On 4/19/2019 12:53 AM, Masahiro Yamada wrote: > On Thu, Apr 11, 2019 at 11:33 PM Steve Sistare > <steven.sistare@xxxxxxxxxx> wrote: >> >> Perform module install actions only if the object area module is newer than >> install area module, and only announce the updated modules. This saves 10's >> of seconds for a single CPU make, saves a few seconds for a parallel make, >> and reduces output clutter for an incremental make. > > This will miss to detect the change of > CONFIG_MODULE_SIG_ALL, CONFIG_MODULE_COMPRESS. Thanks, I will fix those (and detect changes of INSTALL_MOD_STRIP). >> Example: >> % touch drivers/virtio/virtio_ring.c >> % make modules modules_install >> DESCEND objtool >> CALL scripts/atomic/check-atomics.sh >> CALL scripts/checksyscalls.sh >> CC [M] drivers/virtio/virtio_ring.o >> Building modules, stage 2. >> MODPOST 2592 modules >> LD [M] drivers/virtio/virtio_ring.ko >> INSTALL drivers/virtio/virtio_ring.ko >> DEPMOD 5.1.0-rc3 >> >> Signed-off-by: Steve Sistare <steven.sistare@xxxxxxxxxx> >> --- >> Makefile | 1 - >> scripts/Makefile.modinst | 30 +++++++++++++++++++++++------- >> 2 files changed, 23 insertions(+), 8 deletions(-) >> >> diff --git a/Makefile b/Makefile >> index 4988dcc..8f58e27 100644 >> --- a/Makefile >> +++ b/Makefile >> @@ -1286,7 +1286,6 @@ modules_install: _modinst_ _modinst_post >> >> PHONY += _modinst_ >> _modinst_: >> - @rm -rf $(MODLIB)/kernel >> @rm -f $(MODLIB)/source >> @mkdir -p $(MODLIB)/kernel >> @ln -s $(abspath $(srctree)) $(MODLIB)/source >> diff --git a/scripts/Makefile.modinst b/scripts/Makefile.modinst >> index 0dae402..062fdca6 100644 >> --- a/scripts/Makefile.modinst >> +++ b/scripts/Makefile.modinst >> @@ -10,6 +10,12 @@ include scripts/Kbuild.include >> >> # >> >> +ifeq ($(quiet),quiet_) >> + mod_announce_cmd := echo >> +else >> + mod_announce_cmd := true >> +endif > > This opts out of the Kbuild policy. Can you be more specific? I read the rules under Beautify output and verified that make -s, make V=0, make V=1 all work with no change in output format. However, I now see that make V=2 changes a bit. For the install_modules targets, the old code produced INSTALL drivers/virtio/virtio_ring.ko - due to target is PHONY but the new code prints only INSTALL drivers/virtio/virtio_ring.ko That does not seem like a big loss of information :) But I will work on it. Are you more generally concerned that $(quiet) is used here rather than being confined to the implementation of $(call cmd ...) ? - Steve >> __modules := $(sort $(shell grep -h '\.ko$$' /dev/null $(wildcard $(MODVERDIR)/*.mod))) >> modules := $(patsubst %.o,%.ko,$(wildcard $(__modules:.ko=.o))) >> >> @@ -17,14 +23,24 @@ PHONY += $(modules) >> __modinst: $(modules) >> @: >> >> -# Don't stop modules_install if we can't sign external modules. >> -quiet_cmd_modules_install = INSTALL $@ >> +# Perform install actions only if the tree module is newer than the installed >> +# module. Create the latter with a final move command to avoid a broken >> +# installed module if one of the intermediate commands fails. Suppress >> +# quiet_cmd so that only updated modules are announced. Don't stop >> +# modules_install if we can't sign external modules. >> + >> +quiet_cmd_modules_install = >> cmd_modules_install = \ >> - mkdir -p $(2) ; \ >> - cp $@ $(2) ; \ >> - $(mod_strip_cmd) $(2)/$(notdir $@) ; \ >> - $(mod_sign_cmd) $(2)/$(notdir $@) $(patsubst %,|| true,$(KBUILD_EXTMOD)) ; \ >> - $(mod_compress_cmd) $(2)/$(notdir $@) >> + if [[ $@ -nt $(2)/$(notdir $@) ]]; then \ >> + $(mod_announce_cmd) ' INSTALL' $@ ; \ >> + mkdir -p $(2) ; \ >> + tmpfile=$(2)/.$(notdir $@).tmp ; \ >> + cp $@ $$tmpfile ; \ >> + $(mod_strip_cmd) $$tmpfile ; \ >> + $(mod_sign_cmd) $$tmpfile $(patsubst %,|| true,$(KBUILD_EXTMOD)) ; \ >> + $(mod_compress_cmd) $$tmpfile ; \ >> + mv $$tmpfile $(2)/$(notdir $@) ; \ >> + fi >> >> # Modules built outside the kernel source tree go into extra by default >> INSTALL_MOD_DIR ?= extra >> -- >> 1.8.3.1 >> > >
