Re: [PATCH] dm-mpath: Don't grab work_mutex while probing paths

[Benjamin Marzinski <bmarzins@redhat.com>]

On Fri, May 16, 2025 at 01:35:39PM +0200, Mikulas Patocka wrote:
> 
> 
> On Thu, 15 May 2025, Benjamin Marzinski wrote:
> 
> > @@ -2077,35 +2095,55 @@ static int probe_path(struct pgpath *pgpath)
> >  static int probe_active_paths(struct multipath *m)
> >  {
> >  	struct pgpath *pgpath;
> > -	struct priority_group *pg;
> > +	struct priority_group *pg = NULL;
> >  	unsigned long flags;
> >  	int r = 0;
> >  
> > -	mutex_lock(&m->work_mutex);
> > -
> >  	spin_lock_irqsave(&m->lock, flags);
> 
> Hi
> 
> I suggest replacing spin_lock_irqsave/spin_unlock_irqrestore with 
> spin_lock_irq/spin_unlock_irq here and in some other places where it is 
> known that interrupts are enabled (for example __map_bio, 
> process_queued_bios, multipath_ctr, flush_multipath_work, 
> multipath_resume, multipath_status, multipath_prepare_ioctl, ...).
> 
> I accepted this patch, so you can send the spinlock changes in a follow-up 
> patch.

Sure. I can do that.

-Ben

> 
> Mikulas