Assigning PCIe Endpoint Devices

Starting with the Oracle VM Server for SPARC 2.0 release and the Oracle Solaris 10 9/10 OS, you can assign an individual PCIe endpoint (or direct I/O-assignable) device to a domain. This use of PCIe endpoint devices increases the granularity of the device assignment to I/O domains. This capability is delivered by means of the direct I/O (DIO) feature.

The DIO feature enables you to create more I/O domains than the number of PCIe buses in a system. The possible number of I/O domains is now limited only by the number of PCIe endpoint devices.

A PCIe endpoint device can be one of the following:

A PCIe card in a slot
An on-board PCIe device that is identified by the platform

The following diagram shows that the PCIe endpoint device, PCIE3, is assigned to an I/O domain. Both bus pci_0 and the switch in the I/O domain are virtual. The PCIE3 endpoint device is no longer accessible in the primary domain.

In the I/O domain, the pci_0 block and the switch are a virtual root complex and a virtual PCIe switch, respectively. This block and switch are very similar to the pci_0 block and the switch in the primary domain. In the primary domain, the devices in slot PCIE3 are a shadow form of the original devices and are identified as SUNW,assigned.

Caution - You cannot use Oracle Solaris hot-plug operations to “hot remove” a PCIe endpoint device after that device is removed from the primary domain by using the ldm rm-io command. For information about replacing or removing a PCIe endpoint device, see Making PCIe Hardware Changes.

Figure 6-2 Assigning a PCIe Endpoint Device to an I/O Domain

image:Diagram shows how to assign a PCIe endpoint device to an I/O domain.

Use the ldm list-io command to list the PCIe endpoint devices.

Though the DIO feature permits any PCIe card in a slot to be assigned to an I/O domain, only certain PCIe cards are supported. See Direct I/O Hardware and Software Requirements in Oracle VM Server for SPARC 2.2 Release Notes.

Caution - PCIe cards that have a bridge are not supported. PCIe function-level assignment is also not supported. Assigning an unsupported PCIe card to an I/O domain might result in unpredictable behavior.

The following are a few important details about the DIO feature:

This feature is enabled only when all the software requirements are met. See Direct I/O Hardware and Software Requirements in Oracle VM Server for SPARC 2.2 Release Notes.
Only PCIe endpoints that are connected to a PCIe bus assigned to the primary domain can be assigned to another domain with the DIO feature.
I/O domains using DIO have access to the PCIe endpoint devices only when the primary domain is running.
Rebooting the primary domain affects I/O domains that have PCIe endpoint devices. See Rebooting the primary Domain. The primary domain also has the following responsibilities:
- Initializes the PCIe bus and manages the bus.
- Handles all bus errors that are triggered by the PCIe endpoint devices that are assigned to I/O domains. Note that only the primary domain receives all PCIe bus-related errors.

Direct I/O Hardware and Software Requirements

To successfully use the DIO feature, you must be running the appropriate software and assign only the PCIe cards that are supported by the DIO feature to I/O domains. For the hardware and software requirements, see Direct I/O Hardware and Software Requirements in Oracle VM Server for SPARC 2.2 Release Notes.

Note - All PCIe cards that are supported on a platform are supported in the primary domain. See the documentation for your platform for the list of supported PCIe cards. However, only direct I/O-supported PCIe cards can be assigned to I/O domains.

Current Direct I/O Feature Limitations

For information about how to work around the following limitations, see Planning PCIe Endpoint Device Configuration.

A delayed reconfiguration is initiated when you assign or remove a PCIe endpoint device to or from the primary domain, which means that the changes are applied only after the primary domain reboots.

Rebooting the primary domain affects direct I/O, so carefully plan your direct I/O configuration changes to maximize the direct I/O-related changes to the primary domain and to minimize primary domain reboots.
Assignment or removal of a PCIe endpoint device to any other domain is only permitted when that domain is either stopped or inactive.

Planning PCIe Endpoint Device Configuration

Carefully plan ahead when you assign or remove PCIe endpoint devices to avoid primary domain downtime. The reboot of the primary domain not only affects the services that are available on the primary domain itself, but it also affects the I/O domains that have PCIe endpoint devices assigned. Though the changes to each I/O domain do not affect the other domains, planning ahead helps to minimize the consequences on the services that are provided by that domain.

The delayed reconfiguration is initiated the first time you assign or remove a device. As a result, you can continue to add or remove more devices and then reboot the primary domain only one time to make all the changes take effect.

For an example, see How to Create an I/O Domain by Assigning a PCIe Endpoint Device.

The following describes the general steps you must take to plan and perform a DIO device configuration:

Understand and record your system hardware configuration.

Specifically, record information about the part numbers and other details of the PCIe cards in the system.

Use the ldm list-io -l and prtdiag -v commands to obtain the information and save it for future reference.
Determine which PCIe endpoint devices are required to be in the primary domain.

For example, determine the PCIe endpoint devices that provide access to the following:
- Boot disk device
- Network device
- Other devices that the primary domain offers as services
Remove all PCIe endpoint devices that you might use in I/O domains.

This step helps you to avoid performing subsequent reboot operations on the primary domain, as reboots affect I/O domains.

Use the ldm rm-io command to remove the PCIe endpoint devices. Use pseudonyms rather than device paths to specify the devices to the rm-io and add-io subcommands.

Note - Though the first removal of a PCIe endpoint device might initiate a delayed reconfiguration, you can continue to remove devices. After you have removed all the devices you want, you only need to reboot the primary domain one time to make all the changes take effect.
Save this configuration to the service processor (SP).

Use the ldm add-config command.
Reboot the primary domain to release the PCIe endpoint devices that you removed in Step 3.
Confirm that the PCIe endpoint devices you removed are no longer assigned to the primary domain.

Use the ldm list-io -l command to verify that the devices you removed appear as SUNW,assigned-device in the output.
Assign an available PCIe endpoint device to a guest domain to provide direct access to the physical device.

After you make this assignment, you can no longer migrate the guest domain to another physical system by means of the domain migration feature.
Add or remove a PCIe endpoint device to or from a guest domain.

Use the ldm add-io command.

Minimize the changes to I/O domains by reducing the reboot operations and by avoiding downtime of services offered by that domain.
(Optional) Make changes to the PCIe hardware.

See Making PCIe Hardware Changes.

Rebooting the `primary` Domain

The primary domain is the owner of the PCIe bus and is responsible for initializing and managing the bus. The primary domain must be active and running a version of the Oracle Solaris OS that supports the DIO feature. Shutting down, halting, or rebooting the primary domain interrupts access to the PCIe bus. When the PCIe bus is unavailable, the PCIe devices on that bus are affected and might become unavailable.

The behavior of I/O domains with PCIe endpoint devices is unpredictable when the primary domain is rebooted while those I/O domains are running. For instance, I/O domains with PCIe endpoint devices might panic during or after the reboot. Upon reboot of the primary domain, you would need to manually stop and start each domain.

To work around these issues, perform one of the following steps:

Manually shut down any domains on the system that have PCIe endpoint devices assigned to them before you shut down the primary domain.

This step ensures that these domains are cleanly shut down before you shut down, halt, or reboot the primary domain.

To find all the domains that have PCIe endpoint devices assigned to them, run the ldm list-io command. This command enables you to list the PCIe endpoint devices that have been assigned to domains on the system. So, use this information to help you plan. For a detailed description of this command output, see the ldm(1M) man page.

For each domain found, stop the domain by running the ldm stop command.
Configure a domain dependency relationship between the primary domain and the domains that have PCIe endpoint devices assigned to them.

This dependency relationship ensures that domains with PCIe endpoint devices are automatically restarted when the primary domain reboots for any reason.

Note that this dependency relationship forcibly resets those domains, and they cannot cleanly shut down. However, the dependency relationship does not affect any domains that were manually shut down.
```
# ldm set-domain failure-policy=reset primary
# ldm set-domain master=primary ldom
```

Making PCIe Hardware Changes

The following steps help you avoid misconfiguring the PCIe endpoint assignments. For platform-specific information about installing and removing specific hardware, see the documentation for your platform.

No action is required if you are installing a PCIe card into an empty slot. This PCIe card is automatically owned by the domain that owns the PCIe bus.

To assign the new PCIe card to an I/O domain, use the ldm rm-io command to first remove the card from the primary domain. Then, use the ldm add-io command to assign the card to an I/O domain.
No action is required if a PCIe card is removed from the system and assigned to the primary domain.
To remove a PCIe card that is assigned to an I/O domain, first remove the device from the I/O domain. Then, add the device to the primary domain before you physically remove the device from the system.
To replace a PCIe card that is assigned to an I/O domain, verify that the new card is supported by the DIO feature.

If so, no action is required to automatically assign the new card to the current I/O domain.

If not, first remove that PCIe card from the I/O domain by using the ldm rm-io command. Next, use the ldm add-io command to reassign that PCIe card to the primary domain. Then, physically replace the PCIe card you assigned to the primary domain with a different PCIe card. These steps enable you to avoid a configuration that is unsupported by the DIO feature.

How to Create an I/O Domain by Assigning a PCIe Endpoint Device

Plan all DIO deployments ahead of time to minimize downtime.

Caution - The primary domain loses access to the on-board DVD device if you assign the /SYS/MB/SASHBA1 slot on a SPARC T3-1 or a SPARC T4-1 system to a DIO domain.

The SPARC T3-1 and SPARC T4-1 systems include two DIO slots for on-board storage, which are represented by the /SYS/MB/SASHBA0 and /SYS/MB/SASHBA1 paths. In addition to hosting multi-headed on-board disks, the /SYS/MB/SASHBA1 slot hosts the on-board DVD device. So, if you assign /SYS/MB/SASHBA1 to a DIO domain, the primary domain loses access to the on-board DVD device.

The SPARC T3-2 and SPARC T4-2 systems have a single SASHBA slot that hosts all on-board disks as well as the on-board DVD device. So if you assign SASHBA to a DIO domain, the on-board disks and the on-board DVD device are loaned to the DIO domain and unavailable to the primary domain.

For an example of adding a PCIe endpoint device to create an I/O domain, see Planning PCIe Endpoint Device Configuration.

Note - In this release, it is best use the DefaultFixed NCP to configure datalinks and network interfaces on Oracle Solaris 11 systems.

The Oracle Solaris 11 OS includes the following NCPs:

DefaultFixed. Enables you to use the dladm or ipadm command to manage networking.
Automatic. Enables you to use the netcfg or netadm command to manage networking.

Ensure that the DefaultFixed NCP is enabled by using the netadm list command. See Chapter 7, Using Datalink and Interface Configuration Commands on Profiles, in Oracle Solaris Administration: Network Interfaces and Network Virtualization.

Identify and archive the devices that are currently installed on the system.

The output of the ldm list-io -l command shows how the I/O devices are currently configured. You can obtain more detailed information by using the prtdiag -v command.

Note - After the devices are assigned to I/O domains, the identity of the devices can only be determined in the I/O domains.

# ldm list-io -l
NAME                                      TYPE   DOMAIN   STATUS
----                                      ----   ------   ------
pci_0                                     BUS    primary  
[pci@400]
pci_1                                     BUS    primary  
[pci@500]
PCIE1                                     PCIE   -        EMP
[pci@400/pci@0/pci@c]
PCIE2                                     PCIE   primary  OCC
[pci@400/pci@0/pci@9]
    network@0
    network@0,1
    network@0,2
    network@0,3
PCIE3                                     PCIE   primary  OCC
[pci@400/pci@0/pci@d]
    SUNW,emlxs/fp/disk
    SUNW,emlxs@0,1/fp/disk
    SUNW,emlxs@0,1/fp@0,0
MB/SASHBA                                 PCIE   primary  OCC
[pci@400/pci@0/pci@8]
    scsi@0/tape
    scsi@0/disk
    scsi@0/sd@0,0
    scsi@0/sd@1,0
PCIE0                                     PCIE   -        EMP
[pci@500/pci@0/pci@9]
PCIE4                                     PCIE   primary  OCC
[pci@500/pci@0/pci@d]
    network@0
    network@0,1
PCIE5                                     PCIE   primary  OCC
[pci@500/pci@0/pci@c]
    SUNW,qlc@0/fp/disk
    SUNW,qlc@0/fp@0,0
    SUNW,qlc@0,1/fp/disk
    SUNW,qlc@0,1/fp@0,0
MB/NET0                                   PCIE   primary  OCC
[pci@500/pci@0/pci@8]
    network@0
    network@0,1
    network@0,2
    network@0,3

Determine the device path of the boot disk, which needs to be retained.

For UFS file systems, run the df / command to determine the device path of the boot disk.
```
primary# df /
/                  (/dev/dsk/c0t1d0s0 ): 1309384 blocks   457028 files
```

For ZFS file systems, first run the df / command to determine the pool name, and then run the zpool status command to determine the device path of the boot disk.

primary# df /
/                  (rpool/ROOT/s10s_u8wos_08a):245176332 blocks 245176332 files
primary# zpool status rpool
 zpool status rpool
  pool: rpool
 state: ONLINE
 scrub: none requested
config:

        NAME        STATE     READ WRITE CKSUM
        rpool       ONLINE       0     0     0
          c0t1d0s0  ONLINE       0     0     0

Determine the physical device to which the block device is linked.
The following example uses block device c0t1d0s0:
```
primary# ls -l /dev/dsk/c0t1d0s0
lrwxrwxrwx    1 root    root    49 Jul 20 22:17 /dev/dsk/c0t1d0s0 ->
../../devices/pci@400/pci@0/pci@8/scsi@0/sd@0,0:a
```
In this example, the physical device for the primary domain's boot disk is connected to the PCIe endpoint device (pci@400/pci@0/pci@8), which corresponds to the listing of MB/SASHBA in Step 1. Removing this device will prevent the primary domain from booting, so do not remove this device from the primary domain.

Determine the network interface that is used by the system.

Oracle Solaris 10 OS. Use the ifconfig command to determine the network interface that is used by the primary domain, which is nxge0 in this example.

# ifconfig -a
lo0: flags=2001000849<UP,LOOPBACK,RUNNING,MULTICAST,IPv4,VIRTUAL> mtu 8232 index 1
        inet 127.0.0.1 netmask ff000000 
nxge0: flags=1004843<UP,BROADCAST,RUNNING,MULTICAST,DHCP,IPv4> mtu 1500 index 2
        inet 10.6.212.149 netmask fffffe00 broadcast 10.6.213.255
        ether 0:21:28:4:27:cc

Oracle Solaris 11 OS. Use the ipadm command to determine the network interface that is used by the primary domain, which is net0 in this example.

# ipadm show-addr
ADDROBJ     TYPE     STATE     ADDR
lo0/v4         static     ok     127.0.0.1/8
net0/ipv4     static     ok     10.6.212.149/23

Determine the physical device to which the network interface is linked.
The following command uses the nxge0 network interface:
```
primary# ls -l /dev/nxge0
lrwxrwxrwx   1 root     root          46 Jul 30 17:29 /dev/nxge0 ->
../devices/pci@500/pci@0/pci@8/network@0:nxge0
```
In this example, the physical device for the network interface used by the primary domain is connected to the PCIe endpoint device (pci@500/pci@0/pci@8), which corresponds to the listing of MB/NET0 in Step 1. So, you do not want to remove this device from the primary domain. You can safely assign all other PCIe devices to other domains because they are not used by the primary domain.
If the network interface used by the primary domain is on a bus that you want to assign to another domain, the primary domain would need to be reconfigured to use a different network interface.

Remove the PCIe endpoint devices that you might use in I/O domains.

In this example, you can remove the PCIE2, PCIE3, PCIE4, and PCIE5 endpoint devices because they are not being used by the primary domain.

Remove the PCIe endpoint devices.

Caution - Do not remove the devices that are used or required by the primary domain.

If you mistakenly remove the wrong devices, use the ldm cancel-reconf primary command to cancel the delayed reconfiguration on the primary domain.

You can remove multiple devices at one time to avoid multiple reboots.

# ldm rm-io PCIE2 primary
Initiating a delayed reconfiguration operation on the primary domain.
All configuration changes for other domains are disabled until the primary
domain reboots, at which time the new configuration for the primary domain
will also take effect.
# ldm rm-io PCIE3 primary
------------------------------------------------------------------------------
Notice: The primary domain is in the process of a delayed reconfiguration.
Any changes made to the primary domain will only take effect after it reboots.
------------------------------------------------------------------------------
# ldm rm-io PCIE4 primary
------------------------------------------------------------------------------
Notice: The primary domain is in the process of a delayed reconfiguration.
Any changes made to the primary domain will only take effect after it reboots.
------------------------------------------------------------------------------
# ldm rm-io PCIE5 primary
------------------------------------------------------------------------------
Notice: The primary domain is in the process of a delayed reconfiguration.
Any changes made to the primary domain will only take effect after it reboots.
------------------------------------------------------------------------------

Save the new configuration to the service processor (SP).
The following command saves the configuration in a file called dio:
```
# ldm add-config dio
```
Reboot the system to reflect the removal of the PCIe endpoint devices.
```
# reboot -- -r
```

Log in to the primary domain and verify that the PCIe endpoint devices are no longer assigned to the domain.

# ldm list-io
NAME                                      TYPE   DOMAIN   STATUS
----                                      ----   ------   ------
pci_0                                     BUS    primary  
pci_1                                     BUS    primary  
PCIE1                                     PCIE   -        EMP
PCIE2                                     PCIE            OCC
PCIE3                                     PCIE            OCC
MB/SASHBA                                 PCIE   primary  OCC
PCIE0                                     PCIE   -        EMP
PCIE4                                     PCIE            OCC
PCIE5                                     PCIE            OCC
MB/NET0                                   PCIE   primary  OCC

Note - The ldm list-io -l output might show SUNW,assigned-device for the PCIe endpoint devices that were removed. Actual information is no longer available from the primary domain, but the domain to which the device is assigned has this information.

Assign a PCIe endpoint device to a domain.
1. Add the PCIE2 device to the ldg1 domain.
```
# ldm add-io PCIE2 ldg1
```
2. Bind and start the ldg1 domain.
```
# ldm bind ldg1
# ldm start ldg1
LDom ldg1 started
```

Log in to the ldg1 domain and verify that the device is available for use.

Verify that the network device is available and then configure the network device for use in the domain.

Oracle Solaris 10. Run the following command:

# dladm show-dev
vnet0           link: up        speed: 0     Mbps       duplex: unknown
nxge0           link: unknown   speed: 0     Mbps       duplex: unknown
nxge1           link: unknown   speed: 0     Mbps       duplex: unknown
nxge2           link: unknown   speed: 0     Mbps       duplex: unknown
nxge3           link: unknown   speed: 0     Mbps       duplex: unknown

Oracle Solaris 11. Run the following command:

# dladm show-phys
LINK              MEDIA                STATE      SPEED  DUPLEX    DEVICE
net0              Ethernet             up         0      unknown   vnet0
net1              Ethernet             unknown    0      unknown   vsw0
net2              Ethernet             unknown    0      unknown   nxge0

Skip Navigation Links
Exit Print View
	Oracle VM Server for SPARC 2.2 Administration Guide Oracle VM Server for SPARC