Re: [libvirt] [PATCH] esx: Add autodetection for the SCSI controller model
2010/7/7 Daniel P. Berrange <berrange(a)redhat.com>:
On Tue, Jul 06, 2010 at 08:19:55PM +0200, Matthias Bolte wrote:
> This works for file-backed SCSI disk device with a datastore
> related source path.
> ---
> docs/drvesx.html.in | 6 ++
> docs/schemas/domain.rng | 1 +
> src/conf/domain_conf.c | 1 +
> src/conf/domain_conf.h | 1 +
> src/esx/esx_vi_generator.input | 149 +++++++++++++++++++++++++++++++++
> src/esx/esx_vi_generator.py | 3 +
> src/esx/esx_vmx.c | 180 +++++++++++++++++++++++++++++++++++++++-
> src/esx/esx_vmx.h | 8 ++-
> 8 files changed, 344 insertions(+), 5 deletions(-)
>
> diff --git a/docs/drvesx.html.in b/docs/drvesx.html.in
> index f4e7530..75c24cb 100644
> --- a/docs/drvesx.html.in
> +++ b/docs/drvesx.html.in
> @@ -292,6 +292,12 @@ ethernet0.checkMACAddress = "false"
>
> <h4>SCSI controller models</h4>
> <dl>
> + <dt><code>auto</code></dt>
> + <dd>
> + This isn't a actual controller model. If specified the ESX driver
> + tries to detect the SCSI controller model referenced in the
> + <code>.vmdk</code> file and use it. <span
class="since">Since 0.8.3</span>
> + </dd>
> <dt><code>buslogic</code></dt>
> <dd>
> BusLogic SCSI controller for older guests.
> diff --git a/docs/schemas/domain.rng b/docs/schemas/domain.rng
> index b171d01..e090366 100644
> --- a/docs/schemas/domain.rng
> +++ b/docs/schemas/domain.rng
> @@ -673,6 +673,7 @@
> <optional>
> <attribute name="model">
> <choice>
> + <value>auto</value>
> <value>buslogic</value>
> <value>lsilogic</value>
> <value>lsisas1068</value>
I don't really see how this can sensibly work.
A controller can have multiple disks associated with
it. Each associated disk has a drive address that uniquely
identifies its position on the controller.
Now consider a controller of mode 'auto' and attach two
disks to it. One VMDK specifies 'lsilogic' and the other
VMDK specifies 'lsisas1068'. This is not reconcilable
because you have 2 conflicting controller models, but have
associated both disks with the same controller object.
In case of detecting conflicting controller models for one controller
the autodetection has to fail. The current code just picks the
reported model for the first disk and doesn't look at the other disk
on the same controller. That's wrong and I fixed this in v2.
Furthermore both disks could be LUN 1 on their respective
controller, which means you'd now have a uniquness violation
on the drive address.
I don't get what you try to say. The model autodetection doesn't
affect the drive address at all and is a per-controller action.
Matthias