Bart, On 4/22/17 02:15, Bart Van Assche wrote: > On Fri, 2017-04-21 at 18:16 +0900, damien.lemoal@xxxxxxx wrote: >> @@ -250,11 +249,6 @@ int sd_zbc_setup_reset_cmnd(struct scsi_cmnd *cmd) >> /* Unaligned request */ >> return BLKPREP_KILL; >> >> - /* Do not allow concurrent reset and writes */ >> - if (sdkp->zones_wlock && >> - test_and_set_bit(zno, sdkp->zones_wlock)) >> - return BLKPREP_DEFER; >> - > > Hello Damien, > > Since concurrent zone resets and writes are easy to detect and since these > indicate a bug in the application that submits these, should an error > message > be logged if that happens? Otherwise this patch looks fine to me. Yes, we could easily add a warning, but if we consider the 3 possible cases of reset+write combinations: 1) Reset comes in while a write is already in the queue: The write will succeed (assuming it is directed at the zone write pointer) and the reset will also succeed. 2) A write directed at the current write pointer comes in while a reset is in the queue: the write will fail. 3) A write directed at the beginning of the zone comes in while a reset is in the queue: the write will succeed. The warning would trigger only in case (1), but that case may be actually intentional by the user. Case (2) will get an "indirect" warning through the write align error, which clearly signals a bug to the user. For case (3), that may also be intentional by the user (not recommended though). So I am afraid that a warning may trigger false-positives and may not be that useful in the end. Best regards. -- Damien Le Moal, Ph.D. Sr. Manager, System Software Research Group, Western Digital Corporation Damien.LeMoal@xxxxxxx (+81) 0466-98-3593 (ext. 513593) 1 kirihara-cho, Fujisawa, Kanagawa, 252-0888 Japan www.wdc.com, www.hgst.com
