rpms/kernel-srpm-macros
7
0
Fork 0
Code Issues Pull Requests Releases Activity
715df6d241
kernel-srpm-macros/find-provides.ksyms
Eugene Syromiatnikov 148ee639b5 find-provides.ksyms, find-requires.ksyms: unify symbol parsing 
Since the immediate symbols can be retrieved for "objdump -t" output
as well, it makes some sence to use it as the sole data source and
handle both vaiants inside the awk script.  This simplification comes
at some runtime cost for the simpler case of the immediate symbols,
though, and overall it is about 5% faster in the relative symbols case
and about the same 5% slower in the absolute symbols case, with the
latter not as important moving forward, probably:

    $ for i in ./lib/modules/*; do \
        echo "====== $i ====="; \
        diff -u <(find $i | ./find-provides.ksyms.old) <(find $i | ./find-provides.ksyms.new); \
        echo -n "old: "; find $i | time ./find-provides.ksyms.old > /dev/null; \
        echo -n "new: "; find $i | time ./find-provides.ksyms.new > /dev/null; \
    done
    ====== ./lib/modules/4.18.0-372.57.1.el8_6.s390x =====
    old: ./find-provides.ksyms.old > /dev/null  3.98s user 3.04s system 129% cpu 5.411 total
    new: ./find-provides.ksyms.new > /dev/null  3.78s user 3.06s system 132% cpu 5.181 total
    ====== ./lib/modules/4.18.0-372.57.1.el8_6.x86_64 =====
    old: ./find-provides.ksyms.old > /dev/null  6.18s user 4.00s system 124% cpu 8.161 total
    new: ./find-provides.ksyms.new > /dev/null  6.57s user 4.84s system 132% cpu 8.644 total
    ====== ./lib/modules/5.14.0-284.15.1.el9_2.s390x =====
    old: ./find-provides.ksyms.old > /dev/null  4.70s user 2.94s system 126% cpu 6.061 total
    new: ./find-provides.ksyms.new > /dev/null  4.37s user 3.03s system 127% cpu 5.793 total
    ====== ./lib/modules/5.14.0-284.15.1.el9_2.x86_64 =====
    old: ./find-provides.ksyms.old > /dev/null  6.66s user 4.35s system 123% cpu 8.884 total
    new: ./find-provides.ksyms.new > /dev/null  7.07s user 5.00s system 130% cpu 9.218 total
    ====== ./lib/modules/6.4.0-0.rc1.20230511git80e62bc8487b.19.eln126.s390x =====
    old: ./find-provides.ksyms.old > /dev/null  3.81s user 2.62s system 128% cpu 5.018 total
    new: ./find-provides.ksyms.new > /dev/null  3.55s user 2.56s system 128% cpu 4.743 total
    ====== ./lib/modules/6.4.0-0.rc1.20230511git80e62bc8487b.19.eln126.x86_64 =====
    old: ./find-provides.ksyms.old > /dev/null  13.79s user 8.59s system 125% cpu 17.817 total
    new: ./find-provides.ksyms.new > /dev/null  13.18s user 8.78s system 127% cpu 17.247 total

* find-provides.ksyms: Do not perform an intial "nm | awk" run in an attempt
to capture absolute symbols, just parse "objdump -t" output and handle both
absolute and relative symbols in the awk script based on the section name.
* find-requires.ksyms (all_provides): Likewise.

Signed-off-by: Eugene Syromiatnikov <esyr@redhat.com>
2023-06-20 15:55:04 +02:00

139 lines
6.0 KiB
Bash
Executable File
Raw Blame History

#! /bin/bash
IFS=$'\n'
export LC_ALL=C
# Prevent elfutils from trying to download debuginfos
unset DEBUGINFOD_URLS
for module in $(grep -E '/lib/modules/.+\.ko(\.gz|\.bz2|\.xz|\.zst)?$') "$@"; do
dep_pfx="ksym"
# For built-in kmods, "kernel()" syntax is used instead of "ksym()"
printf "%s" "$module" | grep -v "^${RPM_BUILD_ROOT}/\?lib/modules/[1-9][^/]*/kernel" > /dev/null \
|| dep_pfx="kernel"
tmpfile=""
if [ "x${module%.ko}" = "x${module}" ]; then
tmpfile=$(mktemp -t ${0##*/}.XXXXXX.ko)
proc_bin=
case "${module##*.}" in
zst)
proc_bin=zstd
;;
xz)
proc_bin=xz
;;
bz2)
proc_bin=bzip2
;;
gz)
proc_bin=gzip
;;
esac
[ -n "$proc_bin" ] || continue
"$proc_bin" -d -c - < "$module" > "$tmpfile" || continue
module="$tmpfile"
fi
# A modversion can be stored as an ELF symbol in various ways:
# - An immediate symbol whose value is available directly; it shows up
# in the nm or objdump -t output, for example:
# $ nm mlx5_core_5.14.x86_64.ko | grep '__crc_' | head -n 3
# 0000000092f175ca A __crc_mlx5_access_reg
# 000000005b88c9f1 A __crc_mlx5_add_flow_rules
# 00000000e7c0ec8a A __crc_mlx5_alloc_bfreg
# $ objdump -t lib/modules/mlx5_core_5.14.x86_64.ko | grep __crc_ | sort -k 5,5 | head -n 3
# 0000000092f175ca g *ABS* 0000000000000000 __crc_mlx5_access_reg
# 000000005b88c9f1 g *ABS* 0000000000000000 __crc_mlx5_add_flow_rules
# 00000000e7c0ec8a g *ABS* 0000000000000000 __crc_mlx5_alloc_bfreg
# $ zgrep mlx5_access_reg ./lib/modules/5.14.0-284.15.1.el9_2.x86_64/symvers.gz
# 0x92f175ca mlx5_access_reg drivers/net/ethernet/mellanox/mlx5/core/mlx5_core EXPORT_SYMBOL_GPL
# This approach was being used on x86 and arm before Linux 5.19,
# for example.
#
# - A globally or locally visible symbol in a read-only (or not;
# sometimes .rodata is not a read-only section, after all, as binutils
# commit binutils-2_33~1385 has revealed (and binutils-2_35~1768 hasn't
# concealed back)) section (historically .rodata, __kcrctab/__kcrctab_gpl
# since Linux v5.19-rc1~139^2~2):
# $ nm mlx5_core_5.14.s390x.ko | grep '__crc_' | head -n 3
# 0000000000002f7c R __crc_mlx5_access_reg
# 0000000000003304 R __crc_mlx5_add_flow_rules
# 0000000000002d2c R __crc_mlx5_alloc_bfreg
# This layout is used on ppc since Linux v4.10-rc7~15^2~1, for example,
# and on all architectures since Linux 5.19. To extract the symbol
# versions in this case, we get the offset and the section name
# from the "objdump -t" output:
# $ objdump -t lib/modules/mlx5_core_5.14.s390x.ko | grep '__crc_' | sort -k 5,5 | head -n 2
# 0000000000002f7c g .rodata 0000000000000000 __crc_mlx5_access_reg
# 0000000000003304 g .rodata 0000000000000000 __crc_mlx5_add_flow_rules
# and the section contents from the "readelf -R" call:
# $ readelf -R .rodata mlx5_core_5.14.s390x.ko
# [... skipped output ...]
# 0x00002f70 6c6f635f 6e6f6465 00000000 ed6560a8 loc_node.....e`.
# ^^^^^^^^
# comparison with the contents
# of lib/modules/5.14.0-284.15.1.el9_2.s390x/symvers.gz corroborates
# its correctness:
# 0xed6560a8 mlx5_access_reg drivers/net/ethernet/mellanox/mlx5/core/mlx5_core EXPORT_SYMBOL_GPL
# As mentioned earlier, for the later kernel versions, __kcrctab{,_gpl}
# sections are used:
# $ objdump -t lib/modules/mlx5_core_6.4.x86_64.ko | grep '__crc_' | sort -k 5,5 | head -n 2
# 0000000000000000 l __kcrctab_gpl 0000000000000000 __crc_mlx5_access_reg
# 0000000000000090 l __kcrctab 0000000000000000 __crc_mlx5_add_flow_rules
# $ readelf -R __kcrctab_gpl mlx5_core_6.4.x86_64.ko
# 0x00000000 38b0d3c3 1840ce35 b99babc7 70b4700c 8....@.5....p.p.
# ^^^^^^^^
# and in lib/modules/6.4.0-0.rc1.20230511git80e62bc8487b.19.eln126.x86_64/symvers.xz
# we see that it is correct (when accounted for the little endianness):
# 0xc3d3b038 mlx5_access_reg drivers/net/ethernet/mellanox/mlx5/core/mlx5_core EXPORT_SYMBOL_GPL
# This data, after some post-processing, can be used in the awk script
# that extracts parts of the section according to the offsets got
# from the "objdump -t" output.
objdump -t "$module" \
| awk \
-v 'dep_pfx='"$dep_pfx" \
-v 'module='"$module" \
--non-decimal-data \
'BEGIN { revbytes = 0 }
function check_endianness( t) {
if (revbytes) return revbytes;
revbytes = -1;
while (("readelf -h \"" module "\"" | getline t) > 0) {
if (match(t, /^ Data: *2\047s complement, little endian$/)) {
revbytes = 1;
break;
}
}
return revbytes;
}
function readsect(name, a, t) {
a = "";
while (("readelf -R \"" name "\" \"" module "\"" | getline t) > 0) {
if (match(t, /^ 0x[0-9a-f]{8}/))
a = a substr(t, 14, 8) substr(t, 23, 8) substr(t, 32, 8) substr(t, 41, 8);
}
if (check_endianness() == 1)
a = gensub(/(..)(..)(..)(..)/, "\\4\\3\\2\\1", "g", a);
sectdata[name] = a;
}
match($0, /^([0-9a-f]+) [gl]...... (.*) [0-9a-f]+ __crc_(.*)$/, a) {
if (a[2] == "*ABS*") {
printf("%s(%s) = 0x%08x\n", dep_pfx, a[3], strtonum("0x" a[1]));
} else {
if (!(a[2] in sectdata)) { readsect(a[2]) }
printf("%s(%s) = 0x%08s\n", dep_pfx, a[3], substr(sectdata[a[2]], (strtonum("0x" a[1]) * 2) + 1, 8))
}
}'
[ -z "$tmpfile" ] || rm -f -- "$tmpfile"
done \
| sort -k1,1 -u
Reference in New Issue View Git Blame Copy Permalink