Re: [PATCH 1/2] kbuild: remove top-level built-in.a
Masahiro Yamada's on January 15, 2019 7:01 pm: > Hi Nicholas, > > > On Tue, Jan 15, 2019 at 5:07 PM Nicholas Piggin wrote: >> >> Masahiro Yamada's on January 14, 2019 1:27 pm: >> > The symbol table in the final archive is unneeded because it is passed >> > to the linker after the --whole-archive option. Every object file in >> > the archive is included in the link anyway. >> > >> > Pass thin archives from subdirectories directly to the linker, and >> > remove the final archiving step. >> >> This seems like a good improvement. As far as I remember, it was slower >> to do the final link without the index built. Maybe I was testing it >> in a revision before moving those files into --whole-archive? If there >> is no slowdown, then I have no objection. > > > Thanks for the review. > > > I measured the performance of scripts/link-vmlinux.sh with 'perf stat'. > (20 times for each) > > > > Before this patch: > > Performance counter stats for 'make vmlinux' (20 runs): > >5.266796537 seconds time elapsed > ( +- 0.12% ) > > > > After this patch: > > Performance counter stats for 'make vmlinux' (20 runs): > >4.945576395 seconds time elapsed > > > > > So, the performance is slightly better. > > Maybe, does omitting the 'ar' save some time? Wow yeah that's a good speedup and simplification to boot. Thanks for checking it. > Anyway, this commit is good. Yes, it's a good improvement. Thanks, Nick
Re: [PATCH 1/2] kbuild: remove top-level built-in.a
Hi Nicholas, On Tue, Jan 15, 2019 at 5:07 PM Nicholas Piggin wrote: > > Masahiro Yamada's on January 14, 2019 1:27 pm: > > The symbol table in the final archive is unneeded because it is passed > > to the linker after the --whole-archive option. Every object file in > > the archive is included in the link anyway. > > > > Pass thin archives from subdirectories directly to the linker, and > > remove the final archiving step. > > This seems like a good improvement. As far as I remember, it was slower > to do the final link without the index built. Maybe I was testing it > in a revision before moving those files into --whole-archive? If there > is no slowdown, then I have no objection. Thanks for the review. I measured the performance of scripts/link-vmlinux.sh with 'perf stat'. (20 times for each) Before this patch: Performance counter stats for 'make vmlinux' (20 runs): 5.266796537 seconds time elapsed ( +- 0.12% ) After this patch: Performance counter stats for 'make vmlinux' (20 runs): 4.945576395 seconds time elapsed So, the performance is slightly better. Maybe, does omitting the 'ar' save some time? Anyway, this commit is good. -- Best Regards Masahiro Yamada
Re: [PATCH 1/2] kbuild: remove top-level built-in.a
Masahiro Yamada's on January 14, 2019 1:27 pm: > The symbol table in the final archive is unneeded because it is passed > to the linker after the --whole-archive option. Every object file in > the archive is included in the link anyway. > > Pass thin archives from subdirectories directly to the linker, and > remove the final archiving step. This seems like a good improvement. As far as I remember, it was slower to do the final link without the index built. Maybe I was testing it in a revision before moving those files into --whole-archive? If there is no slowdown, then I have no objection. Acked-by: Nicholas Piggin
[PATCH 1/2] kbuild: remove top-level built-in.a
The symbol table in the final archive is unneeded because it is passed
to the linker after the --whole-archive option. Every object file in
the archive is included in the link anyway.
Pass thin archives from subdirectories directly to the linker, and
remove the final archiving step.
Signed-off-by: Masahiro Yamada
---
scripts/link-vmlinux.sh | 30 ++
1 file changed, 6 insertions(+), 24 deletions(-)
diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
index c8cf453..4788def 100755
--- a/scripts/link-vmlinux.sh
+++ b/scripts/link-vmlinux.sh
@@ -44,24 +44,6 @@ info()
fi
}
-# Thin archive build here makes a final archive with symbol table and indexes
-# from vmlinux objects INIT and MAIN, which can be used as input to linker.
-# KBUILD_VMLINUX_LIBS archives should already have symbol table and indexes
-# added.
-#
-# Traditional incremental style of link does not require this step
-#
-# built-in.a output file
-#
-archive_builtin()
-{
- info AR built-in.a
- rm -f built-in.a;
- ${AR} rcsTP${KBUILD_ARFLAGS} built-in.a \
- ${KBUILD_VMLINUX_INIT} \
- ${KBUILD_VMLINUX_MAIN}
-}
-
# Link of vmlinux.o used for section mismatch analysis
# ${1} output file
modpost_link()
@@ -69,7 +51,8 @@ modpost_link()
local objects
objects="--whole-archive\
- built-in.a \
+ ${KBUILD_VMLINUX_INIT} \
+ ${KBUILD_VMLINUX_MAIN} \
--no-whole-archive \
--start-group \
${KBUILD_VMLINUX_LIBS} \
@@ -88,7 +71,8 @@ vmlinux_link()
if [ "${SRCARCH}" != "um" ]; then
objects="--whole-archive\
- built-in.a \
+ ${KBUILD_VMLINUX_INIT} \
+ ${KBUILD_VMLINUX_MAIN} \
--no-whole-archive \
--start-group \
${KBUILD_VMLINUX_LIBS} \
@@ -99,7 +83,8 @@ vmlinux_link()
-T ${lds} ${objects}
else
objects="-Wl,--whole-archive\
- built-in.a \
+ ${KBUILD_VMLINUX_INIT} \
+ ${KBUILD_VMLINUX_MAIN} \
-Wl,--no-whole-archive \
-Wl,--start-group \
${KBUILD_VMLINUX_LIBS} \
@@ -160,7 +145,6 @@ cleanup()
rm -f .tmp_System.map
rm -f .tmp_kallsyms*
rm -f .tmp_vmlinux*
- rm -f built-in.a
rm -f System.map
rm -f vmlinux
rm -f vmlinux.o
@@ -217,8 +201,6 @@ fi;
# final build of init/
${MAKE} -f "${srctree}/scripts/Makefile.build" obj=init
-archive_builtin
-
#link vmlinux.o
info LD vmlinux.o
modpost_link vmlinux.o
--
2.7.4
