On Tue, Jul 30, 2024 at 12:34:08PM -1000, Tejun Heo wrote:
> Hello, Matthew.
>
> On Tue, Jul 30, 2024 at 03:17:40PM -0700, Matthew Brost wrote:
> > +/**
> > + * wq_init_user_lockdep_map - init user lockdep map for workqueue
> > + * @wq: workqueue to init lockdep map for
> > + * @lockdep_map: lockdep map to use for workqueue
> > + *
> > + * Initialize workqueue with a user defined lockdep map. WQ_USER_OWNED_LOCKDEP
> > + * must be set for workqueue.
> > + */
> > +void wq_init_user_lockdep_map(struct workqueue_struct *wq,
> > + struct lockdep_map *lockdep_map)
> > +{
> > + if (WARN_ON_ONCE(!(wq->flags & WQ_USER_OWNED_LOCKDEP)))
> > + return;
> > +
> > + wq->lockdep_map = lockdep_map;
> > +}
> > +EXPORT_SYMBOL_GPL(wq_init_user_lockdep_map);
>
> Would it be possible to make it a one-piece interface - ie. add
> alloc_workqueue_lockdep_map() which takes an external lockdep map rather
> than splitting it over two calls?
>
I didn't want to change the export alloc_workqueue() arguments so I went
with this approach. Are you suggesting export a new function
alloc_workqueue_lockdep_map() which will share an internal
implementation with the existing alloc_workqueue() but passes in a
lockdep map? That could work.
Matt
> Thanks.
>
> --
> tejun
