iflib(4)

NAME

iflib – Network Interface Driver Framework

SYNOPSIS

device pci device iflib

DESCRIPTION

iflib is a framework for network interface drivers for FreeBSD. It is designed to remove a large amount of the boilerplate that is often needed for modern network interface devices, allowing driver authors to focus on the specific code needed for their hardware. This allows for a shared set of sysctl(8) names, rather than each driver naming them individually.

SYSCTL VARIABLES

These variables must be set before loading the driver, either via loader.conf(5) or through the use of kenv(1). They are all prefixed by dev.X.Y.iflib. where X is the driver name, and Y is the instance number.

override_nrxds: Override the number of RX descriptors for each queue. The value is a comma separated list of positive integers. Some drivers only use a single value, but others may use more. These numbers must be powers of two, and zero means to use the default. Individual drivers may have additional restrictions on allowable values. Defaults to all zeros.
override_ntxds: Override the number of TX descriptors for each queue. The value is a comma separated list of positive integers. Some drivers only use a single value, but others may use more. These numbers must be powers of two, and zero means to use the default. Individual drivers may have additional restrictions on allowable values. Defaults to all zeros.
override_qs_enable: When set, allows the number of transmit and receive queues to be different. If not set, the lower of the number of TX or RX queues will be used for both.
override_nrxqs: Set the number of RX queues. If zero, the number of RX queues is derived from the number of cores on the socket connected to the controller. Defaults to 0.
override_ntxqs: Set the number of TX queues. If zero, the number of TX queues is derived from the number of cores on the socket connected to the controller.
disable_msix: Disables MSI-X interrupts for the device.
core_offset: Specifies a starting core offset to assign queues to. If the value is unspecified or 65535, cores are assigned sequentially across controllers.
separate_txrx: Requests that RX and TX queues not be paired on the same core. If this is zero or not set, an RX and TX queue pair will be assigned to each core. When set to a non-zero value, TX queues are assigned to cores following the last RX queue.
simple_tx: When set to one, iflib uses a simple transmit routine that sends packets directly while the transmit queue mutex is available. During contention, packets are deferred through a multi-producer buf_ring(9) and drained by the mutex holder. By default, iflib uses a highly optimized, lockless, transmit queue called mp_ring. This performs well when there are more CPU cores than NIC queues and prevents lock contention for transmit resources. Unfortunately, mp_ring incurs unneeded overheads on workloads where resource contention is not a problem (well behaved applications on systems where there are as many NIC queues as CPU cores). Note that when this is enabled, the tx_abdicate sysctl is no longer applicable and is ignored. By default, this is the inverse of net.iflib.prefer_mpring. An explicitly configured value takes precedence over that global default.

These sysctl(8) variables can be changed at any time:

tx_abdicate: Controls how the transmit ring is serviced. If set to zero, when a frame is submitted to the transmission ring, the same task that is submitting it will service the ring unless there's already a task servicing the TX ring. This ensures that whenever there is a pending transmission, the transmit ring is being serviced. This results in higher transmit throughput. If set to a non-zero value, task returns immediately and the transmit ring is serviced by a different task. This returns control to the caller faster and under high receive load, may result in fewer dropped RX frames.
tx_defer_mfree: Controls the threshold in packets before iflib will free the memory (mbufs) for the packets that it has transmitted. When this is nonzero, mbufs will be freed outside the transmit lock. Setting this can reduce lock contention and CPU use when using simple_tx. Note that this applies only when simple_tx is enabled.
tx_reclaim_thresh: Controls the threshold in packets before iflib will ask the driver how many transmitted packets can be reclaimed. Determining how many many packets can be reclaimed can be expensive on some drivers.
tx_reclaim_ticks: Controls the time in ticks before iflib will ask the driver how many transmitted packets can be reclaimed. Determining how many many packets can be reclaimed can be expensive on some drivers.
rx_budget: Sets the maximum number of frames to be received at a time. Zero (the default) indicates the default (currently 16) should be used.

There are also some global sysctls which can change behaviour for all drivers, and, unless noted otherwise, may be changed at any time.

net.iflib.max_producers: Limits the number of callers that may concurrently enqueue packets to a simple transmit deferral ring without taking the transmit queue mutex. The limit is applied separately to each transmit queue. If lockless producers from one last-level cache are active, a producer from another last-level cache waits for the transmit queue mutex regardless of the limit. Setting this to zero forces all contending producers to wait for the mutex. The default is 8. This setting applies only when simple_tx is enabled.
net.iflib.min_tx_latency: If this is set to a non-zero value, iflib will avoid any attempt to combine multiple transmits, and notify the hardware as quickly as possible of new descriptors. This will lower the maximum throughput, but will also lower transmit latency.
net.iflib.no_tx_batch: Some NICs allow processing completed transmit descriptors in batches. Doing so usually increases the transmit throughput by reducing the number of transmit interrupts. Setting this to a non-zero value will disable the use of this feature.
net.iflib.prefer_mpring: When set, use the mp_ring transmit path by default instead of the simple transmit path. This is a read-only tunable that must be set before iflib is loaded. An explicitly configured per-device simple_tx value takes precedence. The default is one.
net.iflib.simple_drain_quota: Sets the maximum number of packets that the transmit task may remove from a simple transmit deferral ring in one drain operation. The default is 8. This setting applies only when simple_tx is enabled.
net.iflib.simple_drain_quota_thread: Sets the maximum number of packets that a transmitting thread holding the transmit queue mutex may remove from a simple transmit deferral ring in one drain operation. A smaller value bounds the time for which a transmitting thread can hold the mutex while other producers replenish the ring. The default is 65536. This setting applies only when simple_tx is enabled.
net.iflib.simple_txbr_size: Sets the number of entries in each buf_ring(9) used by the simple transmit path to defer packets. The value must be a power of two and at least 64. Invalid values are replaced with the default of 1024. This is a read-only tunable that must be set before iflib is loaded. This setting applies only when simple_tx is enabled.
net.iflib.tx_watchdog_periods: Number of consecutive net.iflib.timer_default intervals for which a transmit queue must remain frozen under demand before iflib asks the hardware whether it has completions pending. A queue is frozen when it holds descriptors the hardware has not reported as completed while none are reclaimed. Demand persists while the outstanding descriptor count grows, the software queue is stalled, or the hardware descriptor ring is at its backpressure threshold. If it reports none, the interface is reset. Setting this to zero disables the check.

These sysctl(8) variables are read-only:

driver_version: A string indicating the internal version of the driver.
tx_watchdog_events: Number of transmit watchdog resets initiated by iflib. This counts recovery events, not failed packets, and is not included in the interface output error counter.

There are a number of queue state sysctl(8) variables as well:

txqZ

The following are repeated for each transmit queue, where Z is the transmit queue instance number:

r_abdications: Number of consumer abdications in the MP ring for this queue. An abdication occurs on every ring submission when tx_abdicate is true.
r_restarts: Number of consumer restarts in the MP ring for this queue. A restart occurs when an attempt to drain a non-empty ring fails, and the ring is already in the STALLED state.
r_stalls: Number of consumer stalls in the MP ring for this queue. A stall occurs when an attempt to drain a non-empty ring fails.
r_starts: Number of normal consumer starts in the MP ring for this queue. A start occurs when the MP ring transitions from IDLE to BUSY.
r_drops: Number of drops in the MP ring for this queue. A drop occurs when there is an attempt to add an entry to an MP ring with no available space.
r_enqueues: Number of entries which have been enqueued to the MP ring for this queue.
ring_state: MP (soft) ring state. This provides a snapshot of the current MP ring state, including the producer head and tail indexes, the consumer index, and the state. The state is one of "IDLE", "BUSY", "STALLED", or "ABDICATED".
txq_cleaned: The number of transmit descriptors which have been reclaimed. Total cleaned.
txq_processed: The number of transmit descriptors which have been processed, but may not yet have been reclaimed.
txq_in_use: Descriptors which have been added to the transmit queue, but have not yet been cleaned. This value will include both untransmitted descriptors as well as descriptors which have been processed.
txq_cidx_processed: The transmit queue consumer index of the next descriptor to process.
txq_cidx: The transmit queue consumer index of the oldest descriptor to reclaim.
txq_pidx: The transmit queue producer index where the next descriptor to transmit will be inserted.
no_tx_dma_setup: Number of times DMA mapping a transmit mbuf failed for reasons other than EFBIG.
txd_encap_efbig: Number of times DMA mapping a transmit mbuf failed due to requiring too many segments.
tx_map_failed: Number of times DMA mapping a transmit mbuf failed for any reason (sum of no_tx_dma_setup and txd_encap_efbig)
no_desc_avail: Number of times a descriptor couldn't be added to the transmit ring because the transmit ring was full.
mbuf_defrag_failed: Number of times both m_collapse(9) and m_defrag(9) failed after an EFBIG error result from DMA mapping a transmit mbuf.
m_pullups: Number of times m_pullup(9) was called attempting to parse a header.
mbuf_defrag: Number of times m_defrag(9) was called.

rxqZ

The following are repeated for each receive queue, where Z is the receive queue instance number:

rxq_fl0.credits: Credits currently available in the receive ring.
rxq_fl0.cidx: Current receive ring consumer index.
rxq_fl0.pidx: Current receive ring producer index.

Additional OIDs useful for driver and iflib development are exposed when the INVARIANTS and/or WITNESS options are enabled in the kernel.

FAILURE INJECTION

Registration failures can be injected for one selected device through the fail(9) framework. Set debug.fail_point.iflib.register_device to the exact device name and configure one of these fail points:

register_before_attach_pre: Before the driver's pre-attach method.
register_after_attach_pre: After the driver's pre-attach method succeeds.
register_after_taskqueue: After the private taskqueue is started.
register_after_interrupts: After interrupt resources are allocated.
register_after_queues: After queue setup and core-offset allocation.
register_after_attach_post: After the interface is attached and the driver's post-attach method succeeds.

The fail points are under debug.fail_point.iflib and accept the syntax described by fail(9). A one-shot failure is safest because it permits the next probe to recover the device:

sysctl debug.fail_point.iflib.register_device=igc1
sysctl debug.fail_point.iflib.register_after_attach_pre='1*return(5)'
devctl detach igc1
devctl attach pci0:4:0:0       # returns EIO
devctl attach pci0:4:0:0       # succeeds

After detach, a device may no longer have its former name. Use a bus-specific address, as shown in the PCI example above, to reprobe it.

The normal detach path can be synchronized with an in-flight admin task to test taskqueue teardown. Set debug.fail_point.iflib.admin_task_device to the exact device name, bring the interface up, and set debug.fail_point.iflib.admin_task_after_detach_check to a one-shot bounded delay, such as ‘1*print(1)->1*delay(5000000)’. The delay keeps the admin task active long enough for detach to begin and remain blocked in the taskqueue drain. As a safety backstop, this fail point is marked nonsleepable and converts a configured sleep action to a delay. Once the execution message appears in the system log, detach should wait for the delay to finish. This exercises the case where an admin task has passed its detach check but has not yet acquired the context lock.

HISTORY

This framework was introduced in FreeBSD 11.0.