scripts/decodecode: add the ability to find code sequence
PATCH 1/1 scripts/decodecode: add the ability to find code sequence
It adds a tool to search code sequence from vmlinux. If additional parameters (vmlinux and kernel build path) are provided to decodecode command, it will try to search the code sequence in the binary, and provide a code block surrounding the target.
An example:
$ ARCH=x86 PC=0xffffffff8141642f ./decodecode <vmlinux> <kernel build path> < test.oops
Code: 04 25 28 00 00 00 74 05 e8 54 85 c5 ff c9 c3 0f 1f 44 00 00 55 48 89 e5 e8 07 31 cb ff c7 05 27 3f ee 00 01 00 00 00 0f ae f8 <c6> 04 25 00 00 00 00 01 5
d c3 0f 1f 44 00 00 55 48 89 e5 fb 66 0f
All code
========
ffffffff81416405: 04 25 add $0x25,%al
...
Code starting with the faulting instruction
===========================================
ffffffff81416405: c6 04 25 00 00 00 00 movb $0x1,0x0
...
Finding code sequences from /home/cansun/src/vmlinux
Failed to find full Code segment, but did find a subsegment.
Surrounding code may have changed.
Warning: PC provided does not match calculated binary offset.
Found code string: 00 00 00 74 05 e8 54 85 c5 ff c9 c3 e8 __ __ __ __ 55 48 89 e5 e8 07 31 cb ff c7 05 27 3f ee 00 01 00 00 00 0f ae f8 c6 04 25 00 00 00 00 0
1 5d c3 e8 __ __ __ __ 55 48 89 e5
============================================================
ffffffff814163dd: 75 0c jne ffffffff814163eb <moom_callback+0x62>
pr_info("OOM request ignored. No task eligible\n");
ffffffff814163df: 48 c7 c7 45 e3 dc 81 mov $0xffffffff81dce345,%rdi
...
ffffffff81416425: 0f ae f8 sfence
*killer = 1;
ffffffff81416428:* c6 04 25 00 00 00 00 movb $0x1,0x0 <-- trapping instruction
ffffffff8141642f: 01
...--- scripts/decodecode | 143 ++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 140 insertions(+), 3 deletions(-)
scripts/decodecode +140 -3
@@ -7,9 +7,11 @@ # e.g., to decode an i386 oops on an x86_64 system, use: # AFLAGS=--32 decodecode < 386.oops # PC=hex - the PC (program counter) the oops points to +# To search code sequences from vmlinux: +# ARCH=x86 decodecode [vmlinux] [kernel build path] < oops.file cleanup() { - rm -f $T $T.s $T.o $T.oo $T.aa $T.dis + rm -f $T $T.s $T.o $T.oo $T.aa $T.ad $T.dis exit 1 }
@@ -51,8 +53,20 @@ if [ -z "$code" ]; then fi echo $code + +# Backup OBJDUMPFLAGS to dump vmlinux later +objdumpflagsbackup=$OBJDUMPFLAGS + +# Encode Code: to \xff\xff format to search in vmlinux +rawcode=`echo $code | sed -e 's/.*Code://; s/\s//g'` + code=`echo $code | sed -e 's/.*Code: //'` +# Get offset of faulting instruction +pcoffset=${rawcode%%<*} +pcoffset=$(( ${#pcoffset} / 2 )) +rawcode=`echo $rawcode | sed -e 's/[<>()]//g; s/.\{2\}/\\\x&/g'` + width=`expr index "$code" ' '` width=$((($width-1)/2)) case $width in
@@ -62,10 +76,11 @@ case $width in esac if [ -z "$ARCH" ]; then - case `uname -m` in + case `uname -m` in aarch64*) ARCH=arm64 ;; arm*) ARCH=arm ;; - esac + x86*) ARCH=x86 ;; + esac fi # Params: (tmp_file, pc_sub)
@@ -144,4 +159,126 @@ faultline=`echo "$faultline" | sed -e 's/\[/\\\[/g; s/\]/\\\]/g'` cat $T.oo | sed -e "${faultlinenum}s/^\([^:]*:\)\(.*\)/\1\*\2\t\t<-- trapping instruction/" echo cat $T.aa + +# If a vmlinux binary is provided, try to search code sequences from it +# It is an x86-specific tool. +if [ "$ARCH" = "x86" ] && ! [ -z "$1" ] && [ -f "$1" ]; then + if [ -z "$2" ] || ! [ -d "$2" ]; then + echo + echo "Usage:" + echo " $0 [vmlinux] [kernel build path]" + echo " To find code sequences from vmlinux" + exit 1 + fi + + echo > $T.ad + echo "Finding code sequences from $1" >> $T.ad + vmlinux=$1 + kernelpath=`readlink -f $2` + kernelpathdepth=`echo $kernelpath | grep -o '/' | wc -l` + minmatchlen=$(( 10 * 4 )) + partialmatch=0 + removefromstart=1 +else + exit +fi + +# __fentry__ (e8 + 4 bytes) is patched by nop ( 0f 1f 44 + 2 bytes ) at runtime +restoredcode="$(echo $rawcode | \ + sed -e 's/\\x0f\\x1f\\x44.\{8\}/\\xe8\.\{4\}/g')" +addressresult=`LANG=C grep -boP --binary --text $restoredcode $vmlinux | \ + LANG=C sed 's/:.*$//'` + +# Try to find binary sequence, while reducing the length of the pattern +while [ -z "$addressresult" ] && [ "${#restoredcode}" -gt "$minmatchlen" ]; do + partialmatch=1 + if [ "$removefromstart" -eq 1 ]; then + restoredcode=${restoredcode#????} + if [ "$pcoffset" -gt 0 ]; then + pcoffset=$(( pcoffset - 1 )) + fi + else + restoredcode=${restoredcode%????} + fi + removefromstart=$((1-removefromstart)) + addressresult=`LANG=C grep -boP --binary --text $restoredcode $vmlinux | \ + LANG=C sed 's/:.*$//' | tr '\n' ' '` +done + +if [ -z "$addressresult" ]; then + cat $T.ad + echo "No matching code sequences found." + echo "=================================" + + exit +fi + +minoffset= +maxoffset= +regex= +trappingregex= +# 0x200000 to cancel the offset in the vmlinux file +# 0xffffffff81000000 to add the offset into memory of the kernel +offsetadjust=`echo "ibase=16;FFFFFFFF81000000 - 200000" | bc` +for address in $addressresult; do + offset=$(( address + pcoffset )) + if [ -z "$minoffset" ]; then + minoffset=$offset + maxoffset=$offset + else + if [ "$minoffset" -gt "$offset" ]; then + minoffset=$offset + fi + if [ "$maxoffset" -lt "$offset" ]; then + maxoffset=$offset + fi + fi + + offset=`echo "obase=16;$offset + $offsetadjust" | bc` + fulloffset=$offset + + # Truncate LSB to match any nearby addresses + offset=`echo $offset | sed 's/.$//'` + if [ -z "$regex" ]; then + regex="($offset.:" + trappingregex="$fulloffset:" + else + regex="$regex|$offset.:" + trappingregex="$trappingregex|$fulloffset:" + fi +done + +regex="$regex)" +minoffset=`echo "obase=16;$minoffset + $offsetadjust - 4096" | bc` +maxoffset=`echo "obase=16;$maxoffset + $offsetadjust + 4096" | bc` +display= +if [ "$pcoffset" -eq -1 ]; then + # No current instruction found, so display the objdump from the beginning + # of the code block + display='-A 60' +else + # Display around the current instruction + display='-C 30' +fi + +OBJDUMPFLAGS="$objdumpflagsbackup -dS --start-address=0x$minoffset --stop-address=0x$maxoffset \ + --prefix=$kernelpath --prefix-strip=$kernelpathdepth $vmlinux" +if [ "$partialmatch" -eq 1 ]; then + echo "Failed to find full Code segment, but did find a subsegment." >> $T.ad + echo "Surrounding code may have changed." >> $T.ad +fi + +if [ "$PC" ]; then + ( echo $PC | grep -Eqi $regex ) || echo "Warning: PC provided does not match \ +calculated binary offset." >> $T.ad +fi + +restoredcode=`echo $restoredcode | sed 's/\\\\\x/ /g; s/.{4}/ __ __ __ __/g'` +echo "Found code string: $restoredcode" >> $T.ad +echo "============================================================" >> $T.ad + +${CROSS_COMPILE}objdump $OBJDUMPFLAGS | grep -Ei $display $regex >> $T.ad + +cat $T.ad | sed -e "s/^\(${trappingregex}\)\(.*\)/\1\*\2\t\t<-- trapping instruction/gI" + cleanup
Signature · 1 line
2.32.1 (Apple Git-133)
Quoted 4 lines · Can Sun
It adds a tool to search code sequence from vmlinux. If additional parameters (vmlinux and kernel build path) are provided to decodecode command, it will try to search the code sequence in the binary, and provide a code block surrounding the target.
The use case being?
objdump -d vmlinux and searching for the rIP is too much typing?
Besides, with the kernel being more and more runtime-patched and rewritten, the live code differs more and more from the actual compiled opcode bytes in vmlinux, as you noticed yourself with restoredcode below.
So maybe I'm missing some important use case but right now this looks pointless to me.
Thx.
Signature · 6 lines
Regards/Gruss,
Boris.
SUSE Software Solutions Germany GmbH
GF: Ivo Totev, Andrew Myers, Andrew McDonald, Martje Boudien Moerman
(HRB 36809, AG Nürnberg)