[NOVA] pci alias device_type and numa_policy and device_type meanings

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

[NOVA] pci alias device_type and numa_policy and device_type meanings

Manuel Sopena Ballesteros

Dear Openstack community.

 

I am setting up pci passthrough for GPUs using aliases.

 

I was wondering the meaning of the fields device_type and numa_policy and how should I use them as I could not find much details in the official documentation.

 

https://docs.openstack.org/nova/rocky/admin/pci-passthrough.html#configure-nova-api-controller

https://docs.openstack.org/nova/rocky/configuration/config.html#pci

 

thank you very much

 

Manuel

 

NOTICE
Please consider the environment before printing this email. This message and any attachments are intended for the addressee named and may contain legally privileged/confidential/copyright information. If you are not the intended recipient, you should not read, use, disclose, copy or distribute this communication. If you have received this message in error please notify us at once by return email and then delete both messages. We accept no liability for the distribution of viruses or similar in electronic communications. This notice should not be removed.

__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: [hidden email]?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

ATT00001.txt (338 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [NOVA] pci alias device_type and numa_policy and device_type meanings

Sean Mooney
On Mon, Nov 5, 2018 at 11:02 PM Manuel Sopena Ballesteros
<[hidden email]> wrote:

>
> Dear Openstack community.
>
>
>
> I am setting up pci passthrough for GPUs using aliases.
>
>
>
> I was wondering the meaning of the fields device_type and numa_policy and how should I use them as I could not find much details in the official documentation.
numa policy is used to configure the affinity policy for the passthrough device.
it was introduced by
https://specs.openstack.org/openstack/nova-specs/specs/queens/implemented/share-pci-between-numa-nodes.html#proposed-change
not that the flavor extra specs and image metadata also descibed by
that spec are stil outstanding and may be added in the future.

the device type is a legacy hack that was added when nova's pci
passthough support was extended to enable neutron sriov.
when using flavor based pci passthough using aliases the type should
always be set to Type-PCI
Type-PF and Type-VF in general should only be used for nics that will
be used with neutron sriov port types.

you technically can us the other device types in the alias but nova
will only record a device as  Type-PF or Type-VF
if physical_network is set in the pci whitelist which is used to
indicate the neutron physnet the device is attached too.

in the case of GPUs you can ignore Type-PF and Type-VF.
you should also know that Type-PCI can refer to either the full pci
device or a virtual function
allocated form a device.

an example alias is available in the code here
https://github.com/openstack/nova/blob/master/nova/pci/request.py#L18
| [pci]
| alias = '{
| "name": "QuickAssist",
| "product_id": "0443",
| "vendor_id": "8086",
| "device_type": "type-PCI",
| "numa_policy": "legacy"
| }'

in this case the alias consumes an entire QuickAssist (QAT) device
with the legacy numa policy.
if the  QAT device was subdivided using sriov VFs you could request
the VF instead of the PF by
simply changing the product_id to that corresponding to the VF.


>
>
>
> https://docs.openstack.org/nova/rocky/admin/pci-passthrough.html#configure-nova-api-controller
>
> https://docs.openstack.org/nova/rocky/configuration/config.html#pci
>
>
>
> thank you very much
>
>
>
> Manuel
>
>
>
> NOTICE
> Please consider the environment before printing this email. This message and any attachments are intended for the addressee named and may contain legally privileged/confidential/copyright information. If you are not the intended recipient, you should not read, use, disclose, copy or distribute this communication. If you have received this message in error please notify us at once by return email and then delete both messages. We accept no liability for the distribution of viruses or similar in electronic communications. This notice should not be removed.
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: [hidden email]?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: [hidden email]?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev