qemu/scripts/simpletrace.py

222 lines
7.0 KiB
Python
Raw Normal View History

#!/usr/bin/env python
#
# Pretty-printer for simple trace backend binary trace files
#
# Copyright IBM, Corp. 2010
#
# This work is licensed under the terms of the GNU GPL, version 2. See
# the COPYING file in the top-level directory.
#
# For help see docs/tracing.txt
import struct
import re
import inspect
from tracetool import read_events, Event
from tracetool.backend.simple import is_string
header_event_id = 0xffffffffffffffff
header_magic = 0xf2b177cb0aa429b4
dropped_event_id = 0xfffffffffffffffe
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
record_type_mapping = 0
record_type_event = 1
log_header_fmt = '=QQQ'
rec_header_fmt = '=QQII'
def read_header(fobj, hfmt):
'''Read a trace record header'''
hlen = struct.calcsize(hfmt)
hdr = fobj.read(hlen)
if len(hdr) != hlen:
return None
return struct.unpack(hfmt, hdr)
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
def get_record(edict, idtoname, rechdr, fobj):
"""Deserialize a trace record from a file into a tuple
(name, timestamp, pid, arg1, ..., arg6)."""
if rechdr is None:
return None
if rechdr[0] != dropped_event_id:
event_id = rechdr[0]
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
name = idtoname[event_id]
rec = (name, rechdr[1], rechdr[3])
event = edict[name]
for type, name in event.args:
if is_string(type):
l = fobj.read(4)
(len,) = struct.unpack('=L', l)
s = fobj.read(len)
rec = rec + (s,)
else:
(value,) = struct.unpack('=Q', fobj.read(8))
rec = rec + (value,)
else:
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
rec = ("dropped", rechdr[1], rechdr[3])
(value,) = struct.unpack('=Q', fobj.read(8))
rec = rec + (value,)
return rec
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
def get_mapping(fobj):
(event_id, ) = struct.unpack('=Q', fobj.read(8))
(len, ) = struct.unpack('=L', fobj.read(4))
name = fobj.read(len)
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
return (event_id, name)
def read_record(edict, idtoname, fobj):
"""Deserialize a trace record from a file into a tuple (event_num, timestamp, pid, arg1, ..., arg6)."""
rechdr = read_header(fobj, rec_header_fmt)
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
return get_record(edict, idtoname, rechdr, fobj)
def read_trace_header(fobj):
"""Read and verify trace file header"""
header = read_header(fobj, log_header_fmt)
if header is None:
raise ValueError('Not a valid trace file!')
if header[0] != header_event_id:
raise ValueError('Not a valid trace file, header id %d != %d' %
(header[0], header_event_id))
if header[1] != header_magic:
raise ValueError('Not a valid trace file, header magic %d != %d' %
(header[1], header_magic))
log_version = header[2]
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
if log_version not in [0, 2, 3, 4]:
raise ValueError('Unknown version of tracelog format!')
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
if log_version != 4:
raise ValueError('Log format %d not supported with this QEMU release!'
% log_version)
def read_trace_records(edict, fobj):
"""Deserialize trace records from a file, yielding record tuples (event_num, timestamp, pid, arg1, ..., arg6)."""
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
idtoname = {
dropped_event_id: "dropped"
}
while True:
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
t = fobj.read(8)
if len(t) == 0:
break
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
(rectype, ) = struct.unpack('=Q', t)
if rectype == record_type_mapping:
event_id, name = get_mapping(fobj)
idtoname[event_id] = name
else:
rec = read_record(edict, idtoname, fobj)
yield rec
class Analyzer(object):
"""A trace file analyzer which processes trace records.
An analyzer can be passed to run() or process(). The begin() method is
invoked, then each trace record is processed, and finally the end() method
is invoked.
If a method matching a trace event name exists, it is invoked to process
that trace record. Otherwise the catchall() method is invoked."""
def begin(self):
"""Called at the start of the trace."""
pass
def catchall(self, event, rec):
"""Called if no specific method for processing a trace event has been found."""
pass
def end(self):
"""Called at the end of the trace."""
pass
def process(events, log, analyzer, read_header=True):
"""Invoke an analyzer on each event in a log."""
if isinstance(events, str):
events = read_events(open(events, 'r'))
if isinstance(log, str):
log = open(log, 'rb')
if read_header:
read_trace_header(log)
dropped_event = Event.build("Dropped_Event(uint64_t num_events_dropped)")
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
edict = {"dropped": dropped_event}
trace: emit name <-> ID mapping in simpletrace header Currently simpletrace assumes that events are given IDs starting from 0, based on the order in which they appear in the trace-events file, with no gaps. When the trace-events file is split up, this assumption becomes problematic. To deal with this, extend the simpletrace format so that it outputs a table of event name <-> ID mappings. That will allow QEMU to assign arbitrary IDs to events without breaking simpletrace parsing. The v3 simple trace format was FILE HEADER EVENT TRACE RECORD 0 EVENT TRACE RECORD 1 ... EVENT TRACE RECORD N The v4 simple trace format is now FILE HEADER EVENT MAPPING RECORD 0 EVENT MAPPING RECORD 1 ... EVENT MAPPING RECORD M EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 ... EVENT TRACE RECORD N Although this shows all the mapping records being emitted upfront, this is not required by the format. While the main simpletrace backend will emit all mappings at startup, the systemtap simpletrace.stp script will emit the mappings at first use. eg FILE HEADER ... EVENT MAPPING RECORD 0 EVENT TRACE RECORD RECORD 0 EVENT TRACE RECORD RECORD 1 EVENT MAPPING RECORD 1 EVENT TRACE RECORD RECORD 2 ... EVENT TRACE RECORD N This is more space efficient given that most trace records only include a subset of events. In modifying the systemtap simpletrace code, a 'begin' probe was added to emit the trace event header, so you no longer need to add '--no-header' when running simpletrace.py for systemtap generated trace files. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Message-id: 1475588159-30598-12-git-send-email-berrange@redhat.com Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2016-10-04 21:35:50 +08:00
for event in events:
edict[event.name] = event
def build_fn(analyzer, event):
if isinstance(event, str):
return analyzer.catchall
fn = getattr(analyzer, event.name, None)
if fn is None:
return analyzer.catchall
event_argcount = len(event.args)
fn_argcount = len(inspect.getargspec(fn)[0]) - 1
if fn_argcount == event_argcount + 1:
# Include timestamp as first argument
return lambda _, rec: fn(*((rec[1:2],) + rec[3:3 + event_argcount]))
elif fn_argcount == event_argcount + 2:
# Include timestamp and pid
return lambda _, rec: fn(*rec[1:3 + event_argcount])
else:
# Just arguments, no timestamp or pid
return lambda _, rec: fn(*rec[3:3 + event_argcount])
analyzer.begin()
fn_cache = {}
for rec in read_trace_records(edict, log):
event_num = rec[0]
event = edict[event_num]
if event_num not in fn_cache:
fn_cache[event_num] = build_fn(analyzer, event)
fn_cache[event_num](event, rec)
analyzer.end()
def run(analyzer):
"""Execute an analyzer on a trace file given on the command-line.
This function is useful as a driver for simple analysis scripts. More
advanced scripts will want to call process() instead."""
import sys
read_header = True
if len(sys.argv) == 4 and sys.argv[1] == '--no-header':
read_header = False
del sys.argv[1]
elif len(sys.argv) != 3:
sys.stderr.write('usage: %s [--no-header] <trace-events> ' \
'<trace-file>\n' % sys.argv[0])
sys.exit(1)
events = read_events(open(sys.argv[1], 'r'))
process(events, sys.argv[2], analyzer, read_header=read_header)
if __name__ == '__main__':
class Formatter(Analyzer):
def __init__(self):
self.last_timestamp = None
def catchall(self, event, rec):
timestamp = rec[1]
if self.last_timestamp is None:
self.last_timestamp = timestamp
delta_ns = timestamp - self.last_timestamp
self.last_timestamp = timestamp
fields = [event.name, '%0.3f' % (delta_ns / 1000.0),
'pid=%d' % rec[2]]
i = 3
for type, name in event.args:
if is_string(type):
fields.append('%s=%s' % (name, rec[i]))
else:
fields.append('%s=0x%x' % (name, rec[i]))
i += 1
print ' '.join(fields)
run(Formatter())