From: Brian Johannesmeyer <bjohannesmeyer@gmail.com>
To: Josh Poimboeuf <jpoimboe@kernel.org>,
linux-kernel@vger.kernel.org,
Brian Johannesmeyer <bjohannesmeyer@gmail.com>
Subject: [PATCH 5/7] scripts/faddr2line: Invoke addr2line as a single long-running process
Date: Mon, 11 Mar 2024 21:40:17 +0100 [thread overview]
Message-ID: <20240311204019.1183634-6-bjohannesmeyer@gmail.com> (raw)
In-Reply-To: <20240311204019.1183634-1-bjohannesmeyer@gmail.com>
Rather than invoking a separate addr2line process for each address, invoke
a single addr2line coprocess, and pass each address to its stdin. Previous
work [0] applied a similar change to perf, leading to a ~60x speed-up [1].
If using an object file that is _not_ vmlinux, faddr2line passes a section
name argument to addr2line. Because we do not know until runtime which
section names will be passed to addr2line, we cannot apply this change to
non-vmlinux object files. Hence, it only applies to vmlinux.
[0] commit be8ecc57f180 ("perf srcline: Use long-running addr2line per
DSO")
[1] Link:
https://eighty-twenty.org/2021/09/09/perf-addr2line-speed-improvement
Signed-off-by: Brian Johannesmeyer <bjohannesmeyer@gmail.com>
---
scripts/faddr2line | 52 +++++++++++++++++++++++++++++++++++++++-------
1 file changed, 45 insertions(+), 7 deletions(-)
diff --git a/scripts/faddr2line b/scripts/faddr2line
index 820680c59a39..48fc8cfc80df 100755
--- a/scripts/faddr2line
+++ b/scripts/faddr2line
@@ -126,6 +126,48 @@ check_vmlinux() {
fi
}
+init_addr2line() {
+ local objfile=$1
+
+ check_vmlinux
+
+ ADDR2LINE_ARGS="--functions --pretty-print --inlines --addresses --exe=$objfile"
+ if [[ $IS_VMLINUX = 1 ]]; then
+ # If the executable file is vmlinux, we don't pass section names to
+ # addr2line, so we can launch it now as a single long-running process.
+ coproc ADDR2LINE_PROC (${ADDR2LINE} ${ADDR2LINE_ARGS})
+ fi
+}
+
+run_addr2line() {
+ local addr=$1
+ local sec_name=$2
+
+ if [[ $IS_VMLINUX = 1 ]]; then
+ # We send to the addr2line process: (1) the address, then (2) a sentinel
+ # value, i.e., something that can't be interpreted as a valid address
+ # (i.e., ","). This causes addr2line to write out: (1) the answer for
+ # our address, then (2) either "?? ??:0" or "0x0...0: ..." (if
+ # using binutils' addr2line), or "," (if using LLVM's addr2line).
+ echo ${addr} >& "${ADDR2LINE_PROC[1]}"
+ echo "," >& "${ADDR2LINE_PROC[1]}"
+ local first_line
+ read -r first_line <& "${ADDR2LINE_PROC[0]}"
+ ADDR2LINE_OUT=$(echo "${first_line}" | sed 's/^0x[0-9a-fA-F]*: //')
+ while read -r line <& "${ADDR2LINE_PROC[0]}"; do
+ if [[ "$line" == "?? ??:0" ]] || [[ "$line" == "," ]] || [[ $(echo "$line" | ${GREP} "^0x00*: ") ]]; then
+ break
+ fi
+ ADDR2LINE_OUT+=$'\n'$(echo "$line" | sed 's/^0x[0-9a-fA-F]*: //')
+ done
+ else
+ # Run addr2line as a single invocation.
+ local sec_arg
+ [[ -z $sec_name ]] && sec_arg="" || sec_arg="--section=${sec_name}"
+ ADDR2LINE_OUT=$(${ADDR2LINE} ${ADDR2LINE_ARGS} ${sec_arg} ${addr} | sed 's/^0x[0-9a-fA-F]*: //')
+ fi
+}
+
__faddr2line() {
local objfile=$1
local func_addr=$2
@@ -260,12 +302,8 @@ __faddr2line() {
# Pass section address to addr2line and strip absolute paths
# from the output:
- local args="--functions --pretty-print --inlines --addresses --exe=$objfile"
- [[ $IS_VMLINUX = 0 ]] && args="$args --section=$sec_name"
- local output_with_addr=$(${ADDR2LINE} $args $addr | sed "s; $dir_prefix\(\./\)*; ;")
- [[ -z $output_with_addr ]] && continue
-
- local output=$(echo "${output_with_addr}" | sed 's/^0x[0-9a-fA-F]*: //')
+ run_addr2line $addr $sec_name
+ local output=$(echo "${ADDR2LINE_OUT}" | sed "s; $dir_prefix\(\./\)*; ;")
[[ -z $output ]] && continue
# Default output (non --list):
@@ -309,7 +347,7 @@ run_readelf $objfile
echo "${ELF_SECHEADERS}" | ${GREP} -q '\.debug_info' || die "CONFIG_DEBUG_INFO not enabled"
-check_vmlinux
+init_addr2line $objfile
DIR_PREFIX=supercalifragilisticexpialidocious
find_dir_prefix $objfile
--
2.34.1
next prev parent reply other threads:[~2024-03-11 20:40 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-03-11 20:40 [PATCH 0/7] Improve performance of 'faddr2line' Brian Johannesmeyer
2024-03-11 20:40 ` [PATCH 1/7] scripts/faddr2line: Reduce number of readelf calls to three Brian Johannesmeyer
2024-03-11 20:40 ` [PATCH 2/7] scripts/faddr2line: Combine three readelf calls into one Brian Johannesmeyer
2024-03-11 20:40 ` [PATCH 3/7] scripts/faddr2line: Check vmlinux only once Brian Johannesmeyer
2024-03-11 20:40 ` [PATCH 4/7] scripts/faddr2line: Pass --addresses argument to addr2line Brian Johannesmeyer
2024-03-11 20:40 ` Brian Johannesmeyer [this message]
2024-03-11 20:40 ` [PATCH 6/7] scripts/faddr2line: Remove call to addr2line from find_dir_prefix() Brian Johannesmeyer
2024-03-11 20:40 ` [PATCH 7/7] scripts/faddr2line: Check only two symbols when calculating symbol size Brian Johannesmeyer
2024-04-11 19:28 ` Brian Johannesmeyer
2024-04-12 20:27 ` Josh Poimboeuf
2024-03-12 23:06 ` [PATCH 0/7] Improve performance of 'faddr2line' Josh Poimboeuf
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20240311204019.1183634-6-bjohannesmeyer@gmail.com \
--to=bjohannesmeyer@gmail.com \
--cc=jpoimboe@kernel.org \
--cc=linux-kernel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®