On 10/3/24 3:30 PM, Tejun Heo wrote: > On Thu, Oct 03, 2024 at 11:22:09AM -1000, Tejun Heo wrote: >> Yeah, that should be spin_lock_irq() for consistency but at the same time it >> doesn't look like anything is actually grabbing that lock (or blkcg->lock >> nesting outside of it) from an IRQ context, so no actual deadlock scenario >> exists and lockdep doesn't trigger. > > Oh, wait, it's not that. blkg_conf_prep() implies queue_lock, so the IRQ is > disabled around it and adding _irq will trigger lockdep. Ah makes sense, didn't realize it was nested under the queue lock. Then it does look like it's just that one spot. -- Jens Axboe
