diff options
Diffstat (limited to 'tools/gator/daemon/PerfBuffer.cpp')
-rw-r--r-- | tools/gator/daemon/PerfBuffer.cpp | 223 |
1 files changed, 223 insertions, 0 deletions
diff --git a/tools/gator/daemon/PerfBuffer.cpp b/tools/gator/daemon/PerfBuffer.cpp new file mode 100644 index 000000000000..3b9da1dc6592 --- /dev/null +++ b/tools/gator/daemon/PerfBuffer.cpp @@ -0,0 +1,223 @@ +/** + * Copyright (C) ARM Limited 2013-2015. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + */ + +#include "PerfBuffer.h" + +#include <sys/ioctl.h> +#include <sys/mman.h> + +#include "Buffer.h" +#include "Logging.h" +#include "Sender.h" +#include "SessionData.h" + +PerfBuffer::PerfBuffer() { + for (int cpu = 0; cpu < ARRAY_LENGTH(mBuf); ++cpu) { + mBuf[cpu] = MAP_FAILED; + mDiscard[cpu] = false; + mFds[cpu] = -1; + } +} + +PerfBuffer::~PerfBuffer() { + for (int cpu = ARRAY_LENGTH(mBuf) - 1; cpu >= 0; --cpu) { + if (mBuf[cpu] != MAP_FAILED) { + munmap(mBuf[cpu], gSessionData->mPageSize + BUF_SIZE); + } + } +} + +bool PerfBuffer::useFd(const int cpu, const int fd) { + if (mFds[cpu] < 0) { + if (mBuf[cpu] != MAP_FAILED) { + logg->logMessage("cpu %i already online or not correctly cleaned up", cpu); + return false; + } + + // The buffer isn't mapped yet + mBuf[cpu] = mmap(NULL, gSessionData->mPageSize + BUF_SIZE, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); + if (mBuf[cpu] == MAP_FAILED) { + logg->logMessage("mmap failed"); + return false; + } + mFds[cpu] = fd; + + // Check the version + struct perf_event_mmap_page *pemp = static_cast<struct perf_event_mmap_page *>(mBuf[cpu]); + if (pemp->compat_version != 0) { + logg->logMessage("Incompatible perf_event_mmap_page compat_version"); + return false; + } + } else { + if (mBuf[cpu] == MAP_FAILED) { + logg->logMessage("cpu already online or not correctly cleaned up"); + return false; + } + + if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, mFds[cpu]) < 0) { + logg->logMessage("ioctl failed"); + return false; + } + } + + return true; +} + +void PerfBuffer::discard(const int cpu) { + if (mBuf[cpu] != MAP_FAILED) { + mDiscard[cpu] = true; + } +} + +bool PerfBuffer::isEmpty() { + for (int cpu = 0; cpu < gSessionData->mCores; ++cpu) { + if (mBuf[cpu] != MAP_FAILED) { + // Take a snapshot of the positions + struct perf_event_mmap_page *pemp = static_cast<struct perf_event_mmap_page *>(mBuf[cpu]); + const __u64 head = ACCESS_ONCE(pemp->data_head); + const __u64 tail = ACCESS_ONCE(pemp->data_tail); + + if (head != tail) { + return false; + } + } + } + + return true; +} + +bool PerfBuffer::isFull() { + for (int cpu = 0; cpu < gSessionData->mCores; ++cpu) { + if (mBuf[cpu] != MAP_FAILED) { + // Take a snapshot of the positions + struct perf_event_mmap_page *pemp = static_cast<struct perf_event_mmap_page *>(mBuf[cpu]); + const __u64 head = ACCESS_ONCE(pemp->data_head); + + if (head + 2000 <= (unsigned int)BUF_SIZE) { + return true; + } + } + } + + return false; +} + +class PerfFrame { +public: + PerfFrame(Sender *const sender) : mSender(sender), mWritePos(-1), mCpuSizePos(-1) {} + + void add(const int cpu, const __u64 head, __u64 tail, const uint8_t *const b) { + cpuHeader(cpu); + + while (head > tail) { + const int count = reinterpret_cast<const struct perf_event_header *>(b + (tail & BUF_MASK))->size/sizeof(uint64_t); + // Can this whole message be written as Streamline assumes events are not split between frames + if (sizeof(mBuf) <= mWritePos + count*Buffer::MAXSIZE_PACK64) { + send(); + cpuHeader(cpu); + } + for (int i = 0; i < count; ++i) { + // Must account for message size + Buffer::packInt64(mBuf, sizeof(mBuf), mWritePos, *reinterpret_cast<const uint64_t *>(b + (tail & BUF_MASK))); + tail += sizeof(uint64_t); + } + } + } + + void send() { + if (mWritePos > 0) { + writeFrameSize(); + mSender->writeData(mBuf, mWritePos, RESPONSE_APC_DATA); + mWritePos = -1; + mCpuSizePos = -1; + } + } + +private: + void writeFrameSize() { + writeCpuSize(); + const int typeLength = gSessionData->mLocalCapture ? 0 : 1; + Buffer::writeLEInt(reinterpret_cast<unsigned char *>(mBuf + typeLength), mWritePos - typeLength - sizeof(uint32_t)); + } + + void frameHeader() { + if (mWritePos < 0) { + mWritePos = 0; + mCpuSizePos = -1; + if (!gSessionData->mLocalCapture) { + mBuf[mWritePos++] = RESPONSE_APC_DATA; + } + // Reserve space for frame size + mWritePos += sizeof(uint32_t); + Buffer::packInt(mBuf, sizeof(mBuf), mWritePos, FRAME_PERF); + } + } + + void writeCpuSize() { + if (mCpuSizePos >= 0) { + Buffer::writeLEInt(reinterpret_cast<unsigned char *>(mBuf + mCpuSizePos), mWritePos - mCpuSizePos - sizeof(uint32_t)); + } + } + + void cpuHeader(const int cpu) { + if (sizeof(mBuf) <= mWritePos + Buffer::MAXSIZE_PACK32 + sizeof(uint32_t)) { + send(); + } + frameHeader(); + writeCpuSize(); + Buffer::packInt(mBuf, sizeof(mBuf), mWritePos, cpu); + mCpuSizePos = mWritePos; + // Reserve space for cpu size + mWritePos += sizeof(uint32_t); + } + + // Pick a big size but something smaller than the chunkSize in Sender::writeData which is 100k + char mBuf[1<<16]; + Sender *const mSender; + int mWritePos; + int mCpuSizePos; + + // Intentionally unimplemented + PerfFrame(const PerfFrame &); + PerfFrame& operator=(const PerfFrame &); +}; + +bool PerfBuffer::send(Sender *const sender) { + PerfFrame frame(sender); + + for (int cpu = 0; cpu < gSessionData->mCores; ++cpu) { + if (mBuf[cpu] == MAP_FAILED) { + continue; + } + + // Take a snapshot of the positions + struct perf_event_mmap_page *pemp = static_cast<struct perf_event_mmap_page *>(mBuf[cpu]); + const __u64 head = ACCESS_ONCE(pemp->data_head); + const __u64 tail = ACCESS_ONCE(pemp->data_tail); + + if (head > tail) { + const uint8_t *const b = static_cast<uint8_t *>(mBuf[cpu]) + gSessionData->mPageSize; + frame.add(cpu, head, tail, b); + + // Update tail with the data read + pemp->data_tail = head; + } + + if (mDiscard[cpu]) { + munmap(mBuf[cpu], gSessionData->mPageSize + BUF_SIZE); + mBuf[cpu] = MAP_FAILED; + mDiscard[cpu] = false; + mFds[cpu] = -1; + logg->logMessage("Unmaped cpu %i", cpu); + } + } + + frame.send(); + + return true; +} |