Re: [PATCH v4 04/10] block: Modify revalidate zones

["hch@infradead.org" <hch@infradead.org>]

On Wed, Apr 08, 2020 at 08:29:13AM +0000, Damien Le Moal wrote:
> (Changed the subject to point to the correct patch)
> 
> Yes. Indeed we can do that. A flag will keep the interface of the report_zones
> method simpler.
> 
> But the second call to the revalidate callback done after the zone report is
> done is still needed so that the wp_ofst array can be updated under the queue
> freeze and avoid races with number of zones, bitmaps etc updates. I have not
> found any good way to avoid that one if we want to preserve
> blk_revalidate_disk_zones() as a generic helper. Of course if we reimplement
> this in sd_zbc, things are simple, but then we will have a lot of code repeated.

True, I missed the second call.  But given that it does something
entirely different multiplexing it to the same interface does indeed seem
rather odd.  So I think we can pass a 'finish' or so callback to
blk_revalidate_disk_zones.  I'm not even sure we need the args or a
flag. Ithink we can just make sure calls to blk_revalidate_disk_zones
are synchronized and then have a pending_wps pointer int the scsi_disk
structure.  If that is set, sd_zbc_report_zones updates it, and then
the finish callback puts it into the real place.