Re: [PATCH v2 13/21] modpost: emit module descriptors as assembly

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

On 9/21/26 2:48 PM, Lorenzo Stoakes (ARM) wrote:
> On Mon, Sep 21, 2026 at 02:36:30PM +0200, Petr Pavlu wrote:
>> On 9/21/26 1:57 PM, Lorenzo Stoakes (ARM) wrote:
>>> (note there's a v3 see [0])
>>>
>>> [0]: https://lore.kernel.org/linux-kbuild/20260917-build-speedup-v3-0-9ecf4163ff36@kernel.org/
>>
>> Thanks, I missed it.
>>
>>> On Mon, Sep 21, 2026 at 01:25:09PM +0200, Petr Pavlu wrote:
>>>> On 9/14/26 11:22 AM, Lorenzo Stoakes (ARM) wrote:
>>>>> The sections were confirmed to be byte-for-byte identical to the C version
>>>>> produced - each of .modinfo, .gnu.linkonce.this_module, __ksymtab*,
>>>>> __ksymtab_strings, __kcrctab*, __kflagstab*, __versions,
>>>>> __version_ext_crcs, __version_ext_names and their relocations - for all
>>>>> 8,135 modules of a clang allmodconfig build with CONFIG_COMPILE_TEST off
>>>>> and CONFIG_MODVERSIONS, CONFIG_EXTENDED_MODVERSIONS and
>>>>> CONFIG_MODULE_SRCVERSION_ALL on, and for a sample built with gcc.
>>>>
>>>> Lowering the data to assembly loses debug information for __this_module,
>>>> specifically that it has type `struct module`. It might be worth
>>>
>>> No, it keeps it - every module includes linux/module.h, and the extern
>>> declaration of __this_module carries the type, so all of the stuff gdb
>>> needs (e.g. ptype, p & field access), DWARF etc. is all there in the same
>>> way as with .mod.c.
>>>
>>> Have confirmed locally.
>>
>> An extern declaration alone shouldn't cause its name and type to be
>> included in debug information. They should appear only when the variable
>> is referenced by some code.
> 
> Ack yeah sorry my mistake, I had the LLM check that and should have known
> better than to trust it without reviewing what it said (as I have done for
> the most part for everything in this series).
> 
>>
>> Testing with a simple module that previously referenced __this_module
>> only in .mod.c and in no other file, such as lib/test_module.c, shows
>> that the .ko no longer contains debug information for this variable.
>>
>> I've verified this behavior with GCC 16 and Clang 19.
> 
> Ack confirmed locally.
> 
> Can put a fix in module-common.c.
> 
> The LLM came up with a reference as an alternative:
> 
> static struct module *const __this_module_dbg __used __section(".discard.this_module") = &__this_module;
> 
> Would that work? Or should we go with your original suggestion? -
> 
> __weak __section(".discard.this_module") struct module __this_module;

With GCC, the LLM's suggestion produces debug information for both
__this_module_dbg and __this_module. However, Clang only generates debug
information for __this_module_dbg, so it is not quite what is needed.

-- 
Thanks,
Petr