Re: [RFC PATCH] kbuild: Make --build-id linker flag configurable

[Naman Jain <namjain@linux.microsoft.com>]

On 2/3/2026 12:31 PM, Thomas Weißschuh wrote:
> On Tue, Feb 03, 2026 at 11:58:11AM +0530, Naman Jain wrote:
>> On 2/2/2026 7:45 PM, Thomas Weißschuh wrote:
>>> Hi Naman,
>>>
>>> On Mon, Feb 02, 2026 at 11:06:31AM +0000, Naman Jain wrote:
>>>> Build ID hashes include file paths, so building the same source from
>>>> different directories produces different binaries. This breaks
>>>> reproducible builds.
>>>>
>>>> Add KBUILD_BUILD_ID variable (default: sha1) to allow overriding:
>>>>
>>>>       make KBUILD_BUILD_ID=none
>>>>
>>>> The variable is exported to VDSO Makefiles which also include a
>>>> fallback default for standalone invocation.
>>>>
>>>> Signed-off-by: Naman Jain <namjain@linux.microsoft.com>
>>>> ---
>>>> Hi,
>>>> Sending this change for RFC, as it is quite possible that this is a
>>>> generic problem and I may be missing something.
>>>>
>>>> I am trying to implement reproducible builds for one of my product
>>>> kernel. I referred https://reproducible-builds.org/docs/build-path/
>>>> and tried to use both -fdebug-prefix-map=OLD=NEW and
>>>> -fmacro-prefix-map=OLD=NEW, but still could not achieve bit by bit
>>>> binary reproducibility without overwriting build-id to none.
>>>> If I move the kernel to same path in other setup, I was able to create
>>>> same binary hash, however, without it, there is some difference in
>>>> build-id hash values.
>>>
>>
>> Hi Thomas,
>> Thanks for looking into this and sharing your inputs.
>>
>>
>>> Can you force the same build path during package building?
>>> That should avoid this issue.
>>
>> Since we can't control where the user would clone their kernel, I was
>> initially skeptical to copy the kernel to a same build path like
>> /tmp/kernel/src directory due to uncertainties related to free space,
>> permissions, but I tried it now and it works fine. It should be OK for my
>> use-case.
>>
>> I am currently using NixOS for reproducible build environment.
> 
> So users are already forced to use a specific distribution for rebuilding.
> Also requiring a specific build path doesn't look like a big step then.

Ack.

> 
>>>> Reproducibility wiki says "In most cases however, post-processing is
>>>> required to either remove the build path or to normalize it to a
>>>> predefined value.". I have tried that, and it works, but wanted to
>>>> conclude if that is my last option here.
>>>
>>> I am not a fan of this aproach. The build id should stay usable.
>>> Can you figure out where the build paths are used?
>>> You may need to also compare the debug symbols.
>>>
>>>> Thanks.
>>
>> I agree.
>> We did not have any use of these build paths, but some vendors may be using
>> it to fetch the build information from the binaries.
>> If your comment was about in-kernel usage of these build paths, I'll look
>> into it.
> 
> I'd like to know where the build paths in the binary are coming from.
> So we can fix the issue properly instead of working around it.
> You said you are using -fmacro-prefix-map and -fdebug-prefix-map to avoid them.
> (There is also -ffile-prefix-map which should be more robust and easy to use)

I was suspecting these are coming from linker scripts, in VDSO 
compilation but I could not pin-point to the real issue here. I will 
check more to find out what exactly is missing here.

> 
>>>> ---
>>>>    Makefile                          | 8 ++++++--
>>>>    arch/arm64/kernel/vdso/Makefile   | 5 ++++-
>>>>    arch/arm64/kernel/vdso32/Makefile | 5 ++++-
>>>>    arch/x86/entry/vdso/Makefile      | 5 ++++-
>>>>    4 files changed, 18 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/Makefile b/Makefile
>>>> index 3373308d2217c..3fcff4af200d7 100644
>>>> --- a/Makefile
>>>> +++ b/Makefile
>>>> @@ -1132,8 +1132,12 @@ KBUILD_AFLAGS   += $(KAFLAGS)
>>>>    KBUILD_CFLAGS   += $(KCFLAGS)
>>>>    KBUILD_RUSTFLAGS += $(KRUSTFLAGS)
>>>> -KBUILD_LDFLAGS_MODULE += --build-id=sha1
>>>> -LDFLAGS_vmlinux += --build-id=sha1
>>>> +# Can be overridden for reproducible builds by using "make KBUILD_BUILD_ID=none"
>>>> +KBUILD_BUILD_ID ?= sha1
>>>> +export KBUILD_BUILD_ID
>>>> +
>>>> +KBUILD_LDFLAGS_MODULE += --build-id=$(KBUILD_BUILD_ID)
>>>> +LDFLAGS_vmlinux += --build-id=$(KBUILD_BUILD_ID)
>>>>    KBUILD_LDFLAGS	+= -z noexecstack
>>>>    ifeq ($(CONFIG_LD_IS_BFD),y)
>>>> diff --git a/arch/arm64/kernel/vdso/Makefile b/arch/arm64/kernel/vdso/Makefile
>>>> index 7dec05dd33b70..b3ee5982b4676 100644
>>>> --- a/arch/arm64/kernel/vdso/Makefile
>>>> +++ b/arch/arm64/kernel/vdso/Makefile
>>>> @@ -9,6 +9,9 @@
>>>>    # Include the generic Makefile to check the built vdso.
>>>>    include $(srctree)/lib/vdso/Makefile.include
>>>> +# Fallback for standalone builds, normally inherited from top-level Makefile
>>>> +KBUILD_BUILD_ID ?= sha1
>>>> +
>>>
>>> What kind of standalone builds?
>>> This doesn't look like it belongs into this patch.
>>>
>>> (...)
>>
>> The case I was trying to cover here was when we try to compile
>> arch/x86/entry/vdso/ separately, without the KBUILD_BUILD_ID coming from
>> main build scripts, "--build-id=" would be left empty, while we may want to
>> retain original value i.e. sha1.
>>
>>      make ARCH=x86_64 arch/x86/entry/vdso/
> 
> I don't think this is or should be supported.

Ack.

> 
>> arch/x86/entry/vdso/Makefile:
>> -VDSO_LDFLAGS = -shared --hash-style=both --build-id=sha1 --no-undefined \
>> +VDSO_LDFLAGS = -shared --hash-style=both --build-id=$(KBUILD_BUILD_ID)
>> --no-undefined \
>>
>> Anyways, this may not be required now.
> 
> 
> Thomas

Regards,
Naman