xref: /linux/scripts/faddr2line (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
1#!/bin/bash
2# SPDX-License-Identifier: GPL-2.0
3#
4# Translate stack dump function offsets.
5#
6# addr2line doesn't work with KASLR addresses.  This works similarly to
7# addr2line, but instead takes the 'func+0x123' format as input:
8#
9#   $ ./scripts/faddr2line ~/k/vmlinux meminfo_proc_show+0x5/0x568
10#   meminfo_proc_show+0x5/0x568:
11#   meminfo_proc_show at fs/proc/meminfo.c:27
12#
13# If the address is part of an inlined function, the full inline call chain is
14# printed:
15#
16#   $ ./scripts/faddr2line ~/k/vmlinux native_write_msr+0x6/0x27
17#   native_write_msr+0x6/0x27:
18#   arch_static_branch at arch/x86/include/asm/msr.h:121
19#    (inlined by) static_key_false at include/linux/jump_label.h:125
20#    (inlined by) native_write_msr at arch/x86/include/asm/msr.h:125
21#
22# The function size after the '/' in the input is optional, but recommended.
23# It's used to help disambiguate any duplicate symbol names, which can occur
24# rarely.  If the size is omitted for a duplicate symbol then it's possible for
25# multiple code sites to be printed:
26#
27#   $ ./scripts/faddr2line ~/k/vmlinux raw_ioctl+0x5
28#   raw_ioctl+0x5/0x20:
29#   raw_ioctl at drivers/char/raw.c:122
30#
31#   raw_ioctl+0x5/0xb1:
32#   raw_ioctl at net/ipv4/raw.c:876
33#
34# Multiple addresses can be specified on a single command line:
35#
36#   $ ./scripts/faddr2line ~/k/vmlinux type_show+0x10/45 free_reserved_area+0x90
37#   type_show+0x10/0x2d:
38#   type_show at drivers/video/backlight/backlight.c:213
39#
40#   free_reserved_area+0x90/0x123:
41#   free_reserved_area at mm/page_alloc.c:6429 (discriminator 2)
42
43
44set -o errexit
45set -o nounset
46
47usage() {
48	echo "usage: faddr2line [--list] <object file> <func+offset> <func+offset>..." >&2
49	exit 1
50}
51
52warn() {
53	echo "$1" >&2
54}
55
56die() {
57	echo "ERROR: $1" >&2
58	exit 1
59}
60
61UTIL_SUFFIX=""
62if [[ "${LLVM:-}" == "" ]]; then
63	UTIL_PREFIX=${CROSS_COMPILE:-}
64else
65	UTIL_PREFIX=llvm-
66
67	if [[ "${LLVM}" == *"/" ]]; then
68		UTIL_PREFIX=${LLVM}${UTIL_PREFIX}
69	elif [[ "${LLVM}" == "-"* ]]; then
70		UTIL_SUFFIX=${LLVM}
71	fi
72fi
73
74READELF="${UTIL_PREFIX}readelf${UTIL_SUFFIX}"
75ADDR2LINE="${UTIL_PREFIX}addr2line${UTIL_SUFFIX}"
76AWK="awk"
77GREP="grep"
78
79command -v ${AWK} >/dev/null 2>&1 || die "${AWK} isn't installed"
80command -v ${READELF} >/dev/null 2>&1 || die "${READELF} isn't installed"
81command -v ${ADDR2LINE} >/dev/null 2>&1 || die "${ADDR2LINE} isn't installed"
82
83# Try to figure out the source directory prefix so we can remove it from the
84# addr2line output.  HACK ALERT: This assumes that start_kernel() is in
85# init/main.c!  This only works for vmlinux.  Otherwise it falls back to
86# printing the absolute path.
87find_dir_prefix() {
88	local start_kernel_addr=$(echo "${ELF_SYMS}" | sed 's/\[.*\]//' |
89		${AWK} '$8 == "start_kernel" {printf "0x%s", $2}')
90	[[ -z $start_kernel_addr ]] && return
91
92	run_addr2line ${start_kernel_addr} ""
93	[[ -z $ADDR2LINE_OUT ]] && return
94
95	local file_line=${ADDR2LINE_OUT#* at }
96	if [[ -z $file_line ]] || [[ $file_line = $ADDR2LINE_OUT ]]; then
97		return
98	fi
99	local prefix=${file_line%init/main.c:*}
100	if [[ -z $prefix ]] || [[ $prefix = $file_line ]]; then
101		return
102	fi
103
104	DIR_PREFIX=$prefix
105	return 0
106}
107
108run_readelf() {
109	local objfile=$1
110	local out=$(${READELF} --file-header --section-headers --symbols --wide $objfile)
111
112	# This assumes that readelf first prints the file header, then the section headers, then the symbols.
113	# Note: It seems that GNU readelf does not prefix section headers with the "There are X section headers"
114	# line when multiple options are given, so let's also match with the "Section Headers:" line.
115	ELF_FILEHEADER=$(echo "${out}" | sed -n '/There are [0-9]* section headers, starting at offset\|Section Headers:/q;p')
116	ELF_SECHEADERS=$(echo "${out}" | sed -n '/There are [0-9]* section headers, starting at offset\|Section Headers:/,$p' | sed -n '/Symbol table .* contains [0-9]* entries:/q;p')
117	ELF_SYMS=$(echo "${out}" | sed -n '/Symbol table .* contains [0-9]* entries:/,$p')
118}
119
120check_vmlinux() {
121	# vmlinux uses absolute addresses in the section table rather than
122	# section offsets.
123	IS_VMLINUX=0
124	local file_type=$(echo "${ELF_FILEHEADER}" |
125		${AWK} '$1 == "Type:" { print $2; exit }')
126	if [[ $file_type = "EXEC" ]] || [[ $file_type == "DYN" ]]; then
127		IS_VMLINUX=1
128	fi
129}
130
131init_addr2line() {
132	local objfile=$1
133
134	check_vmlinux
135
136	ADDR2LINE_ARGS="--functions --pretty-print --inlines --addresses --exe=$objfile"
137	if [[ $IS_VMLINUX = 1 ]]; then
138		# If the executable file is vmlinux, we don't pass section names to
139		# addr2line, so we can launch it now as a single long-running process.
140		coproc ADDR2LINE_PROC (${ADDR2LINE} ${ADDR2LINE_ARGS})
141	fi
142}
143
144run_addr2line() {
145	local addr=$1
146	local sec_name=$2
147
148	if [[ $IS_VMLINUX = 1 ]]; then
149		# We send to the addr2line process: (1) the address, then (2) a sentinel
150		# value, i.e., something that can't be interpreted as a valid address
151		# (i.e., ","). This causes addr2line to write out: (1) the answer for
152		# our address, then (2) either "?? ??:0" or "0x0...0: ..." (if
153		# using binutils' addr2line), or "," (if using LLVM's addr2line).
154		echo ${addr} >& "${ADDR2LINE_PROC[1]}"
155		echo "," >& "${ADDR2LINE_PROC[1]}"
156		local first_line
157		read -r first_line <& "${ADDR2LINE_PROC[0]}"
158		ADDR2LINE_OUT=$(echo "${first_line}" | sed 's/^0x[0-9a-fA-F]*: //')
159		while read -r line <& "${ADDR2LINE_PROC[0]}"; do
160			if [[ "$line" == "?? ??:0" ]] || [[ "$line" == "," ]] || [[ $(echo "$line" | ${GREP} "^0x00*: ") ]]; then
161				break
162			fi
163			ADDR2LINE_OUT+=$'\n'$(echo "$line" | sed 's/^0x[0-9a-fA-F]*: //')
164		done
165	else
166		# Run addr2line as a single invocation.
167		local sec_arg
168		[[ -z $sec_name ]] && sec_arg="" || sec_arg="--section=${sec_name}"
169		ADDR2LINE_OUT=$(${ADDR2LINE} ${ADDR2LINE_ARGS} ${sec_arg} ${addr} | sed 's/^0x[0-9a-fA-F]*: //')
170	fi
171}
172
173__faddr2line() {
174	local objfile=$1
175	local func_addr=$2
176	local dir_prefix=$3
177	local print_warnings=$4
178
179	local sym_name=${func_addr%+*}
180	local func_offset=${func_addr#*+}
181	func_offset=${func_offset%/*}
182	local user_size=
183	[[ $func_addr =~ "/" ]] && user_size=${func_addr#*/}
184
185	if [[ -z $sym_name ]] || [[ -z $func_offset ]] || [[ $sym_name = $func_addr ]]; then
186		warn "bad func+offset $func_addr"
187		DONE=1
188		return
189	fi
190
191	# Go through each of the object's symbols which match the func name.
192	# In rare cases there might be duplicates, in which case we print all
193	# matches.
194	while read line; do
195		local fields=($line)
196		local sym_addr=0x${fields[1]}
197		local sym_elf_size=${fields[2]}
198		local sym_sec=${fields[6]}
199		local sec_size
200		local sec_name
201
202		# Get the section size:
203		sec_size=$(echo "${ELF_SECHEADERS}" | sed 's/\[ /\[/' |
204			${AWK} -v sec=$sym_sec '$1 == "[" sec "]" { print "0x" $6; exit }')
205
206		if [[ -z $sec_size ]]; then
207			warn "bad section size: section: $sym_sec"
208			DONE=1
209			return
210		fi
211
212		# Get the section name:
213		sec_name=$(echo "${ELF_SECHEADERS}" | sed 's/\[ /\[/' |
214			${AWK} -v sec=$sym_sec '$1 == "[" sec "]" { print $2; exit }')
215
216		if [[ -z $sec_name ]]; then
217			warn "bad section name: section: $sym_sec"
218			DONE=1
219			return
220		fi
221
222		# Calculate the symbol size.
223		#
224		# Unfortunately we can't use the ELF size, because kallsyms
225		# also includes the padding bytes in its size calculation.  For
226		# kallsyms, the size calculation is the distance between the
227		# symbol and the next symbol in a sorted list.
228		local sym_size
229		local cur_sym_addr
230		local found=0
231		while read line; do
232			local fields=($line)
233			cur_sym_addr=0x${fields[1]}
234			local cur_sym_elf_size=${fields[2]}
235			local cur_sym_name=${fields[7]:-}
236
237			# is_mapping_symbol(cur_sym_name)
238			if [[ ${cur_sym_name} =~ ^(\.L|L0|\$) ]]; then
239				continue
240			fi
241
242			if [[ $cur_sym_addr = $sym_addr ]] &&
243			   [[ $cur_sym_elf_size = $sym_elf_size ]] &&
244			   [[ $cur_sym_name = $sym_name ]]; then
245				found=1
246				continue
247			fi
248
249			if [[ $found = 1 ]]; then
250				sym_size=$(($cur_sym_addr - $sym_addr))
251				[[ $sym_size -lt $sym_elf_size ]] && continue;
252				found=2
253				break
254			fi
255		done < <(echo "${ELF_SYMS}" | sed 's/\[.*\]//' | ${AWK} -v sec=$sym_sec '$7 == sec' | sort --key=2 | ${GREP} -A1 --no-group-separator " ${sym_name}$")
256
257		if [[ $found = 0 ]]; then
258			warn "can't find symbol: sym_name: $sym_name sym_sec: $sym_sec sym_addr: $sym_addr sym_elf_size: $sym_elf_size"
259			DONE=1
260			return
261		fi
262
263		# If nothing was found after the symbol, assume it's the last
264		# symbol in the section.
265		[[ $found = 1 ]] && sym_size=$(($sec_size - $sym_addr))
266
267		if [[ -z $sym_size ]] || [[ $sym_size -le 0 ]]; then
268			warn "bad symbol size: sym_addr: $sym_addr cur_sym_addr: $cur_sym_addr"
269			DONE=1
270			return
271		fi
272
273		sym_size=0x$(printf %x $sym_size)
274
275		# Calculate the address from user-supplied offset:
276		local addr=$(($sym_addr + $func_offset))
277		if [[ -z $addr ]] || [[ $addr = 0 ]]; then
278			warn "bad address: $sym_addr + $func_offset"
279			DONE=1
280			return
281		fi
282		addr=0x$(printf %x $addr)
283
284		# If the user provided a size, make sure it matches the symbol's size:
285		if [[ -n $user_size ]] && [[ $user_size -ne $sym_size ]]; then
286			[[ $print_warnings = 1 ]] &&
287				echo "skipping $sym_name address at $addr due to size mismatch ($user_size != $sym_size)"
288			continue;
289		fi
290
291		# Make sure the provided offset is within the symbol's range:
292		if [[ $func_offset -gt $sym_size ]]; then
293			[[ $print_warnings = 1 ]] &&
294				echo "skipping $sym_name address at $addr due to size mismatch ($func_offset > $sym_size)"
295			continue
296		fi
297
298		# In case of duplicates or multiple addresses specified on the
299		# cmdline, separate multiple entries with a blank line:
300		[[ $FIRST = 0 ]] && echo
301		FIRST=0
302
303		echo "$sym_name+$func_offset/$sym_size:"
304
305		# Pass section address to addr2line and strip absolute paths
306		# from the output:
307		run_addr2line $addr $sec_name
308		local output=$(echo "${ADDR2LINE_OUT}" | sed "s; $dir_prefix\(\./\)*; ;")
309		[[ -z $output ]] && continue
310
311		# Default output (non --list):
312		if [[ $LIST = 0 ]]; then
313			echo "$output" | while read -r line
314			do
315				echo $line
316			done
317			DONE=1;
318			continue
319		fi
320
321		# For --list, show each line with its corresponding source code:
322		echo "$output" | while read -r line
323		do
324			echo
325			echo $line
326			n=$(echo $line | sed 's/.*:\([0-9]\+\).*/\1/g')
327			n1=$[$n-5]
328			n2=$[$n+5]
329			f=$(echo $line | sed 's/.*at \(.\+\):.*/\1/g')
330			${AWK} 'NR>=strtonum("'$n1'") && NR<=strtonum("'$n2'") { if (NR=='$n') printf(">%d<", NR); else printf(" %d ", NR); printf("\t%s\n", $0)}' $f
331		done
332
333		DONE=1
334
335	done < <(echo "${ELF_SYMS}" | sed 's/\[.*\]//' | ${AWK} -v fn=$sym_name '$8 == fn')
336}
337
338[[ $# -lt 2 ]] && usage
339
340objfile=$1
341
342LIST=0
343[[ "$objfile" == "--list" ]] && LIST=1 && shift && objfile=$1
344
345[[ ! -f $objfile ]] && die "can't find objfile $objfile"
346shift
347
348run_readelf $objfile
349
350echo "${ELF_SECHEADERS}" | ${GREP} -q '\.debug_info' || die "CONFIG_DEBUG_INFO not enabled"
351
352init_addr2line $objfile
353
354DIR_PREFIX=supercalifragilisticexpialidocious
355find_dir_prefix
356
357FIRST=1
358while [[ $# -gt 0 ]]; do
359	func_addr=$1
360	shift
361
362	# print any matches found
363	DONE=0
364	__faddr2line $objfile $func_addr $DIR_PREFIX 0
365
366	# if no match was found, print warnings
367	if [[ $DONE = 0 ]]; then
368		__faddr2line $objfile $func_addr $DIR_PREFIX 1
369		warn "no match for $func_addr"
370	fi
371done
372