|Main index||Section 4||Options|
To load the driver as a module at boot time, place the following line in loader.conf(5):
The driver allocates 1 receive queue, transmit queue, event queue and IRQ per CPU up to a maximum of 64. IRQ affinities should be spread out using cpuset(1). Interrupt moderation may be controlled through the sysctl dev.sfxge.%d.int_mod (units are microseconds).
For more information on configuring this device, see ifconfig(8).
A large number of MAC, PHY and data path statistics are available under the sysctl dev.sfxge.%d.stats. The adapter's VPD fields including its serial number are available under the sysctl dev.sfxge.%d.vpd.
|The maximum number of descriptors in a receive queue ring. Supported values are: 512, 1024, 2048 and 4096.|
|The maximum number of descriptors in a transmit queue ring. Supported values are: 512, 1024, 2048 and 4096.|
|The maximum length of the deferred packet "get-list" for queued transmit packets (TCP and non-TCP), used only if the transmit queue lock can be acquired. If a packet is dropped, the tx_get_overflow counter is incremented and the local sender receives ENOBUFS. The value must be greater than 0.|
|The maximum number of non-TCP packets in the deferred packet "get-list" , used only if the transmit queue lock can be acquired. If a packet is dropped, the tx_get_non_tcp_overflow counter is incremented and the local sender receives ENOBUFS. The value must be greater than 0.|
|The maximum length of the deferred packet "put-list" for queued transmit packets, used if the transmit queue lock cannot be acquired. If a packet is dropped, the tx_put_overflow counter is incremented and the local sender receives ENOBUFS. The value must be greater than or equal to 0.|
|Bitmask to enable/disable usage of FW-assisted TSO version if supported by NIC firmware. FATSOv1 (bit 0) and FATSOv2 (bit 1) are supported. All enabled by default.|
|The maximum number of allocated RSS channels for the Nth adapter. If set to 0 or unset, the number of channels is determined by the number of CPU cores.|
|Size of the LRO hash table. Must be a power of 2. A larger table means we can accelerate a larger number of streams.|
|The maximum length of a hash chain. If chains get too long then the lookup time increases and may exceed the benefit of LRO.|
|The maximum time (in ticks) that a connection can be idle before it's LRO state is discarded.|
|Number of packets with payload that must arrive in-order before a connection is eligible for LRO. The idea is we should avoid coalescing segments when the sender is in slow-start because reducing the ACK rate can damage performance.|
|Number of packets with payload that must arrive in-order following loss before a connection is eligible for LRO. The idea is we should avoid coalescing segments when the sender is recovering from loss, because reducing the ACK rate can damage performance.|
|Enable logging of MCDI protocol messages (only available if enabled at compile-time).|
|Enable or disable logging of MCDI protocol messages on a per-port basis. The default for each port will be the value of hw.sfxge.mcdi_logging. The logging may also be enabled or disabled after the driver is loaded using the sysctl dev.sfxge.%d.mcdi_logging.|
|Period in milliseconds to refresh interface statistics from hardware. The accepted range is 0 to 65535, the default is 1000 (1 second). Use zero value to disable periodic statistics update. Supported on SFN8xxx series adapters with firmware v126.96.36.1993 and later and SFN5xxx and SFN6xxx series adapters. SFN7xxx series adapters and SFN8xxx series with earlier firmware use a fixed 1000 milliseconds statistics update period. The period may also be changed after the driver is loaded using the sysctl dev.sfxge.%d.stats_update_period_ms.|
|SFXGE (4)||February 22, 2015|
|Main index||Section 4||Options|
|“||To err is human...to really foul up requires the root password.||”|