On Fri, Sep 18, 2015 at 2:09 PM, Nicholas A. Bellinger <nab@xxxxxxxxxxxxxxx> wrote: > On Fri, 2015-09-18 at 11:12 -0700, Ming Lin wrote: >> On Thu, 2015-09-17 at 17:55 -0700, Nicholas A. Bellinger wrote: >> > On Thu, 2015-09-17 at 16:31 -0700, Ming Lin wrote: >> > > On Wed, 2015-09-16 at 23:10 -0700, Nicholas A. Bellinger wrote: >> > > > Hi Ming & Co, > > <SNIP> > >> > > > > I think the future "LIO NVMe target" only speaks NVMe protocol. >> > > > > >> > > > > Nick(CCed), could you correct me if I'm wrong? >> > > > > >> > > > > For SCSI stack, we have: >> > > > > virtio-scsi(guest) >> > > > > tcm_vhost(or vhost_scsi, host) >> > > > > LIO-scsi-target >> > > > > >> > > > > For NVMe stack, we'll have similar components: >> > > > > virtio-nvme(guest) >> > > > > vhost_nvme(host) >> > > > > LIO-NVMe-target >> > > > > >> > > > >> > > > I think it's more interesting to consider a 'vhost style' driver that >> > > > can be used with unmodified nvme host OS drivers. >> > > > >> > > > Dr. Hannes (CC'ed) had done something like this for megasas a few years >> > > > back using specialized QEMU emulation + eventfd based LIO fabric driver, >> > > > and got it working with Linux + MSFT guests. >> > > > >> > > > Doing something similar for nvme would (potentially) be on par with >> > > > current virtio-scsi+vhost-scsi small-block performance for scsi-mq >> > > > guests, without the extra burden of a new command set specific virtio >> > > > driver. >> > > >> > > Trying to understand it. >> > > Is it like below? >> > > >> > > .------------------------. MMIO .---------------------------------------. >> > > | Guest |--------> | Qemu | >> > > | Unmodified NVMe driver |<-------- | NVMe device simulation(eventfd based) | >> > > '------------------------' '---------------------------------------' >> > > | ^ >> > > write NVMe | | notify command >> > > command | | completion >> > > to eventfd | | to eventfd >> > > v | >> > > .--------------------------------------. >> > > | Host: | >> > > | eventfd based LIO NVMe fabric driver | >> > > '--------------------------------------' >> > > | >> > > | nvme_queue_rq() >> > > v >> > > .--------------------------------------. >> > > | NVMe driver | >> > > '--------------------------------------' >> > > | >> > > | >> > > v >> > > .-------------------------------------. >> > > | NVMe device | >> > > '-------------------------------------' >> > > >> > >> > Correct. The LIO driver on KVM host would be handling some amount of >> > NVMe host interface emulation in kernel code, and would be able to >> > decode nvme Read/Write/Flush operations and translate -> submit to >> > existing backend drivers. >> >> Let me call the "eventfd based LIO NVMe fabric driver" as >> "tcm_eventfd_nvme" >> >> Currently, LIO frontend driver(iscsi, fc, vhost-scsi etc) talk to LIO >> backend driver(fileio, iblock etc) with SCSI commands. >> >> Did you mean the "tcm_eventfd_nvme" driver need to translate NVMe >> commands to SCSI commands and then submit to backend driver? >> > > IBLOCK + FILEIO + RD_MCP don't speak SCSI, they simply process I/Os with > LBA + length based on SGL memory or pass along a FLUSH with LBA + > length. > > So once the 'tcm_eventfd_nvme' driver on KVM host receives a nvme host > hardware frame via eventfd, it would decode the frame and send along the > Read/Write/Flush when exposing existing (non nvme native) backend > drivers. > > This doesn't apply to PSCSI backend driver of course, because it expects > to process actual SCSI CDBs. > >> But I thought the future "LIO NVMe target" can support frontend driver >> talk to backend driver directly with NVMe commands without translation. >> > > The native target_core_nvme backend driver is not processing nvme > commands per-say, but simply exposing nvme hardware queue resources to a > frontend fabric driver. > > For the nvme-over-fabrics case, backend nvme submission/complete queues > are mapped to RDMA hardware queues. So essentially the nvme physical > region page (PRP) is mapped to ib_sgl->addr. > > For a 'tcm_eventfd_nvme' style host-paravirt fabric case, it's less > clear how exposing native nvme backend hardware resources would work, or > if there is a significant performance benefit over just using > submit_bio() for Read/Write/Flush. Now it's much more clear. I'll do a tcm_eventfd_nvme prototype. Thanks for all the detail explanation. > > --nab > _______________________________________________ Virtualization mailing list Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linuxfoundation.org/mailman/listinfo/virtualization
