Re: bhyve NVMe 1.4 support
- In reply to: Mario Marietto : "Re: bhyve NVMe 1.4 support"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Sun, 20 Mar 2022 22:34:15 UTC
On Mar 20, 2022, at 3:11 PM, Mario Marietto <marietto2008@gmail.com> wrote: > > -s 2,nvme,/dev/nvd0 \ AIUI, this 'nvme' means bhyve will *emulate* an NVME device and it will treat whatever is given to it (/dev/nvd0) as just dumb storage. It doesn't care if /dev/nvd0 is an NVME device. So commands such as nvmecontrol identify etc. in the *guest* won't be passed through to /dev/nvd0 through to it. If you run nvmecontrol identify under ktrace in the guest, you will see something like 3657 nvmecontrol NAMI "/dev/nvme0ns1" 3657 nvmecontrol RET openat 3 3657 nvmecontrol CALL ioctl(0x3,NVME_GET_NSID,0x1245f31731a8) 3657 nvmecontrol RET ioctl 0 3657 nvmecontrol CALL close(0x3) 3657 nvmecontrol RET close 0 3657 nvmecontrol CALL openat(AT_FDCWD,0x1245f3173260,0<O_RDONLY>) 3657 nvmecontrol NAMI "/dev/nvme0" 3657 nvmecontrol RET openat 3 3657 nvmecontrol CALL ioctl(0x3,NVME_PASSTHROUGH_CMD,0x1245f3173250) 3657 nvmecontrol RET ioctl 0 3657 nvmecontrol CALL close(0x3) These ioctls will *not* be passed on the real /dev/nvd0 in the host so they are *not* going to return the same information. If you want to talk to the real nvme device from a guest, you have to use a passthrough device. Though I don't know how well this works (or if at all).