move some tools/lkl/Makefile.autoconf logic to kconfig

[ddiss]

This is (currently) one extra change atop the #602 s390 fixes.
I think it makes sense to move the Makefile.autoconf logic closer to where it's used, so arch specific kernel default settings can be moved into arch/lkl/Kconfig . Kconfig can call external shell scripts, which should allow us to share some logic between arch and tools while avoiding the cat kernel.config >> ../../.config gymnastics.

The following changes since commit d2df129665a47f8a2ebedcee45fb6e8172ee8170:

  Merge pull request #605 from ddiss/windows_ci (2025-06-09 11:43:50 -0700)

are available in the Git repository at:

  https://github.com/ddiss/linux lkl_mv_autoconf_to_kconfig

for you to fetch changes up to c9f0472ec6a65fe619db0b4b953385cd5f80c873:

  lkl: drop explicit CONFIG_SHMEM=n setting (2025-06-12 19:56:13 +1000)

----------------------------------------------------------------
David Disseldorp (9):
      arch/lkl: fix Kconfig whitespace to match rest of kernel
      lkl: remove obsolete STACK_HASH_ORDER setting
      arch/lkl: set OUTPUT_FORMAT, 64BIT, CPU_BIG_ENDIAN defaults
      arch/lkl: use a common helper to determine OUTPUT_FORMAT
      lkl: set CONFIG_LKL_FUZZING=y in fuzzing_defconfig
      lkl: set host memcpy/set/move defaults in Kconfig
      lkl: move KASAN_SHADOW_OFFSET/_SIZE defaults to Kconfig
      lkl: drop LKL_MEMORY_START and LKL_TASK_UNMAPPED_BASE autoconf
      lkl: drop explicit CONFIG_SHMEM=n setting

Sumanth Korikkar (1):
      mm/shmem: inline shmem_is_huge() for disabled transparent hugepages

 arch/lkl/Kconfig                           | 174 +++++++++++++++--------------
 arch/lkl/configs/fuzzing_defconfig         |   1 +
 arch/lkl/scripts/cc-objdump-file-format.sh |   8 ++
 include/linux/shmem_fs.h                   |   9 ++
 mm/shmem.c                                 |   6 -
 tools/lkl/Makefile.autoconf                |  35 ------
 6 files changed, 107 insertions(+), 126 deletions(-)
 create mode 100755 arch/lkl/scripts/cc-objdump-file-format.sh

[ddiss]

the clang build is throwing

*** Default configuration is based on 'defconfig'
ld.lld: error: unknown argument '-print-output-format'
arch/lkl/Kconfig:42: syntax error
arch/lkl/Kconfig:41: invalid statement
arch/lkl/Kconfig:42: invalid statement

I need to revisit the logic at:

config OUTPUT_FORMAT
       string "Output format"
       default $(shell,$(LD) -r -print-output-format) if LD_IS_BFD
       default "elf64-x86-64" if LD_IS_LLD && $(success,test "$(CROSS_COMPILE)" = "x86_64-linux-gnu")

[ddiss]

...

I need to revisit the logic at:

config OUTPUT_FORMAT
       string "Output format"
       default $(shell,$(LD) -r -print-output-format) if LD_IS_BFD
       default "elf64-x86-64" if LD_IS_LLD && $(success,test "$(CROSS_COMPILE)" = "x86_64-linux-gnu")

I wonder whether we could use something like this instead:

$ echo 'void foo(void) {}'| gcc -x c - -c -o out && objdump -p out|awk '/file format/ {print $4}'
elf64-x86-64
$ echo 'void foo(void) {}'| clang -x c - -c -o out && llvm-objdump -p out|awk '/file format/ {print $4}'
elf64-x86-64

i.e. with $(CC) and $(OBJDUMP) it'd be common for GCC and LLVM. The existing behaviour of only allowing elf64-x86-64 with LLVM appears a bit broken. Either way this would come as change separate to the Makefile.autoconf -> Kconfig move.

[ddiss]

v2:

• use tabs in Kconfig like the rest of the kernel
• use a gcc/llvm common helper script to determine OUTPUT_FORMAT

[ddiss]

This is a scheduled Windows Server 2019 brownout. The Windows Server 2019 image will be removed on 2025-06-30.

It'd be nice to get Windows builds tested for this PR. Does anyone know how to upgrade the workflow to target a more recent (e.g. Server 2025) version?

[ddiss]

Windows 2022 build / tests are passing 🎉 . I'll keep the draft tag for now while I look at what other kernel .config autoconf options can be easily moved into arch/lkl/Kconfig. I also think it makes sense to first merge the s390 fixes via #602 .

[ddiss]

update:

• moved the cc-objdump-file-format.sh helper under the existing arch/lkl/scripts instead of creating a new arch/lkl/tools dir.
• moved CONFIG_LKL_FUZZING and CONFIG_LKL_HOST_MEM* default logic from Makefile.autoconf into arch/lkl/Kconfig

[ddiss]

update:

• rebase against master branch and reorder
• add drop LKL_MEMORY_START and LKL_TASK_UNMAPPED_BASE autoconf change

[ddiss]

update:

• cherry-pick mainline commit 1f73784 ("mm/shmem: inline shmem_is_huge() for disabled transparent hugepages")
• drop the explicit CONFIG_SHMEM=n setting from Makefile.autoconf, now that it builds successfully

The changes so here far are all minor and incremental. I'll drop the WIP flag for now.
I think my only real (minor) concern with the series is the multiple cc-objdump-file-format.sh calls, but I don't expect any significant built time increase.

[ddiss]

Windows CI failure appears unrelated to the changes here:

  TEST       *        nanosleep 
  TEST       not ok   nanosleep 
sleep 78570700 (ret=0), expected sleep 87654321

[tavip]

This seems reasonable and a good improvement IMO, thanks @ddiss for doing all this work! Lets wait a couple of days to get @rodionov a chance to take a look as well.

[rodionov]

Thank you for looping me in and doing this great clean up, folks! LGTM!