On 3/29/24 07:29, Bart Van Assche wrote: > On 3/27/24 5:43 PM, Damien Le Moal wrote: >> Allocating zone write plugs using kmalloc() does not guarantee that >> enough write plugs can be allocated to simultaneously write up to >> the maximum number of active zones or maximum number of open zones of >> a zoned block device. >> >> Avoid any issue with memory allocation by pre-allocating zone write >> plugs up to the disk maximum number of open zones or maximum number of >> active zones, whichever is larger. For zoned devices that do not have >> open or active zone limits, the default 128 is used as the number of >> write plugs to pre-allocate. >> >> Pre-allocated zone write plugs are managed using a free list. If a >> change to the device zone limits is detected, the disk free list is >> grown if needed when blk_revalidate_disk_zones() is executed. > > Is there a way to retry bio submission if allocating a zone write plug > fails? Would that make it possible to drop this patch? This patch is merged into the main zone write plugging patch in v4 (about to post it) and the free list is replaced with a mempool. Note that for BIOs that do not have REQ_NOWAIT, the allocation is done with GFP_NIO. If that fails, the OOM killer is probably already wreaking the system... > > Thanks, > > Bart. > -- Damien Le Moal Western Digital Research
