On Sun, 09 Apr 2017 14:33:26 +0200, Takashi Sakamoto wrote:
Hi,
ALSA IEC 61883-1/6 engine already supports some tracepoints to probe events for a part of parameters about incoming/outgoing packets. The main purpose of the tracepoints to assists users/developers for further work to the engine. For example, users can gather the parameters and developers use it to investigate quirks of actual units. Actually, as I reported [0], some drivers in ALSA firewire stack seems to have timestamping issue which have never resolved by any Free Software [1], and further investigation is required for users/developers to improves the engine.
My recent works add support for some variants of IEC 61883-1/6. Existent tracepoints are unavailable for a part of the variants. Furthermore, in a part of the variants, presentation timestamp is transferred by a different way against IEC 61883-1/6. It's better to trace packet parameters for the variants, as well as for compliant protocols to IEC 61883-1/6.
This patchset adds some tracepoints for the purpose, mainly for RME Fireface series and MOTU FireWire series. Below code/command is a sample to work with perf(1) for tracepoints on ALSA driver for MOTU FireWire series, added at third patch. The final command outputs contents of data chunks for message from units. Please install or build perf command with python 2 support, in advance.
$ cat ./motu.py import os import sys
sys.path.append(os.environ['PERF_EXEC_PATH'] + \ '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
from perf_trace_context import * from Core import *
# For a case that several MOTU FireWire units on the same bus. cycles = {}
def snd_firewire_lib__in_packet(name, context, cpu, secs, nsecs, pid, comm, callchain, second, cycle, channel, src, dest, cip_header0, cip_header1, payload_quadlets, packet_index, irq, index): global cycles
if src not in cycles: cycles[src] = {} cycles[src]['sec'] = second cycles[src]['cycle'] = cycledef snd_firewire_motu__in_data_block_message(name, context, cpu, secs, nsecs, pid, comm, callchain, src, dst, data_blocks, messages): global cycles
if src not in cycles: return print('{0} sec {1} cycle'.format(cycles[src]['sec'], cycles[src]['cycle'])) for i in range(data_blocks): offset = i * 8 msg = messages[offset:offset + 8] print(' {0}: {1[5]:02x}{1[4]:02x}{1[3]:02x}{1[2]:02x}{1[1]:02x}{1[0]:02x}'.format(i, msg))$ ~/linux.git/tools/perf/perf record -a -e snd_firewire_motu:in_data_block_message -e snd_firewire_lib:in_packet ^C $ PERF_EXEC_PATH=~/linux.git/tools/perf/ ~/linux.git/tools/perf/perf script -s ./motu.py ... 7 sec 7999 cycle 0: 000000805854 1: 0000008050fc 2: 0000808050a4 3: 00003f80504c 4: 0000008048f4 5: 00000080489c 6: 000080804844 7: 00003f8040ec 0 sec 0 cycle 0 sec 1 cycle 0: 000000804094 1: 0000008050fc 2: 0000808050a4 3: 00003f80504c 4: 0000008048f4 5: 00000080489c 6: 000080804844 7: 00003f8058ac 0 sec 2 cycle 0: 000000805854 1: 0000008050fc 2: 0000808050a4 3: 00003f80504c 4: 0000008048f4 5: 00000080605c 6: 000080806004 7: 00003f8058ac ...
[0] [alsa-devel] Dice packet sequence quirk and ALSA firewire stack in Linux 4.6 http://mailman.alsa-project.org/pipermail/alsa-devel/2016-May/107715.html [1] This is not resolved in FFADO, user space library to drive the units, too. Some workarounds are applied to the library, but it can not work well depending on units and protocols.
Takashi Sakamoto (3): ALSA: firewire_lib: add tracepoints for packets without CIP headers ALSA: firewire-motu: add tracepoints for SPH in IEC 61883-1 fashion ALSA: firewire-motu: add tracepoints for messages for unique protocol
Applied all three patches now. Thanks.
Takashi