BXE(4) | FreeBSD Kernel Interfaces Manual | BXE(4) |
NAME
bxe — QLogic NetXtreme II Ethernet 10Gb PCIe adapter driverSYNOPSIS
To compile this driver into the kernel, place the following lines in your kernel configuration file:
Alternatively, to load the driver as a module at boot time, place the following line in loader.conf(5):
if_bxe_load="YES"
DESCRIPTION
The bxe driver provides support for PCIe 10Gb Ethernet adapters based on the QLogic NetXtreme II family of 10Gb chips. The driver supports Jumbo Frames, VLAN tagging, checksum offload (IPv4, TCP, UDP, IPv6-TCP, IPv6-UDP), MSI-X interrupts, TCP Segmentation Offload (TSO), Large Receive Offload (LRO), and Receive Side Scaling (RSS).HARDWARE
The bxe driver provides support for various NICs based on the QLogic NetXtreme II family of 10Gb Ethernet controller chips, including the following:
- QLogic NetXtreme II BCM57710 10Gb
- QLogic NetXtreme II BCM57711 10Gb
- QLogic NetXtreme II BCM57711E 10Gb
- QLogic NetXtreme II BCM57712 10Gb
- QLogic NetXtreme II BCM57712-MF 10Gb
- QLogic NetXtreme II BCM57800 10Gb
- QLogic NetXtreme II BCM57800-MF 10Gb
- QLogic NetXtreme II BCM57810 10Gb
- QLogic NetXtreme II BCM57810-MF 10Gb
- QLogic NetXtreme II BCM57840 10Gb / 20Gb
- QLogic NetXtreme II BCM57840-MF 10Gb
CONFIGURATION
There a number of configuration parameters that can be set to tweak the driver's behavior. These parameters can be set via the loader.conf(5) file to take affect during the next system boot. The following parameters affect ALL instances of the driver.- hw.bxe.debug
-
DEFAULT = 0
- hw.bxe.interrupt_mode
-
DEFAULT = 2
- hw.bxe.queue_count
-
DEFAULT = 4
- hw.bxe.max_rx_bufs
-
DEFAULT = 0
- hw.bxe.hc_rx_ticks
-
DEFAULT = 25
- hw.bxe.hc_tx_ticks
-
DEFAULT = 50
- hw.bxe.rx_budget
-
DEFAULT = 0xffffffff
- hw.bxe.max_aggregation_size
-
DEFAULT = 32768
- hw.bxe.mrrs
-
DEFAULT = -1
- hw.bxe.autogreeen
-
DEFAULT = 0
- hw.bxe.udp_rss
-
DEFAULT = 0
Special care must be taken when modifying the number of queues and receive buffers. FreeBSD imposes a limit on the maximum number of mbuf(9) allocations. If buffer allocations fail, the interface initialization will fail and the interface will not be usable. The driver does not make a best effort for buffer allocations. It is an all or nothing effort.
You can tweak the mbuf(9) allocation limit using sysctl(8) and view the current usage with netstat(1) as follows:
# netstat -m # sysctl kern.ipc.nmbclusters # sysctl kern.ipc.nmbclusters=<#>
There are additional configuration parameters that can be set on a per-instance basis to dynamically override the default configuration. The '#' below must be replaced with the driver instance / interface unit number:
- dev.bxe.#.debug
-
DEFAULT = 0
- dev.bxe.#.rx_budget
-
DEFAULT = 0xffffffff
Additional items can be configured using ifconfig(8):
- MTU - Maximum Transmission Unit
-
DEFAULT = 1500
- Promiscuous Mode
-
DEFAULT = OFF
- Rx/Tx Checksum Offload
-
DEFAULT = RX/TX CSUM ON
- TSO - TCP Segmentation Offload
-
DEFAULT = ON
- LRO - TCP Large Receive Offload
-
DEFAULT = ON
DIAGNOSTICS AND DEBUGGING
There are many statistics exposed by bxe via sysctl(8).To dump the default driver configuration:
# sysctl -a | grep hw.bxe
To dump every instance's configuration and detailed statistics:
# sysctl -a | grep dev.bxe
To dump information for a single instance (replace the '#' with the driver instance / interface unit number):
# sysctl -a | grep dev.bxe.#
To dump information for all the queues of a single instance:
# sysctl -a | grep dev.bxe.#.queue
To dump information for a single queue of a single instance (replace the additional '#' with the queue number):
# sysctl -a | grep dev.bxe.#.queue.#
The bxe driver has the ability to dump a ton of debug messages to the system log. The default level of logging can be set with the hw.bxe.debug sysctl(8). Take care with this setting as it can result in too many logs being dumped. Since this parameter is the default one, it affects every instance and will dramatically change the timing in the driver. A better alternative to aid in debugging is to dynamically change the debug level of a specific instance with the dev.bxe.#.debug sysctl(8). This allows you to turn on/off logging of various debug groups on-the-fly.
The different debug groups that can be toggled are:
DBG_LOAD 0x00000001 /* load and unload */ DBG_INTR 0x00000002 /* interrupt handling */ DBG_SP 0x00000004 /* slowpath handling */ DBG_STATS 0x00000008 /* stats updates */ DBG_TX 0x00000010 /* packet transmit */ DBG_RX 0x00000020 /* packet receive */ DBG_PHY 0x00000040 /* phy/link handling */ DBG_IOCTL 0x00000080 /* ioctl handling */ DBG_MBUF 0x00000100 /* dumping mbuf info */ DBG_REGS 0x00000200 /* register access */ DBG_LRO 0x00000400 /* lro processing */ DBG_ASSERT 0x80000000 /* debug assert */ DBG_ALL 0xFFFFFFFF /* flying monkeys */
For example, to debug an issue in the receive path on bxe0:
# sysctl dev.bxe.0.debug=0x22
When finished turn the logging back off:
# sysctl dev.bxe.0.debug=0
SUPPORT
For support questions please contact your QLogic approved reseller or QLogic Technical Support at http://support.qlogic.com, or by E-mail at <support@qlogic.com>.HISTORY
The bxe device driver first appeared in FreeBSD 9.0.AUTHORS
The bxe driver was written by <edavis@broadcom.com>, <davidch@broadcom.com>, and <zambrano@broadcom.com>.April 29, 2012 | FreeBSD |