This commit adds python script to parse CoreSight tracing event and print out source line and disassembly, it generates readable program execution flow for easier humans inspecting. The script receives CoreSight tracing packet with below format: +------------+------------+------------+ packet(n): | addr | ip | cpu | +------------+------------+------------+ packet(n+1): | addr | ip | cpu | +------------+------------+------------+ packet::addr presents the start address of the coming branch sample, and packet::ip is the last address of the branch smple. Therefore, a code section between branches starts from packet(n)::addr and it stops at packet(n+1)::ip. As results we combines the two continuous packets to generate the address range for instructions: [ sample(n)::addr .. sample(n+1)::ip ] The script supports both objdump or llvm-objdump for disassembly with specifying option '-d'. If doesn't specify option '-d', the script simply outputs source lines and symbols. Below shows usages with llvm-objdump or objdump to output disassembly. # perf script -s scripts/python/arm-cs-trace-disasm.py -- -d llvm-objdump-11 -k ./vmlinux ARM CoreSight Trace Data Assembler Dump ffff800008eb3198 <etm4_enable_hw>: ffff800008eb3310: c0 38 00 35 cbnz w0, 0xffff800008eb3a28 <etm4_enable_hw+0x890> ffff800008eb3314: 9f 3f 03 d5 dsb sy ffff800008eb3318: df 3f 03 d5 isb ffff800008eb331c: f5 5b 42 a9 ldp x21, x22, [sp, #32] ffff800008eb3320: fb 73 45 a9 ldp x27, x28, [sp, #80] ffff800008eb3324: e0 82 40 39 ldrb w0, [x23, #32] ffff800008eb3328: 60 00 00 34 cbz w0, 0xffff800008eb3334 <etm4_enable_hw+0x19c> ffff800008eb332c: e0 03 19 aa mov x0, x25 ffff800008eb3330: 8c fe ff 97 bl 0xffff800008eb2d60 <etm4_cs_lock.isra.0.part.0> main 6728/6728 [0004] 0.000000000 etm4_enable_hw+0x198 [kernel.kallsyms] ffff800008eb2d60 <etm4_cs_lock.isra.0.part.0>: ffff800008eb2d60: 1f 20 03 d5 nop ffff800008eb2d64: 1f 20 03 d5 nop ffff800008eb2d68: 3f 23 03 d5 hint #25 ffff800008eb2d6c: 00 00 40 f9 ldr x0, [x0] ffff800008eb2d70: 9f 3f 03 d5 dsb sy ffff800008eb2d74: 00 c0 3e 91 add x0, x0, #4016 ffff800008eb2d78: 1f 00 00 b9 str wzr, [x0] ffff800008eb2d7c: bf 23 03 d5 hint #29 ffff800008eb2d80: c0 03 5f d6 ret main 6728/6728 [0004] 0.000000000 etm4_cs_lock.isra.0.part.0+0x20 # perf script -s scripts/python/arm-cs-trace-disasm.py -- -d objdump -k ./vmlinux ARM CoreSight Trace Data Assembler Dump ffff800008eb3310 <etm4_enable_hw+0x178>: ffff800008eb3310: 350038c0 cbnz w0, ffff800008eb3a28 <etm4_enable_hw+0x890> ffff800008eb3314: d5033f9f dsb sy ffff800008eb3318: d5033fdf isb ffff800008eb331c: a9425bf5 ldp x21, x22, [sp, #32] ffff800008eb3320: a94573fb ldp x27, x28, [sp, #80] ffff800008eb3324: 394082e0 ldrb w0, [x23, #32] ffff800008eb3328: 34000060 cbz w0, ffff800008eb3334 <etm4_enable_hw+0x19c> ffff800008eb332c: aa1903e0 mov x0, x25 ffff800008eb3330: 97fffe8c bl ffff800008eb2d60 <etm4_cs_lock.isra.0.part.0> main 6728/6728 [0004] 0.000000000 etm4_enable_hw+0x198 [kernel.kallsyms] ffff800008eb2d60 <etm4_cs_lock.isra.0.part.0>: ffff800008eb2d60: d503201f nop ffff800008eb2d64: d503201f nop ffff800008eb2d68: d503233f paciasp ffff800008eb2d6c: f9400000 ldr x0, [x0] ffff800008eb2d70: d5033f9f dsb sy ffff800008eb2d74: 913ec000 add x0, x0, #0xfb0 ffff800008eb2d78: b900001f str wzr, [x0] ffff800008eb2d7c: d50323bf autiasp ffff800008eb2d80: d65f03c0 ret main 6728/6728 [0004] 0.000000000 etm4_cs_lock.isra.0.part.0+0x20
Showing the changes to tools/perf/scripts/python (1 of 1 file). Show all
Unified · Split
| @@ -0,0 +1,272 @@ | |||
| 1 | # SPDX-License-Identifier: GPL-2.0 | ||
| 2 | # arm-cs-trace-disasm.py: ARM CoreSight Trace Dump With Disassember | ||
| 3 | # | ||
| 4 | # Author: Tor Jeremiassen <tor@ti.com> | ||
| 5 | # Mathieu Poirier <mathieu.poirier@linaro.org> | ||
| 6 | # Leo Yan <leo.yan@linaro.org> | ||
| 7 | # Al Grant <Al.Grant@arm.com> | ||
| 8 | |||
| 9 | from __future__ import print_function | ||
| 10 | import os | ||
| 11 | from os import path | ||
| 12 | import sys | ||
| 13 | import re | ||
| 14 | from subprocess import * | ||
| 15 | from optparse import OptionParser, make_option | ||
| 16 | |||
| 17 | from perf_trace_context import perf_set_itrace_options, \ | ||
| 18 | perf_sample_insn, perf_sample_srccode | ||
| 19 | |||
| 20 | # Below are some example commands for using this script. | ||
| 21 | # | ||
| 22 | # Output disassembly with objdump: | ||
| 23 | # perf script -s scripts/python/arm-cs-trace-disasm.py \ | ||
| 24 | # -- -d objdump -k path/to/vmlinux | ||
| 25 | # Output disassembly with llvm-objdump: | ||
| 26 | # perf script -s scripts/python/arm-cs-trace-disasm.py \ | ||
| 27 | # -- -d llvm-objdump-11 -k path/to/vmlinux | ||
| 28 | # Output only source line and symbols: | ||
| 29 | # perf script -s scripts/python/arm-cs-trace-disasm.py | ||
| 30 | |||
| 31 | # Command line parsing. | ||
| 32 | option_list = [ | ||
| 33 | # formatting options for the bottom entry of the stack | ||
| 34 | make_option("-k", "--vmlinux", dest="vmlinux_name", | ||
| 35 | help="Set path to vmlinux file"), | ||
| 36 | make_option("-d", "--objdump", dest="objdump_name", | ||
| 37 | help="Set path to objdump executable file"), | ||
| 38 | make_option("-v", "--verbose", dest="verbose", | ||
| 39 | action="store_true", default=False, | ||
| 40 | help="Enable debugging log") | ||
| 41 | ] | ||
| 42 | |||
| 43 | parser = OptionParser(option_list=option_list) | ||
| 44 | (options, args) = parser.parse_args() | ||
| 45 | |||
| 46 | # Initialize global dicts and regular expression | ||
| 47 | disasm_cache = dict() | ||
| 48 | cpu_data = dict() | ||
| 49 | disasm_re = re.compile("^\s*([0-9a-fA-F]+):") | ||
| 50 | disasm_func_re = re.compile("^\s*([0-9a-fA-F]+)\s.*:") | ||
| 51 | cache_size = 64*1024 | ||
| 52 | |||
| 53 | glb_source_file_name = None | ||
| 54 | glb_line_number = None | ||
| 55 | glb_dso = None | ||
| 56 | |||
| 57 | def get_optional(perf_dict, field): | ||
| 58 | if field in perf_dict: | ||
| 59 | return perf_dict[field] | ||
| 60 | return "[unknown]" | ||
| 61 | |||
| 62 | def get_offset(perf_dict, field): | ||
| 63 | if field in perf_dict: | ||
| 64 | return f"+0x{perf_dict[field]:x}" | ||
| 65 | return "" | ||
| 66 | |||
| 67 | def get_dso_file_path(dso_name, dso_build_id): | ||
| 68 | if (dso_name == "[kernel.kallsyms]" or dso_name == "vmlinux"): | ||
| 69 | if (options.vmlinux_name): | ||
| 70 | return options.vmlinux_name; | ||
| 71 | else: | ||
| 72 | return dso_name | ||
| 73 | |||
| 74 | if (dso_name == "[vdso]") : | ||
| 75 | append = "/vdso" | ||
| 76 | else: | ||
| 77 | append = "/elf" | ||
| 78 | |||
| 79 | dso_path = f"{os.environ['PERF_BUILDID_DIR']}/{dso_name}/{dso_build_id}{append}" | ||
| 80 | # Replace duplicate slash chars to single slash char | ||
| 81 | dso_path = dso_path.replace('//', '/', 1) | ||
| 82 | return dso_path | ||
| 83 | |||
| 84 | def read_disam(dso_fname, dso_start, start_addr, stop_addr): | ||
| 85 | addr_range = str(start_addr) + ":" + str(stop_addr) + ":" + dso_fname | ||
| 86 | |||
| 87 | # Don't let the cache get too big, clear it when it hits max size | ||
| 88 | if (len(disasm_cache) > cache_size): | ||
| 89 | disasm_cache.clear(); | ||
| 90 | |||
| 91 | if addr_range in disasm_cache: | ||
| 92 | disasm_output = disasm_cache[addr_range]; | ||
| 93 | else: | ||
| 94 | start_addr = start_addr - dso_start; | ||
| 95 | stop_addr = stop_addr - dso_start; | ||
| 96 | disasm = [ options.objdump_name, "-d", "-z", | ||
| 97 | f"--start-address=0x{start_addr:x}", | ||
| 98 | f"--stop-address=0x{stop_addr:x}" ] | ||
| 99 | disasm += [ dso_fname ] | ||
| 100 | disasm_output = check_output(disasm).decode('utf-8').split('\n') | ||
| 101 | disasm_cache[addr_range] = disasm_output | ||
| 102 | |||
| 103 | return disasm_output | ||
| 104 | |||
| 105 | def print_disam(dso_fname, dso_start, start_addr, stop_addr): | ||
| 106 | for line in read_disam(dso_fname, dso_start, start_addr, stop_addr): | ||
| 107 | m = disasm_func_re.search(line) | ||
| 108 | if m is None: | ||
| 109 | m = disasm_re.search(line) | ||
| 110 | if m is None: | ||
| 111 | continue | ||
| 112 | print(f"\t{line}") | ||
| 113 | |||
| 114 | def print_sample(sample): | ||
| 115 | print(f"Sample = {{ cpu: {sample['cpu']:04} addr: 0x{sample['addr']:016x} " \ | ||
| 116 | f"phys_addr: 0x{sample['phys_addr']:016x} ip: 0x{sample['ip']:016x} " \ | ||
| 117 | f"pid: {sample['pid']} tid: {sample['tid']} period: {sample['period']} time: {sample['time']} }}") | ||
| 118 | |||
| 119 | def trace_begin(): | ||
| 120 | print('ARM CoreSight Trace Data Assembler Dump') | ||
| 121 | |||
| 122 | def trace_end(): | ||
| 123 | print('End') | ||
| 124 | |||
| 125 | def trace_unhandled(event_name, context, event_fields_dict): | ||
| 126 | print(' '.join(['%s=%s'%(k,str(v))for k,v in sorted(event_fields_dict.items())])) | ||
| 127 | |||
| 128 | def common_start_str(comm, sample): | ||
| 129 | sec = int(sample["time"] / 1000000000) | ||
| 130 | ns = sample["time"] % 1000000000 | ||
| 131 | cpu = sample["cpu"] | ||
| 132 | pid = sample["pid"] | ||
| 133 | tid = sample["tid"] | ||
| 134 | return f"{comm:>16} {pid:>5}/{tid:<5} [{cpu:04}] {sec:9}.{ns:09} " | ||
| 135 | |||
| 136 | # This code is copied from intel-pt-events.py for printing source code | ||
| 137 | # line and symbols. | ||
| 138 | def print_srccode(comm, param_dict, sample, symbol, dso): | ||
| 139 | ip = sample["ip"] | ||
| 140 | if symbol == "[unknown]": | ||
| 141 | start_str = common_start_str(comm, sample) + ("%x" % ip).rjust(16).ljust(40) | ||
| 142 | else: | ||
| 143 | offs = get_offset(param_dict, "symoff") | ||
| 144 | start_str = common_start_str(comm, sample) + (symbol + offs).ljust(40) | ||
| 145 | |||
| 146 | global glb_source_file_name | ||
| 147 | global glb_line_number | ||
| 148 | global glb_dso | ||
| 149 | |||
| 150 | source_file_name, line_number, source_line = perf_sample_srccode(perf_script_context) | ||
| 151 | if source_file_name: | ||
| 152 | if glb_line_number == line_number and glb_source_file_name == source_file_name: | ||
| 153 | src_str = "" | ||
| 154 | else: | ||
| 155 | if len(source_file_name) > 40: | ||
| 156 | src_file = ("..." + source_file_name[-37:]) + " " | ||
| 157 | else: | ||
| 158 | src_file = source_file_name.ljust(41) | ||
| 159 | |||
| 160 | if source_line is None: | ||
| 161 | src_str = src_file + str(line_number).rjust(4) + " <source not found>" | ||
| 162 | else: | ||
| 163 | src_str = src_file + str(line_number).rjust(4) + " " + source_line | ||
| 164 | glb_dso = None | ||
| 165 | elif dso == glb_dso: | ||
| 166 | src_str = "" | ||
| 167 | else: | ||
| 168 | src_str = dso | ||
| 169 | glb_dso = dso | ||
| 170 | |||
| 171 | glb_line_number = line_number | ||
| 172 | glb_source_file_name = source_file_name | ||
| 173 | |||
| 174 | print(f"{start_str}{src_str}") | ||
| 175 | |||
| 176 | def process_event(param_dict): | ||
| 177 | global cache_size | ||
| 178 | global options | ||
| 179 | |||
| 180 | sample = param_dict["sample"] | ||
| 181 | comm = param_dict["comm"] | ||
| 182 | |||
| 183 | name = param_dict["ev_name"] | ||
| 184 | dso = get_optional(param_dict, "dso") | ||
| 185 | dso_bid = get_optional(param_dict, "dso_bid") | ||
| 186 | dso_start = get_optional(param_dict, "dso_map_start") | ||
| 187 | dso_end = get_optional(param_dict, "dso_map_end") | ||
| 188 | symbol = get_optional(param_dict, "symbol") | ||
| 189 | |||
| 190 | if (options.verbose == True): | ||
| 191 | print(f"Event type: {name}") | ||
| 192 | print_sample(sample) | ||
| 193 | |||
| 194 | # If cannot find dso so cannot dump assembler, bail out | ||
| 195 | if (dso == '[unknown]'): | ||
| 196 | return | ||
| 197 | |||
| 198 | # Validate dso start and end addresses | ||
| 199 | if ((dso_start == '[unknown]') or (dso_end == '[unknown]')): | ||
| 200 | print(f"Failed to find valid dso map for dso {dso}") | ||
| 201 | return | ||
| 202 | |||
| 203 | if (name[0:12] == "instructions"): | ||
| 204 | print_srccode(comm, param_dict, sample, symbol, dso) | ||
| 205 | return | ||
| 206 | |||
| 207 | # Don't proceed if this event is not a branch sample, . | ||
| 208 | if (name[0:8] != "branches"): | ||
| 209 | return | ||
| 210 | |||
| 211 | cpu = sample["cpu"] | ||
| 212 | ip = sample["ip"] | ||
| 213 | addr = sample["addr"] | ||
| 214 | |||
| 215 | # Initialize CPU data if it's empty, and directly return back | ||
| 216 | # if this is the first tracing event for this CPU. | ||
| 217 | if (cpu_data.get(str(cpu) + 'addr') == None): | ||
| 218 | cpu_data[str(cpu) + 'addr'] = addr | ||
| 219 | return | ||
| 220 | |||
| 221 | # The format for packet is: | ||
| 222 | # | ||
| 223 | # +------------+------------+------------+ | ||
| 224 | # sample_prev: | addr | ip | cpu | | ||
| 225 | # +------------+------------+------------+ | ||
| 226 | # sample_next: | addr | ip | cpu | | ||
| 227 | # +------------+------------+------------+ | ||
| 228 | # | ||
| 229 | # We need to combine the two continuous packets to get the instruction | ||
| 230 | # range for sample_prev::cpu: | ||
| 231 | # | ||
| 232 | # [ sample_prev::addr .. sample_next::ip ] | ||
| 233 | # | ||
| 234 | # For this purose, sample_prev::addr is stored into cpu_data structure | ||
| 235 | # and read back for 'start_addr' when the new packet comes, and we need | ||
| 236 | # to use sample_next::ip to calculate 'stop_addr', plusing extra 4 for | ||
| 237 | # 'stop_addr' is for the sake of objdump so the final assembler dump can | ||
| 238 | # include last instruction for sample_next::ip. | ||
| 239 | start_addr = cpu_data[str(cpu) + 'addr'] | ||
| 240 | stop_addr = ip + 4 | ||
| 241 | |||
| 242 | # Record for previous sample packet | ||
| 243 | cpu_data[str(cpu) + 'addr'] = addr | ||
| 244 | |||
| 245 | # Handle CS_ETM_TRACE_ON packet if start_addr=0 and stop_addr=4 | ||
| 246 | if (start_addr == 0 and stop_addr == 4): | ||
| 247 | print(f"CPU{cpu}: CS_ETM_TRACE_ON packet is inserted") | ||
| 248 | return | ||
| 249 | |||
| 250 | if (start_addr < int(dso_start) or start_addr > int(dso_end)): | ||
| 251 | print(f"Start address 0x{start_addr:x} is out of range [ 0x{dso_start:x} .. 0x{dso_end:x} ] for dso {dso}") | ||
| 252 | return | ||
| 253 | |||
| 254 | if (stop_addr < int(dso_start) or stop_addr > int(dso_end)): | ||
| 255 | print(f"Stop address 0x{stop_addr:x} is out of range [ 0x{dso_start:x} .. 0x{dso_end:x} ] for dso {dso}") | ||
| 256 | return | ||
| 257 | |||
| 258 | if (options.objdump_name != None): | ||
| 259 | # It doesn't need to decrease virtual memory offset for disassembly | ||
| 260 | # for kernel dso, so in this case we set vm_start to zero. | ||
| 261 | if (dso == "[kernel.kallsyms]"): | ||
| 262 | dso_vm_start = 0 | ||
| 263 | else: | ||
| 264 | dso_vm_start = int(dso_start) | ||
| 265 | |||
| 266 | dso_fname = get_dso_file_path(dso, dso_bid) | ||
| 267 | if path.exists(dso_fname): | ||
| 268 | print_disam(dso_fname, dso_vm_start, start_addr, stop_addr) | ||
| 269 | else: | ||
| 270 | print(f"Failed to find dso {dso} for address range [ 0x{start_addr:x} .. 0x{stop_addr:x} ]") | ||
| 271 | |||
| 272 | print_srccode(comm, param_dict, sample, symbol, dso) | ||