Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Leo Yan <leo.yan@linaro.org>
To:     Arnaldo Carvalho de Melo <acme@kernel.org>,
        Mathieu Poirier <mathieu.poirier@linaro.org>,
        Jonathan Corbet <corbet@lwn.net>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@redhat.com>,
        Namhyung Kim <namhyung@kernel.org>,
        linux-arm-kernel@lists.infradead.org, linux-doc@vger.kernel.org,
        linux-kernel@vger.kernel.org, Tor Jeremiassen <tor@ti.com>,
        mike.leach@linaro.org, kim.phillips@arm.com,
        Robert Walker <Robert.Walker@arm.com>,
        coresight@lists.linaro.org
Cc:     Leo Yan <leo.yan@linaro.org>
Subject: [RFT v2 3/4] perf script python: Add script for CoreSight trace disassembler
Date:   Mon, 21 May 2018 16:52:27 +0800
Message-Id: <1526892748-326-4-git-send-email-leo.yan@linaro.org>
In-Reply-To: <1526892748-326-1-git-send-email-leo.yan@linaro.org>
References: <1526892748-326-1-git-send-email-leo.yan@linaro.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

This commit adds python script to parse CoreSight tracing event and
use command 'objdump' for disassembled lines, finally we can generate
readable program execution flow for reviewing tracing data.

The script receives CoreSight tracing packet with below format:

                +------------+------------+------------+
  packet(n):    |    addr    |    ip      |    cpu     |
                +------------+------------+------------+
  packet(n+1):  |    addr    |    ip      |    cpu     |
                +------------+------------+------------+

packet::ip is the last address of current branch instruction and
packet::addr presents the start address of the next coming branch
instruction.  So for one branch instruction which starts in packet(n),
its execution flow starts from packet(n)::addr and it stops at
packet(n+1)::ip.  As results we need to combine the two continuous
packets to generate the instruction range, this is the rationale for the
script implementation:

  [ sample(n)::addr .. sample(n+1)::ip ]

Credits to Tor Jeremiassen who have written the script skeleton and
provides the ideas for reading symbol file according to build-id,
creating memory map for dso and basic packet handling.  Mathieu Poirier
contributed fixes for build-id and memory map bugs.  The detailed
development history for this script you can find from [1].  Based on Tor
and Mathieu work, the script is updated samples handling for the
corrected sample format.  Another minor enhancement is to support for
without build-id case, the script can parse kernel symbols with option
'-k' for vmlinux file path.

[1] https://github.com/Linaro/perf-opencsd/commits/perf-opencsd-v4.15/tools/perf/scripts/python/cs-trace-disasm.py

Co-authored-by: Tor Jeremiassen <tor@ti.com>
Co-authored-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 tools/perf/scripts/python/arm-cs-trace-disasm.py | 234 +++++++++++++++++++++++
 1 file changed, 234 insertions(+)
 create mode 100644 tools/perf/scripts/python/arm-cs-trace-disasm.py

diff --git a/tools/perf/scripts/python/arm-cs-trace-disasm.py b/tools/perf/scripts/python/arm-cs-trace-disasm.py
new file mode 100644
index 0000000..58de36f
--- /dev/null
+++ b/tools/perf/scripts/python/arm-cs-trace-disasm.py
@@ -0,0 +1,234 @@
+# arm-cs-trace-disasm.py: ARM CoreSight Trace Dump With Disassember
+# SPDX-License-Identifier: GPL-2.0
+#
+# Tor Jeremiassen <tor@ti.com> is original author who wrote script
+# skeleton, Mathieu Poirier <mathieu.poirier@linaro.org> contributed
+# fixes for build-id and memory map; Leo Yan <leo.yan@linaro.org>
+# updated the packet parsing with new samples format.
+
+import os
+import sys
+import re
+from subprocess import *
+from optparse import OptionParser, make_option
+
+# Command line parsing
+
+option_list = [
+	# formatting options for the bottom entry of the stack
+	make_option("-k", "--vmlinux", dest="vmlinux_name",
+		    help="Set path to vmlinux file"),
+	make_option("-d", "--objdump", dest="objdump_name",
+		    help="Set path to objdump executable file"),
+	make_option("-v", "--verbose", dest="verbose",
+		    action="store_true", default=False,
+		    help="Enable debugging log")
+]
+
+parser = OptionParser(option_list=option_list)
+(options, args) = parser.parse_args()
+
+if (options.objdump_name == None):
+	sys.exit("No objdump executable file specified - use -d or --objdump option")
+
+# Initialize global dicts and regular expression
+
+build_ids = dict()
+mmaps = dict()
+disasm_cache = dict()
+cpu_data = dict()
+disasm_re = re.compile("^\s*([0-9a-fA-F]+):")
+disasm_func_re = re.compile("^\s*([0-9a-fA-F]+)\s\<.*\>:")
+cache_size = 32*1024
+prev_cpu = -1
+
+def parse_buildid():
+	global build_ids
+
+	buildid_regex = "([a-fA-f0-9]+)[ \t]([^ \n]+)"
+	buildid_re = re.compile(buildid_regex)
+
+	results = check_output(["perf", "buildid-list"]).split('\n');
+	for line in results:
+		m = buildid_re.search(line)
+		if (m == None):
+			continue;
+
+		id_name = m.group(2)
+		id_num  = m.group(1)
+
+		if (id_name == "[kernel.kallsyms]") :
+			append = "/kallsyms"
+		elif (id_name == "[vdso]") :
+			append = "/vdso"
+		else:
+			append = "/elf"
+
+		build_ids[id_name] = os.environ['PERF_BUILDID_DIR'] + \
+					"/" + id_name + "/" + id_num + append;
+		# Replace duplicate slash chars to single slash char
+		build_ids[id_name] = build_ids[id_name].replace('//', '/', 1)
+
+	if ((options.vmlinux_name == None) and ("[kernel.kallsyms]" in build_ids)):
+		print "kallsyms cannot be used to dump assembler"
+
+	# Set vmlinux path to replace kallsyms file, if without buildid we still
+	# can use vmlinux to prase kernel symbols
+	if ((options.vmlinux_name != None)):
+		build_ids['[kernel.kallsyms]'] = options.vmlinux_name;
+
+def parse_mmap():
+	global mmaps
+
+	# Check mmap for PERF_RECORD_MMAP and PERF_RECORD_MMAP2
+	mmap_regex = "PERF_RECORD_MMAP.* -?[0-9]+/[0-9]+: \[(0x[0-9a-fA-F]+)\((0x[0-9a-fA-F]+)\).*:\s.*\s(\S*)"
+	mmap_re = re.compile(mmap_regex)
+
+	results = check_output("perf script --show-mmap-events | fgrep PERF_RECORD_MMAP", shell=True).split('\n')
+	for line in results:
+		m = mmap_re.search(line)
+		if (m != None):
+			if (m.group(3) == '[kernel.kallsyms]_text'):
+				dso = '[kernel.kallsyms]'
+			else:
+				dso = m.group(3)
+
+			start = int(m.group(1),0)
+			end   = int(m.group(1),0) + int(m.group(2),0)
+			mmaps[dso] = [start, end]
+
+def find_dso_mmap(addr):
+	global mmaps
+
+	for key, value in mmaps.items():
+		if (addr >= value[0] and addr < value[1]):
+			return key
+
+	return None
+
+def read_disam(dso, start_addr, stop_addr):
+	global mmaps
+	global build_ids
+
+	addr_range = start_addr  + ":" + stop_addr;
+
+	# Don't let the cache get too big, clear it when it hits max size
+	if (len(disasm_cache) > cache_size):
+		disasm_cache.clear();
+
+	try:
+		disasm_output = disasm_cache[addr_range];
+	except:
+		try:
+			fname = build_ids[dso];
+		except KeyError:
+			sys.exit("cannot find symbol file for " + dso)
+
+		disasm = [ options.objdump_name, "-d", "-z",
+			   "--start-address="+start_addr,
+			   "--stop-address="+stop_addr, fname ]
+
+		disasm_output = check_output(disasm).split('\n')
+		disasm_cache[addr_range] = disasm_output;
+
+	return disasm_output
+
+def dump_disam(dso, start_addr, stop_addr, check_svc):
+	for line in read_disam(dso, start_addr, stop_addr):
+		m = disasm_func_re.search(line)
+		if (m != None):
+			print "\t",line
+			continue
+
+		m = disasm_re.search(line)
+		if (m == None):
+			continue;
+
+		print "\t",line
+
+		if ((check_svc == True) and "svc" in line):
+			return
+
+def dump_packet(sample):
+	print "Packet = { cpu: 0x%d addr: 0x%x phys_addr: 0x%x ip: 0x%x " \
+	      "pid: %d tid: %d period: %d time: %d }" % \
+	      (sample['cpu'], sample['addr'], sample['phys_addr'], \
+	       sample['ip'], sample['pid'], sample['tid'], \
+	       sample['period'], sample['time'])
+
+def trace_begin():
+	print 'ARM CoreSight Trace Data Assembler Dump'
+	parse_buildid()
+	parse_mmap()
+
+def trace_end():
+	print 'End'
+
+def trace_unhandled(event_name, context, event_fields_dict):
+	print ' '.join(['%s=%s'%(k,str(v))for k,v in sorted(event_fields_dict.items())])
+
+def process_event(param_dict):
+	global cache_size
+	global options
+	global prev_cpu
+
+	sample = param_dict["sample"]
+
+	if (options.verbose == True):
+		dump_packet(sample)
+
+	# If period doesn't equal to 1, this packet is for instruction sample
+	# packet, we need drop this synthetic packet.
+	if (sample['period'] != 1):
+		print "Skip synthetic instruction sample"
+		return
+
+	cpu = format(sample['cpu'], "d");
+
+	# Initialize CPU data if it's empty, and directly return back
+	# if this is the first tracing event for this CPU.
+	if (cpu_data.get(str(cpu) + 'addr') == None):
+		cpu_data[str(cpu) + 'addr'] = format(sample['addr'], "#x")
+		prev_cpu = cpu
+		return
+
+	# The format for packet is:
+	#
+	#                 +------------+------------+------------+
+	#  sample_prev:   |    addr    |    ip      |    cpu     |
+	#                 +------------+------------+------------+
+	#  sample_next:   |    addr    |    ip      |    cpu     |
+	#                 +------------+------------+------------+
+	#
+	# We need to combine the two continuous packets to get the instruction
+	# range for sample_prev::cpu:
+	#
+	#     [ sample_prev::addr .. sample_next::ip ]
+	#
+	# For this purose, sample_prev::addr is stored into cpu_data structure
+	# and read back for 'start_addr' when the new packet comes, and we need
+	# to use sample_next::ip to calculate 'stop_addr', plusing extra 4 for
+	# 'stop_addr' is for the sake of objdump so the final assembler dump can
+	# include last instruction for sample_next::ip.
+
+	start_addr = cpu_data[str(prev_cpu) + 'addr']
+	stop_addr  = format(sample['ip'] + 4, "#x")
+
+	# Sanity checking dso for start_addr and stop_addr
+	prev_dso = find_dso_mmap(int(start_addr, 0))
+	next_dso = find_dso_mmap(int(stop_addr, 0))
+
+	# If cannot find dso so cannot dump assembler, bail out
+	if (prev_dso == None or next_dso == None):
+		print "Address range [ %s .. %s ]: failed to find dso" % (start_addr, stop_addr)
+		prev_cpu = cpu
+		return
+	elif (prev_dso != next_dso):
+		print "Address range [ %s .. %s ]: isn't in same dso" % (start_addr, stop_addr)
+		prev_cpu = cpu
+		return
+
+	dump_disam(prev_dso, start_addr, stop_addr, False)
+
+	cpu_data[str(cpu) + 'addr'] = format(sample['addr'], "#x")
+	prev_cpu = cpu
-- 
2.7.4