I am trying to learn linux kernel module building and kbuild by following https://www.tldp.org/LDP/lkmpg/2.6/lkmpg.pdf and reading GNU make manual.
Here is the Makefile of the first example, Hello-1, on The Linux Kernel Module Programming Guide:
obj-m += hello-1.o
all:
make -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules
clean:
make -C /lib/modules/$(shell uname -r)/build M=$(PWD) clean
AFAIK, obj-m should be read by kbuild. However according to GNU Make manual, I understand that obj-m shouldn't be exported.
Except by explicit request, make exports a variable only if it is
either defined in the environment initially or set on the command
line, and if its name consists only of letters, numbers, and
underscores. Some shells cannot cope with environment variable names
consisting of characters other than letters, numbers, and underscores.
https://www.gnu.org/software/make/manual/html_node/Variables_002fRecursion.html
obj-m is neither defined in the environment initially nor set on command line. So I expect that it shouldn't exported to recipe of target all. How does kbuild access obj-m?
Your makefile calls a submake and passes it the make variable M that points to the current directory. The submake is invoked with the -C option such that it is run as if it was invoked from the specified directory, that is the kernel source directory. It is thus the kernel build system with its own makefiles that is used. Thanks to the M variable the kernel makefiles know where they can find your makefile and include it with its obj-m definition.
Note: the makefile you show should probably be modified a bit with conditionals such that only the obj-m variable definition is visible from the Linux build system. Else, there is a risk of collision between the all and clean targets of your makefile and targets with the same names in the kernel makefiles. And, as noted by MadScientist, using make is not a good idea; $(MAKE) is preferable. You should probably use something like:
ifeq ($(KERNELRELEASE),)
all:
$(MAKE) -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules
clean:
$(MAKE) -C /lib/modules/$(shell uname -r)/build M=$(PWD) clean
else
obj-m += hello-1.o
endif
Related
This is the Makefile that I'm using for cross-buildig a kernel module.
export CROSS_COMPILE:=aarch64-none-linux-gnu-
export ARCH:=arm64
obj-m += chr_drv_ex1.o
export KDIR:=linux-source-5.4.0
#EXTRA_CFLAGS=-I../../qemu-5.1.0/hw/misc
#ccflags-y=-I../../qemu-5.1.0/hw/misc
all: test_chr_drv map_hugetlb test_ioctl_drv
make ARCH=$(ARCH) CROSS_COMPILE=$(CROSS_COMPILE) -C $(KDIR) M=$(PWD) modules
clean:
make ARCH=$(ARCH) CROSS_COMPILE=$(CROSS_COMPILE) -C $(KDIR) M=$(PWD) clean
rm -f test_chr_drv map_hugetlb test_ioctl_drv
%: %.c
$(CROSS_COMPILE)gcc $^ -o $#
Now in my chr_drv_ex1.c file, I want to include a header file placed in ../../qemu-5.1.0/hw/misc directory. What is the correct method to add this path? I saw this and tried setting EXTRA_CFLAGS and ccflags-y but none of them works(those are commented out above). Of course if I use #include "../../qemu-5.1.0/hw/misc/axpu_regs.h" in the chr_drv_ex1.c source I can compile it. But I want to use #include <axpu_regs.h>.
I changed the ccflags-y from
ccflags-y := -I../../qemu-5.1.0/hw/misc
to
ccflags-y := -I../../../qemu-5.1.0/hw/misc
So the include path should be specified as seen from the kernel make directory (where the Makefile for linux kernel is located. in this case linux-source-5.4.0 directory which is one step below from where I am now).
having this makefile:
# If KERNELRELEASE is defined, we've been invoked from the
# kernel build system and can use its language.
ifneq ($(KERNELRELEASE),)
obj-m := module.o
# Otherwise we were called directly from the command
# line; invoke the kernel build system.
else
KERNELDIR ?= /lib/modules/$(shell uname -r)/build
PWD := $(shell pwd)
default:
$(MAKE) -C $(KERNELDIR) M=$(PWD) modules
endif
I have just one question. This makefile is read 2 times. In the first time, it sets the $KERNELDIR and PWD - the else branch. On the second read - there comes me question - it only sets $obj-m, but how? By the time the 'second' read, the script should invoke Makefile in source directory (as called with -C options) and therefor could not set obj-m variable from within my current directory (the one before, it changed to source directory via -C). Or does the second invoking by $(MAKE) inherit variables?
Note the option passed to the sub-make: M=$(PWD)
This means the sub-make knows the working directory where your makefile exists (in its $(M) variable) and can then do something like include $(M)/Makefile to have your obj-m variable set.
I have a kernel module I'd like to build with any of make, make debug, make test, where the only difference between each one is a -D option to the compiler. This is essentially the same question as Creating a debug target in Linux 2.6 driver module makefile, but that one was marked as answered, and my question remains, after trying a few other things as well.
I've tried the deprecated EXTRA_CFLAGS option in my makefile:
debug:
$(MAKE) -C $(KDIR) M=$(PWD) EXTRA_CFLAGS="-DDEBUG" modules
as well as the newer ccflags-y option (doesn't seem to work even outside of the debug target):
ccflags-y := -DDEBUG
debug:
$(MAKE) -C $(KDIR) M=$(PWD) modules
and setting CFLAGS just before the debug target:
debug: CFLAGS_main.o=-DDEBUG
debug:
$(MAKE) -C $(KDIR) M=$(PWD) modules
but the only way I've found to accomplish what I want is with a separate build script:
#!/bin/sh
case "$1" in
debug)
make CFLAGS_main.o=-DDEBUG
;;
*)
make
;;
esac
Is there no way to do this directly in the makefile when building a kernel module??
You should be able to use your original version that used EXTRA_CFLAGS, but just replace EXTRA_CFLAGS with ccflags-y:
debug:
$(MAKE) -C $(KDIR) M=$(PWD) ccflags-y="-DDEBUG" modules
or replace it with CFLAGS_main.o to apply the CFLAGS to a single object:
debug:
$(MAKE) -C $(KDIR) M=$(PWD) CFLAGS_main.o="-DDEBUG" modules
EDIT
As mentioned by the OP Roger Dueck, setting variables on the make command line has a global effect. It overrides any setting of the same variables within the makefiles which may be undesirable, especially for a globally used variable such as ccflags-y. To avoid this, use your own makefile variable. In the "normal" part of the Makefile that invokes $(MAKE) on the "KBuild" part, change the debug: target to the following, using a custom variable of your choice (I used FOO_CFLAGS here):
debug:
$(MAKE) -C $(KDIR) M=$(PWD) FOO_CFLAGS="-DDEBUG" modules
In the "KBuild" part of the Makefile invoked by the above rule, use the following to append the custom CFLAGS from FOO_CFLAGS to ccflags-y:
ccflags-y += $(FOO_CFLAGS)
I'm writing a character driver to sit on top of a modified version of ahci in the source tree. I basically have something that looks like this:
topdir
|
|- Makfile
|
|- mod_ahci
| | - Makefile, codefiles
|
|- char_interface
| | - Makefile, codefiles
now, char_interface requires symbols from mod_ahci. I have the appropriate EXPORT_SYMBOL() macro use for the symbols I need to export. However, I'm having trouble getting the makefiles right to pick up the header file in mod_ahci from char_interface. My toplevel Makefile
ifneq ($(KERNELRELEASE),)
obj-y := mod_ahci/ char_interface/
else
KERNELDIR ?= /lib/modules/$(shell uname -r)/build
PWD := $(shell pwd)
default:
$(MAKE) -C $(KERNELDIR) M=$(PWD) modules
modules_install:
$(MAKE) -C $(KERNELDIR) M=$(PWD) modules_install
endif
The makefile for char_interface (because the other builds just fine)
ifneq ($(KERNELRELEASE),)
ccflags-y += -I../mod_ahci
obj-m := char_interface.o
else
KERNELDIR ?= /lib/modules/$(shell uname -r)/build
default:
$(MAKE) -C $(KERNELDIR) M=$(shell pwd) modules
modules_install:
$(MAKE) -C $(KERNELDIR) M=$(shell pwd) modules_install
endif
clean:
-sudo rmmod ahcip
-rm -f *.ko* *.mod.* *.o modules.order Modules.symvers
I have referenced various text files in the kernel documentation. For example, I'm referring to .../Documentation/kbuild/makefiles.txt right now as well as .../Documentation/kbuild/modules.txt. Whenever I build, I'm getting /home/captaink/devel/kmodtests/char_interface/char_interface.c:2:22: error: mod_ahci.h: No such file or directory. There is a file named mod_ahci.h in the directory ../mod_ahci. What am I getting wrong with my use of ccflags-y in the makefile for the char driver?
Thanks
After some digging, I found the answer to the problem. I was misunderstanding what was happening with the makefile's that I was viewing from LDD3 and the kernel documentation (which is, apparently, exactly where O'Reilly took their examples). The build system actually changes directories into /usr/src/kernels/$(uname -r)/build (or similar) because this is why my header file wasn't being found by the compiler.
I'm not saying this is an elegant way of doing this, but here's how I fixed it. The makefile in the top directory now looks like:
ifneq ($(KERNELRELEASE),)
obj-y := mod_ahci/ char_interface/
else
KERNELDIR ?= /lib/modules/$(shell uname -r)/build
PWD := $(shell pwd)
default:
$(MAKE) -C $(KERNELDIR) M=$(PWD) TOP_DIR=$(PWD) modules
modules_install:
$(MAKE) -C $(KERNELDIR) M=$(PWD) modules_install
endif
And the makefile in the subdirectory containing the char driver interface looks like:
ifneq ($(KERNELRELEASE),)
ccflags-y += -I$(TOP_DIR)/mod_ahci
obj-m := char_interface.o
else
KERNELDIR ?= /lib/modules/$(shell uname -r)/build
default:
$(MAKE) -C $(KERNELDIR) M=$(shell pwd) modules
modules_install:
$(MAKE) -C $(KERNELDIR) M=$(shell pwd) modules_install
endif
clean:
-sudo rmmod ahcip
-rm -f *.ko* *.mod.* *.o modules.order Modules.symvers
As you can tell, the makefiles have been copied extensively. The subdirectories probably don't need the "shared" makefile stuff in them because that's taken care of by the higher level makefile. Nevertheless, the modules now build and the character driver I have knows of the exported symbols I made in my modified ahci driver.
I hope this may help someone who's a neophyte, like myself, to the Linux kernel build world and Linux kernel drivers.
Eg: a common device module's Makefile
obj-m:=jc.o
default:
$(MAKE) -C /lib/modules/$(shell uname -r)/build M=$(shell pwd) modules
clean:
$(MAKE) -C /lib/modules/$(shell uname -r)/build M=$(shell pwd) modules clean
I consider if I can set CFLAGS to the file. When I change default section to
$(MAKE) -O2 -C /lib/modules/$(shell uname -r)/build M=$(shell pwd) modules
But it didn't work.
Any help? Thanks a lot.
-O2 would be an option to make (or $(MAKE), as you're using it) in what you tried. Obviously, the compiler (probably gcc) needs this flag, not make.
Kbuild understands a make variable named CFLAGS_modulename.o to add specific C flags when compiling this unit. In your case, your module object will be jc.o, so you can specify:
CFLAGS_jc.o := -O2
and it should work. Add V=1 to your $(MAKE) lines to get a verbose output and you should see -O2 when jc.c is being compiled.
You can find more about compiling modules in the official documentation.
You can also use
ccflags-y := -O2
This will be applied to all of the source files compiled for your module with the Makefile. This is indirectly documented in the link provided by eepp in Section 4.2