[libvirt] [PATCH v4 9/9] docs: Add vhost-scsi

Jim Fehlig jfehlig at suse.com
Thu Dec 8 22:23:42 UTC 2016

Eric Farman wrote:
> On 12/07/2016 08:16 PM, Jim Fehlig wrote:
>> On 11/22/2016 02:16 PM, John Ferlan wrote:
>>> On 11/21/2016 10:58 PM, Eric Farman wrote:
>>>> Signed-off-by: Eric Farman <farman at linux.vnet.ibm.com>
>>>> ---
>>>>  docs/formatdomain.html.in | 24 ++++++++++++++++++++++++
>>>>  1 file changed, 24 insertions(+)
>>> This will get squashed in with the conf patch. I'll also generate the
>>> news.html.in entry:
>>>           <li>vhost-scsi: Add support scsi_host hostdev passthrough<br/>
>>>           Add the capability to pass through a scsi_host HBA and the
>>>           associated LUNs to the guest.
>>>           </li>
>>> ACK -
>>> I'll push these in a bit, just seeing what I get on my test system... Of
>>> course I'm also curious what would happen if I try to pass through a
>>> vHBA ;-)...
>> I have unsuccessfully tried it, but not sure how to correctly specify
>> the vHBA in domXML.
>> First, create a vHBA:
>> cat vhba.xml
>> <device>
>>    <parent_wwpn>2100001b32847342</parent_wwpn>
>>    <parent_wwnn>2000001b32847342</parent_wwnn>
>>    <capability type='scsi_host'>
>>      <capability type='fc_host'>
>>       <wwnn>9990100000000001</wwnn>
>>       <wwpn>9990000000000001</wwpn>
>>       </capability>
>>    </capability>
>> </device>
>> virsh nodedev-create vhba.xml
>> (@John: Notice I'm using your "Allow creation of vHBA by
>> parent_wwnn/wwpn or fabric_name" series. I'll try to respond with
>> review comments on that series tomorrow.)
> (@John:  Me too :)
>> Add the vHBA to domain config using a variant of Eric's example:
>>   <hostdev mode='subsystem' type='scsi_host'>
>>     <source protocol='vhost' wwpn='naa.9990000000000001'/>
>>     <address type='ccw' cssid='0xfe' ssid='0x0' devno='0x1000'/>
>>   </hostdev>
>> Start the VM:
>> virsh start test
>> error: Failed to start domain test
>> error: Path '/sys/kernel/config/target/vhost//naa.5001405df3e54061' is
>> not accessible: No such file or directory
> Eh?  Where did "5001..." come from?  Given the above hostdev snippet, I
> would've expected this to show "99900..."
> Also, if you're not using an s390 machine, the address should probably
> be type='pci' or omitted altogether.

Sorry, too much copy and paste between your example and the various configs I
tried. Also too hasty in trying to get the mail out before knocking off for the day.

The actual hostdev config is

  <hostdev mode='subsystem' type='scsi_host'>
    <source protocol='vhost' wwpn='naa.9990000000000001'/>
    <address type='pci' domain='0x0000' bus='0x00' slot='0x0b' function='0x0'/>

and the error

error: Path '/sys/kernel/config/target/vhost//naa.9990000000000001' is not
accessible: No such file or directory

>> It looks like this patch series only handles scsi_host created with
>> targetcli, but I didn't look at all the patches closely. 
> Today, yes.  But more specifically, it's only handling scsi_host that
> connects to the vhost_scsi target, thus the
> "/sys/kernel/config/target/vhost/" prefix.  The NPIV vport is at a
> different locale.
>> Is it possible to use vhost-scsi with NPIV vport (vHBA)? If so, any
>> pointers on how to specify the vHBA in <hostdev> element?
> This vhost-scsi series had to do some trickery rather than relying on
> virsh nodedev-list for data.  Looking at John's series, I suspect
> there's some additional work necessary for the vport-capable sysfs
> entries than what exists with this series.  Some of which I deferred
> from an earlier review comment.

Ok, thanks. I'll see if I can figure out how to make the vports work with


More information about the libvir-list mailing list