Re: [PATCH] kbuild/btf: Avoid relinking modules when only vmlinux changes

[Petr Pavlu <petr.pavlu@suse.com>]

On 4/7/26 1:30 PM, Petr Pavlu wrote:
> On 4/2/26 4:17 PM, Petr Pavlu wrote:
>> Commit 5f9ae91f7c0d ("kbuild: Build kernel module BTFs if BTF is enabled
>> and pahole supports it") in 2020 introduced CONFIG_DEBUG_INFO_BTF_MODULES
>> to enable generation of split BTF for kernel modules. This change required
>> the %.ko Makefile rule to additionally depend on vmlinux, which is used as
>> a base for deduplication. The regular ld_ko_o command executed by the rule
>> was then modified to be skipped if only vmlinux changes. This was done by
>> introducing a new if_changed_except command and updating the original call
>> to '+$(call if_changed_except,ld_ko_o,vmlinux)'.
>>
>> Later, commit 214c0eea43b2 ("kbuild: add $(objtree)/ prefix to some
>> in-kernel build artifacts") in 2024 updated the rule's reference to vmlinux
>> from 'vmlinux' to '$(objtree)/vmlinux'. This accidentally broke the
>> previous logic to skip relinking modules if only vmlinux changes. The issue
>> is that '$(objtree)' is typically '.' and GNU Make normalizes the resulting
>> prerequisite './vmlinux' to just 'vmlinux', while the exclusion logic
>> retains the raw './vmlinux'. As a result, if_changed_except doesn't
>> correctly filter out vmlinux. Consequently, with
>> CONFIG_DEBUG_INFO_BTF_MODULES=y, modules are relinked even if only vmlinux
>> changes.
>>
>> Additionally, commit 522397d05e7d ("resolve_btfids: Change in-place update
>> with raw binary output") in 2025 reworked the method for patching BTF data
>> into the resulting modules by using 'objcopy --add-section'. This command
>> fails if a section already exists.
>>
>> Fix the unnecessary relinking issue by also excluding the normalized form
>> 'vmlinux' when invoking ld_ko_o. Adjust embed_btf_data() to first use the
>> --remove-section option to remove the patched BTF section if it is already
>> present.
> 
> I noticed that sorting id+flags in BTF_SET8 by resolve_btfids doesn't
> seem to be idempotent, so this requires additional work.

It is possible to make sorting id+flags in BTF_SET8 by resolve_btfids
idempotent. One approach would be to also update the offsets (st_value)
of the __BTF_ID__* symbols so that they reflect the result after
sorting.

However, I don't think this is worth doing. Since this logic would be
relevant in specific cases when vmlinux changes and only the BTF data
needs to be regenerated, it would have limited usage and testing.
Importantly, always relinking the modules adds only about 6% to the
rebuild time of the modules target on my system when vmlinux is touched.
The work required for BTF and Makefile processing dominates this target.
When developing the kernel locally, it's common to use a custom config
with a limited amount of modules. In such a case, avoiding the relinking
of modules saves very little.

I plan to instead send a patch to replace the current condition that
invokes ld_ko_o from if_changed_except to if_changed, and remove the
if_changed_except logic.

-- Petr