Cross compiling , yes but with `gcc` or `clang` ?

 I need to build a driver for a graphic card of the powerpc architecture.

 Build with `gcc`

The steps to do that seems more of less easy: add toolchain, run make with options to point new toolchain.

First: 

Here in opensuse with `zypper` we added  the `ppc64le` toolchain.
sudo zypper in cross-ppc64le-linux-glibc-devel cross-ppc64le-gcc15-bootstrap 

Second:

Clean previous configs with `mrproper` and then run `all` (maybe you need to run also `menuconfig`) 
make -j$(nproc) O=~/LF/cross/powerpc ARCH=powerpc HOSTCC=gcc CROSS_COMPILE=powerpc64le-suse-linux- mrproper
make ARCH=powerpc mrproper
make -j$(nproc) V=1 O=~/LF/cross/powerpc ARCH=powerpc HOSTCC=gcc CROSS_COMPILE=powerpc64le-suse-linux-  all
 

Log:

This is a line from the output, we see how the compiler name is composed with $CROSS_COMPILE plus the compiler name `gcc` in this case.
  powerpc64le-suse-linux-gcc -Wp,-MMD,lib/.iomem_copy.o.d -nostdinc -I/home/jg/LF/linux_work/linux_mainline/arch/powerpc/include -I./arch/powerpc/include/generated -I/home/jg/LF/linux_work/linux_mainline/include -I./include -I/home/jg/LF/linux_work/linux_mainline/arch/powerpc/include/uapi -I./arch/powerpc/include/generated/uapi -I/home/jg/LF/linux_work/linux_mainline/include/uapi -I./include/generated/uapi -include /home/jg/LF/linux_work/linux_mainline/include/linux/compiler-version.h -include /home/jg/LF/linux_work/linux_mainline/include/linux/kconfig.h -include /home/jg/LF/linux_work/linux_mainline/include/linux/compiler_types.h -D__KERNEL__ -mbig-endian -m32 -I /home/jg/LF/linux_work/linux_mainline/arch/powerpc -fmacro-prefix-map=/home/jg/LF/linux_work/linux_mainline/= -std=gnu11 -fshort-wchar -funsigned-char -fno-common -fno-PIE -fno-strict-aliasing -msoft-float -ffixed-r2 -mmultiple -mno-readonly-in-sdata -mcpu=powerpc -mno-prefixed -mno-pcrel -mno-altivec -mno-vsx -mno-mma -fno-asynchronous-unwind-tables -mno-string -mbig-endian -fno-delete-null-pointer-checks -O2 -fno-allow-store-data-races -fstack-protector-strong -fomit-frame-pointer -ftrivial-auto-var-init=zero -fzero-init-padding-bits=all -fno-stack-clash-protection -fmin-function-alignment=4 -fstrict-flex-arrays=3 -fno-strict-overflow -fno-stack-check -fconserve-stack -fno-builtin-wcslen -Wall -Wextra -Wundef -Werror=implicit-function-declaration -Werror=implicit-int -Werror=return-type -Werror=strict-prototypes -Wno-format-security -Wno-trigraphs -Wno-frame-address -Wno-address-of-packed-member -Wmissing-declarations -Wmissing-prototypes -Wframe-larger-than=1024 -Wno-main -Wno-dangling-pointer -Wvla-larger-than=1 -Wno-pointer-sign -Wcast-function-type -Wno-unterminated-string-initialization -Wno-array-bounds -Wno-stringop-overflow -Wno-alloc-size-larger-than -Wimplicit-fallthrough=5 -Werror=date-time -Werror=incompatible-pointer-types -Werror=designated-init -Wenum-conversion -Wunused -Wno-unused-but-set-variable -Wno-unused-const-variable -Wno-packed-not-aligned -Wno-format-overflow -Wno-format-truncation -Wno-stringop-truncation -Wno-override-init -Wno-missing-field-initializers -Wno-type-limits -Wno-shift-negative-value -Wno-maybe-uninitialized -Wno-sign-compare -Wno-unused-parameter -mstack-protector-guard=tls -mstack-protector-guard-reg=r2 -mstack-protector-guard-offset=776 -I/home/jg/LF/linux_work/linux_mainline/lib -Ilib    -DKBUILD_MODFILE='"lib/iomem_copy"' -DKBUILD_BASENAME='"iomem_copy"' -DKBUILD_MODNAME='"iomem_copy"' -D__KBUILD_MODNAME=kmod_iomem_copy -c -o lib/iomem_copy.o /home/jg/LF/linux_work/linux_mainline/lib/iomem_copy.c   ; ./tools/objtool/objtool --static-call   lib/iomem_copy.o  

 

 gcc or clang

But,but, but ... what if you've been playing also with `clang` and you have a leftover like `LLVM=0` in your environment which seems harmless and it seems to say "don't used LLVM (clang)" ;-)  .

That will trigger `clang` event if you have the `make` command with options that points to`gcc` toolchain like `CROSS_COMPILE` . 

In the case of `clang` there is no such different binaries for different architectures , there is only one `clang` binary.

So, if even you have not installed anything to support cross compile with clang  (because in clang the only you need is the  target <triple> option) you can cross compile.

The kernel utilities (makefile) is able to setup the needed options to build based in `CROSS_COMPILE` parameter. 

 


Comments

Post a Comment

Popular posts from this blog

I need a i386 kernel.

 Long time ago everything changed to 64bits for the most common architectures like x86 and arm.  We just don't remember i386 but sometimes we need to build an old 32bits .   What I need to build an i386 (how I did it) The toolchain for the i386 Tell the kernel build system that we want an architecture different from default (which now is 64bits)   That's it , in fact the linux kernel build systems is used to build several different architectures so this exercise is nothing for this beast.   I'll add the commands for my OpenSuse , for other distros it should be very similar sudo zypper in cross-i386-binutils   export ARCH=i386 ;time make LLVM=1 -j$(nproc) all CC='ccache clang'   You can check that the binary is one of 32bits architecture with the file command   $file arch/x86/boot/bzImage arch/x86/boot/bzImage: Linux kernel x86 boot executable, bzImage, version 6.16.0-rc1-dirty (jg@dell) #28 SMP PREEMPT_DYNAMIC Sat Oct ...
Read more

End of the Mentorship start of new journey

 The short Mentorship reaches its end , it used to be 6 months long but in this case it was a compressed 3 months one. The experience started with an initial filter based in a very good training available freely in the Linux Foundation site,  I think it worth mention because that Beginners' Guide to Linux Kernel  has everything you need to do your first commit. Even there are others resources in Linux Foundation that are not free, I think this is the only one you need to start.  Then started the real mentorship and the hurry to get your target of at least 5 accepted patches and that's a bit stressful:   first you start with  syzkaller   a great tool leverage from Google to run fuzzy testing. It's a really astonishing thing because give you the kernel and the C code to reproduce the issue ... or it should because not always you get the `reproducer` (how it's called that C code to reproduce the issue) . Check this blog entry for a really good guide on h...
Read more

My kernel is building ... after 1h `make` exits with no error

Image
 arghhhhh ... after so much time building all the sources the vmlinux is not created.   I did repeat the process after a make clean and an extra hour ... but again no vmlinux     The kernel build process has several steps like creating .o from .c and .h files but then objtool will do an static analysis of the generated object.   The problem comes at the end when the biggest vmlinux object must be analyzed with objtool and if options like  CONFIG_DEBUG_INFO the number DWARF data increases and eveything blows.   How much memory you'll need ... in my case with a 16GB laptop needed to add swap to allow generate finally the vmlinux  .   In the image you can see objtool with 79.93% of the memory , so aprox 12GB of ram     
Read more