On 10/29/20 5:04 AM, Damien Le Moal wrote: > Jens, > > Here is a couple of fixes for nullblk for this cycle. > > The first patch fixes tracing of zone condition in the case of a > REQ_OP_ZONE_RESET_ALL operation. > > The second patch fixes a more serious problem introduced with the recent > modification for protecting zone information using a spinlock. A > spinlock cannot be used when memory backing is turned on as that results > in potential memory allocations with the lock held and IRQ disabled. > The fix changes the locking method to using a bit, with the spinlock > retained only for function local protection. This new locking was > extensively tested with xfstests/btrfs and zonefs tests runs. Applied, thanks. -- Jens Axboe
