hiperf/src/perf_event_record.cpp

/*
 * Copyright (c) 2021-2022 Huawei Device Co., Ltd.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
#define HILOG_TAG "PerfRecord"

#include "perf_event_record.h"
#include "spe_decoder.h"
#include <cinttypes>

#include "utilities.h"

namespace OHOS {
namespace Developtools {
namespace HiPerf {

bool PerfRecordSample::dumpRemoveStack_ = false;
std::vector<u64> PerfRecordSample::ips_ = {};
std::vector<DfxFrame> PerfRecordSample::callFrames_ = {};
std::vector<pid_t> PerfRecordSample::serverPidMap_ = {};
thread_local std::unordered_map<PerfRecordType, PerfEventRecord*> PerfEventRecordFactory::recordMap_ = {};

static PerfEventRecord* CreatePerfEventRecord(PerfRecordType type)
{
    switch (type) {
        case PERF_RECORD_SAMPLE:
            return new PerfRecordSample();
        case PERF_RECORD_MMAP:
            return new PerfRecordMmap();
        case PERF_RECORD_MMAP2:
            return new PerfRecordMmap2();
        case PERF_RECORD_LOST:
            return new PerfRecordLost();
        case PERF_RECORD_COMM:
            return new PerfRecordComm();
        case PERF_RECORD_EXIT:
            return new PerfRecordExit();
        case PERF_RECORD_THROTTLE:
            return new PerfRecordThrottle();
        case PERF_RECORD_UNTHROTTLE:
            return new PerfRecordUnthrottle();
        case PERF_RECORD_FORK:
            return new PerfRecordFork();
        case PERF_RECORD_READ:
            return new PerfRecordRead();
        case PERF_RECORD_AUX:
            return new PerfRecordAux();
        case PERF_RECORD_AUXTRACE:
            return new PerfRecordAuxtrace();
        case PERF_RECORD_AUXTRACE_INFO:
            return new PerfRecordAuxTraceInfo();
        case PERF_RECORD_TIME_CONV:
            return new PerfRecordTimeConv();
        case PERF_RECORD_CPU_MAP:
            return new PerfRecordCpuMap();
        case PERF_RECORD_ITRACE_START:
            return new PerfRecordItraceStart();
        case PERF_RECORD_LOST_SAMPLES:
            return new PerfRecordLostSamples();
        case PERF_RECORD_SWITCH:
            return new PerfRecordSwitch();
        case PERF_RECORD_SWITCH_CPU_WIDE:
            return new PerfRecordSwitchCpuWide();
        default:
            HLOGE("unknown record type %d\n", type);
            return new PerfRecordNull();
    }
}

template<typename T>
inline void PushToBinary(const bool condition, uint8_t *&p, const T &v)
{
    if (condition) {
        *(reinterpret_cast<T *>(p)) = v;
        p += sizeof(T);
    }
}

template<typename T1, typename T2>
inline void PushToBinary2(const bool condition, uint8_t *&p, const T1 &v1, const T2 &v2)
{
    if (condition) {
        *(reinterpret_cast<T1 *>(p)) = v1;
        p += sizeof(T1);
        *(reinterpret_cast<T2 *>(p)) = v2;
        p += sizeof(T2);
    }
}

template<typename T>
inline void PopFromBinary(const bool condition, uint8_t*& p, T& v, u64& size)
{
    HIPERF_ASSERT(sizeof(T) <= size, "PopFromBinary error\n");
    if (condition) {
        v = *(reinterpret_cast<const T *>(p));
        p += sizeof(T);
        size -= sizeof(T);
    }
}

template<typename T1, typename T2>
inline void PopFromBinary2(const bool condition, uint8_t*& p, T1& v1, T2& v2, u64& size)
{
    HIPERF_ASSERT(sizeof(T1) + sizeof(T2) <= size, "PopFromBinary2 error\n");
    if (condition) {
        v1 = *(reinterpret_cast<const T1 *>(p));
        p += sizeof(T1);
        v2 = *(reinterpret_cast<const T2 *>(p));
        p += sizeof(T2);
        size -= (sizeof(T1) + sizeof(T2));
    }
}

inline void SetPointerOffset(uint8_t*& p, u64 offset, u64& size)
{
    HIPERF_ASSERT(offset <= size && offset <= RECORD_SIZE_LIMIT, "SetPointerOffset error\n");
    size -= offset;
    p += offset;
}

void PerfEventRecord::GetHeaderBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetHeaderSize()) {
        buf.resize(GetHeaderSize());
    }
    uint8_t *p = buf.data();
    *(reinterpret_cast<perf_event_header*>(p)) = header_;
}

void PerfEventRecord::Dump(const int indent, std::string outputFilename, FILE *outputDump) const
{
    if (outputDump != nullptr) {
        g_outputDump = outputDump;
    } else if (!outputFilename.empty() && g_outputDump == nullptr) {
        std::string resolvedPath = CanonicalizeSpecPath(outputFilename.c_str());
        g_outputDump = fopen(resolvedPath.c_str(), "w");
        if (g_outputDump == nullptr) {
            printf("unable open file to '%s' because '%d'\n", outputFilename.c_str(), errno);
            return;
        }
    }
    PRINT_INDENT(indent, "\n");
    PRINT_INDENT(indent, "record %s: type %u, misc %u, size %zu\n", GetName(), GetType(),
                 GetMisc(), GetSize());
    DumpData(indent + 1);
}

void PerfEventRecord::DumpLog(const std::string &prefix) const
{
    HLOGV("%s: record %s: type %u, misc %u, size %zu\n", prefix.c_str(), GetName(),
          GetType(), GetMisc(), GetSize());
}

void PerfEventRecord::Init(const perf_event_type type, const bool inKernel)
{
    header_.type = type;
    header_.misc = inKernel ? PERF_RECORD_MISC_KERNEL : PERF_RECORD_MISC_USER;
    header_.size = sizeof(header_);
};

void PerfEventRecord::Init(const perf_event_hiperf_ext_type type)
{
    header_.type = type;
    header_.misc = PERF_RECORD_MISC_USER;
    header_.size = sizeof(header_);
}

void PerfEventRecord::InitHeader(uint8_t* p)
{
    if (p == nullptr) {
        header_.type = PERF_RECORD_MMAP;
        header_.misc = PERF_RECORD_MISC_USER;
        header_.size = 0;
        return;
    }
    header_ = *(reinterpret_cast<perf_event_header*>(p));
}

void PerfRecordAuxtrace::Init(uint8_t* data, const perf_event_attr& attr)
{
    data_ = {};
    rawData_ = nullptr;
    if (data == nullptr) {
        PerfEventRecord::InitHeader(data);
        HLOGE("Init failed");
        return;
    }
    PerfEventRecordTemplate::Init(data);
    if (header_.size != sizeof(header_) + sizeof(data_)) {
        HLOGE("header_.size invalid");
        return;
    }
    rawData_ = data + header_.size;
}

PerfRecordAuxtrace::PerfRecordAuxtrace(const u64 size, const u64 offset, const u64 reference, const u32 idx,
                                       const u32 tid, const u32 cpu, const u32 pid)
{
    PerfEventRecord::Init(PERF_RECORD_AUXTRACE);
    data_.size = size;
    data_.offset = offset;
    data_.reference = reference;
    data_.idx = idx;
    data_.tid = tid;
    data_.cpu = cpu;
    data_.reserved__ = pid;

    header_.size = sizeof(header_) + sizeof(data_);
}

bool PerfRecordAuxtrace::GetBinary1(std::vector<uint8_t> &buf) const
{
    if (buf.size() < header_.size) {
        buf.resize(header_.size);
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    size_t copySize = header_.size - GetHeaderSize();
    if (memcpy_s(p, buf.size() - GetHeaderSize(), reinterpret_cast<const uint8_t *>(&data_), copySize) != 0) {
        HLOGE("memcpy_s return failed in GetBinary1");
        return false;
    }
    return true;
}

bool PerfRecordAuxtrace::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    size_t copySize = header_.size - GetHeaderSize();
    if (memcpy_s(p, buf.size() - GetHeaderSize(), reinterpret_cast<const uint8_t *>(&data_), copySize) != 0) {
        HLOGE("memcpy_s return failed in GetBinary with data_");
        return false;
    }
    p += header_.size - GetHeaderSize();
    if (memcpy_s(p, buf.size() - header_.size, static_cast<uint8_t *>(rawData_), data_.size) != 0) {
        HLOGE("memcpy_s return failed in GetBinary with rawData_");
        return false;
    }
    return true;
}

void PerfRecordAuxtrace::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "size 0x%llx, offset 0x%llx, reference 0x%llx, idx %u, tid %u, cpu %u, pid %u\n",
                 data_.size, data_.offset, data_.reference, data_.idx, data_.tid, data_.cpu, data_.reserved__);
#if defined(is_ohos) && is_ohos
    if (!SpeDumpRawData(rawData_, data_.size, indent, g_outputDump)) {
        HLOGE("SpeDumpRawData failed");
    }
#endif
}

void PerfRecordAuxtrace::DumpLog(const std::string &prefix) const
{
    HLOGV("size %llu, offset 0x%llx, reference 0x%llx, idx %u, tid %u, cpu %u\n",
          data_.size, data_.offset, data_.reference, data_.idx, data_.tid, data_.cpu);
}

size_t PerfRecordAuxtrace::GetSize() const
{
    return header_.size + data_.size;
}

void PerfRecordMmap::Init(uint8_t* data, const perf_event_attr&)
{
    PerfEventRecordTemplate::Init(data);
    data_.filename[KILO - 1] = '\0';
}

PerfRecordMmap::PerfRecordMmap(bool inKernel, u32 pid, u32 tid, u64 addr, u64 len, u64 pgoff,
                               const std::string &filename)
{
    PerfEventRecord::Init(PERF_RECORD_MMAP, inKernel);

    data_.pid = pid;
    data_.tid = tid;
    data_.addr = addr;
    data_.len = len;
    data_.pgoff = pgoff;
    if (strncpy_s(data_.filename, KILO, filename.c_str(), filename.size()) != 0) {
        HLOGE("strncpy_s failed");
    }

    header_.size = sizeof(header_) + sizeof(data_) - KILO + filename.size() + 1;
}

bool PerfRecordMmap::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    // data_.filename[] is variable-length
    std::copy(reinterpret_cast<const uint8_t *>(&data_),
              reinterpret_cast<const uint8_t *>(&data_) + GetSize() - GetHeaderSize(), p);
    return true;
}

void PerfRecordMmap::DumpData(const int indent) const
{
#if defined(is_ohos) && is_ohos
    if (IsRoot()) {
        PRINT_INDENT(indent, "pid %u, tid %u, addr 0x%llx, len 0x%llx\n", data_.pid, data_.tid,
                     data_.addr, data_.len);
        PRINT_INDENT(indent, "pgoff 0x%llx, filename %s\n", data_.pgoff, data_.filename);
    }
#endif
}

void PerfRecordMmap::DumpLog(const std::string &prefix) const
{
    HLOGV("%s:  MMAP: size %d pid %u tid %u dso '%s' (0x%llx-0x%llx)@0x%llx", prefix.c_str(),
          header_.size, data_.pid, data_.tid, data_.filename, data_.addr, data_.addr + data_.len, data_.pgoff);
}

void PerfRecordMmap2::Init(uint8_t* data, const perf_event_attr&)
{
    PerfEventRecordTemplate::Init(data);
    discard_ = false;
}

PerfRecordMmap2::PerfRecordMmap2(const bool inKernel, const u32 pid, const u32 tid, const u64 addr, const u64 len,
                                 const u64 pgoff, const u32 maj, const u32 min, const u64 ino, const u32 prot,
                                 const u32 flags, const std::string &filename)
{
    PerfEventRecord::Init(PERF_RECORD_MMAP2, inKernel);
    data_.pid = pid;
    data_.tid = tid;
    data_.addr = addr;
    data_.len = len;
    data_.pgoff = pgoff;
    data_.maj = maj;
    data_.min = min;
    data_.ino = ino;
    data_.ino_generation = 0;
    data_.prot = prot;
    data_.flags = flags;
    if (strncpy_s(data_.filename, KILO, filename.c_str(), filename.size()) != 0) {
        HLOGE("strncpy_s failed");
    }

    header_.size = sizeof(header_) + sizeof(data_) - KILO + filename.size() + 1;
}

PerfRecordMmap2::PerfRecordMmap2(const bool inKernel, const u32 pid, const u32 tid, std::shared_ptr<DfxMap> item)
{
    PerfEventRecord::Init(PERF_RECORD_MMAP2, inKernel);
    data_.pid = pid;
    data_.tid = tid;
    if (item != nullptr) {
        data_.addr = item->begin;
        data_.len = item->end - item->begin;
        data_.pgoff = item->offset;
        data_.maj = item->major;
        data_.min = item->minor;
        data_.ino = item->inode;
        data_.ino_generation = 0;
        // r--p 00000000 103:3e 12307                         /data/storage/el1/bundle/entry.hap
        // why prot get from this is 7. rwxp
        DfxMap::PermsToProts(item->perms, data_.prot, data_.flags);
        if (strncpy_s(data_.filename, KILO, item->name.c_str(), item->name.size()) != 0) {
            HLOGE("strncpy_s failed");
        }

        header_.size = sizeof(header_) + sizeof(data_) - KILO + item->name.size() + 1;
    } else {
        data_.addr = 0;
        data_.len = 0;
        data_.pgoff = 0;
        data_.maj = 0;
        data_.min = 0;
        data_.ino = 0;
        data_.ino_generation = 0;
        if (memset_s(data_.filename, KILO, 0, KILO) != EOK) {
            HLOGE("memset_s failed");
        }
    }
}

bool PerfRecordMmap2::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    // data_.filename[] is variable-length
    std::copy(reinterpret_cast<const uint8_t *>(&data_),
              reinterpret_cast<const uint8_t *>(&data_) + GetSize() - GetHeaderSize(), p);
    return true;
}

void PerfRecordMmap2::DumpData(const int indent) const
{
#if defined(is_ohos) && is_ohos
    if (IsRoot()) {
        PRINT_INDENT(indent, "pid %u, tid %u, addr 0x%llx, len 0x%llx\n", data_.pid, data_.tid,
                     data_.addr, data_.len);
        PRINT_INDENT(indent, "pgoff 0x%llx, maj %u, min %u, ino %llu, ino_generation %llu\n",
                     data_.pgoff, data_.maj, data_.min, data_.ino, data_.ino_generation);
        PRINT_INDENT(indent, "prot %u, flags %u, filename %s\n", data_.prot, data_.flags,
                     data_.filename);
    }
#endif
}

void PerfRecordMmap2::DumpLog(const std::string &prefix) const
{
    HLOGV("%s:  MMAP2: size %d pid %u tid %u dso '%s' (0x%llx-0x%llx)@0x%llx", prefix.c_str(),
          header_.size, data_.pid, data_.tid, data_.filename, data_.addr, data_.addr + data_.len,
          data_.pgoff);
}

bool PerfRecordLost::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordLostData *>(p);
    *pDest = data_;

    return true;
}

void PerfRecordLost::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "id %llu, lost %llu\n", data_.id, data_.lost);
}

void PerfRecordComm::Init(uint8_t* data, const perf_event_attr&)
{
    PerfEventRecordTemplate::Init(data);
    data_.comm[KILO - 1] = '\0';
}

PerfRecordComm::PerfRecordComm(const bool inKernel, const u32 pid, const u32 tid, const std::string &comm)
{
    PerfEventRecord::Init(PERF_RECORD_COMM, inKernel);
    data_.pid = pid;
    data_.tid = tid;
    if (strncpy_s(data_.comm, KILO, comm.c_str(), comm.size()) != 0) {
        HLOGE("strncpy_s failed !!!");
    }

    header_.size = sizeof(header_) + sizeof(data_) - KILO + comm.size() + 1;
}

bool PerfRecordComm::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    // data_.comm[] is variable-length
    std::copy(reinterpret_cast<const uint8_t *>(&data_),
              reinterpret_cast<const uint8_t *>(&data_) + GetSize() - GetHeaderSize(), p);

    return true;
}

void PerfRecordComm::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "pid %u, tid %u, comm %s\n", data_.pid, data_.tid, data_.comm);
}

void PerfRecordComm::DumpLog(const std::string &prefix) const
{
    HLOGV("pid %u, tid %u, comm %s\n", data_.pid, data_.tid, data_.comm);
}

PerfRecordSample::PerfRecordSample(const PerfRecordSample& sample)
{
    header_ = sample.header_;
    data_ = sample.data_;

    sampleType_ = sample.sampleType_;
    skipKernel_ = sample.skipKernel_;
    skipPid_ = sample.skipPid_;

    ips_ = sample.ips_;
    callFrames_ = sample.callFrames_;
    serverPidMap_ = sample.serverPidMap_;

    stackId_ = sample.stackId_;
    removeStack_ = sample.removeStack_;
}

void PerfRecordSample::Init(uint8_t *p, const perf_event_attr &attr)
{
    PerfEventRecord::InitHeader(p);

    HLOG_ASSERT(p != nullptr);
    // clear the vector data
    Clean();
    sampleType_ = attr.sample_type;
    skipKernel_ = 0;
    skipPid_ = 0;
    stackId_ = {0};
    removeStack_ = false;
    data_ = {};
    uint8_t *start = p;
    u64 dataSize = static_cast<u64>(RECORD_SIZE_LIMIT);
    SetPointerOffset(p, sizeof(header_), dataSize);

    // parse record according SAMPLE_TYPE
    PopFromBinary(sampleType_ & PERF_SAMPLE_IDENTIFIER, p, data_.sample_id, dataSize);
    PopFromBinary(sampleType_ & PERF_SAMPLE_IP, p, data_.ip, dataSize);
    PopFromBinary2(sampleType_ & PERF_SAMPLE_TID, p, data_.pid, data_.tid, dataSize);
    PopFromBinary(sampleType_ & PERF_SAMPLE_TIME, p, data_.time, dataSize);
    PopFromBinary(sampleType_ & PERF_SAMPLE_ADDR, p, data_.addr, dataSize);
    PopFromBinary(sampleType_ & PERF_SAMPLE_ID, p, data_.id, dataSize);
    PopFromBinary(sampleType_ & PERF_SAMPLE_STREAM_ID, p, data_.stream_id, dataSize);
    PopFromBinary2(sampleType_ & PERF_SAMPLE_CPU, p, data_.cpu, data_.res, dataSize);
    PopFromBinary(sampleType_ & PERF_SAMPLE_PERIOD, p, data_.period, dataSize);
    PopFromBinary(sampleType_ & PERF_SAMPLE_CALLCHAIN, p, data_.nr, dataSize);
    if (data_.nr > 0) {
        // the pointer is from input(p), require caller keep input(p) with *this together
        // think it in next time
        data_.ips = reinterpret_cast<u64 *>(p);
        SetPointerOffset(p, data_.nr * sizeof(u64), dataSize);
    }
    PopFromBinary(sampleType_ & PERF_SAMPLE_RAW, p, data_.raw_size, dataSize);
    if (data_.raw_size > 0) {
        data_.raw_data = p;
        SetPointerOffset(p, data_.raw_size * sizeof(u8), dataSize);
    }
    PopFromBinary(sampleType_ & PERF_SAMPLE_BRANCH_STACK, p, data_.bnr, dataSize);
    if (data_.bnr > 0) {
        data_.lbr = reinterpret_cast<PerfBranchEntry *>(p);
        SetPointerOffset(p, data_.bnr * sizeof(PerfBranchEntry), dataSize);
    }
    PopFromBinary(sampleType_ & PERF_SAMPLE_REGS_USER, p, data_.user_abi, dataSize);
    if (data_.user_abi > 0) {
        data_.reg_mask = attr.sample_regs_user;
        data_.reg_nr = __builtin_popcountll(data_.reg_mask);
        data_.user_regs = reinterpret_cast<u64 *>(p);
        SetPointerOffset(p, data_.reg_nr * sizeof(u64), dataSize);
    }
    PopFromBinary(sampleType_ & PERF_SAMPLE_SERVER_PID, p, data_.server_nr, dataSize);
    if (data_.server_nr > 0) {
        data_.server_pids = reinterpret_cast<u64 *>(p);
        SetPointerOffset(p, data_.server_nr * sizeof(u64), dataSize);
    }
    PopFromBinary(sampleType_ & PERF_SAMPLE_STACK_USER, p, data_.stack_size, dataSize);
    if (data_.stack_size > 0) {
        data_.stack_data = p;
        SetPointerOffset(p, data_.stack_size, dataSize);
        PopFromBinary(true, p, data_.dyn_size, dataSize);
    }
    uint32_t remain = header_.size - (p - start);
    if (data_.nr == 0 && dumpRemoveStack_ && remain == sizeof(stackId_)) {
        PopFromBinary(true, p, stackId_.value, dataSize);
    }
}

void PerfRecordSample::SetDumpRemoveStack(bool dumpRemoveStack)
{
    dumpRemoveStack_ = dumpRemoveStack;
}

bool PerfRecordSample::IsDumpRemoveStack()
{
    return dumpRemoveStack_;
}

bool PerfRecordSample::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    PushToBinary(sampleType_ & PERF_SAMPLE_IDENTIFIER, p, data_.sample_id);
    PushToBinary(sampleType_ & PERF_SAMPLE_IP, p, data_.ip);
    PushToBinary2(sampleType_ & PERF_SAMPLE_TID, p, data_.pid, data_.tid);
    PushToBinary(sampleType_ & PERF_SAMPLE_TIME, p, data_.time);
    PushToBinary(sampleType_ & PERF_SAMPLE_ADDR, p, data_.addr);
    PushToBinary(sampleType_ & PERF_SAMPLE_ID, p, data_.id);
    PushToBinary(sampleType_ & PERF_SAMPLE_STREAM_ID, p, data_.stream_id);
    PushToBinary2(sampleType_ & PERF_SAMPLE_CPU, p, data_.cpu, data_.res);
    PushToBinary(sampleType_ & PERF_SAMPLE_PERIOD, p, data_.period);
    PushToBinary(sampleType_ & PERF_SAMPLE_CALLCHAIN, p, data_.nr);
    if (data_.nr > 0 && !removeStack_) {
        std::copy(data_.ips + skipKernel_, data_.ips + data_.nr + skipKernel_,
                  reinterpret_cast<u64 *>(p));
        p += data_.nr * sizeof(u64);
    }
    PushToBinary(sampleType_ & PERF_SAMPLE_RAW, p, data_.raw_size);
    if (data_.raw_size > 0) {
        std::copy(data_.raw_data, data_.raw_data + data_.raw_size, p);
        p += data_.raw_size * sizeof(u8);
    }
    PushToBinary(sampleType_ & PERF_SAMPLE_BRANCH_STACK, p, data_.bnr);
    if (data_.bnr > 0) {
        std::copy(data_.lbr, data_.lbr + data_.bnr, reinterpret_cast<PerfBranchEntry *>(p));
        p += data_.bnr * sizeof(PerfBranchEntry);
    }
    PushToBinary(sampleType_ & PERF_SAMPLE_REGS_USER, p, data_.user_abi);
    if (data_.user_abi > 0 && data_.reg_nr > 0) {
        std::copy(data_.user_regs, data_.user_regs + data_.reg_nr, reinterpret_cast<u64 *>(p));
        p += data_.reg_nr * sizeof(u64);
    }
    PushToBinary(sampleType_ & PERF_SAMPLE_SERVER_PID, p, data_.server_nr);
    if (data_.server_nr > 0) {
        std::copy(data_.server_pids + skipPid_, data_.server_pids + data_.server_nr + skipPid_,
                  reinterpret_cast<u64 *>(p));
        p += data_.server_nr * sizeof(u64);
    }
    PushToBinary(sampleType_ & PERF_SAMPLE_STACK_USER, p, data_.stack_size);
    if (data_.stack_size > 0) {
        std::copy(data_.stack_data, data_.stack_data + data_.stack_size, p);
        p += data_.stack_size * sizeof(u8);
        PushToBinary(true, p, data_.dyn_size);
    }
    PushToBinary(removeStack_, p, stackId_.value);
    return true;
}

void PerfRecordSample::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "sample_type: 0x%" PRIx64 "\n", sampleType_);

    // dump record according sampleType
    if (sampleType_ & (PERF_SAMPLE_ID | PERF_SAMPLE_IDENTIFIER)) {
        PRINT_INDENT(indent, "ID %" PRIu64 "\n", static_cast<uint64_t>(data_.sample_id));
    }
    if (sampleType_ & PERF_SAMPLE_IP) {
        PRINT_INDENT(indent, "ip %llx\n", data_.ip);
    }
    if (sampleType_ & PERF_SAMPLE_TID) {
        PRINT_INDENT(indent, "pid %u, tid %u\n", data_.pid, data_.tid);
    }
    if (sampleType_ & PERF_SAMPLE_TIME) {
        PRINT_INDENT(indent, "time %llu\n", data_.time);
    }
    // PERF_SAMPLE_ADDR is not enabled, delete output
    if (sampleType_ & PERF_SAMPLE_ADDR) {
        PRINT_INDENT(indent, "addr\n");
    }
    if (sampleType_ & PERF_SAMPLE_STREAM_ID) {
        PRINT_INDENT(indent, "stream_id %" PRIu64 "\n", static_cast<uint64_t>(data_.stream_id));
    }
    if (sampleType_ & PERF_SAMPLE_CPU) {
        PRINT_INDENT(indent, "cpu %u, res %u\n", data_.cpu, data_.res);
    }
    if (sampleType_ & PERF_SAMPLE_PERIOD) {
        PRINT_INDENT(indent, "period %" PRIu64 "\n", static_cast<uint64_t>(data_.period));
    }
    if (stackId_.section.id > 0) {
        PRINT_INDENT(indent, "stackid %" PRIu64 "\n", static_cast<uint64_t>(stackId_.section.id));
    }
    if (sampleType_ & PERF_SAMPLE_CALLCHAIN) {
        bool userContext = false;
        PRINT_INDENT(indent, "callchain nr=%lld\n", data_.nr);
        for (uint64_t i = 0; i < data_.nr; ++i) {
            std::string_view supplement = "";
            if ((sampleType_ & PERF_SAMPLE_STACK_USER) == 0 || data_.ips[i] != PERF_CONTEXT_USER) {
                PRINT_INDENT(indent + 1, "0x%llx%s\n", data_.ips[i], supplement.data());
                continue;
            }
            // is PERF_SAMPLE_STACK_USER type and is PERF_CONTEXT_USER
            if (!userContext) {
                userContext = true;
                supplement = " <unwind callstack>";
            } else {
                supplement = " <expand callstack>";
            }
            PRINT_INDENT(indent + 1, "0x%llx%s\n", data_.ips[i], supplement.data());
        }
    }
    if (sampleType_ & PERF_SAMPLE_RAW) {
        PRINT_INDENT(indent, "raw size=%u\n", data_.raw_size);
        const uint32_t *data = reinterpret_cast<const uint32_t *>(data_.raw_data);
        size_t size = data_.raw_size / sizeof(uint32_t);
        for (size_t i = 0; i < size; ++i) {
            PRINT_INDENT(indent + 1, "0x%08x (%x)\n", data[i], data[i]);
        }
    }
    if (sampleType_ & PERF_SAMPLE_BRANCH_STACK) {
        PRINT_INDENT(indent, "branch_stack nr=%lld\n", data_.bnr);
        for (uint64_t i = 0; i < data_.bnr; ++i) {
            auto &item = data_.lbr[i];
            PRINT_INDENT(indent + 1, "from 0x%llx, to 0x%llx, flags 0x%llx\n", item.from, item.to, item.flags);
        }
    }
    if (sampleType_ & PERF_SAMPLE_REGS_USER) {
        PRINT_INDENT(indent, "user regs: abi=%lld, reg_nr=%lld\n", data_.user_abi, data_.reg_nr);
#if defined(is_ohos) && is_ohos
        if (IsRoot()) {
            for (uint64_t i = 0; i < data_.reg_nr; ++i) {
                PRINT_INDENT(indent + 1, "0x%llx\n", data_.user_regs[i]);
            }
        }
#endif
    }
    if (sampleType_ & PERF_SAMPLE_SERVER_PID) {
        PRINT_INDENT(indent, "server nr=%lld\n", data_.server_nr);
        for (uint64_t i = 0; i < data_.server_nr; ++i) {
            PRINT_INDENT(indent + 1, "pid: %llu\n", data_.server_pids[i]);
        }
    }
    if (sampleType_ & PERF_SAMPLE_STACK_USER) {
        PRINT_INDENT(indent, "user stack: size %llu dyn_size %lld\n", data_.stack_size,
                     data_.dyn_size);
    }
}

void PerfRecordSample::DumpLog(const std::string &prefix) const
{
    HLOGV("%s: SAMPLE: id= %llu size %d pid %u tid %u ips %llu regs %llu, stacks %llu time %llu",
          prefix.c_str(), data_.sample_id, header_.size, data_.pid, data_.tid, data_.nr,
          data_.reg_nr, data_.dyn_size, data_.time);
}

void PerfRecordSample::RecoverCallStack()
{
    data_.ips = ips_.data();
    data_.nr = ips_.size();
    removeStack_ = true;
}

void PerfRecordSample::ReplaceWithCallStack(const size_t originalSize)
{
    // first we check if we have some user unwind stack need to merge ?
    if (callFrames_.size() != 0) {
        // when we have some kernel ips , we cp it first
        // new size is user call frames + kernel call frames
        // + PERF_CONTEXT_USER(last + 1) + expand mark(also PERF_CONTEXT_USER)
        const unsigned int perfContextSize = 2;
        ips_.reserve(data_.nr + callFrames_.size() + perfContextSize);
        if (data_.nr > 0) {
            ips_.assign(data_.ips, data_.ips + data_.nr);
        }
        // add user context mark
        ips_.emplace_back(PERF_CONTEXT_USER);
        // we also need make a expand mark just for debug only
        const size_t beginIpsSize = ips_.size();
        bool ret = std::all_of(callFrames_.begin(), callFrames_.end(), [&](const DfxFrame &frame) {
            ips_.emplace_back(frame.pc);
            if (originalSize != 0 && (originalSize != callFrames_.size()) &&
                ips_.size() == (originalSize + beginIpsSize)) {
                // just for debug
                // so we can see which frame begin is expand call frames
                ips_.emplace_back(PERF_CONTEXT_USER);
            }
            return true;
        });
        if (ret) {
            HLOGV("combed %zu", callFrames_.size());
        } else {
            HLOGV("failed to combed %zu", callFrames_.size());
        }

        if (sampleType_ & PERF_SAMPLE_REGS_USER) {
            header_.size -= data_.reg_nr * sizeof(u64);
            data_.reg_nr = 0;
            data_.user_abi = 0;
        }

        if (sampleType_ & PERF_SAMPLE_STACK_USER) {
            // 1. remove the user stack
            header_.size -= data_.stack_size;
            header_.size -= sizeof(data_.dyn_size);

            // 2. clean the size
            data_.stack_size = 0;
            data_.dyn_size = 0;
        }

        if (sampleType_ & PERF_SAMPLE_CALLCHAIN) {
            HLOGV("ips change from %llu -> %zu", data_.nr, ips_.size());

            // 3. remove the nr size
            header_.size -= data_.nr * sizeof(u64);

            // 4. add new nr size
            data_.nr = ips_.size();
            header_.size += data_.nr * sizeof(u64);

            // 5. change ips potin to our ips array and hold it.
            data_.ips = ips_.data();
        }
    }
}

inline pid_t PerfRecordSample::GetPid() const
{
    return data_.pid;
}

uint64_t PerfRecordSample::GetTime() const
{
    return data_.time;
}

void PerfRecordSample::Clean()
{
    ips_.clear();
    callFrames_.clear();
    serverPidMap_.clear();
}

pid_t PerfRecordSample::GetUstackServerPid()
{
    if (!data_.server_nr) {
        return data_.pid;
    }

    size_t currServer = 0;
    // ipNr == 1...nr: server_pid of data_.ips[nr]
    for (size_t i = 0; i < data_.nr; i++) {
        // context change, use next server pid
        if (data_.ips[i] >= PERF_CONTEXT_MAX) {
            currServer++;
        }
    }
    // ipNr == nr + 1: server_pid of ustack
    if (currServer > 0) {
        currServer++;
    }
    if (currServer >= data_.server_nr) {
        HLOGE("ustack server pid nr %zu out of range", currServer);
        return data_.pid;
    }

    // return server pid
    return data_.server_pids[currServer];
}

pid_t PerfRecordSample::GetServerPidof(unsigned int ipNr)
{
    if (!data_.server_nr) {
        return data_.pid;
    }

    // init serverPidMap_
    if (!serverPidMap_.size()) {
        size_t currServer = 0;
        // ipNr == 0: server_pid of data_.ip
        serverPidMap_.emplace_back(data_.server_pids[currServer]);
        // ipNr == 1...nr: server_pid of data_.ips[nr]
        for (size_t i = 1; i < data_.nr; i++) {
            // context change, use next server pid
            if (data_.ips[i] >= PERF_CONTEXT_MAX) {
                currServer++;
            }
            if (currServer >= data_.server_nr) {
                HLOGE("callchain server pid nr %zu out of range", currServer);
                break;
            }
            serverPidMap_.emplace_back(data_.server_pids[currServer]);
        }
    }

    // return server pid
    if (ipNr >= serverPidMap_.size()) {
        return data_.pid;
    } else {
        return serverPidMap_[ipNr];
    }
}

bool PerfRecordExit::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordExitData *>(p);
    *pDest = data_;
    return true;
}

void PerfRecordExit::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "pid %u, ppid %u, tid %u, ptid %u time 0x%llx\n", data_.pid, data_.ppid,
                 data_.tid, data_.ptid, data_.time);
}

bool PerfRecordThrottle::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordThrottleData *>(p);
    *pDest = data_;
    return true;
}

void PerfRecordThrottle::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "time 0x%llx, id %llx, stream_id %llx\n", data_.time, data_.id,
                 data_.stream_id);
}

bool PerfRecordUnthrottle::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordThrottleData *>(p);
    *pDest = data_;
    return true;
}

void PerfRecordUnthrottle::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "time 0x%llx, id %llx, stream_id %llx\n", data_.time, data_.id,
                 data_.stream_id);
}

bool PerfRecordFork::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordForkData *>(p);
    *pDest = data_;
    return true;
}

void PerfRecordFork::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "pid %u, ppid %u, tid %u, ptid %u\n", data_.pid, data_.ppid, data_.tid,
                 data_.ptid);
}

bool PerfRecordRead::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordReadData *>(p);
    *pDest = data_;
    return true;
}

void PerfRecordRead::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "pid %u, tid %u\n", data_.pid, data_.tid);
    PRINT_INDENT(indent, "values: value %llx, timeEnabled %llx, timeRunning %llx, id %llx\n",
                 data_.values.value, data_.values.timeEnabled, data_.values.timeRunning, data_.values.id);
}

bool PerfRecordAux::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    PushToBinary(true, p, data_.aux_offset);
    PushToBinary(true, p, data_.aux_size);
    PushToBinary(true, p, data_.flags);
    PushToBinary2(sampleType_ & PERF_SAMPLE_TID, p, data_.sample_id.pid, data_.sample_id.tid);
    PushToBinary(sampleType_ & PERF_SAMPLE_TIME, p, data_.sample_id.time);
    PushToBinary(sampleType_ & PERF_SAMPLE_ID, p, data_.sample_id.id);
    PushToBinary(sampleType_ & PERF_SAMPLE_STREAM_ID, p, data_.sample_id.stream_id);

    PushToBinary2(sampleType_ & PERF_SAMPLE_CPU, p, data_.sample_id.cpu, data_.sample_id.res);
    PushToBinary(sampleType_ & PERF_SAMPLE_IDENTIFIER, p, data_.sample_id.id2);
    return true;
}

void PerfRecordAux::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "aux_offset 0x%llx aux_size 0x%llx flags 0x%llx pid %u tid %u time %llu",
                 data_.aux_offset, data_.aux_size, data_.flags, data_.sample_id.pid, data_.sample_id.tid,
                 data_.sample_id.time);
}

bool PerfRecordAuxTraceInfo::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();
    std::copy(reinterpret_cast<const uint8_t *>(&data_),
              reinterpret_cast<const uint8_t *>(&data_) + GetSize() - GetHeaderSize(), p);
    return true;
}

void PerfRecordAuxTraceInfo::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "aux_trace_event");
}

bool PerfRecordTimeConv::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();
    std::copy(reinterpret_cast<const uint8_t *>(&data_),
              reinterpret_cast<const uint8_t *>(&data_) + GetSize() - GetHeaderSize(), p);
    return true;
}

void PerfRecordTimeConv::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "aux_time_event");
}

bool PerfRecordCpuMap::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();
    std::copy(reinterpret_cast<const uint8_t *>(&data_),
              reinterpret_cast<const uint8_t *>(&data_) + GetSize() - GetHeaderSize(), p);
    return true;
}

void PerfRecordCpuMap::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "cpu_map_event");
}

bool PerfRecordItraceStart::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordItraceStartData *>(p);
    *pDest = data_;
    return true;
}

void PerfRecordItraceStart::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "pid %u, tid %u\n", data_.pid, data_.tid);
}

bool PerfRecordLostSamples::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordLostSamplesData *>(p);
    *pDest = data_;
    return true;
}

void PerfRecordLostSamples::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "lost %llu\n", data_.lost);
}

bool PerfRecordSwitch::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordSwitchData *>(p);
    *pDest = data_;
    return true;
}

bool PerfRecordSwitchCpuWide::GetBinary(std::vector<uint8_t> &buf) const
{
    if (buf.size() < GetSize()) {
        buf.resize(GetSize());
    }

    GetHeaderBinary(buf);
    uint8_t *p = buf.data() + GetHeaderSize();

    auto pDest = reinterpret_cast<PerfRecordSwitchCpuWideData *>(p);
    *pDest = data_;
    return true;
}

void PerfRecordSwitchCpuWide::DumpData(const int indent) const
{
    PRINT_INDENT(indent, "next_prev_pid %u, next_prev_tid %u\n", data_.next_prev_pid,
                 data_.next_prev_tid);
}

PerfEventRecord& PerfEventRecordFactory::GetPerfEventRecord(PerfRecordType type, uint8_t* data,
                                                            const perf_event_attr &attr)
{
    HLOG_ASSERT(data != nullptr);
    PerfEventRecord* record = nullptr;
    auto it = recordMap_.find(type);
    if (it == recordMap_.end()) {
        record = CreatePerfEventRecord(type);
        if (record == nullptr) {
            record = new PerfRecordNull();
        }
        recordMap_.emplace(type, record);
    } else {
        record = it->second;
    }
    record->Init(data, attr);
    return *record;
}
} // namespace HiPerf
} // namespace Developtools
} // namespace OHOS