The swap_lock used by sbitmap has a chain with locks taken from softirq,
but the swap_lock is not protected from being preempted by softirqs.
A chain exists of:
sbq->ws[i].wait -> dispatch_wait_lock -> swap_lock
Where the sbq->ws[i].wait lock can be taken from softirq context, which
means all locks below it in the chain must also be protected from
softirqs.
Reported-by: Clark Williams <williams@redhat.com>
Fixes: cd80ddc212fa ("sbitmap: silence bogus lockdep IRQ warning")
Fixes: 5cd47079aec1 ("sbitmap: amortize cost of clearing bits")
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
static inline bool sbitmap_deferred_clear(struct sbitmap *sb, int index)
{
unsigned long mask, val;
- unsigned long __maybe_unused flags;
bool ret = false;
- /* Silence bogus lockdep warning */
-#if defined(CONFIG_LOCKDEP)
- local_irq_save(flags);
-#endif
- spin_lock(&sb->map[index].swap_lock);
+ spin_lock_bh(&sb->map[index].swap_lock);
if (!sb->map[index].cleared)
goto out_unlock;
ret = true;
out_unlock:
- spin_unlock(&sb->map[index].swap_lock);
-#if defined(CONFIG_LOCKDEP)
- local_irq_restore(flags);
-#endif
+ spin_unlock_bh(&sb->map[index].swap_lock);
return ret;
}