[PATCH] kbuild: Avoid race generating / using MODLIB

[PATCH] kbuild: Avoid race generating / using MODLIB

[Doug Anderson]

The $(MODLIB) variable is recursively expanded and is:
  $(INSTALL_MOD_PATH)/lib/modules/$(KERNELRELEASE)

The $(KERNELRELEASE) variable is also recursively expanded:
  $(shell cat include/config/kernel.release 2> /dev/null)

The file being read is a generated by the Makefile itself.  When the
file is generated it is removed and then recreated.  This leaves a
period of time when the file doesn't exist.  If we happen to use
$(MODLIB) during this time then we'll end up with a bad value.

This is fairly easy for me to reproduce by:
* Add a "sleep 2" between the 'rm -f $@' and the 'echo' in the rule
  to build 'include/config/kernel.release'
* make -j32 ARCH=arm CROSS_COMPILE=armv7a-cros-linux-gnueabi- \
  zImage dtbs modules
* make -j32 ARCH=arm CROSS_COMPILE=armv7a-cros-linux-gnueabi- \
  INSTALL_PATH=foo INSTALL_MOD_PATH=foomod \
  install firmware_install modules_install

Without trying to rethink the whole KERNELRELEASE variable, the simple
fix is to just add a dependency.

Signed-off-by: Doug Anderson <dianders@chromium.org>
---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Makefile b/Makefile
index 0142c93..9cf2b24 100644
--- a/Makefile
+++ b/Makefile
@@ -964,7 +964,7 @@ PHONY += modules_install
 modules_install: _modinst_ _modinst_post
 
 PHONY += _modinst_
-_modinst_:
+_modinst_: include/config/kernel.release
 	@rm -rf $(MODLIB)/kernel
 	@rm -f $(MODLIB)/source
 	@mkdir -p $(MODLIB)/kernel
@@ -1232,7 +1232,7 @@ modules_install: _emodinst_ _emodinst_post
 
 install-dir := $(if $(INSTALL_MOD_DIR),$(INSTALL_MOD_DIR),extra)
 PHONY += _emodinst_
-_emodinst_:
+_emodinst_: include/config/kernel.release
 	$(Q)mkdir -p $(MODLIB)/$(install-dir)
 	$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modinst
 
-- 
1.8.3