On Wed, Jul 18, 2018 at 8:43 PM, Hannes Reinecke <hare@xxxxxxx> wrote: > On 07/18/2018 02:06 PM, Ming Lei wrote: >> On Wed, Jul 18, 2018 at 5:49 AM, Jens Axboe <axboe@xxxxxxxxx> wrote: >>> On 7/17/18 9:38 AM, Ming Lei wrote: >>>> On Tue, Jul 17, 2018 at 03:34:35PM +0000, Bart Van Assche wrote: >>>>> On Tue, 2018-07-17 at 23:30 +0800, Ming Lei wrote: >>>>>> On Tue, Jul 17, 2018 at 03:24:11PM +0200, Christoph Hellwig wrote: >>>>>>> On Fri, Jul 13, 2018 at 04:06:02PM +0800, Ming Lei wrote: >>>>>>>> Usually SCSI supports runtime PM, so pass BLK_MQ_F_SUPPORT_RPM to blk-mq >>>>>>>> core for enabling block runtime PM. >>>>>>> >>>>>>> I still think enabling this unconditionally for any SCSI device was >>>>>>> a mistake, and it is even more so for blk-mq. >>>>>>> >>>>>>> Please only opt in for ufs, ATA first, adding others if wanted by >>>>>>> maintainers. >>>>>> >>>>>> No, this way will cause regression because runtime PM works for >>>>>> all sd/sr device actually, and it isn't related with scsi host. >>>>> >>>>> For which SCSI devices other than ufs and ATA do we need PM support? >>>> >>>> As I said, it is any sd/sr device, which can be put down by runtime PM >>>> via START_STOP command if it isn't used for a while. >>> >>> Christoph is basically echoing my concerns. Why don't we just enable >>> it on slower devices, similarly to what we do for adding >>> randomness? Nobody wants to pay this overhead for faster devices, >>> since most people won't care. >> >> IMO the problem isn't related with slow or quick device, it is related with >> the system, especially when it cares about power consumption, such as >> mobile phone, or laptop or servers with lots of disks attached. And we know >> it is often to see some fast disks shipped in laptop, such as NVMe, or other >> SSD. >> > But those typically have dedicated (transport/driver) mechanism to put > the device to sleep, _and_ START STOP UNIT will be emulated for those > devices anyway. > I'd rather advocate to move runtime PM into the individual > subsystems/drivers, and _not_ abuse START STOP UNIT here. Firstly START_STOP isn't a abuse for PM purpose, please see the description in t10.org: "The START STOP UNIT command (see 5.17) allows an application client to control the power condition of a logical unit. This method includes specifying that the logical unit transition to a power condition." Secondly the runtime PM on LUN level isn't contradictory with subsystem/driver's runtime PM. The two can co-exits, for examples, USB storage host, ufshcd, ... Thanks, Ming Lei
