xref: /linux/scripts/faddr2line (revision 567f9c428f99560fe14e647def9f42f5344ebde9)
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
79# Enforce ASCII-only output from tools like readelf
80# ensuring sed processes strings correctly.
81export LANG=C
82
83command -v ${AWK} >/dev/null 2>&1 || die "${AWK} isn't installed"
84command -v ${READELF} >/dev/null 2>&1 || die "${READELF} isn't installed"
85command -v ${ADDR2LINE} >/dev/null 2>&1 || die "${ADDR2LINE} isn't installed"
86
87# Try to figure out the source directory prefix so we can remove it from the
88# addr2line output.  HACK ALERT: This assumes that start_kernel() is in
89# init/main.c!  This only works for vmlinux.  Otherwise it falls back to
90# printing the absolute path.
91find_dir_prefix() {
92	local start_kernel_addr=$(echo "${ELF_SYMS}" | sed 's/\[.*\]//' |
93		${AWK} '$8 == "start_kernel" {printf "0x%s", $2}')
94	[[ -z $start_kernel_addr ]] && return
95
96	run_addr2line ${start_kernel_addr} ""
97	[[ -z $ADDR2LINE_OUT ]] && return
98
99	local file_line=${ADDR2LINE_OUT#* at }
100	if [[ -z $file_line ]] || [[ $file_line = $ADDR2LINE_OUT ]]; then
101		return
102	fi
103	local prefix=${file_line%init/main.c:*}
104	if [[ -z $prefix ]] || [[ $prefix = $file_line ]]; then
105		return
106	fi
107
108	DIR_PREFIX=$prefix
109	return 0
110}
111
112run_readelf() {
113	local objfile=$1
114	local out=$(${READELF} --file-header --section-headers --symbols --wide $objfile)
115
116	# This assumes that readelf first prints the file header, then the section headers, then the symbols.
117	# Note: It seems that GNU readelf does not prefix section headers with the "There are X section headers"
118	# line when multiple options are given, so let's also match with the "Section Headers:" line.
119	ELF_FILEHEADER=$(echo "${out}" | sed -n '/There are [0-9]* section headers, starting at offset\|Section Headers:/q;p')
120	ELF_SECHEADERS=$(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')
121	ELF_SYMS=$(echo "${out}" | sed -n '/Symbol table .* contains [0-9]* entries:/,$p')
122}
123
124check_vmlinux() {
125	# vmlinux uses absolute addresses in the section table rather than
126	# section offsets.
127	IS_VMLINUX=0
128	local file_type=$(echo "${ELF_FILEHEADER}" |
129		${AWK} '$1 == "Type:" { print $2; exit }')
130	if [[ $file_type = "EXEC" ]] || [[ $file_type == "DYN" ]]; then
131		IS_VMLINUX=1
132	fi
133}
134
135init_addr2line() {
136	local objfile=$1
137
138	check_vmlinux
139
140	ADDR2LINE_ARGS="--functions --pretty-print --inlines --addresses --exe=$objfile"
141	if [[ $IS_VMLINUX = 1 ]]; then
142		# If the executable file is vmlinux, we don't pass section names to
143		# addr2line, so we can launch it now as a single long-running process.
144		coproc ADDR2LINE_PROC (${ADDR2LINE} ${ADDR2LINE_ARGS})
145	fi
146}
147
148run_addr2line() {
149	local addr=$1
150	local sec_name=$2
151
152	if [[ $IS_VMLINUX = 1 ]]; then
153		# We send to the addr2line process: (1) the address, then (2) a sentinel
154		# value, i.e., something that can't be interpreted as a valid address
155		# (i.e., ","). This causes addr2line to write out: (1) the answer for
156		# our address, then (2) either "?? ??:0" or "0x0...0: ..." (if
157		# using binutils' addr2line), or "," (if using LLVM's addr2line).
158		echo ${addr} >& "${ADDR2LINE_PROC[1]}"
159		echo "," >& "${ADDR2LINE_PROC[1]}"
160		local first_line
161		read -r first_line <& "${ADDR2LINE_PROC[0]}"
162		ADDR2LINE_OUT=$(echo "${first_line}" | sed 's/^0x[0-9a-fA-F]*: //')
163		while read -r line <& "${ADDR2LINE_PROC[0]}"; do
164			if [[ "$line" == "?? ??:0" ]] || [[ "$line" == "," ]] || [[ $(echo "$line" | ${GREP} "^0x00*: ") ]]; then
165				break
166			fi
167			ADDR2LINE_OUT+=$'\n'$(echo "$line" | sed 's/^0x[0-9a-fA-F]*: //')
168		done
169	else
170		# Run addr2line as a single invocation.
171		local sec_arg
172		[[ -z $sec_name ]] && sec_arg="" || sec_arg="--section=${sec_name}"
173		ADDR2LINE_OUT=$(${ADDR2LINE} ${ADDR2LINE_ARGS} ${sec_arg} ${addr} | sed 's/^0x[0-9a-fA-F]*: //')
174	fi
175}
176
177__faddr2line() {
178	local objfile=$1
179	local func_addr=$2
180	local dir_prefix=$3
181	local print_warnings=$4
182
183	local sym_name=${func_addr%+*}
184	local func_offset=${func_addr#*+}
185	func_offset=${func_offset%/*}
186	local user_size=
187	[[ $func_addr =~ "/" ]] && user_size=${func_addr#*/}
188
189	if [[ -z $sym_name ]] || [[ -z $func_offset ]] || [[ $sym_name = $func_addr ]]; then
190		warn "bad func+offset $func_addr"
191		DONE=1
192		return
193	fi
194
195	# Go through each of the object's symbols which match the func name.
196	# In rare cases there might be duplicates, in which case we print all
197	# matches.
198	while read line; do
199		local fields=($line)
200		local sym_addr=0x${fields[1]}
201		local sym_elf_size=${fields[2]}
202		local sym_sec=${fields[6]}
203		local sec_size
204		local sec_name
205
206		# Get the section size:
207		sec_size=$(echo "${ELF_SECHEADERS}" | sed 's/\[ /\[/' |
208			${AWK} -v sec=$sym_sec '$1 == "[" sec "]" { print "0x" $6; exit }')
209
210		if [[ -z $sec_size ]]; then
211			warn "bad section size: section: $sym_sec"
212			DONE=1
213			return
214		fi
215
216		# Get the section name:
217		sec_name=$(echo "${ELF_SECHEADERS}" | sed 's/\[ /\[/' |
218			${AWK} -v sec=$sym_sec '$1 == "[" sec "]" { print $2; exit }')
219
220		if [[ -z $sec_name ]]; then
221			warn "bad section name: section: $sym_sec"
222			DONE=1
223			return
224		fi
225
226		# Calculate the symbol size.
227		#
228		# Unfortunately we can't use the ELF size, because kallsyms
229		# also includes the padding bytes in its size calculation.  For
230		# kallsyms, the size calculation is the distance between the
231		# symbol and the next symbol in a sorted list.
232		local sym_size
233		local cur_sym_addr
234		local found=0
235		while read line; do
236			local fields=($line)
237			cur_sym_addr=0x${fields[1]}
238			local cur_sym_elf_size=${fields[2]}
239			local cur_sym_name=${fields[7]:-}
240
241			# is_mapping_symbol(cur_sym_name)
242			if [[ ${cur_sym_name} =~ ^(\.L|L0|\$) ]]; then
243				continue
244			fi
245
246			if [[ $cur_sym_addr = $sym_addr ]] &&
247			   [[ $cur_sym_elf_size = $sym_elf_size ]] &&
248			   [[ $cur_sym_name = $sym_name ]]; then
249				found=1
250				continue
251			fi
252
253			if [[ $found = 1 ]]; then
254				sym_size=$(($cur_sym_addr - $sym_addr))
255				[[ $sym_size -lt $sym_elf_size ]] && continue;
256				found=2
257				break
258			fi
259		done < <(echo "${ELF_SYMS}" | sed 's/\[.*\]//' | ${AWK} -v sec=$sym_sec '$7 == sec' | sort --key=2)
260
261		if [[ $found = 0 ]]; then
262			warn "can't find symbol: sym_name: $sym_name sym_sec: $sym_sec sym_addr: $sym_addr sym_elf_size: $sym_elf_size"
263			DONE=1
264			return
265		fi
266
267		# If nothing was found after the symbol, assume it's the last
268		# symbol in the section.
269		[[ $found = 1 ]] && sym_size=$(($sec_size - $sym_addr))
270
271		if [[ -z $sym_size ]] || [[ $sym_size -le 0 ]]; then
272			warn "bad symbol size: sym_addr: $sym_addr cur_sym_addr: $cur_sym_addr"
273			DONE=1
274			return
275		fi
276
277		sym_size=0x$(printf %x $sym_size)
278
279		# Calculate the address from user-supplied offset:
280		local addr=$(($sym_addr + $func_offset))
281		if [[ -z $addr ]] || [[ $addr = 0 ]]; then
282			warn "bad address: $sym_addr + $func_offset"
283			DONE=1
284			return
285		fi
286		addr=0x$(printf %x $addr)
287
288		# If the user provided a size, make sure it matches the symbol's size:
289		if [[ -n $user_size ]] && [[ $user_size -ne $sym_size ]]; then
290			[[ $print_warnings = 1 ]] &&
291				echo "skipping $sym_name address at $addr due to size mismatch ($user_size != $sym_size)"
292			continue;
293		fi
294
295		# Make sure the provided offset is within the symbol's range:
296		if [[ $func_offset -gt $sym_size ]]; then
297			[[ $print_warnings = 1 ]] &&
298				echo "skipping $sym_name address at $addr due to size mismatch ($func_offset > $sym_size)"
299			continue
300		fi
301
302		# In case of duplicates or multiple addresses specified on the
303		# cmdline, separate multiple entries with a blank line:
304		[[ $FIRST = 0 ]] && echo
305		FIRST=0
306
307		echo "$sym_name+$func_offset/$sym_size:"
308
309		# Pass section address to addr2line and strip absolute paths
310		# from the output:
311		run_addr2line $addr $sec_name
312		local output=$(echo "${ADDR2LINE_OUT}" | sed "s; $dir_prefix\(\./\)*; ;")
313		[[ -z $output ]] && continue
314
315		# Default output (non --list):
316		if [[ $LIST = 0 ]]; then
317			echo "$output" | while read -r line
318			do
319				echo $line
320			done
321			DONE=1;
322			continue
323		fi
324
325		# For --list, show each line with its corresponding source code:
326		echo "$output" | while read -r line
327		do
328			echo
329			echo $line
330			n=$(echo $line | sed 's/.*:\([0-9]\+\).*/\1/g')
331			n1=$[$n-5]
332			n2=$[$n+5]
333			f=$(echo $line | sed 's/.*at \(.\+\):.*/\1/g')
334			${AWK} 'NR>=strtonum("'$n1'") && NR<=strtonum("'$n2'") { if (NR=='$n') printf(">%d<", NR); else printf(" %d ", NR); printf("\t%s\n", $0)}' $f
335		done
336
337		DONE=1
338
339	done < <(echo "${ELF_SYMS}" | sed 's/\[.*\]//' | ${AWK} -v fn=$sym_name '$8 == fn')
340}
341
342[[ $# -lt 2 ]] && usage
343
344objfile=$1
345
346LIST=0
347[[ "$objfile" == "--list" ]] && LIST=1 && shift && objfile=$1
348
349[[ ! -f $objfile ]] && die "can't find objfile $objfile"
350shift
351
352run_readelf $objfile
353
354echo "${ELF_SECHEADERS}" | ${GREP} -q '\.debug_info' || die "CONFIG_DEBUG_INFO not enabled"
355
356init_addr2line $objfile
357
358DIR_PREFIX=supercalifragilisticexpialidocious
359find_dir_prefix
360
361FIRST=1
362while [[ $# -gt 0 ]]; do
363	func_addr=$1
364	shift
365
366	# print any matches found
367	DONE=0
368	__faddr2line $objfile $func_addr $DIR_PREFIX 0
369
370	# if no match was found, print warnings
371	if [[ $DONE = 0 ]]; then
372		__faddr2line $objfile $func_addr $DIR_PREFIX 1
373		warn "no match for $func_addr"
374	fi
375done
376