Re: [PATCH 3/4 v2] cgroup: separate rstat locks for subsystems

["Michal Koutný" <mkoutny@suse.com>]

[-- Attachment #1: Type: text/plain, Size: 1192 bytes --]

On Mon, Mar 10, 2025 at 10:59:30AM -0700, JP Kobryn <inwardvessel@gmail.com> wrote:
> > DEFINE_PER_CPU(raw_spinlock_t, cgroup_rstat_base_cpu_lock) =
> > 	__RAW_SPIN_LOCK_INITIALIZER(cgroup_rstat_base_cpu_lock);
> > 
> > (I see other places in kernel that assign into the per-cpu definition
> > but I have no idea whether that does expands and links to what's
> > expected. Neglecting the fact that the lock initializer is apparently
> > not for external use.)
> 
> I gave this a try. Using lockdep fields to verify, it expanded as
> intended:
> [    1.442498] [ss_rstat_init] cpu:0, lock.magic:dead4ead,
> lock.owner_cpu:-1, lock.owner:ffffffffffffffff
> [    1.443027] [ss_rstat_init] cpu:1, lock.magic:dead4ead,
> lock.owner_cpu:-1, lock.owner:ffffffffffffffff
> ...
> 
> Unless anyone has objections on using the double under macro, I will use
> this in v3.

Actually, I have the objection (to the underscored macro).
It may be work today but it may break subtly in the future.

Maybe add a separate patch that introduces a proper (non-underscore)
initializer (or percpu wrapped initializer) macro and people Cc'd on
that may evaluate wiseness of that.

Michal

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]