Re: [PATCH 10/10] rust: xarray: fix false positive lockdep warnings

[Andreas Hindborg <a.hindborg@kernel.org>]

"Gary Guo" <gary@garyguo.net> writes:

> On Wed Jan 21, 2026 at 7:01 PM GMT, Andreas Hindborg wrote:
>> "Gary Guo" <gary@garyguo.net> writes:
>>
>>> On Wed Dec 3, 2025 at 10:26 PM GMT, Andreas Hindborg wrote:
>>>> Replace the `xa_init_flags` helper with direct initialization of XArray
>>>> structures using `__spin_lock_init`. This allows each XArray to have
>>>> its own lock class key for lockdep, preventing false positive warnings
>>>> about lock ordering violations.
>>>>
>>>
>>> Isn't this potentially a problem on the C side as well? `xa_init_flags` is a
>>> static inline function, which means that the lock class is going to be the same
>>> if a single C compilation unit initializes multiple xarrays -- unlike when you
>>> use spin_lock_init, where each callsite gets a different lock class.
>>>
>>> Best,
>>> Gary
>>>
>>>> Add a `new_xarray!` macro that automatically generates a unique lock
>>>> class key for each XArray instantiation site. The macro accepts an
>>>> optional name parameter and uses the `optional_name!` and
>>>> `static_lock_class!` macros to generate appropriate names and lock
>>>> classes.
>>
>> My intuition about this was that when the C static function is inlined, a
>> new static address is used for each place the function is inlined. Is
>> this not correct?
>
> No, the static inside static functions still have the normal static semantics.
> There's a single copy regardless how many times the function is inlined.
>
> So you have a unique copy per compilation unit.

I see. A any rate I get into trouble with lockdep when I lock multiple
different xarrays concurrently in the same code gen unit. I am not sure
how they solve that in C, or if it will also be a problem there.


Best regards,
Andreas Hindborg