Re: pvSCSI test

[Juergen Gross <jgross@suse.com>]

On 02/06/2015 03:02 PM, Kristian Hagsted Rasmussen wrote:
>
> On Friday, February 6, 2015 10:57, Juergen Gross <jgross@suse.com> wrote:
>> To: Kristian Hagsted Rasmussen; Olaf Hering; xen-devel@lists.xensource.com
>> Subject: Re: pvSCSI test
>>
>> On 02/06/2015 10:32 AM, Kristian Hagsted Rasmussen wrote:
>>> On Wednesday, February 4, 2015 05:41, Juergen Gross <jgross@suse.com> wrote:
>>>> To: Kristian Hagsted Rasmussen; Olaf Hering; xen-devel@lists.xensource.com
>>>> Subject: Re: pvSCSI test
>>>>
>>>> On 02/03/2015 07:16 PM, Kristian Hagsted Rasmussen wrote:
>>>>>
>>>>> Hi Olaf and Juergen
>>>>>
>>>>>
>>>>> I am interested in testing pvSCSI as I have a system were it would be ideal.
>>>>> I have tried to apply this patch "http://marc.info/?l=xen-devel&m=139885599019457&w=2" called "libbxl: add support for pvscsi, iteration 1", to my xen-4.5 tree.
>>>>> I am using kernel 3.18.4 compiled with xen-scsiback and xen-scsifront compiled into the kernel, for both Dom0 and DomU.
>
> I have by the way changed my kernel config, so that xen-scsiback, xen-scsifront, target_core_mod is compiled as modules.
>
> snip
>
>>
>> I've used xm to get the drivers working, not xl. I wanted to have only
>> one changed component during tests in order to know which component is
>> failing. :-)
>>
>> What are the contents of
>>
>> /sys/kernel/config/target/xen-pvscsi/$TARG/tpgt_0/param/alias
>>
>> ($TARG replaced by the UUID generated above, of course)?
>>
>> This should be "3:0:0:0" in your case. That's where the backend is
>> looking for the match from xenstore.
>>
>> And the final symlink in the script is required. $NAME can be anything,
>> but has to match the pscsi name, of course.
>
> After some more fiddling around, I believe the configuration should be okay. My /etc/target/xen-pvscsi_start.sh look like this:
>
> modprobe xen-scsiback
> mkdir /sys/kernel/config/target/xen-pvscsi
> mkdir -p /sys/kernel/config/target/xen-pvscsi/naa.600140512a981c66/tpgt_0
> echo naa.6001405708ab297e > /sys/kernel/config/target/xen-pvscsi/naa.600140512a981c66/tpgt_0/nexus
> #### xen-pvscsi Target Ports
> mkdir -p /sys/kernel/config/target/xen-pvscsi/naa.600140512a981c66/tpgt_0/lun/lun_0
> ln -s /sys/kernel/config/target/xen-pvscsi/naa.600140512a981c66/tpgt_0/lun/lun_0/../../../../../../target/core/iblock_0/3_0_0_0 /sys/kernel/config/target/xen-pvscsi/naa.600140512a981c66/tpgt_0/lun/lun_0/xen-pvscsi_port

iblock_0?

You are not using pscsi, but iblock. Is that on purpose? I have tested
pscsi and fileio only.

What does lsscsi tell you after adding the device via targetcli? I
suppose you see a new scsi target you should use instead of 3:0:0:0
(that's what I did in the fileio case).

> #### Attributes for xen-pvscsi Target Portal Group
> #### Parameters for xen-pvscsi Target Portal Group
> echo "3:0:0:0" > /sys/kernel/config/target/xen-pvscsi/naa.600140512a981c66/tpgt_0/param/alias
>
> And
> cat /sys/kernel/config/target/xen-pvscsi/naa.600140512a981c66/tpgt_0/param/alias
> returns 3:0:0:0 as expected.
>
> If I understand this correctly, it is xen-scsiback that reads the target configuration from ConfigFS and hence the problem that xen-pvscsi cannot find the device has nothing to do with which toolstack is used?

Yes and no. In theory the backend would accept anything from xenstore
which it can find in configfs. The toolstack however will only write
values into the xenstore it believes are valid SCSI devices.

> If you are willing to part with the script you used together with targetcli, I would be more then happy to try that out.

I did my first tests with fileio on a machine I have no longer access
to. after I got it running I changed to my local test machine and pscsi.
Here I did the targetcli stuff manually and verified afterwards that
the single steps I put in my script were working. So the script I gave
you is basically the documentation of the manual setup I used.

Juergen