NVMe PCI-passthrough

Manuel Sopena Ballesteros manuel.sb at garvan.org.au
Wed Apr 3 23:46:13 UTC 2019


Thank you this was the actual problem.

Now it works

# cat /proc/cmdline
BOOT_IMAGE=/vmlinuz-3.10.0-862.el7.x86_64 root=UUID=da7b03a2-d0f5-42f2-a660-861fcb5c8488 ro rootflags=subvol=root crashkernel=auto rhgb quiet intel_iommu=on iommu=pt

Manuel


-----Original Message-----
From: Stephen Finucane [mailto:sfinucan at redhat.com]
Sent: Wednesday, April 3, 2019 8:20 PM
To: Manuel Sopena Ballesteros; openstack at lists.openstack.org
Subject: Re: NVMe PCI-passthrough

On Wed, 2019-04-03 at 06:18 +0000, Manuel Sopena Ballesteros wrote:
> Dear Openstack community,
>
> I am trying to attach NVMe drives to my vms and would like to check the configuration I have because for some reason it is not working.
>
> This is my physical host details:
>
> [root at zeus-59 ~]# lspci -nn -D | grep -i ssd
> 0000:03:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:04:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:06:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:07:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:08:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:09:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:84:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:85:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:86:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
> 0000:87:00.0 Non-Volatile memory controller [0108]: Intel Corporation PCIe Data Center SSD [8086:0953] (rev 01)
>
> My idea is to attach either 0000:03:00.0 or 0000:04:00.0 to a vm
>
> This is how I identify the block devices for each pci device:
>
> [root at zeus-59 ~]# ls -l /sys/bus/pci/devices/0000\:03\:00.0/nvme
> total 0
> drwxr-xr-x. 4 root root 0 Apr  3 14:02 nvme0
> [root at zeus-59 ~]# ls -l /sys/bus/pci/devices/0000\:04\:00.0/nvme
> total 0
> drwxr-xr-x. 4 root root 0 Apr  3 14:02 nvme1
>
> I installed the OS on the physical machine leaving those 2 drives intact:
>
> [root at zeus-59 ~]# lsblk
> NAME        MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
> sda           8:0    0 59.6G  0 disk
> ├─sda1        8:1    0    1G  0 part /boot
> └─sda2        8:2    0    4G  0 part [SWAP]
> sr0          11:0    1  906M  0 rom
> nvme0n1     259:15   0  1.8T  0 disk
> nvme1n1     259:0    0  1.8T  0 disk
> nvme2n1     259:11   0  1.8T  0 disk
> └─nvme2n1p1 259:12   0  1.8T  0 part /var/lib/docker/btrfs
> nvme3n1     259:6    0  1.8T  0 disk
> └─nvme3n1p1 259:7    0  1.8T  0 part
> nvme4n1     259:13   0  1.8T  0 disk
> └─nvme4n1p1 259:14   0  1.8T  0 part
> nvme5n1     259:5    0  1.8T  0 disk
> └─nvme5n1p1 259:8    0  1.8T  0 part
> nvme6n1     259:17   0  1.8T  0 disk
> └─nvme6n1p1 259:18   0  1.8T  0 part
> nvme7n1     259:9    0  1.8T  0 disk
> └─nvme7n1p1 259:10   0  1.8T  0 part
> nvme8n1     259:3    0  1.8T  0 disk
> └─nvme8n1p1 259:4    0  1.8T  0 part
> nvme9n1     259:1    0  1.8T  0 disk
> └─nvme9n1p1 259:2    0  1.8T  0 part
>
> The next thing is to configure Openstack
>
> nova api config
>
> [pci]
> alias = { "vendor_id":"8086", "product_id":"0953", "device_type":"type-PCI", "name":"nvme"}
>
>
> nova scheduler config
>
> [filter_scheduler]
> enabled_filters = RetryFilter, AvailabilityZoneFilter, ComputeFilter, ComputeCapabilitiesFilter, ImagePropertiesFilter, ServerGroupAntiAffinityFilter, ServerGroupAffinityFilter, PciPassthroughFilter
> available_filters = nova.scheduler.filters.all_filters
>
>
> nova compute config
>
> [filter_scheduler]
> enabled_filters = RetryFilter, AvailabilityZoneFilter, ComputeFilter, ComputeCapabilitiesFilter, ImagePropertiesFilter, ServerGroupAntiAffinityFilter, ServerGroupAffinityFilter, PciPassthroughFilter
> available_filters = nova.scheduler.filters.all_filters
>
> [pci]
> passthrough_whitelist = [ {"address":"0000:03:00.0"}, {"address":"0000:04:00.0"} ]
> alias = { "vendor_id":"8086", "product_id":"0953", "device_type":"type-PCI", "name":"nvme"}
>
>
> Then I create my flavor
> openstack flavor create nvme.small --ram 64000 --disk 10 --vcpus 7 --property "pci_passthrough:alias"="nvme:1"
>
>
> I created the vm towards the host with the free nvme drives
>
> [root at openstack-deployment ~]# openstack server create --flavor nvme.small --image centos7.5-image --nic net-id=hpc --security-group admin --key-name mykey --availability-zone nova:zeus-59.localdomain test_nvme_small
>
>
> Vm creation fails and I can see this error in the nova compute logs
>
> 2019-04-03 16:50:35.512 7 ERROR nova.virt.libvirt.guest [req-8478a99f-bad0-43d5-b405-55c45e8d8cae 91e83343e9834c8ba0172ff369c8acac b91520cff5bd45c59a8de07c38641582 - default default] Error launching a defined domain with XML: <domain type='kvm'>
>   <name>instance-00000090</name>
>>   <devices>
>     <emulator>/usr/libexec/qemu-kvm</emulator>
>     <disk type='file' device='disk'>
>       <driver name='qemu' type='qcow2' cache='none'/>
>       <source file='/var/lib/nova/instances/a261fcd4-eca3-4982-8b2a-1df33087ab40/disk'/>
>       <target dev='vda' bus='virtio'/>
>       <address type='pci' domain='0x0000' bus='0x00' slot='0x04' function='0x0'/>
>     </disk>
>> 2019-04-03 16:50:35.513 7 ERROR nova.virt.libvirt.driver [req-8478a99f-bad0-43d5-b405-55c45e8d8cae 91e83343e9834c8ba0172ff369c8acac b91520cff5bd45c59a8de07c38641582 - default default] [instance: a261fcd4-eca3-4982-8b2a-1df33087ab40] Failed to start libvirt guest: libvirtError: unsupported configuration: host doesn't support passthrough of host PCI devices

This error suggests you haven't enabled IOMMU. See the troubleshooting
section of this page for more information:

  http://ask.xmodulo.com/pci-passthrough-virt-manager.html

If this isn't called out in the docs, it should be.

Stephen

> Any idea or advice would be very helpful

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.


More information about the openstack-discuss mailing list