It turns out that cgroup filtering is relatively trivial and works
really nicely. Thid diffs adds automatic cgroup filtering when in
per-cpu mode, unless a new --disable-cgroup-filtering flag is passed in
the start command. At least on Meta machines, all processes are spawned
inside a cgroup by default, which comes super handy, because per cpu
tracing is now much more precise.
A manual test gave me this result
- Without filtering:
Total number of trace items: 36083
Total number of continuous executions found: 229
Number of continuous executions for this thread: 2
Total number of PSB blocks found: 98
Number of PSB blocks for this thread 2
Total number of unattributed PSB blocks found: 38
- With filtering:
Total number of trace items: 87756
Total number of continuous executions found: 123
Number of continuous executions for this thread: 2
Total number of PSB blocks found: 10
Number of PSB blocks for this thread 3
Total number of unattributed PSB blocks found: 2
Filtering gives us great results. The number of instructions collected
more than double (probalby because we have less noise in the trace), and
we have much less unattributed PSBs blocks and unrelated PSBs in
general. The ones that are unrelated probably belong to other processes
in the same cgroup.
Differential Revision: https://reviews.llvm.org/D129257
290 lines
8.9 KiB
C++
290 lines
8.9 KiB
C++
//===-- IntelPTSingleBufferTrace.cpp --------------------------------------===//
|
|
//
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#include "IntelPTSingleBufferTrace.h"
|
|
|
|
#include "Plugins/Process/POSIX/ProcessPOSIXLog.h"
|
|
#include "lldb/Utility/Status.h"
|
|
#include "lldb/Utility/StreamString.h"
|
|
|
|
#include "llvm/Support/Host.h"
|
|
#include "llvm/Support/MemoryBuffer.h"
|
|
|
|
#include <sstream>
|
|
|
|
#include <linux/perf_event.h>
|
|
#include <sys/syscall.h>
|
|
#include <unistd.h>
|
|
|
|
using namespace lldb;
|
|
using namespace lldb_private;
|
|
using namespace process_linux;
|
|
using namespace llvm;
|
|
|
|
const char *kOSEventIntelPTTypeFile =
|
|
"/sys/bus/event_source/devices/intel_pt/type";
|
|
|
|
const char *kPSBPeriodCapFile =
|
|
"/sys/bus/event_source/devices/intel_pt/caps/psb_cyc";
|
|
|
|
const char *kPSBPeriodValidValuesFile =
|
|
"/sys/bus/event_source/devices/intel_pt/caps/psb_periods";
|
|
|
|
const char *kPSBPeriodBitOffsetFile =
|
|
"/sys/bus/event_source/devices/intel_pt/format/psb_period";
|
|
|
|
const char *kTSCBitOffsetFile =
|
|
"/sys/bus/event_source/devices/intel_pt/format/tsc";
|
|
|
|
enum IntelPTConfigFileType {
|
|
Hex = 0,
|
|
// 0 or 1
|
|
ZeroOne,
|
|
Decimal,
|
|
// a bit index file always starts with the prefix config: following by an int,
|
|
// which represents the offset of the perf_event_attr.config value where to
|
|
// store a given configuration.
|
|
BitOffset
|
|
};
|
|
|
|
static Expected<uint32_t> ReadIntelPTConfigFile(const char *file,
|
|
IntelPTConfigFileType type) {
|
|
ErrorOr<std::unique_ptr<MemoryBuffer>> stream =
|
|
MemoryBuffer::getFileAsStream(file);
|
|
|
|
if (!stream)
|
|
return createStringError(inconvertibleErrorCode(),
|
|
"Can't open the file '%s'", file);
|
|
|
|
uint32_t value = 0;
|
|
StringRef text_buffer = stream.get()->getBuffer();
|
|
|
|
if (type == BitOffset) {
|
|
const char *prefix = "config:";
|
|
if (!text_buffer.startswith(prefix))
|
|
return createStringError(inconvertibleErrorCode(),
|
|
"The file '%s' contents doesn't start with '%s'",
|
|
file, prefix);
|
|
text_buffer = text_buffer.substr(strlen(prefix));
|
|
}
|
|
|
|
auto getRadix = [&]() {
|
|
switch (type) {
|
|
case Hex:
|
|
return 16;
|
|
case ZeroOne:
|
|
case Decimal:
|
|
case BitOffset:
|
|
return 10;
|
|
}
|
|
llvm_unreachable("Fully covered switch above!");
|
|
};
|
|
|
|
auto createError = [&](const char *expected_value_message) {
|
|
return createStringError(
|
|
inconvertibleErrorCode(),
|
|
"The file '%s' has an invalid value. It should be %s.", file,
|
|
expected_value_message);
|
|
};
|
|
|
|
if (text_buffer.trim().consumeInteger(getRadix(), value) ||
|
|
(type == ZeroOne && value != 0 && value != 1)) {
|
|
switch (type) {
|
|
case Hex:
|
|
return createError("an unsigned hexadecimal int");
|
|
case ZeroOne:
|
|
return createError("0 or 1");
|
|
case Decimal:
|
|
case BitOffset:
|
|
return createError("an unsigned decimal int");
|
|
}
|
|
}
|
|
return value;
|
|
}
|
|
|
|
/// Return the Linux perf event type for Intel PT.
|
|
Expected<uint32_t> process_linux::GetIntelPTOSEventType() {
|
|
return ReadIntelPTConfigFile(kOSEventIntelPTTypeFile,
|
|
IntelPTConfigFileType::Decimal);
|
|
}
|
|
|
|
static Error CheckPsbPeriod(size_t psb_period) {
|
|
Expected<uint32_t> cap =
|
|
ReadIntelPTConfigFile(kPSBPeriodCapFile, IntelPTConfigFileType::ZeroOne);
|
|
if (!cap)
|
|
return cap.takeError();
|
|
if (*cap == 0)
|
|
return createStringError(inconvertibleErrorCode(),
|
|
"psb_period is unsupported in the system.");
|
|
|
|
Expected<uint32_t> valid_values = ReadIntelPTConfigFile(
|
|
kPSBPeriodValidValuesFile, IntelPTConfigFileType::Hex);
|
|
if (!valid_values)
|
|
return valid_values.takeError();
|
|
|
|
if (valid_values.get() & (1 << psb_period))
|
|
return Error::success();
|
|
|
|
std::ostringstream error;
|
|
// 0 is always a valid value
|
|
error << "Invalid psb_period. Valid values are: 0";
|
|
uint32_t mask = valid_values.get();
|
|
while (mask) {
|
|
int index = __builtin_ctz(mask);
|
|
if (index > 0)
|
|
error << ", " << index;
|
|
// clear the lowest bit
|
|
mask &= mask - 1;
|
|
}
|
|
error << ".";
|
|
return createStringError(inconvertibleErrorCode(), error.str().c_str());
|
|
}
|
|
|
|
#ifdef PERF_ATTR_SIZE_VER5
|
|
static Expected<uint64_t>
|
|
GeneratePerfEventConfigValue(bool enable_tsc, Optional<uint64_t> psb_period) {
|
|
uint64_t config = 0;
|
|
// tsc is always supported
|
|
if (enable_tsc) {
|
|
if (Expected<uint32_t> offset = ReadIntelPTConfigFile(
|
|
kTSCBitOffsetFile, IntelPTConfigFileType::BitOffset))
|
|
config |= 1 << *offset;
|
|
else
|
|
return offset.takeError();
|
|
}
|
|
if (psb_period) {
|
|
if (Error error = CheckPsbPeriod(*psb_period))
|
|
return std::move(error);
|
|
|
|
if (Expected<uint32_t> offset = ReadIntelPTConfigFile(
|
|
kPSBPeriodBitOffsetFile, IntelPTConfigFileType::BitOffset))
|
|
config |= *psb_period << *offset;
|
|
else
|
|
return offset.takeError();
|
|
}
|
|
return config;
|
|
}
|
|
|
|
/// Create a \a perf_event_attr configured for
|
|
/// an IntelPT event.
|
|
///
|
|
/// \return
|
|
/// A \a perf_event_attr if successful,
|
|
/// or an \a llvm::Error otherwise.
|
|
static Expected<perf_event_attr>
|
|
CreateIntelPTPerfEventConfiguration(bool enable_tsc,
|
|
llvm::Optional<uint64_t> psb_period) {
|
|
perf_event_attr attr;
|
|
memset(&attr, 0, sizeof(attr));
|
|
attr.size = sizeof(attr);
|
|
attr.exclude_kernel = 1;
|
|
attr.exclude_hv = 1;
|
|
attr.exclude_idle = 1;
|
|
|
|
if (Expected<uint64_t> config_value =
|
|
GeneratePerfEventConfigValue(enable_tsc, psb_period))
|
|
attr.config = *config_value;
|
|
else
|
|
return config_value.takeError();
|
|
|
|
if (Expected<uint32_t> intel_pt_type = GetIntelPTOSEventType())
|
|
attr.type = *intel_pt_type;
|
|
else
|
|
return intel_pt_type.takeError();
|
|
|
|
return attr;
|
|
}
|
|
#endif
|
|
|
|
size_t IntelPTSingleBufferTrace::GetIptTraceSize() const {
|
|
return m_perf_event.GetAuxBuffer().size();
|
|
}
|
|
|
|
Error IntelPTSingleBufferTrace::Pause() {
|
|
return m_perf_event.DisableWithIoctl();
|
|
}
|
|
|
|
Error IntelPTSingleBufferTrace::Resume() {
|
|
return m_perf_event.EnableWithIoctl();
|
|
}
|
|
|
|
Expected<std::vector<uint8_t>> IntelPTSingleBufferTrace::GetIptTrace() {
|
|
// Disable the perf event to force a flush out of the CPU's internal buffer.
|
|
// Besides, we can guarantee that the CPU won't override any data as we are
|
|
// reading the buffer.
|
|
// The Intel documentation says:
|
|
//
|
|
// Packets are first buffered internally and then written out
|
|
// asynchronously. To collect packet output for postprocessing, a collector
|
|
// needs first to ensure that all packet data has been flushed from internal
|
|
// buffers. Software can ensure this by stopping packet generation by
|
|
// clearing IA32_RTIT_CTL.TraceEn (see “Disabling Packet Generation” in
|
|
// Section 35.2.7.2).
|
|
//
|
|
// This is achieved by the PERF_EVENT_IOC_DISABLE ioctl request, as
|
|
// mentioned in the man page of perf_event_open.
|
|
return m_perf_event.GetReadOnlyAuxBuffer();
|
|
}
|
|
|
|
Expected<IntelPTSingleBufferTrace> IntelPTSingleBufferTrace::Start(
|
|
const TraceIntelPTStartRequest &request, Optional<lldb::tid_t> tid,
|
|
Optional<cpu_id_t> cpu_id, bool disabled, Optional<int> cgroup_fd) {
|
|
#ifndef PERF_ATTR_SIZE_VER5
|
|
return createStringError(inconvertibleErrorCode(),
|
|
"Intel PT Linux perf event not supported");
|
|
#else
|
|
Log *log = GetLog(POSIXLog::Trace);
|
|
|
|
LLDB_LOG(log, "Will start tracing thread id {0} and cpu id {1}", tid, cpu_id);
|
|
|
|
if (__builtin_popcount(request.ipt_trace_size) != 1 ||
|
|
request.ipt_trace_size < 4096) {
|
|
return createStringError(
|
|
inconvertibleErrorCode(),
|
|
"The intel pt trace size must be a power of 2 greater than or equal to "
|
|
"4096 (2^12) bytes. It was %" PRIu64 ".",
|
|
request.ipt_trace_size);
|
|
}
|
|
uint64_t page_size = getpagesize();
|
|
uint64_t aux_buffer_numpages = static_cast<uint64_t>(llvm::PowerOf2Floor(
|
|
(request.ipt_trace_size + page_size - 1) / page_size));
|
|
|
|
Expected<perf_event_attr> attr = CreateIntelPTPerfEventConfiguration(
|
|
request.enable_tsc, request.psb_period.map([](int value) {
|
|
return static_cast<uint64_t>(value);
|
|
}));
|
|
if (!attr)
|
|
return attr.takeError();
|
|
attr->disabled = disabled;
|
|
|
|
LLDB_LOG(log, "Will create intel pt trace buffer of size {0}",
|
|
request.ipt_trace_size);
|
|
unsigned long flags = 0;
|
|
if (cgroup_fd) {
|
|
tid = *cgroup_fd;
|
|
flags |= PERF_FLAG_PID_CGROUP;
|
|
}
|
|
|
|
if (Expected<PerfEvent> perf_event =
|
|
PerfEvent::Init(*attr, tid, cpu_id, -1, flags)) {
|
|
if (Error mmap_err = perf_event->MmapMetadataAndBuffers(
|
|
/*num_data_pages=*/0, aux_buffer_numpages,
|
|
/*data_buffer_write=*/true)) {
|
|
return std::move(mmap_err);
|
|
}
|
|
return IntelPTSingleBufferTrace(std::move(*perf_event));
|
|
} else {
|
|
return perf_event.takeError();
|
|
}
|
|
#endif
|
|
}
|
|
|
|
const PerfEvent &IntelPTSingleBufferTrace::GetPerfEvent() const {
|
|
return m_perf_event;
|
|
}
|