On Wed, Oct 19, 2022 at 10:57:11AM +0300, Sagi Grimberg wrote: > > > On 10/19/22 10:32, Christoph Hellwig wrote: >> On Wed, Oct 19, 2022 at 10:30:44AM +0300, Sagi Grimberg wrote: >>> iscsi defers network sends to a workqueue anyways, and no scsi lld sets >>> BLK_MQ_F_BLOCKING. >> >> Yes, but Mike had a series to implement a similar fast path as nvme-tcp >> as while ago that included supporting BLK_MQ_F_BLOCKING in the scsi >> midlayer. > > Hmmm, didn't see that, pointer? https://www.spinics.net/lists/linux-scsi/msg170849.html >> We'll also need it to convert the paride drivers to libata, >> but those are single request_queue per tag_set as far as I know. > > didn't see paride even attempting to quiesce... The block layer might. But it really does not matter for our considerations here..
