ixgbe: Apply DMAC changes through an if-up reset request
Serialize cached DMA coalescing policy with initialization and schedule
its application through the admin task. The old running check
preceded if_init() acquiring the context lock, so an intervening down
operation could be followed by an unconditional initialization.
Use the deferred if-up request so restart permission is checked when the
task runs. Changes made while stopped or suspended remain cached for
the next initialization. Do not schedule a reset for an unchanged value.
MFC after: 2 weeks
Sponsored by: BBOX.io
enic: Report device enable failures to iflib
Do not publish a running interface after vnic_dev_enable_wait() fails.
Run the existing stop cleanup for the queues configured before the enable
request, then report initialization failure to iflib.
Mark those queues as needing cleanup before submitting the enable request
so the stopped-state shortcut cannot skip the unwind.
MFC after: 2 weeks
Sponsored by: BBOX.io
enic: Clean the completion ring matching each queue
The transmit stop loop cleaned the receive completion ring, and the
receive stop loop cleaned the transmit completion ring. Use enic_cq_wq()
for transmit queues and enic_cq_rq() for receive queues, matching the
queue indices used by the remaining ring bookkeeping.
MFC after: 2 weeks
Sponsored by: BBOX.io
ice: Honor iflib transmit completion batching
ice marked every transmitted packet RS and recorded every last
descriptor in its report-status queue. Hardware therefore wrote
descriptor status for every packet, and the driver traversed every
packet while reclaiming completed descriptors.
iflib marks selected packets with IPI_TX_INTR as completion
checkpoints. It forces a checkpoint as deferred work or ring pressure
grows. Retain EOP on every packet, but set RS and record the descriptor
only at those checkpoints.
DPDK uses the same sparse-RS design and defaults tx_rs_thresh to 32.
Let iflib choose the adaptive interval for FreeBSD. This reduces PCIe
and memory traffic while preserving bounded descriptor reclamation.
Validated on an E810-XXV in an A-B-A test with five matched
four-stream, TSO-disabled transmit runs per phase. Median throughput
was 9.413, 9.413, and 9.414 Gbps. Median whole-system CPU was 21.54%,
[11 lines not shown]
ice: Defer RDMA critical error notifications
ice_msix_admin() runs as an interrupt filter inside a critical section.
ice_rdma_notify_pe_intr() acquires the global RDMA sx and invokes the
client event handler, both of which require sleepable thread context. A
PE or HMC critical error could therefore panic under WITNESS or sleep
from interrupt context.
Accumulate OICR causes atomically in the interrupt filter and mark them
pending in the driver state. Deliver the notification from the iflib
admin task before processing reset events. This preserves the existing
ordering, lets an iRDMA-requested reset run in the same admin pass, and
coalesces causes from multiple interrupts.
MFC after: 2 weeks
Sponsored by: BBOX.io
Differential Revision: https://reviews.freebsd.org/D59340
ping: do a better job with what we received from the net, part 2
When we see a difference between the payload we sent and what we
received, we dump both but we were not prepared for the case when
the received payload is less than we sent. In this case we were trying
to dump more than needed.
Funny enough, 23 years ago I already fixed a similar issue here but
didn't pay attention to this small dumping loop.
Test written by jlduran.
Reviewed by: jlduran
MFC after: 1 month
Found with: Claude Code Sonnet 5
Differential Revision: https://reviews.freebsd.org/D59556
Differential Revision: https://reviews.freebsd.org/D59582
ping: do a better job checking what we receive from the net
- zero out a buffer for the incoming icmp message that
we are about to parse
- for ICMP_MASKREPLY and ICMP_TSTAMPREPLY responses check their length
and warn and reject them if they are truncated
Without these checks a part of stack allocated struct icmp icp could
be printed out which seems low severity since we already dropped root
privileges by the time icp is allocated.
Reviewed by: glebius
MFC after: 1 month
Found with: Claude Code Sonnet 5
Differential Revision: https://reviews.freebsd.org/D59555
iavf: Honor PF-negotiated RSS sizes
Use the key and lookup table lengths returned by GET_VF_RESOURCES when
configuring RSS through virtchnl, as DPDK does. The Windows E835 PF
advertises a 40-byte key and rejects our fixed 52-byte CONFIG_RSS_KEY
request, leaving receive traffic on queue zero.
Validate the negotiated lengths before constructing AdminQ messages and
publish the lookup table size to iflib. Preserve register-mode RSS
selection and its fixed hardware sizes. Use aligned, zero initialized
key storage so an RSS kernel's 40-byte key does not leave an uninitialized
tail when the PF requests 52 bytes.
Validation: normal and RSS enabled iavf module builds passed. On an E835
VF under Hyper-V Server 2025, repeated IPv4 and IPv6 receive tests used all
three configured guest RX queues in both transparent hn and non-transparent
lagg modes. The RSS key rejection disappeared, IPv4 transmit tests passed,
and no TX watchdog fired. Each traffic case used three runs of 16 streams.
[3 lines not shown]
thunderbolt: Account for CRC in router config write message size
Fixes: 9c6e9bfb3474 ("thunderbolt: Support writing to router config space")
Sponsored by: The FreeBSD Foundation
hn: Forward capability changes to transparent VFs
The transparent VF capability handler ignored the requested change and
only copied the VF's enabled capabilities. Forward SIOCSIFCAP to the VF
and return its result. Preserve VF capabilities which hn does not expose.
Limit advertised capabilities to those supported by the transparent
packet path and VLAN relay. Do not inherit VF services such as send tags
or the extended capability ioctl when hn has no corresponding methods.
At handoff, adopt the VF's enabled offloads without reconfiguring it.
Mark the datapath as switching while the VF applies a capability change,
since its ioctl may reinitialize the device. If the association is still
ready and unchanged afterwards, synchronize hn with the actual VF state
even on error and restrict checksum assistance to the forwarded offloads.
Republish link state suppressed during the transition when the VF is
still ready. Refresh VLAN child capabilities after adoption and when
restoring the synthetic path.
[9 lines not shown]
hn: Relay VLAN membership to transparent VFs
VLANs configured on the synthetic interface do not otherwise reach the
accelerating VF's VLAN callbacks. Those callbacks can be needed for
hardware filter membership or for interpreting stripped VLAN tags, even
though no vlan interface is attached directly to the VF.
Subscribe to VLAN events only in transparent mode and schedule the
existing VF task. Snapshot the synthetic interface's VLAN topology under
network epoch, then leave epoch before invoking the VF callbacks, which
may sleep. Do not acquire hn_lock or configure the VF from a VLAN event
handler; the worker applies membership outside the VLAN configuration
lock.
Keep an applied-VID bitmap under hn_lock and relay only changes. Replay
VLANs configured before VF arrival, reconcile changes while acceleration
is active, and preserve membership across temporary datapath switches.
This relays guest intent; it does not configure host access VLAN policy
or overcome PF restrictions on tagged traffic.
[7 lines not shown]
ice: Use sleepable locks in shared code
The Intel shared code can wait for firmware resources while holding its
OS abstraction locks. FreeBSD mapped these locks to mutexes, which
cannot be held across a voluntary sleep. Concurrent PF rebuilds
therefore trigger WITNESS when RSS profile updates contend for the
firmware change lock.
Map the shared-code lock abstraction to exclusive sx locks. This also
covers tunnel and flow-profile operations which can reach the same
firmware wait while serialized.
Validated with WITNESS on a dual port Intel E835. Sixteen CORE resets
rebuilt both PFs without lock warnings, reset failures, or watchdogs.
Ten interface down/up cycles and twenty promiscuous-filter cycles also
completed cleanly.
Reviewed by: erj
MFC after: 2 weeks
[2 lines not shown]
libbsdconf: independent version macros
sysconf(8) --version now prints the library version alongside its
own so each can move on its own clock. Assigning a bitmask to
bool already converts zero/nonzero; drop the redundant != 0 (fuz).
Reviewed by: fuz, kfv
Differential Revision: https://reviews.freebsd.org/D59720
Add sysconf(8) and libbsdconf(3)
Complete the native configuration trinity: sysctl(8) for live kernel
state, sysrc(8) for rc.conf(5), and sysconf(8) for the remaining base
configuration -- loader.conf(5), sysctl.conf(5), and the make.conf(5)
family -- atop libbsdconf(3).
libbsdconf resurrects figpar as a unified reader/writer. Callbacks own
semantics; statements may span multiple lines via backslash continuation;
non-seekable input is spooled; writes are atomic (mkstemp, fsync, rename)
with mode/owner preservation. Format descriptors name each target, its
files, and quoting rules without private parsers. Multi-file targets
follow boot sourcing order; loader chases loader_conf_files as the boot
loader does.
sysconf(8) is the operator-facing tool: name / name=value on a required
target, sysrc-style list edits, make append and list-strike where they
belong, jail/altroot, and a capsicum sandbox for read-only use.
[21 lines not shown]
igbv: Recover disabled Hyper-V transmit queues
The Windows PF can disable a VF transmit queue while continuing to
report carrier up. Link polling alone then leaves the VF operationally
up even though it cannot transmit. The reproduced VLAN failure shows
this state with PF driver 14.1.5.0 and an MDD indication in the host trace.
Check queue zero from the admin path only while the Hyper-V VF is
running with sanitized queues and a completed host handshake. Report
operational link down and invalidate the statistics baseline when the
queue is disabled. Request recovery through the normal iflib stop/init
path only when a fresh, accessible STATUS read reports carrier up.
Rate limit requests if the host continues to hold the queue disabled,
and leave recovery pending while carrier is down.
Document the recovery behavior and clarify why the Hyper-V reset retains
the VF-local software reset before its host reset/MAC exchange.
Sponsored by: BBOX.io
cxgbe: Use the correct GHASH offset for a GMAC from a full TLS record
If a TLS request transmits all but a part of the GMAC at the end of a
TLS record, the work request asks the crypto engine to return the
calculated GMAC to the driver so it can be sent in a simple TCP packet
when the rest of the TLS record is transmitted in the future.
However, the offset of the returned GHASH offset was calculated
incorrectly in this case causing the driver to not recognize the
cached GMAC and instead use a more wasteful work request in the future
that encrypted the entire TLS record discarding all but the needed
bytes of the trailer.
Note that this does not effect correctness, just efficiency.
Reviewed by: np
Fixes: 9e269eafebfc ("cxgbe: Use partial GCM mode for partial TLS records on T7")
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D59711
vlan: Notify the parent when replacing a VLAN ID
Changing the VID of an existing VLAN interface rehashes the interface and
announces the new VID, but does not unregister the old VID. Parent
drivers and VLAN event consumers can consequently retain stale filter
membership.
After successfully inserting the new VID, emit vlan_unconfig for the old
VID before the existing vlan_config notification. Do not unregister
anything if insertion fails and the old VID is restored.
MFC after: 2 weeks
Sponsored by: BBOX.io
igbv: Support Hyper-V virtual functions
Use the Hyper-V reset/MAC exchange for 82576 and I350 VFs instead of the
native posted mailbox protocol, which the Windows PF does not service.
Read the host assigned address through configuration bytes 0x201 through
0x206 only during reset, and use it to identify the matching synthetic
hn(4) interface. The operations are local to the VF frontend.
Poll hardware link status rather than retaining a native mailbox link
handshake. Leave MAC, multicast, promiscuous-mode, and VLAN membership
policy with the host. Disable guest VLAN registration and native receive
limit requests, and limit the VF to an MTU of 1500 bytes.
Preserve accumulated statistics across host resets without counting a
counter clear as a wrap. Reject inaccessible register samples and rebase
after a reset indication or a disabled transmit queue, including when the
PF blocks the queue for malicious driver detection.
Document single queue support and host assigned access VLANs. Guest VLAN
[9 lines not shown]
man: Link mlx5en.4 also to if_mce.4
For consistency, create a symbolic link from mlx5en.4 to also if_mce.4
Reviewed by: ziaee, #manpages
Event: EuroBSDCon 2026
Differential Revision: https://reviews.freebsd.org/D59610
MFC after: 3 days
(cherry picked from commit e46a7d842a7572cc7ccef88a463a9af1a725fefc)
acpi: Don't compile CPU_VENDOR_{AMD,HYGON} cases on non-x86_64
Fixes build on aarch64.
Fixes: 5f68acc931a4 ("acpi: Warn if no amdsmu(4) loaded after suspend-to-idle resume")
Sponsored by: The FreeBSD Foundation
iflib: add a new global tunable "prefer_mpring"
net.iflib.prefer_mpring can be used to control whether or
not all iflib driver instances default to mp_ring or simple_tx.
This is intended to be temporary, to allow easy testing (now) of
simple_tx, and to allow an easy fallback to the legacy path
(later) once the default is switched to simple_tx
acpi: Warn if no amdsmu(4) loaded after suspend-to-idle resume
If amdsmu(4) is not loaded when entering suspend-to-idle and on an AMD
CPU, emit a warning.
FreeBSD currently only supports S0ix on AMD CPUs through the SMU. When
Intel support is completed, we should check the equivalent for Intel
(intelpmc).
Reviewed by: olce
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D59672
iflib: implement driver-provided queue selection for simple tx
Until now, simple tx has used its own queue selector, and has
ignored isc_txq_select and isc_txq_select_v2 (not this only
seems to matter for ice(4) with dcb enabled). This change
makes simple-tx use isc_txq_select* when present.
The implementation is defined to be efficient, with a
transmit routine chosen up-front that hard-codes the
queue selection and calls an always-inlined body. This
avoids a useless test per packet in the hotpath, and
also may avoid speculation into header parsing.
Note that this was designed for readability and efficiency
in the common case (interface up, not ALTQ). That's why
we do queue selection without duplicating nic-running and
altq checks, leaving them to the common implmentation.
Sponsored by: Netflix
[2 lines not shown]
proc: free kstack buffers when debug permission changes
The kern.proc.kstack handler allocates its output and stack buffers
before deliberately checking p_candebug again under the process lock.
If trace-control state changes between authorization checks, the failure
path balances the process and exec state but leaks both buffers.
Submitted by calif.io for the OpenAI Patch The Planet program
Signed-off-by: Andrew Griffiths <andrew at calif.io>
Fixes: 8b5abd9027b8 ("kern_proc.c: disallow execve around sysctl kern.proc.kstacks")
Reviewed by: markj
MFC after: 1 week
