$insmod helloworld module generates the error message "Invalid module format".
$dmesg outputs:
overflow in relocation type 10 val ffffffff88640070
'hello' likely not compiled with -mcmodel=kernel
The Makefile is a mix of tradition format (using (CC)) and module build system format "make -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules".
The system is 1.6.18-194.el5 x86_64. The same Makefile works fine when it is used in i386 machine.
Any idea of what to look into? Thanks.
#Makefile, mix of (CC) and kernel module build system
CFLAGS+=-D"KBUILD_STR(s)=\#s" -D"KBUILD_BASENAME=KBUILD_STR(hello)"
CFLAGS+=-D__KERNEL__ -DMODULE -I$(KERNEL_BUILD_DIR)/include
KERNEL_BUILD_DIR=/lib/modules/2.6.18-194.el5/build
TARGETNAME=hello
BUILD_ALT_DIR=linux
# The main target (note that both library and driver are .ko files
#
all: $(BUILD_ALT_DIR)/$(TARGETNAME).ko
$(BUILD_ALT_DIR)/_$(TARGETNAME).o: hello.o
#echo Linking objects to loadable module
#mkdir -p $(BUILD_ALT_DIR)
#echo $(CURDIR)/$#
#$(LD) -Map=$#.map -r -o $# $^
#echo " LD_D [$#]"
$(BUILD_ALT_DIR)/$(TARGETNAME).ko: $(BUILD_ALT_DIR)/_$(TARGETNAME).o
#rm -f $(BUILD_ALT_DIR)/$(TARGETNAME).o
#echo create Makefile
#$(SHELL) -c 'echo "obj-m := $(TARGETNAME).o" > $(BUILD_ALT_DIR)/Makefile'
#$(SHELL) -c 'echo "$(TARGETNAME)-objs := _$(TARGETNAME).o" >> $(BUILD_ALT_DIR)/Makefile'
#$(SHELL) -c 'echo ".PHONY: `pwd`/$(BUILD_ALT_DIR)/_$(TARGETNAME).o" >> $(BUILD_ALT_DIR)/Makefile'
#$(SHELL) -c 'cd $(BUILD_ALT_DIR); $(MAKE) -C $(KERNEL_BUILD_DIR) M=`pwd`'
#echo " KO_D [$#]"
$(BUILD_ALT_DIR)/%.o: %.c
#echo Compiling C source to object file:
#mkdir -p $(BUILD_ALT_DIR)
# #echo $(CURDIR)/$#
#$(CC) -c -Wall $(CFLAGS) $(CFLAGS) $< -o $#
#echo " CC_D [$#]"
clean:
rm -f $(BUILD_ALT_DIR)/*.o $(BUILD_ALT_DIR)/*.d $(BUILD_ALT_DIR)/core $(BUILD_ALT_DIR)/*.map
hello.c
#include <linux/autoconf.h> // this is needed
#include <linux/init.h>
#include <linux/module.h>
static int hello_init(void)
{
printk(KERN_ALERT "Hello, world\n");
return 0;
}
static void hello_exit(void)
{
printk(KERN_ALERT "Goodbye, cruel world\n");
}
module_init(hello_init);
module_exit(hello_exit);
The dmesg output tells you what is wrong:
'hello' likely not compiled with -mcmodel=kernel
On the x86-64 architecture, all code running in the kernel (including modules) must be compiled with a special flag, which tells the compiler to generate code which will run in the top half of the virtual address space (user mode programs run on the bottom half of the address space).
If I am reading your Makefile correctly, you are compiling the C code outside the kernel build system, and calling into the kernel build system only for the final linking. Do not do this. It is not just the memory model, there are several other flags which must be added when compiling the C source code. These flags can change with the kernel version, or even with the kernel configuration — you do not know, and you are not supposed to have to know, since the kernel build system deals with it all for you.
It is not just -mcmodel=kernel. There are many others, and getting them wrong can and will cause problems.
The fact that you are missing the correct flags is quite visible in your code:
#include <linux/autoconf.h> // this is needed
No, this is not needed. If you need it, you are doing it wrong. Take a look at one of the flags the kernel build system passes to the C compiler:
-include include/linux/autoconf.h
This flags tells the C compiler to always implicitly include the linux/autoconf.h header. Since it is always implicitly included, you never have to include it manually. And since you never include it manually, its location is allowed to change — it moved to generated/autoconf.h and later to linux/kconfig.h, and who knows where else it will end up next.
The fact that it worked at all for you in 32-bit x86 is just luck. You should do it the correct way even on 32-bit.
Second iteration:
All right, you've confirmed that Make builds _helloworld.o and $(BUILD_ALT_DIR)/Makefile. Now could you post the makefile that's in $(KERNEL_BUILD_DIR) (on the x86_64)?
(I'll go out on a limb and suggest that maybe $(BUILD_ALT_DIR) and $(KERNEL_BUILD_DIR) are the same on the i386 but not on the x84_64.)
EDIT: (third iteration)
1400 lines? Most of that in one branch? Half a dozen include directives? Architecture-specific branches? Autoconf?
Are you allowed to modify this nightmare in place, or must you tinker with the higher-level tools like automake? (I ask only because in the latter case it won't do much good to figure out what the problem is, you'll probably have to start over from scratch.)
1) From $(KERNEL_BUILD_DIR), try make M=$(BUILD_ALT_DIR) (whatever $(BUILD_ALT_DIR) is). Verify that that gives you the same error message. If it doesn't then ignore what follows (in this iteration).
2) From $(BUILD_ALT_DIR), try make, and see if it gives the same error. If it does, try replacing the makefile (in $(BUILD_ALT_DIR)) with a trivial makefile like
all:
#echo trivial makefile working
and test it, then try step 1 again. If it just says "trivial makefile working" and doesn't give the error, then things look good.
EDIT: (fourth iteration)
All right, we can't change $(KERNEL_BUILD_DIR)/Makefile. That doesn't necessarily sink us.
"Your suggestion (1) would not work, since $(BUILD_ALT_DIR) is a output directory, no source is in it."
Make does not require the presence of source code, and whether or not this would work (whatever "work" means), it is what your makefile appears to be attempting:
#$(SHELL) -c 'cd $(BUILD_ALT_DIR); $(MAKE) -C $(KERNEL_BUILD_DIR) M=`pwd`'
Try it and see what happens.
"Is there a possibility that x86_64 doesn't support tradition format (CC) for the kernel module?"
I'm not entirely sure what you mean, but we can probably answer your question by experiment.
Related
I want to create a Makefile which will build a program which when run's output will be used as a value for a variable in the same Makefile.
Consider the program:
print_name.c:
#include <stdio.h>
int main() {
printf("foo\n");
return 0;
}
I'd like to use the standard output of this program to then determine a directory name to be stored in a variable used in the makefile and used for other rules, so far I've tried:
all: $(MY_DIRECTORY_NAME)/my_program
print_name: print_name.c | print_name
gcc $^ -o $#
MY_DIRECTORY_NAME:=$(shell ./print_name)
$(MY_DIRECTORY_NAME)/my_program: my_program.c
mkdir -p $(MY_DIRECTORY_NAME)
gcc $^ -o $#
However when I run:
make all
I get:
make: *** No rule to make target '/my_program', needed by 'all'. Stop
Whereas I'd like:
mkdir -p foo
gcc my_program.c -o foo/my_program
So basically, I want to compile print_name before the assignment to MY_DIRECTORY_NAME, and then that variable in then used in the following rules. Is this possible, and if not are there any workarounds for this?
The basic problem is that you need to build (and run) print_name before you can parse the rules that use it -- but all rules are read and parsed before anything is built. So you need to run make twice -- first to build print_name and second to build everything that depends on it. Something like
all: print_name
$(MAKE) `./print_name`/my_program
should do the trick
In a simple embedded project, i have two files main.rs and module.rs. To build the project, I use something similar to this:
all: main.o
$(CC) main.o $(LDFLAGS)
%.o: %.rs
$(RUSTC) $(RUSTFLAGS) -o ${#} ${<}
If only module.rs is changed, make all won't recompile my Rust code. How can I fix this?
I'm posting a suboptimal self-answer as a first step, but would love to see better ways.
The best way to use Make is to encode every single dependency into the Makefile. That's what gives Make the power to know what to rebuild in order to reach a goal state.
To do this for a C project, you'll often use something like the GCC command line option -M. This brings the compiler into the mix as it's the best tool to parse C source code and understand the dependencies between the
files.
There is actually an equivalent switch for rustc, the Rust compiler: --emit=dep-info. When you run this on a source file, it will output a file with the extension .d, which contains an almost-Makefile-compatible list of dependencies. If you had a main.rs that referenced the module foo.rs, it would output something like:
main.d: main.rs foo.rs
With a bit of sed tweaking you can get this to play nicely. You can then include this in your Makefile:
main.o:
rustc -o $# $<
main.d: main.rs
rustc --emit=dep-info $<
# Add the object file as a rule
gsed 's/:/ $(#:.d=.o):/' -i $#
-include main.d
Here, I've specified main in a few parts, but I believe that you can easily modify them into pattern rules.
The pragmatic solution is to just use Cargo, the Rust build tool and package manager. Let it deal with dependencies (both local modules and other crates).
libbar.dylib: target/debug/libbar.dylib
cp $< $#
.PHONY: target/debug/libbar.dylib
target/debug/libbar.dylib:
cargo build --verbose
Here, I've marked the rule as PHONY, which says "always run this rule". I've added --verbose to have Cargo print out what it is doing so you can verify when things are rebuilt.
I'd recommend dropping off the cp step if you can and instead just use the nested path, but the copy might be needed if other things rely on the current location.
The pattern
%.o: %.rs
is familiar from building C projects, but that's not the only way a target can be written. Specific to the setup above, this would fix the situation:
main.o: main.rs module.rs
$(RUSTC) $(RUSTFLAGS) -o main.o main.rs
A noteworthy difference to the original code is that the names of the inputs is not really what matters for the command. We can generalize this as follows:
main.o: $(wildcard *.rs)
$(RUSTC) $(RUSTFLAGS) -o ${#} ${#:.o=.rs}
This is a start, but it still has some downsides I couldn't get rid of:
The main.o: part is hardcoded. If there are multiple top-level modules to compile, there would be code duplication
All Rust files will be considered for all top-level modules, due to the wildcard. In other words, changing any Rust file will require a full recompilation.
I am a newbie in Kernel Development. I was trying to understand the following makefile for Hello World! program. But I am not able to figure it out completely.
obj-m += hello.o
all:
sudo make -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules
clean:
sudo make -C /lib/modules/$(shell uname -r)/build M=$(PWD) clean
I am not able to understand what is meant by 'obj-m += hello.o' . I know m here means module and thats it.
Also why are we not defining the dependencies of hello.o
And lastly I am not able to figure out completely the compiling rules defined under all: and clean:
Any help would be highly appreciated.!!
obj-m is a Makefile variable. It actually consists of 2 parts: 'obj' means that the referred target is a kernel object, while 'm' part means that the object is to be build like a module.
The variable is considered by kernel build rules. As kernel modules follow a certain convention, running your Makefile will result in creation of module hello.ko from source file hello.c (if everything works properly).
The 'obj' variable may take different suffixes as well. For example 'obj-y' will try to link the referred object into the main kernel image, instead of creating a module. The suffix may also refer to a kernel .config file variable, like this:
obj-$(CONFIG_HOTPLUG) += hotplug.o
In this case, if CONFIG_HOTPLUG is set to 'y' the hoplug object will be compiled into the main kernel; if set to 'm' then a separate hotplug.ko loadable module will be created. If not set to anything (resulting in 'obj-'), hotplug will be omitted outright.
Ive got some large make files for a third party project that are not building due to linker issues.
From looking at the make files, I think it should be executing something like:
LIBS = -lm
CC = gcc
bin = bin
myapp: $(bin)/main.o $(bin)/other.o $(bin)/etc.o
$(CC) $(bin)/main.o $(bin)/other.o $(bin)/etc.o $(LIBS) -o myapp
gcc bin/main.o bin/other.o bin/etc.o -lm -o myapp
Instead from the error it seems to be failing on something like: It also didn't put any of the .o files in the expected bin/ location, but just left them in the source directory...
cc main.o -o myapp
But I cant locate anywhere that might come from. Is there some way to get some kind of stacktrace through the make files?
I am aware of -n and -d, but neither seems to tell me what target line and file yeilded that command, or which series of targets led there and the values of any $() expansions (The one im expecting is the only myapp: I can find in any of the makefiles...)
Check out the --debug option. From my manpage:
--debug[=FLAGS]
Print debugging information in addition to normal processing. If the
FLAGS are omitted, then the behavior is the same as if -d was specified.
FLAGS may be a for all debugging output (same as using -d), b for basic
debugging, v for more verbose basic debugging, i for showing implicit
rules, j for details on invocation of commands, and m for debugging
while remaking makefiles.
remake is a very good choice but in a pinch something like the following (saved as debug.mk) can be a good help too. It won't tell you as much as remake but it might tell you enough to start with.
# Use as: MAKEFILES=debug.mk make
OLD_SHELL := $(SHELL)
ifneq (undefined,$(origin X))
override X = -x
endif
SHELL = $(if $#,$(warning Running $#$(if $<, (from: $<))$(if $?, (newer: $?))))$(OLD_SHELL) $(X)
You can print out the other automatic variables there too if you wanted to see a bit more about what was going on.
Let's say you have a Makefile with two pseudo-targets, 'all' and 'debug'. The 'debug' target is meant to build the same project as 'all', except with some different compile switches (like -ggdb, for example). Since the targets use different compile switches, you obviously need to rebuild the entire project if you switch between the two. But GNUmake doesn't naturally recognize this.
So if you type make all you'll get
Building ...
...
Then if you type make debug, you get
make: Nothing to be done for `debug'.
So my question is: how do you implement a clean solution in the Makefile to notice that the last build used a different pseudo-target, or different compile switches, than the one you want currently? If they are different, the Makefile would rebuild everything.
Put the build products into different directory trees (whilst keeping one copy of the source of course). That way you are always just a short compile from an up-to-date build, be it debug or release (or even others). No possibility of confusion either.
EDIT
Sketch of the above.
src := 1.c 2.c 3.c
bare-objs := ${src:%.c=%.o}
release-objs := ${bare-objs:%=Release/%}
debug-objs := ${bare-objs:%=Debug/%}
Release/prog: ${release-objs}
Debug/prog: ${debug-objs}
${release-objs}: Release/%.o: %.c # You gotta lurve static pattern rules
gcc -c $< -o $#
${debug-objs}: Debug/%.o: %.c
gcc -c $< -o $#
Release/prog Debug/prog:
gcc $^ -o $#
.PHONY: all
all: Release/prog ; echo $# Success
.PHONY: debug
debug: Debug/prog ; echo $# Success
(Disclaimer: not tested, nor even run through make.)
There you go. It's even -j safe so you can do make -j5 all debug. There is a lot of obvious boiler plate just crying out for tidying up.
Keeping variant sets of object files (as in bobbogo's solution) is probably the best way, but if for some reason you don't want to do that, you can use empty files as markers, to indicate which way you last built the executable:
%-marker:
#rm -f $(OBJECTS) *-marker
#touch $#
debug: GCCFLAGS += -ggdb
debug: SOMEOTHERFLAG = WHATEVER
all debug: % : %-marker
#echo making $#
#$(MAKE) -S GCCFLAGS='$(GCCFLAGS)' SOMEOTHERFLAG='$(SOMEOTHERFLAG)' main
There are other variants on this idea; you could have a small file containing the flag settings, which the makefile would build and include. That would be clever, but not really any cleaner than this.
The only clean solution is to incorporate the difference into the target names.
E.g. you can define a variable $(DEBUG) and consistently use it in all targets that depend on the compile step.