336 void CloseClientThread(); 337 std::string HandleAppInfo(); 338 339 bool PreparePerfEvent(); 340 bool PrepareSysKernel(); 341 void PrepareKernelMaps(); 342 bool PrepareVirtualRuntime(); 343 344 size_t recordSamples_ = 0; 345 size_t recordNoSamples_ = 0; 346 347 bool isNeedSetPerfHarden_ = false; 348 bool isSpe_ = false; 349 350 // callback to process record 351 bool ProcessRecord(PerfEventRecord& record); 352 bool SaveRecord(const PerfEventRecord& record); 353 354 // file format like as 0,1-3,4-6,7,8 355 uint32_t GetCountFromFile(const std::string &fileName); 356 std::string GetCpuDescFromFile(); 357 bool AddCpuFeature(); 358 void AddMemTotalFeature(); 359 void AddEventDescFeature(); 360 void AddRecordTimeFeature(); 361 void AddWorkloadCmdFeature(); 362 void AddCommandLineFeature(); 363 void AddCpuOffFeature(); 364 void AddDevhostFeature(); 365 bool AddFeatureRecordFile(); 366 367 bool CreateInitRecordFile(bool compressData = false); 368 bool FinishWriteRecordFile(); 369 bool PostProcessRecordFile(); 370 bool RecordCompleted(); 371 #ifdef HIPERF_DEBUG_TIME 372 void ReportTime(); 373 #endif 374 375 bool CollectionSymbol(PerfEventRecord& record); 376 void CollectSymbol(PerfRecordSample *sample); 377 bool SetPerfLimit(const std::string& file, int value, std::function<bool (int, int)> const& cmd, 378 const std::string& param); 379 bool SetPerfCpuMaxPercent(); 380 bool SetPerfMaxSampleRate(); 381 bool SetPerfEventMlock(); 382 bool SetPerfHarden(); 383 384 bool TraceOffCpu(); 385 bool ParseCallStackOption(const std::vector<std::string> &callStackType); 386 bool ParseDataLimitOption(const std::string &str); 387 bool ParseBranchSampleType(const std::vector<std::string> &vecBranchSampleTypes); 388 bool ParseControlCmd(const std::string cmd); 389 bool CheckTargetProcessOptions(); 390 bool CheckTargetPids(); 391 bool CheckReportOption(); 392 bool CheckBacktrackOption(); 393 bool CheckSpeOption(); 394 void WriteCommEventBeforeSampling(); 395 void RemoveVdsoTmpFile(); 396 397 VirtualRuntime virtualRuntime_; 398 #if USE_COLLECT_SYMBOLIC 399 std::unordered_map<pid_t, std::unordered_set<uint64_t>> kernelThreadSymbolsHits_; 400 kSymbolsHits kernelSymbolsHits_; 401 uSymbolsHits userSymbolsHits_; 402 void SymbolicHits(); 403 #endif 404 405 #ifdef HIPERF_DEBUG_TIME 406 std::chrono::microseconds prcessRecordTimes_ = std::chrono::microseconds::zero(); 407 std::chrono::microseconds saveRecordTimes_ = std::chrono::microseconds::zero(); 408 std::chrono::microseconds saveFeatureTimes_ = std::chrono::microseconds::zero(); 409 #endif 410 std::chrono::time_point<std::chrono::steady_clock> startSaveFileTimes_; 411 412 void SetHM(); 413 void SetSavedCmdlinesSize(); 414 void RecoverSavedCmdlinesSize(); 415 bool OnlineReportData(); 416 417 // only used in UT 418 using CheckRecordCallBack = std::function<void(const PerfEventRecord&)>; 419 void SetCheckRecordCallback(CheckRecordCallBack callback); 420 CheckRecordCallBack checkCallback_ = nullptr; 421 }; 422 } // namespace HiPerf 423 } // namespace Developtools 424 } // namespace OHOS 425 #endif // SUBCOMMAND_RECORD_H 426