>> Either of the changes should not have led to a bug in zoned devices: >> >> - blk_execute_rq_nowait: >> Only passthrough requests can use this function, and plugging can be >> performed on those requests in zoned devices. So no issues directly >> accessing the plug. >> >> - blk_flush_plug in bio_poll: >> As we don't plug the requests that require a zone lock in the first >> place, flushing should not have any impact. So no issues directly >> accessing the plug. >> >> This is just a cleanup patch to use this wrapper to get the plug >> consistently across the block layer. > > While I did suggest to make this consistent and in principle it's > the right thing to do, it also irks me to add extra checks to paths > where we know that it's just extra pointless code. Maybe we can > just comment these two spots? Basically each of the sections above > could just go into the appropriate file. > The checks should go away, and the plug could be inlined by the compiler if we don't have CONFIG_BLK_DEV_ZONED. Otherwise, I do agree with you that it is a pointless check. I am fine with either, though I prefer what this patch is doing. So if you feel strongly against calling the blk_mq_plug function, I can turn this patch into just adding comments as you suggested.
