Jump to: navigation, search

Difference between revisions of "PCI passthrough SRIOV support"

(→‎=define the flavors on Control node)
m (→‎Common PCI SRIOV design)
 
(68 intermediate revisions by 6 users not shown)
Line 1: Line 1:
 +
'''!!!This is a design discussion document, not for end user reference!!! '''
  
−
===background===
+
===Related Resource===
−
this design based on this PCI passthrough meeting:
+
This design is based on the PCI pass-through IRC meetings, provide common support for PCI SRIOV:
−
https://wiki.openstack.org/wiki/Meetings/Passthrough
+
*https://wiki.openstack.org/wiki/Meetings/Passthrough
−
https://docs.google.com/document/d/1vadqmurlnlvZ5bv3BlUbFeXRS_wh-dsgi5plSjimWjU/edit#
+
This document was used to finalise the design:
 +
*https://docs.google.com/document/d/1vadqmurlnlvZ5bv3BlUbFeXRS_wh-dsgi5plSjimWjU/edit#
 
link back to bp:
 
link back to bp:
−
https://blueprints.launchpad.net/nova/+spec/pci-extra-info
+
*https://blueprints.launchpad.net/nova/+spec/pci-extra-info
  
−
PCI devices has PCI standard property like address(BDF), vendor_id, product_id, etc, and PCI property like every Virtual Function's physical address  , it also should can be attach some application specific extra information, like physical network connectivity used by neutron  SRIOV, or any other property.
+
===Common PCI SRIOV design ===
  
−
All kind of these PCI property should be well classify, and for every property what's the scope it belong to should be well defined.   our design will focus on several PCI modules to provide the PCI pass-through SRIOV support,  it's current functionality is:
+
PCI devices have PCI standard properties like address (BDF), vendor_id, product_id, etc, Virtual functions also have a property referring to the function's physical address.  Application specific or installation specific extra information can be attached to PCI device, like physical network connectivity using by Neutron SRIOV.
  
−
* on compute node the whit-list define a spec which filter the PCI device, given a set of PCI device which is available for allocation.
+
This bp focus on functionality to provide the common PCI SRIOV  support.
−
* PCI compute also report the PCI stats information to scheduler. PCI stats contain several pools. each pool defined by several PCI property (vendor_id, product_id, extra_info).
 
−
* PCI alias define the user point of view PCI device selector: alias provide a set of (k,v) to form specs to select the available device(filter by white list).
 
  
−
===PCI NEXT over all design ===
+
* on compute node the pci_information/white-list define a set of filter. PCI device passed filter will be available for allocation. at same time extra information attached to the PCI device.
 +
* PCI compute report the PCI stats information to scheduler. PCI stats contain several pools. each pool defined by several PCI property, control by the local configuration item: pci_flavor_attrs , default value is vendor_id, product_id, extra_info.
 +
* PCI flavor define the user point of view PCI device selector: PCI flavor provide a set of (k,v) to form specs to select the available device(filtered by pci_information/white-list).
  
−
====PCI flavor used to select available device====
+
the new PCI design based on 2 key change: the PCI flavor and the extra information attache to pci device. the following diagram is a summary of the design:
−
User use the PCI flavor to select available device. and the PCI flavor is global to all cloud, and can be configuration via API.  user treat the PCI flavor as  an resealable name like 'oldGPU', 'FastGPU',  '10GNIC', 'SSD'. 
 
  
 +
[[Image:PCI-sriov.jpg]]
  
−
====define the  flavors on Control node ====
+
====Design Choice ====
−
Control node has flavors which allow the administrator to package up devices for users. flavors have a name, and matching expression that selects available(offer by white list) devices. flavors can overlap - that is, the same device on the same machine may be matched by multiple flavors.
+
   
 +
      PCI flavor:  User don't want to know details of a pci device and all of it's attrs and extra information attached to it.
 +
      (pci_flavor_attr:  admin need to know all PCI attrs to define flavor for user/tenant.)
 +
      PCI Stats:  compute node might have many devices and most of device properties are same,  summary a stats to scheduler can reduce DB load, simply schedule.
 +
      PCI extra info:  a pci device might attach to a specific network, a specific resource, with can be attach to device and schedule base on it.
 +
 
 +
 
 +
====PCI flavor====
  
−
PCI flavor defined by a set of (k,v), the k is the *well defined* PCI property. not every PCI property is available to PCI flavor,  only a specific set of PCI property can used to define the PCI flavor. these PCI property is defined via a global configuration :
+
For OS users, PCI flavor is a reasonable name like 'oldGPU', 'FastGPU',  '10GNIC', 'SSD', describe one kind of PCI device. User use the PCI flavor to select available pci devices. Internally the PCI flavor created by a set of API and saved in a DB table, keep PCI flavor available for all cloud.  
  
−
    pci_flavor_attrs = [ vendor_id, product_id, ...]
 
  
−
Only the global attrs should be one of pci_falvor_attrs, like the vendor, product, etc, the 'host', and 'BDF' of a pci device should not be used as pci_flavor_attrs. this is explicitly an optimization to simplify scheduling complexity, we may change this to an API-changeable value in the future, though we believe it will rarely be necessary to change its value.
+
Administrator define the PCI flavors via matching expression that selects available(offer by white list) devices， and a reasonable name. PCI flavor  matching expression is a set of (k,v), the k is the PCI property, v is its value. not every PCI property is available to PCI flavor, only a selected set of PCI property can used to define the PCI flavor, the selected property should be global to cloud like vendor/product_id, can not be BDF or host of a PCI device. these selected PCI property is defined via compute local configuration :
 +
 
 +
    pci_flavor_attrs = vendor_id, product_id, ...
 +
 
 +
a important behavior is the PCI flavors could overlap - that is, the same device on the same machine may be matched by multiple flavors.
 +
 
 +
====Use PCI flavor in instance flavor extra info====
 +
 
 +
user set pci flavor into instance flavor's extra info to specify how many PCI device/and what type PCI flavor the VM want to boot with.
 +
 
 +
    nova flavor-key m1.small set pci_passthrough:pci_flavor= <pci flavor spec list>
 +
 
 +
    pci flavor spec:
 +
                    num1:flavor1,flavor2
 +
    mean: want <number>'s pci devices from flavor or flavor
 +
   
 +
    pci flavor spec list:
 +
                <pci flavor spec1>; <pci flavor spec2>
 +
 
 +
 
 +
for example:
 +
   
 +
    nova flavor-key m1.small set pci_passthrough:pci_flavor= 1:IntelGPU,NvGPU;1:intelQuickAssist;
 +
   
 +
    which define requirements:
 +
          boot with 1 of IntelGPU or NvGPU, and 1 IntelQuickAssist card.
 +
 
 +
====PCI pci_flavor_attrs ====
 +
 
 +
this configuration is keep local to every compute node, this will make deploy process can locally decide what PCI properties this node will exposed.
 +
 
 +
    pci_flavor_attrs = vendor_id, product_id, ...
 +
 
 +
compute node update local pci extra properties to PCI flavor Database, which is accessible by flavor API, provide PCI properties to define flavor.
 +
 
 +
pci_flavor_attrs store in flavor DB as a normal flavor, it's name "_flavor_attrs", it's UUID use "0":
 +
 
 +
    {"vendor_id":"Ture", "product_id":"True", ... }
 +
 
 +
this flavor contain all available attrs can be used to define pci flavor, list it's content use:
 +
 
 +
    nova pci-flavor-show  0
 +
    GET v2/​{tenant_id}​/os-pci-flavor/<0>
 +
    data:
 +
        os-pci-flavor: {
 +
                                'UUID':'0' ,
 +
                                'description':'Available flavor attrs '
 +
                                'name':'_flavor_attrs',
 +
                                'vendor_id": "True",
 +
                                'product_id": "True",
 +
                                ....
 +
          }
 +
 
 +
====PCI request====
 +
 
 +
PCI request is a internal structure to represent all PCI devices a VM want to have.
 +
 
 +
    request = {'count': int(count),
 +
              'spec': [{"vendor_id":"8086", "phynetwork":"phy1"}, ...],
 +
              'alias_name': "Intel.NIC"}
 +
[[Image:PCI-Request.jpg]]
 +
 
 +
====Extra information of PCI device  ====
  
−
====Compute node offers up devices via local config ====
 
 
the compute nodes offer available PCI devices for pass-through,  since the list of devices doesn't usually change unless someone tinkers with the hardware, this matching expression used to create this list of offered devices is stored in compute node config.
 
the compute nodes offer available PCI devices for pass-through,  since the list of devices doesn't usually change unless someone tinkers with the hardware, this matching expression used to create this list of offered devices is stored in compute node config.
 +
 +
 
     *the device information (device ID, vendor ID, BDF etc.) is discovered from the device and stored as part of the PCI device, same as current implement.
 
     *the device information (device ID, vendor ID, BDF etc.) is discovered from the device and stored as part of the PCI device, same as current implement.
−
     *on the compute node, additional arbitrary information, in the form of key-value pairs, can be added to the config and is included in the PCI device
+
     *on the compute node, additional arbitrary extra information, in the form of key-value pairs, can be added to the config and is included in the PCI device
 +
 
 +
this is achieved by extend the pci white-list to:   
 +
 
 +
 
 +
  *pci_information = [ { pci-regex } ,{pci-extra-attrs } ]
 +
  *pci-regex is a dict of { string-key: string-value } pairs , it can only match device properties, like vendor_id, address, product_id,etc.
 +
  *pci-extra-attrs is a dict of { string-key: string-value } pairs.  The values can be arbitrary  The total size of the extra attrs may be restricted. all this extra attrs will be store in the pci device table's extra info field. and the extra attrs should use this naming schema: e.attrname
 +
 
 +
====PCI stats grouping device base on pci_flavor_attrs ====
 +
       
 +
PCI stats pool summary the devices of a compute node, and the scheduler use flavor's matching specs select the available host for VM. The stats pool must contain the PCI properties used by PCI flavor.
 +
 
 +
  *  current grouping is based on  [vendor_id, product_id, extra_info]
 +
  *  going to group by keys specified by  pci_flavor_attrs.
 +
 +
The algorithm for stats report should meet this request: the one device should only be in one pci stats pool, this mean pci stats can not overlap. this simplifies the scheduler design.
 +
 
 +
on computer node the pci_flavor_attrs provide the specs for pci stats to group its pool. and the pci_flavor_attrs on control node collection the attrs which can be used to define the pci_flavor. the definition of pci_flavor_attrs on controller should contain all the pci_flavor_attrs's content on every compute node.
 +
 
 +
*but compute node report stats pool by a subset of controller's pci_flavor_attrs is acceptable, in such scenario, this means the compute node can only provide the devices with these propertes.*
 +
 
 +
===Use cases ===
 +
 
 +
==== General PCI pass through  ====
 +
given compute nodes contain 1 GPU with vendor:device 8086:0001
 +
 
 +
*on the compute nodes, config the pci_information
 +
    pci_information =  { { 'device_id': "8086", 'vendor_id': "0001" }, {} }
 +
 
 +
    pci_flavor_attrs ='device_id','vendor_id'
 +
 
 +
the compute node would report PCI stats group by ('device_id', 'vendor_id').
 +
pci stats will report one pool:
 +
  {'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
  
−
this is achived by extend the pci white-list to:     
+
* create PCI flavor
−
pci_information = { pci-regex,pci-extra-attrs }
+
 
−
pci-regex is a dict of { string-key: string-value } pairs , it can only match device properties, like vendor_id, address, product_id,etc.
+
  nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
−
pci-extra-attrs is a dict of { string-key: string-value } pairs.  The values can be arbitrary The total size of the extra attrs may be restricted.
+
  nova pci-flavor-update  name 'bigGPU'  set    'vendor_id'='8086'  'product_id': '0001'
−
   
+
 
−
===PCI NEXT Config  ===
+
* create flavor and boot with it ( same as current PCI passthrough)
 +
 
 +
  nova flavor-key m1.small set pci_passthrough:pci_flavor= 1:bigGPU;
 +
  nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec
 +
 
 +
==== General PCI pass through with multi PCI flavor candidate ====
 +
 
 +
given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002
 +
 
 +
*on the compute nodes, config the pci_information
 +
    pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, {} }
 +
 
 +
* on controller
 +
  pci_flavor_attrs = ['device_id', 'vendor_id']
 +
 
 +
the compute node would report PCI stats group by ('device_id', 'vendor_id').
 +
pci stats will report 2 pool:
 +
 
 +
  {'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
 +
  {'device_id':'0002', 'vendor_id':'8086', 'count': 1 }
 +
 
 +
* create PCI flavor
 +
  nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 +
  nova pci-flavor-update  name 'bigGPU'  set    'vendor_id'='8086'  'product_id': '0001'
 +
  nova pci-flavor-create  name 'bigGPU2'  description 'passthrough Intel's on-die GPU'
 +
  nova pci-flavor-update  name 'bigGPU2'  set   'vendor_id'='8086'  'product_id': '0002'
 +
 
 +
* create flavor and boot with it
 +
  nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU,bigGPU2;'
 +
  nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec
 +
 
 +
==== General PCI pass through wild-cast PCI flavor ====
 +
 
 +
given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002
 +
 
 +
*on the compute nodes, config the pci_information
 +
    pci_information = { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, {} }
 +
    pci_flavor_attrs = ['device_id', 'vendor_id']
 +
 
 +
the compute node would report PCI stats group by ('device_id', 'vendor_id').
 +
pci stats will report 2 pool:
 +
 
 +
  {'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
 +
  {'device_id':'0002', 'vendor_id':'8086', 'count': 1 }
 +
 
 +
* create PCI flavor
 +
 
 +
  nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 +
  nova pci-flavor-update  name 'bigGPU'  set    'vendor_id'='8086'  'product_id': '000[1-2]'
 +
 
 +
* create flavor and boot with it
 +
 
 +
  nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU;'
 +
  nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec
 +
 
 +
====  PCI pass through support grouping tag ====
 +
 
 +
given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002
 +
 
 +
*on the compute nodes, config the pci_information
 +
    pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, { 'e.group':'gpu' } }  
 +
 
 +
    pci_flavor_attrs = ['e.group']
 +
 
 +
the compute node would report PCI stats group by ('e.group').
 +
pci stats will report 1 pool:
 +
{'e.group':'gpu', 'count': 2 }
 +
 
 +
 
 +
* create PCI flavor
 +
 +
  nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 +
  nova pci-flavor-update  name 'bigGPU'  set    'e.group'='gpu'
 +
 
 +
* create flavor and boot with it
 +
 
 +
  nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU;'
 +
  nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec
 +
 
 +
====  PCI SRIOV with tagged flavor ====
 +
given compute nodes contain 5 PCI NIC , vendor:device 8086:0022, and it connect to physical network "X".
 +
 
 +
*on the compute nodes, config the pci_information
 +
 
 +
    pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, { 'e.physical_netowrk': 'X' } }
 +
 
 +
    pci_flavor_attrs = 'e.physical_netowrk'
 +
 
 +
the compute node would report PCI stats group by ('e.physical_netowrk').
 +
pci stats will report 1 pool:
 +
 
 +
 
 +
  {'e.physical_netowrk':'X', 'count': 1 }
 +
 
 +
 
 +
* create PCI flavor
 +
  nova pci-flavor-create  name 'phyX_NIC'  description 'passthrough NIC connect to physical network X'
 +
  nova pci-flavor-update  name 'bigGPU'  set    'e.physical_netowrk'='X'
 +
 
 +
 
 +
* create flavor and boot with it
 +
  nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec  --nic  net-id=network_X  pci_flavor= '1:phyX_NIC;'
 +
 
 +
====encryption card use case ====
 +
 
 +
there is 3 encryption card:  [ V1-3 here means a vendor_id number.]
 +
 
 +
  card 1 (vendor_id is V1, device_id =0xa) 
 +
  card 2 (vendor_id is V1, device_id=0xb)
 +
  card 3 (vendor_id is V2, device_id=0xb)
 +
suppose there is  two images. One image only support Card 1 and another image support Card 1/3 (or any other combination of the 3 card type).
 +
 
 +
*on the compute nodes, config the pci_information
 +
 
 +
    pci_information =  { { 'device_id': "0xa", 'vendor_id': "v1" }, { 'e.QAclass':'1' } }
 +
    pci_information = { { 'device_id': "0xb", 'vendor_id': "v1" }, { 'e.QAclass':'2' } }
 +
    pci_information = { { 'device_id': "0xb", 'vendor_id': "v2" }, { 'e.QAclass':'3' }}
 +
 
 +
    pci_flavor_attrs = ['e.QAclass']
 +
 
 +
the compute node would report PCI stats group by (['e.QAclass']).
 +
pci stats will report 3 pool:
 +
 
 +
 
 +
  { 'e.QAclass":"1" ,  'count': 1 }
 +
  { 'e.QAclass":"2" ,  'count': 1 }
 +
  { 'e.QAclass":"3" ,  'count': 1 }
 +
 
 +
* create PCI flavor
 +
 
 +
  nova pci-flavor-create  name 'QA1'  description  'QuickAssist card version 1'
 +
  nova pci-flavor-update  name 'QA1'    set 'e.QAclass"="1"
 +
  nova pci-flavor-create  name 'QA13'  description 'QuickAssist card version 1 and version 3'
 +
  nova pci-flavor-update  name 'QA13'    set 'e.QAclass"="(1|3)"
 +
 
 +
* create flavor and boot with it
 +
  nova boot  mytest  --flavor m1.tiny  --image=QA1_image  --nic  net-id=network_X  pci_flavor= '1:QA1;'
 +
  nova boot  mytest  --flavor m1.tiny  --image=QA13_image  --nic  net-id=network_X  pci_flavor= '3:QA13;'
 +
 
 +
===Common PCI SRIOV Configuration detail ===
  
 
====Compute host====
 
====Compute host====
−
pci_information = { pci-regex,pci-extra-attrs }
+
pci_information = [ {pci-regex},{pci-extra-attrs} ]
 +
pci_flavor_attrs=attr,attr,attr
  
−
====Control node====
 
−
pci_flavor_attrs=[attr,attr,attr]
 
 
For instance, when using device and vendor ID this would read:
 
For instance, when using device and vendor ID this would read:
 
     pci_flavor_attrs=device_id,vendor_id
 
     pci_flavor_attrs=device_id,vendor_id
−
When the backend adds an arbitrary ‘group’ attribute to all PCI devices:
+
When the back end adds an arbitrary ‘group’ attribute to all PCI devices:
 
     pci_flavor_attrs=e.group
 
     pci_flavor_attrs=e.group
 
When you wish to find an appropriate device and perhaps also filter by the connection tagged on that device, which you use an extra-info attribute to specify on the compute node config:
 
When you wish to find an appropriate device and perhaps also filter by the connection tagged on that device, which you use an extra-info attribute to specify on the compute node config:
 
pci_flavor_attrs=device_id,vendor_id,e.connection
 
pci_flavor_attrs=device_id,vendor_id,e.connection
 +
  
 
====flavor API====
 
====flavor API====
Line 82: Line 326:
  
 
* get detailed information about one pci-flavor:  
 
* get detailed information about one pci-flavor:  
−
      nova pci-flavor-show  <UUID>
+
    nova pci-flavor-show  <UUID|name>
−
     GET v2/​{tenant_id}​/os-pci-flavor/<UUID>
+
     GET v2/​{tenant_id}​/os-pci-flavor/<UUID|name>
 
     data:
 
     data:
 
         os-pci-flavor: {  
 
         os-pci-flavor: {  
Line 123: Line 367:
 
   DELETE v2/​{tenant_id}​/os-pci-flavor/<UUID>
 
   DELETE v2/​{tenant_id}​/os-pci-flavor/<UUID>
  
−
====nova command extension : --nic with pci-flavor ====
+
=== Current PCI implementation gaps ===
  
−
attaches a virtual NIC to the Neutron network and the VM
+
concept introduce here:
−
nova boot --nic net-id=neutron-network,vnic-type=macvtap,pci-flavor=xxx
+
spec: a filter defined by (k,v) pairs, which k in the pci object fields, this means those (k,v) is the pci device property like: vendor_id, 'address', pci-type etc.
−
nova boot --nic net-id=neutron-network,vnic-type=macvtap,pci-flavor=xxx
+
extra_spec: the filter defined by (k, v) and k not in the pci object fields.
−
 
 
−
===Use cases ===
 
  
−
==== General PCI pass through  =====
 
−
given compute nodes contain 1 GPU with vendor:device 8086:0001
 
  
−
*on the compute nodes, config the pci_information
+
====pci utils support extra property ====
−
    pci_information = { { 'device_id': "8086", 'vendor_id': "0001" }, {} }
+
      * pci utils k,v match support the address reduce regular expression
−
 
+
      * uitils provide a global extract interface to extract base property and extra property of a pci device.
−
* on controller
+
      * extra information also should use schema 'e.name'
−
  pci_flavor_attrs = ['device_id', 'vendor_id']
 
−
 
 
−
the compute node would report PCI stats group by ('device_id', 'vendor_id').
 
−
pci stats will report one pool:  {'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
 
−
 
 
−
* create PCI flavor
 
−
nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 
−
nova pci-flavor-update  name 'bigGPU'  set    'vendor_id'='8086'  'product_id': '0001'
 
−
 
 
−
* create flavor and boot with it
 
−
nova flavor-key m1.small set pci_passthrough:pci_flavor= 1:bigGPU
 
−
nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec
 
−
 
 
−
==== General PCI pass through with multi PCI flavor candidate =====
 
−
 
 
−
given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002
 
−
 
 
−
*on the compute nodes, config the pci_information
 
−
    pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, {} }
 
−
 
 
−
* on controller
 
−
  pci_flavor_attrs = ['device_id', 'vendor_id']
 
−
 
 
−
the compute node would report PCI stats group by ('device_id', 'vendor_id').
 
−
pci stats will report 2 pool:
 
−
{'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
 
−
{'device_id':'0002', 'vendor_id':'8086', 'count': 1 }
 
−
 
 
−
* create PCI flavor
 
−
nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 
−
nova pci-flavor-update  name 'bigGPU'  set    'vendor_id'='8086'  'product_id': '0001'
 
−
 
 
−
nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 
−
nova pci-flavor-update  name 'bigGPU2'  set    'vendor_id'='8086'  'product_id': '0002'
 
−
 
 
−
* create flavor and boot with it
 
−
nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU,bigGPU2;'
 
−
nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec
 
−
 
 
−
==== General PCI pass through wild-cast PCI flavor =====
 
−
 
 
−
given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002
 
−
 
 
−
*on the compute nodes, config the pci_information
 
−
    pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, {} }
 
−
 
 
−
* on controller
 
−
  pci_flavor_attrs = ['device_id', 'vendor_id']
 
−
 
 
−
the compute node would report PCI stats group by ('device_id', 'vendor_id').
 
−
pci stats will report 2 pool:
 
−
{'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
 
−
{'device_id':'0002', 'vendor_id':'8086', 'count': 1 }
 
−
 
 
−
* create PCI flavor
 
−
nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 
−
nova pci-flavor-update  name 'bigGPU'  set    'vendor_id'='8086'  'product_id': '000[1-2]'
 
−
 
 
−
* create flavor and boot with it
 
−
nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU;'
 
−
nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec
 
−
 
 
−
====  PCI pass through support grouping tag =====
 
−
 
 
−
given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002
 
−
 
 
−
*on the compute nodes, config the pci_information
 
−
    pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, {} }
 
−
 
 
−
* on controller
 
−
  pci_flavor_attrs = ['device_id', 'vendor_id']
 
−
 
 
−
the compute node would report PCI stats group by ('device_id', 'vendor_id').
 
−
pci stats will report 2 pool:
 
−
{'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
 
−
{'device_id':'0002', 'vendor_id':'8086', 'count': 1 }
 
  
−
* create PCI flavor
+
====PCI information(extended the white-list) support extra tag====
−
nova pci-flavor-create name 'bigGPU' description 'passthrough Intel's on-die GPU'
+
      * PCI information support  reduce regular expression compare, match the pci device
−
nova pci-flavor-update  name 'bigGPU'  set    'vendor_id'='8086'  'product_id': '000[1-2]'
+
      * PCI information support store any other (k,v) pair pci device's extra info
 +
      * any extra tag's k, v is string.
  
−
* create flavor and boot with it
+
====pci_flavor_attrs====
−
nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU;'
+
      * implement the attrs parser, updated to flavor Database
−
nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec
 
  
 +
====support pci-flavor ====
 +
      * pci-flavor store in DB
 +
      * pci-flavor config via API
 +
      * pci manager use extract method extract the specs and extra_specs, match them against  the pci object & object.extra_info.
  
−
<ijw> In this, pci_attrs is 'group' and we're adding a 'group' attr at the pci_information stage
+
====PCI scheduler: PCI filter ====
−
* sarob 已退出(Ping timeout: 252 seconds)
+
    When scheduling, marcher should  applied  regular expression stored in the named flavor, this read out from DB.  
−
<ijw> So, pci_information={ { device_id: 1, vendor_id: 0x8086}, { group => 'gpu' } }
 
−
<baoli> what is device_id?
 
−
<ijw> The device ID of the whitelisted PCI device
 
−
<ijw> Do I mean device ID?
 
−
<ijw> Yes, that's the right term
 
−
* ruhe 改名为 ruhe_away
 
−
<baoli> but we don't have a device_id
 
−
<heyongli> product_id , i think it is
 
−
<heyongli> the standard pci property
 
−
<ijw> OK, sorry - I'm looking at a (non-Openstack) page which calls it device ID, my apologies
 
−
<ijw> So, compute host gets 'group' for attrs when it requests it
 
−
<ijw> And makes one PCI stat per 'group' value, rather than unique (vendor_id, product_id) tuple and reports that to the scheduler in this case
 
−
<ijw> PCI flavor is going to be name gpu match expression { group: 'gpu' }
 
−
<ijw> Scheduling is as before
 
−
<ijw> Sorry, it will be 'e.group': 'gpu' per the spec about marking extra info when we use it
 
−
* ruhe_away 改名为 ruhe
 
−
<ijw> Difficult cases.
 
−
<ijw> Scheduling is harder when PCI flavors are vague and match multiple pci_stats rows, and particularly when they overlap
 
−
* coolsvap (~SwapnilK@14.97.34.59) 进入了 #openstack-meeting-alt
 
−
<ijw> So, if I have two flavors, one is vendor:8086, device: 1 or 2 and one of which is vendor: 8086, device: 1 then you have more problems and the worst case is when you use both flavors in a single instance
 
−
<ijw> So, say I have the above two flavors, 'vague' and 'specific' for the sake of argument, and I want to start an instance with one device from each
 
−
<irenab> ijw: can we go over networking case taking into account phy_net connectivity?
 
−
<ijw> OK, in a sec, I'll finish this case first
 
−
<irenab> ijw: afraid to be out of time
 
−
<ijw> When I try and schedule, let's assume I have one product_id of each type
 
−
* amrith 已退出(Ping timeout: 245 seconds)
 
−
<ijw> Then there's a case that I can't use the device '1' for 'vague' because I would not be able to allocate anything for 'specific'.  The problem is hard, it's not insoluble, you just have to try multiple combinations until you succeed.
 
  
  
−
====transite config file to API ====
+
==== convert pci flavor from SRIOV ====
−
#  the config file for alias and whitelist defination is going to deprecated.
+
in API stage the network parser should convert the pci flavor in the --nic option to pci request and save them into instance meta data.  
−
#  if database is not NULL , configration is ommit and given deprecated  warning.
 
−
#  if database is NULL, config if read from the file,
 
−
    *white list/alias schema still work
 
−
    * And also  given a deprecated notice, alias will fade out  which will be remove start from next release.
 
  
−
with this solution, we move pci config from file to API.
+
1. translate pci flavor spec to request
 +
    *translate_pci_flavor_to_requests(flavor_spec)
 +
    * input flavor_spce is a list of pci flavor and requested number:  "flavor_name:number, flavor_name2:number2,..." if number is 1, it can be omit.
 +
    * output is pci_request, a internal represents data structure
  
−
===DB for pci configration===
+
2. save request to instance meta data:
−
each pci flavor will be a set of (k,v), and the pci flavor don't need to contain same k, v pair. another problem this define try to slove is, i,.e  SRIOV also want feature autodiscovery(under discuss), with this, the flavor might need a 'feature' key to be added if not store it as (k,v) pair.  the (k,v) paire define let more extra infomation can be store in the pci device.
+
    *update_pci_request_to_metadata(metadata, new_requests, prefix='')
  
−
  talbe: pci_flavor{
+
==== find specific device of a instance based on request ====
−
                id  :  data base of this k,v pair
 
−
                UUID:  which pci-flavor the  k,v belong to
 
−
                key
 
−
                value (might be a simple value or reduce Regular express)
 
−
            }
 
  
 +
to boot VM with PCI SRIOV devices, there might need more configuration action to pci device instead of just a pci host dev. to achieve this, common SRIOV need a interface to query the device allocated to a specific usage, like the SRIOV network.
  
−
===Requirements from SRIOV===
+
3 steps facility to achive this:
−
*group device
 
−
  for SRIOV, all VFs belong to same PF share same physical network reachability. so if you want, say, deploy a vlan network, you need choose the right PF's VF, otherwise network does not work for you.  the pci flavor does this work well.
 
−
*mark the device alloced to the the flavor
 
−
  networking or other special deive is not as simple as pass though to the VM, there is need more configration. to acheive this, SRIOV must know the device infomation allocation to the specific flavor.
 
  
−
=== Implement the grouping===
+
    * Mark the PCI request use UUID, so the pci request is distinguishable.
−
concept introduce here:
+
    * remember the devices allocated the this request.
−
spec: a filter defined by (k,v) paris, which k in the pci object fileds, this means those (k,v) is the pci device property like: vendor_id, 'address', pci-type etc.
+
    * a interface function provide to get these device from a instance.
−
extra_spec: the filter defined by (k, v) and k not in the pci object fileds.  
 
  
−
====pci utils/objects support grouping ====
+
====DB for PCI flavor ====
−
      * pci utils k,v match support the address reduce regular expression
+
each pci flavor will be a set of (k,v), store the (k,v) pair in DB.  both k, v is string, and the value could be a simple  regular expression, support wild-cast, address range operators.
−
      * objects provide a class level extrac interface to extract base spec and extra spec
 
  
−
====pci-flavor(white list) support address set====
 
−
      * white list support 'address' reduce regular expresion compare.
 
−
      * white list support  any other (k,v) pair to group or store special infomation
 
−
      * object extrac specs and extra_info, specs use as whitelist spec, extra info will be updated to device's extra_info fields
 
  
−
====enable flavor support pci-flavor ====
 
−
      * pci-flavor's name set in the extra spec in the instance type
 
−
      * pci manager use extrac method extrac the specs and extra_specs, match them agains  the pci object & object.extra_info.
 
  
−
====pci stats grouping device base on pci flavor ====
+
Talbe: pci_flavor
−
        *  current gourping base on  [vendor_id, product_id, extra_info]
 
−
        *  going to use 'pci-flavor' grouping the device.
 
−
        * still keep compatible by default, via a new config option switch to new grouping policy.
 
  
−
=== Implement device mark from the pci-flavor===
+
  {
 +
              UUID: which pci-flavor the k,v belong to
 +
              name: the pci flavor's name, we need this filed to index the DB with flavor's name
 +
              key
 +
              value (might be a simple string value or reduce Regular express)
 +
  }
  
−
here is the idea how user can identify which device allocated for the pci-flavor.
+
DB interface:
  
−
    *while define the flavor, put a  marker(network uuid)  into the flavor then store in the device's extra_info fileds
+
        get_pci_flavor_by_name
−
    *after finished allocation, user can seach a instance's pci devices to find the specific device do further configration
+
        get_pci_flavor_by_UUID
 +
        get_pci_flavor_all
 +
        update_pci_flavor_by_name
 +
        update_pci_flavor_by_UUID
 +
        delete_pci_flavor_by_name
 +
        delete_pci_flavor_by_UUID
  
−
    the way marker data transfer from user to device utilize the pci_request, which convert from the pci-flavor.
+
===transient config file to API ===
 +
    *the config file for alias and white-list definition is going to deprecated.
 +
    *new config pci_information will replace white-list
 +
    *pci flavor will replace alias
 +
    *white list/alias schema still work, and given a deprecated notice, will fade out  which will be remove start from next release.
 +
    *if pci_flavor_attrs is not defined it will default to vendor_id, product_id and extra_info. this is keep compatible with old system.

Latest revision as of 03:21, 27 July 2016

!!!This is a design discussion document, not for end user reference!!!

Related Resource

This design is based on the PCI pass-through IRC meetings, provide common support for PCI SRIOV:

This document was used to finalise the design:

link back to bp:

Common PCI SRIOV design

PCI devices have PCI standard properties like address (BDF), vendor_id, product_id, etc, Virtual functions also have a property referring to the function's physical address. Application specific or installation specific extra information can be attached to PCI device, like physical network connectivity using by Neutron SRIOV.

This bp focus on functionality to provide the common PCI SRIOV support.

  • on compute node the pci_information/white-list define a set of filter. PCI device passed filter will be available for allocation. at same time extra information attached to the PCI device.
  • PCI compute report the PCI stats information to scheduler. PCI stats contain several pools. each pool defined by several PCI property, control by the local configuration item: pci_flavor_attrs , default value is vendor_id, product_id, extra_info.
  • PCI flavor define the user point of view PCI device selector: PCI flavor provide a set of (k,v) to form specs to select the available device(filtered by pci_information/white-list).

the new PCI design based on 2 key change: the PCI flavor and the extra information attache to pci device. the following diagram is a summary of the design:

PCI-sriov.jpg

Design Choice

     PCI flavor:  User don't want to know details of a pci device and all of it's attrs and extra information attached to it.
     (pci_flavor_attr:  admin need to know all PCI attrs to define flavor for user/tenant.)
     PCI Stats:   compute node might have many devices and most of device properties are same,  summary a stats to scheduler can reduce DB load, simply schedule.
     PCI extra info:  a pci device might attach to a specific network, a specific resource, with can be attach to device and schedule base on it.


PCI flavor

For OS users, PCI flavor is a reasonable name like 'oldGPU', 'FastGPU', '10GNIC', 'SSD', describe one kind of PCI device. User use the PCI flavor to select available pci devices. Internally the PCI flavor created by a set of API and saved in a DB table, keep PCI flavor available for all cloud.


Administrator define the PCI flavors via matching expression that selects available(offer by white list) devices， and a reasonable name. PCI flavor matching expression is a set of (k,v), the k is the PCI property, v is its value. not every PCI property is available to PCI flavor, only a selected set of PCI property can used to define the PCI flavor, the selected property should be global to cloud like vendor/product_id, can not be BDF or host of a PCI device. these selected PCI property is defined via compute local configuration : 

    pci_flavor_attrs = vendor_id, product_id, ...

a important behavior is the PCI flavors could overlap - that is, the same device on the same machine may be matched by multiple flavors.

Use PCI flavor in instance flavor extra info

user set pci flavor into instance flavor's extra info to specify how many PCI device/and what type PCI flavor the VM want to boot with. 

   nova flavor-key m1.small set pci_passthrough:pci_flavor= <pci flavor spec list>

   pci flavor spec:
                   num1:flavor1,flavor2
   mean: want <number>'s pci devices from flavor or flavor 
   
   pci flavor spec list:
               <pci flavor spec1>; <pci flavor spec2>


for example: 

   nova flavor-key m1.small set pci_passthrough:pci_flavor= 1:IntelGPU,NvGPU;1:intelQuickAssist;
   
   which define requirements: 
         boot with 1 of IntelGPU or NvGPU, and 1 IntelQuickAssist card.

PCI pci_flavor_attrs

this configuration is keep local to every compute node, this will make deploy process can locally decide what PCI properties this node will exposed. 

   pci_flavor_attrs = vendor_id, product_id, ...

compute node update local pci extra properties to PCI flavor Database, which is accessible by flavor API, provide PCI properties to define flavor.

pci_flavor_attrs store in flavor DB as a normal flavor, it's name "_flavor_attrs", it's UUID use "0": 

   {"vendor_id":"Ture", "product_id":"True", ... }

this flavor contain all available attrs can be used to define pci flavor, list it's content use: 

    nova pci-flavor-show  0
    GET v2/​{tenant_id}​/os-pci-flavor/<0>
    data:
       os-pci-flavor: { 
                               'UUID':'0' , 
                               'description':'Available flavor attrs ' 
                               'name':'_flavor_attrs', 
                               'vendor_id": "True",
                               'product_id": "True",
                                ....
         }

PCI request

PCI request is a internal structure to represent all PCI devices a VM want to have. 

   request = {'count': int(count),
             'spec': [{"vendor_id":"8086", "phynetwork":"phy1"}, ...],
             'alias_name': "Intel.NIC"}

PCI-Request.jpg

Extra information of PCI device

the compute nodes offer available PCI devices for pass-through, since the list of devices doesn't usually change unless someone tinkers with the hardware, this matching expression used to create this list of offered devices is stored in compute node config. 


   *the device information (device ID, vendor ID, BDF etc.) is discovered from the device and stored as part of the PCI device, same as current implement.
   *on the compute node, additional arbitrary extra information, in the form of key-value pairs, can be added to the config and is included in the PCI device

this is achieved by extend the pci white-list to: 


  *pci_information = [ { pci-regex } ,{pci-extra-attrs } ]
  *pci-regex is a dict of { string-key: string-value } pairs , it can only match device properties, like vendor_id, address, product_id,etc.
  *pci-extra-attrs is a dict of { string-key: string-value } pairs.  The values can be arbitrary  The total size of the extra attrs may be restricted. all this extra attrs will be store in the pci device table's extra info field. and the extra attrs should use this naming schema: e.attrname

PCI stats grouping device base on pci_flavor_attrs

PCI stats pool summary the devices of a compute node, and the scheduler use flavor's matching specs select the available host for VM. The stats pool must contain the PCI properties used by PCI flavor. 

  *  current grouping is based on  [vendor_id, product_id, extra_info]
  *  going to group by keys specified by   pci_flavor_attrs.

The algorithm for stats report should meet this request: the one device should only be in one pci stats pool, this mean pci stats can not overlap. this simplifies the scheduler design.

on computer node the pci_flavor_attrs provide the specs for pci stats to group its pool. and the pci_flavor_attrs on control node collection the attrs which can be used to define the pci_flavor. the definition of pci_flavor_attrs on controller should contain all the pci_flavor_attrs's content on every compute node.

  • but compute node report stats pool by a subset of controller's pci_flavor_attrs is acceptable, in such scenario, this means the compute node can only provide the devices with these propertes.*

Use cases

General PCI pass through

given compute nodes contain 1 GPU with vendor:device 8086:0001

  • on the compute nodes, config the pci_information 
   pci_information =  { { 'device_id': "8086", 'vendor_id': "0001" }, {} }

   pci_flavor_attrs ='device_id','vendor_id'

the compute node would report PCI stats group by ('device_id', 'vendor_id'). pci stats will report one pool: 

 {'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
  • create PCI flavor 
 nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 nova pci-flavor-update  name 'bigGPU'   set    'vendor_id'='8086'   'product_id': '0001'
  • create flavor and boot with it ( same as current PCI passthrough)
 nova flavor-key m1.small set pci_passthrough:pci_flavor= 1:bigGPU;
 nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec

General PCI pass through with multi PCI flavor candidate

given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002

  • on the compute nodes, config the pci_information 
   pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, {} }
  • on controller 
  pci_flavor_attrs = ['device_id', 'vendor_id']

the compute node would report PCI stats group by ('device_id', 'vendor_id'). pci stats will report 2 pool: 

 {'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
 {'device_id':'0002', 'vendor_id':'8086', 'count': 1 }
  • create PCI flavor 
 nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 nova pci-flavor-update  name 'bigGPU'   set    'vendor_id'='8086'   'product_id': '0001'
 nova pci-flavor-create  name 'bigGPU2'  description 'passthrough Intel's on-die GPU'
 nova pci-flavor-update  name 'bigGPU2'   set    'vendor_id'='8086'   'product_id': '0002'
  • create flavor and boot with it
 nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU,bigGPU2;'
 nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec

General PCI pass through wild-cast PCI flavor

given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002

  • on the compute nodes, config the pci_information 
   pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, {} }
   pci_flavor_attrs = ['device_id', 'vendor_id']

the compute node would report PCI stats group by ('device_id', 'vendor_id'). pci stats will report 2 pool: 

 {'device_id':'0001', 'vendor_id':'8086', 'count': 1 }
 {'device_id':'0002', 'vendor_id':'8086', 'count': 1 }
  • create PCI flavor 
 nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 nova pci-flavor-update  name 'bigGPU'   set    'vendor_id'='8086'   'product_id': '000[1-2]'
  • create flavor and boot with it
 nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU;'
 nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec

PCI pass through support grouping tag

given compute nodes contain 2 type GPU with , vendor:device 8086:0001, or vendor:device 8086:0002

  • on the compute nodes, config the pci_information 
   pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, { 'e.group':'gpu' } } 

   pci_flavor_attrs = ['e.group']

the compute node would report PCI stats group by ('e.group'). pci stats will report 1 pool: 

{'e.group':'gpu', 'count': 2 }


  • create PCI flavor
 nova pci-flavor-create  name 'bigGPU'  description 'passthrough Intel's on-die GPU'
 nova pci-flavor-update  name 'bigGPU'   set    'e.group'='gpu'
  • create flavor and boot with it
 nova flavor-key m1.small set pci_passthrough:pci_flavor= '1:bigGPU;'
 nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec

PCI SRIOV with tagged flavor

given compute nodes contain 5 PCI NIC , vendor:device 8086:0022, and it connect to physical network "X".

  • on the compute nodes, config the pci_information 
   pci_information =  { { 'device_id': "8086", 'vendor_id': "000[1-2]" }, { 'e.physical_netowrk': 'X' } }

   pci_flavor_attrs = 'e.physical_netowrk'

the compute node would report PCI stats group by ('e.physical_netowrk'). pci stats will report 1 pool: 


 {'e.physical_netowrk':'X', 'count': 1 }


  • create PCI flavor 
 nova pci-flavor-create  name 'phyX_NIC'  description 'passthrough NIC connect to physical network X'
 nova pci-flavor-update  name 'bigGPU'   set    'e.physical_netowrk'='X'


  • create flavor and boot with it
 nova boot  mytest  --flavor m1.tiny  --image=cirros-0.3.1-x86_64-uec  --nic  net-id=network_X  pci_flavor= '1:phyX_NIC;'

encryption card use case

there is 3 encryption card: [ V1-3 here means a vendor_id number.] 

  card 1 (vendor_id is V1, device_id =0xa)  
  card 2 (vendor_id is V1, device_id=0xb)
  card 3 (vendor_id is V2, device_id=0xb)

suppose there is two images. One image only support Card 1 and another image support Card 1/3 (or any other combination of the 3 card type).

  • on the compute nodes, config the pci_information 
   pci_information =  { { 'device_id': "0xa", 'vendor_id': "v1" }, { 'e.QAclass':'1' } }
   pci_information =  { { 'device_id': "0xb", 'vendor_id': "v1" }, { 'e.QAclass':'2' } }
   pci_information =  { { 'device_id': "0xb", 'vendor_id': "v2" }, { 'e.QAclass':'3' }}

   pci_flavor_attrs = ['e.QAclass']

the compute node would report PCI stats group by (['e.QAclass']). pci stats will report 3 pool: 


 { 'e.QAclass":"1" ,  'count': 1 }
 { 'e.QAclass":"2" ,  'count': 1 }
 { 'e.QAclass":"3" ,  'count': 1 }
  • create PCI flavor 
 nova pci-flavor-create  name 'QA1'  description  'QuickAssist card version 1'
 nova pci-flavor-update  name 'QA1'    set 'e.QAclass"="1"
 nova pci-flavor-create  name 'QA13'   description 'QuickAssist card version 1 and version 3'
 nova pci-flavor-update  name 'QA13'    set 'e.QAclass"="(1|3)"
  • create flavor and boot with it
 nova boot  mytest  --flavor m1.tiny  --image=QA1_image  --nic  net-id=network_X  pci_flavor= '1:QA1;'
 nova boot  mytest  --flavor m1.tiny  --image=QA13_image  --nic  net-id=network_X  pci_flavor= '3:QA13;'

Common PCI SRIOV Configuration detail

Compute host

pci_information = [ {pci-regex},{pci-extra-attrs} ] pci_flavor_attrs=attr,attr,attr

For instance, when using device and vendor ID this would read: 

    pci_flavor_attrs=device_id,vendor_id

When the back end adds an arbitrary ‘group’ attribute to all PCI devices: 

    pci_flavor_attrs=e.group

When you wish to find an appropriate device and perhaps also filter by the connection tagged on that device, which you use an extra-info attribute to specify on the compute node config: pci_flavor_attrs=device_id,vendor_id,e.connection


flavor API

  • overall

nova pci-flavor-list nova pci-flavor-show name|UUID <name|UUID> nova pci-flavor-create name|UUID <name|UUID> description <desc> nova pci-flavor-update name|UUID <name|UUID> set 'description'='xxxx' 'e.group'= 'A' nova pci-flavor-delete <name|UUID> name|UUID 


* list available pci flavor  (white list)
   nova pci-flavor-list 
   GET v2/​{tenant_id}​/os-pci-flavors
   data:
    os-pci-flavors{
                [ 
                           { 
                               'UUID':'xxxx-xx-xx' , 
                               'description':'xxxx' 
                               'vendor_id':'8086', 
                                 ....
                                'name':'xxx', 
                              } ,
               ]
    }


  • get detailed information about one pci-flavor: 
    nova pci-flavor-show  <UUID|name>
    GET v2/​{tenant_id}​/os-pci-flavor/<UUID|name>
    data:
       os-pci-flavor: { 
                               'UUID':'xxxx-xx-xx' , 
                               'description':'xxxx' 
                                  ....
                                'name':'xxx', 
         }
  • create pci flavor 
 nova pci-flavor-create  name 'GetMePowerfulldevice'  description "xxxxx"
 API:
 POST  v2/​{tenant_id}​/os-pci-flavors
 data: 
     pci-flavor: { 
            'name':'GetMePowerfulldevice',
             description: "xxxxx" 
     }
 action:  create database entry for this flavor.


  • update the pci flavor 
    nova pci-flavor-update UUID  set    'description'='xxxx'   'e.group'= 'A'
    PUT v2/​{tenant_id}​/os-pci-flavors/<UUID>
    with data  :
        { 'action': "update", 
          'pci-flavor':
                         { 
                            'description':'xxxx',
                            'vendor': '8086',
                            'e.group': 'A',
                             ....
                         }
        }
   action: set this as the new definition of the pci flavor.
  • delete a pci flavor
  nova pci-flavor-delete <UUID>
  DELETE v2/​{tenant_id}​/os-pci-flavor/<UUID>

Current PCI implementation gaps

concept introduce here: spec: a filter defined by (k,v) pairs, which k in the pci object fields, this means those (k,v) is the pci device property like: vendor_id, 'address', pci-type etc. extra_spec: the filter defined by (k, v) and k not in the pci object fields.


pci utils support extra property

      * pci utils k,v match support the address reduce regular expression
      * uitils provide a global extract interface to extract base property and extra property of a pci device.
      * extra information also should use schema 'e.name'

PCI information(extended the white-list) support extra tag

      * PCI information support  reduce regular expression compare, match the pci device 
      * PCI information  support  store any other (k,v) pair pci device's extra info
      * any extra tag's k, v is string.

pci_flavor_attrs

      * implement the attrs parser, updated to flavor Database

support pci-flavor

      * pci-flavor store in DB
      * pci-flavor config via API
      * pci manager use extract method extract the specs and extra_specs, match them against  the pci object & object.extra_info.

PCI scheduler: PCI filter

   When scheduling,  marcher should  applied  regular expression stored in the named flavor, this read out from DB.


convert pci flavor from SRIOV

in API stage the network parser should convert the pci flavor in the --nic option to pci request and save them into instance meta data.

1. translate pci flavor spec to request 

   *translate_pci_flavor_to_requests(flavor_spec)
   * input flavor_spce is a list of pci flavor and requested number:  "flavor_name:number, flavor_name2:number2,..." if number is 1, it can be omit.
   * output is pci_request, a internal represents data structure

2. save request to instance meta data: 

    *update_pci_request_to_metadata(metadata, new_requests, prefix=)

find specific device of a instance based on request

to boot VM with PCI SRIOV devices, there might need more configuration action to pci device instead of just a pci host dev. to achieve this, common SRIOV need a interface to query the device allocated to a specific usage, like the SRIOV network.

3 steps facility to achive this: 

   * Mark the PCI request use UUID, so the pci request is distinguishable.
   * remember the devices allocated the this request.
   * a interface function provide to get these device from a instance.

DB for PCI flavor

each pci flavor will be a set of (k,v), store the (k,v) pair in DB. both k, v is string, and the value could be a simple regular expression, support wild-cast, address range operators.


Talbe: pci_flavor 

 {
              UUID:  which pci-flavor the  k,v belong to
              name: the pci flavor's name, we need this filed to index the DB with flavor's name
              key 
              value (might be a simple string value or reduce Regular express)
  }

DB interface: 

        get_pci_flavor_by_name
        get_pci_flavor_by_UUID
        get_pci_flavor_all
        update_pci_flavor_by_name
        update_pci_flavor_by_UUID
        delete_pci_flavor_by_name
        delete_pci_flavor_by_UUID

transient config file to API

   *the config file for alias and white-list definition is going to deprecated.
   *new config pci_information will replace white-list 
   *pci flavor will replace alias
   *white list/alias schema still work, and given a deprecated notice, will fade out  which will be remove start from next release.
   *if pci_flavor_attrs is not defined it will default to vendor_id, product_id and extra_info. this is keep compatible with old system.
Retrieved from "https://wiki.openstack.org/w/index.php?title=PCI_passthrough_SRIOV_support&oldid=129110"