On Tue, 2017-12-12 at 09:57 +0100, Hannes Reinecke wrote: > From: Hannes Reinecke <hare@xxxxxxxx> > > When a device is probed asynchronously del_gendisk() might be called > before the async probing was run, causing del_gendisk() to crash > due to uninitialized sysfs objects. > > Signed-off-by: Hannes Reinecke <hare@xxxxxxxx> > --- > block/genhd.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/block/genhd.c b/block/genhd.c > index c2223f1..cc40d95 100644 > --- a/block/genhd.c > +++ b/block/genhd.c > @@ -697,6 +697,9 @@ void del_gendisk(struct gendisk *disk) > struct disk_part_iter piter; > struct hd_struct *part; > > + if (!(disk->flags & GENHD_FL_UP)) > + return; > + > blk_integrity_del(disk); > disk_del_events(disk); Hello Hannes, Thank you for having published your approach for increasing disk probing concurrency. Your approach looks interesting to me. However, I don't think that patches 1/4..3/4 are sufficient to avoid races between e.g. device_add_disk() and del_gendisk(). As far as I know no locks are held around the device_add_disk() and del_gendisk() calls. Does that mean that del_gendisk() can call e.g. blk_integrity_del() before device_add_disk() has called blk_integrity_add()? Bart.
