APIC ID 255 and above require x2APIC and DMAR interrupt remapping.
FreeBSD is starting to be tested on high core count Intel systems that
meet this criteria. We're going to enable DMAR by default to support
this, so default hw.iommu.dma to 0 to avoid a significant performance
regression.
Reviewed by: kib, jhb
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D42435
bitstring.h includes a definition of bit_foreach, for iterating over
the set bits of a bitstring. axgbe implements its own version of this
for bitstrings. Drop it, and use the bitstring method.
Reviewed by: des
Differential Revision: https://reviews.freebsd.org/D46037
Push the #ifdefs down into the function body instead of defining
functions conditionally for ease of readability. These aren't
critical paths, so one extra branch in the !PCI_IOV case is not a big
deal.
Requested by: jrtc27
Differential Revision: https://reviews.freebsd.org/D45877
The qlnxe driver always sets the NIC's rx mac address to the card's
default MAC, ignoring any MAC address set via ifconfig.
PR: 278085
MFC after: 1 week
When the kernel is compiled with options RATELIMIT, the
mlx5en driver cannot detach. It gets stuck waiting for all
kernel users of its rates to drop to zero before finally calling
ether_ifdetach.
The tcp ratelimit code has an eventhandler for ifnet departure
which causes rates to be released. However, this is called as an
ifnet departure eventhandler, which is invoked as part of
ifdetach(), via either_ifdetach(). This means that the tcp
ratelimit code holds down many hw rates when the mlx5en driver
is waiting for the rate count to go to 0. Thus devctl detach
will deadlock on mlx5 with this stack:
mi_switch+0xcf sleepq_timedwait+0x2f _sleep+0x1a3 pause_sbt+0x77 mlx5e_destroy_ifp+0xaf mlx5_remove_device+0xa7 mlx5_unregister_device+0x78 mlx5_unload_one+0x10a remove_one+0x1e linux_pci_detach_device+0x36 linux_pci_detach+0x24 device_detach+0x180 devctl2_ioctl+0x3dc devfs_ioctl+0xbb vn_ioctl+0xca devfs_ioctl_f+0x1e kern_ioctl+0x1c3 sys_ioctl+0x10a
To fix this, provide an explicit API for a driver to call the tcp
ratelimit code telling it to detach itself from an ifnet. This
allows the mlx5 driver to unload cleanly. I considered adding an
ifnet pre-departure eventhandler. However, that would need to be
invoked by the driver, so a simple function call seemed better.
The mlx5en driver has been updated to call this function.
Reviewed by: kib, rrs
Differential Revision: https://reviews.freebsd.org/D46221
Sponsored by: Netflix
1. Remove toepcb from the toep_list on active open failure.
2. Purge the wr_list for an L2T entry on an adpater stop.
Fixes: c1c524852f cxgbe/t4_tom: Implement uld_stop and uld_restart for ULD_TOM.
MFC after: 2 weeks
Sponsored by: Chelsio Communications
netmap's generic mode tries to improve performance by minimizing mbuf
allocations. In service of this goal, it maintains an extra reference
to the mbuf and polls the counter to see if the driver has released its
reference by calling m_freem(). As a result, the extref destructor is
not called when expected by the netfront driver, and mbufs tags are not
freed.
Modify the tx path to release its mbuf tags promptly when reclaiming tx
descriptors. They are drawn from a fixed-size pool, so otherwise are
quickly exhausted when a netfront interface is in netmap generic mode.
Co-authored by: royger
MFC after: 2 weeks
Fixes: dabb3db7a8 ("xen/netfront: deal with mbuf data crossing a page boundary")
Sponsored by: Cloud Software Group
Sponsored by: Klara, Inc.
Sponsored by: Zenarmor
sndstat(4) falsely reports "hardware" as the starting point of
recording, and ending point of playback VCHANs. Recording VCHANs get
their input from the primary recording channel, and playback VCHANs send
their input to the primary playback channel.
Sponsored by: The FreeBSD Foundation
MFC after: 2 days
Reviewed by: dev_submerge.ch, markj
Differential Revision: https://reviews.freebsd.org/D46177
This fixes a number of clang 19 warnings:
sys/dev/qat/qat_api/common/compression/dc_session.c:154:15: error: comparison of different enumeration types ('enum _CpaBoolean' and 'icp_qat_hw_compression_delayed_match_t') [-Werror,-Wenum-compare]
154 | if (CPA_TRUE == pService->comp_device_data.enableDmm) {
| ~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
sys/dev/qat/qat_api/common/compression/dc_session.c:285:17: error: comparison of different enumeration types ('enum _CpaBoolean' and 'icp_qat_hw_compression_delayed_match_t') [-Werror,-Wenum-compare]
285 | (CPA_TRUE == pService->comp_device_data.enableDmm) ?
| ~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The `enableDmm` field of variable `comp_device_data` is of type
`icp_qat_hw_compression_delayed_match_t`, not `CpaBoolean`. In this
case, we can seamlessly replace the value with
`ICP_QAT_HW_COMPRESSION_DELAYED_MATCH_ENABLED`, which is numerically
equal to `CPA_TRUE`.
MFC after: 3 days
This fixes a clang 19 warning:
sys/dev/isci/scil/scif_sas_smp_remote_device.c:197:26: error: comparison of different enumeration types ('SCI_IO_STATUS' (aka 'enum _SCI_IO_STATUS') and 'enum _SCI_STATUS') [-Werror,-Wenum-compare]
197 | if (completion_status == SCI_FAILURE_RETRY_REQUIRED)
| ~~~~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~
The `completion_status` variable is of type `SCI_IO_STATUS`, not
`SCI_STATUS`. In this case, we can seamlessly replace the value with
`SCI_IO_FAILURE_RETRY_REQUIRED`, which is numerically equal to
`SCI_FAILURE_RETRY_REQUIRED`.
MFC after: 3 days
This fixes a clang 19 warning:
sys/dev/iavf/iavf_lib.c:514:39: error: comparison of different enumeration types ('enum virtchnl_vsi_type' and 'enum iavf_vsi_type') [-Werror,-Wenum-compare]
514 | if (sc->vf_res->vsi_res[i].vsi_type == IAVF_VSI_SRIOV)
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~
The `vsi_type` field of `struct virtchnl_vsi_resource` is of type `enum
virtchnl_vsi_type`, not `enum iavf_vsi_type`. In this case, we can
seamlessly replace the value with `VIRTCHNL_VSI_SRIOV`, which is
numerically equal to `IAVF_VSI_SRIOV`.
MFC after: 3 days
Right now, only IPv4 transport mode, with aes-gcm ESP, is supported.
Driver also cooperates with NAT-T, and obeys socket policies, which
makes IKEd like StrongSwan working.
Sponsored by: NVIDIA networking
The increment of 1 was intended to convert qp->maxr2t from 0's based
to 1 based before multiplying by the queue length.
Sponsored by: Chelsio Communications
If a queue pair is destroyed (e.g. due to the TCP connection dropping)
while a host to controller data transfer is in progress, the
pending_r2ts counter can be non-zero. This can later trigger an
assertion failure when the capsule is freed. To fix, update the
relevant R2T accounting stats when aborting active command buffers
during queue pair destruction.
Sponsored by: Chelsio Communications
The current approach is to create foreign mappings in any unpopulated address
on the memory map. This however can cause issues, as late-loaded drivers could
then found their MMIO region has been stolen to be used as foreign mapping
scratch space (due to the Xen drivers having started first).
Add a local resource manager to the xenpv bus driver, architectures can add
suitable resources to that resource manager at boot in order to be used in
preference of the generic MMIO resource allocator.
No functional change, as the introduced scratch mapping range is not yet
populated.
Sponsored by: Cloud Software Group
MFC after: 1 week
Reviewed by: Elliott Mitchell <ehem+freebsd@m5p.com>
Differential revision: https://reviews.freebsd.org/D46122
Remove some uses of PHOLD which were there only to prevent the process'
threads from being swapped out.
Tested by: pho
Reviewed by: imp, kib
Differential Revision: https://reviews.freebsd.org/D46118
- Remove unnecessary CHN_[UN]LOCKs.
- Improve device_printf() messages.
- Remove redundant checks.
- Remove 0 assignments, since the channel is allocated with M_ZERO.
- Re-organize sections in a more coherent way.
- Remove "out1" label, just return NULL directly.
- Rename "out2" to "fail" and simplify logic.
- Do not check the return value of dsp_unit2name(), as it is guaranteed
not to fail (we pass a valid channel type).
Sponsored by: The FreeBSD Foundation
MFC after: 2 days
Reviewed by: dev_submerge.ch
Differential Revision: https://reviews.freebsd.org/D45985
No functional change intended.
Sponsored by: The FreeBSD Foundation
MFC after: 2 days
Reviewed by: dev_submerge.ch, markj
Differential Revision: https://reviews.freebsd.org/D45984
In case chn_init() fails, we are leaking memory by not freeing the
feeder.
Sponsored by: The FreeBSD Foundation
MFC after: 2 days
Reviewed by: dev_submerge.ch, markj
Differential Revision: https://reviews.freebsd.org/D45979
Assignment taken from dsp_oss_engineinfo().
Sponsored by: The FreeBSD Foundation
MFC after: 2 days
Reviewed by: dev_submerge.ch
Differential Revision: https://reviews.freebsd.org/D46166
The pre-freebsd7 stuff is labeled _old right now. Relabel it as
_freebsd6 instead. No functional change intended. I selected freebsd6
because that's the last version to have the ioctls, although I know
there were changes in the interface for 4, 5 and 6. This broadly batches
what we do with system calls that accumulate changes until we break and
use a new one, we use the last major the system call was in.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D45867
As of 9e6544dd6e contigfree(9) is no longer
needed and should not be used anymore. We leave a wrapper for 3rd party
code in at least 15.x but remove (almost) all other cases from the tree.
This leaves one use of contigfree(9) untouched; that was the original
trigger for 9e6544dd6e and is handled in D45813 (to be committed
seperately later).
Sponsored by: The FreeBSD Foundation
Reviewed by: markj, kib
Tested by: pho (10h stress test run)
Differential Revision: https://reviews.freebsd.org/D46099
This sysctl sets a cap on the maximum payload of transmitted data PDUs
including both C2H_DATA and H2C_DATA PDUs, not just C2H_DATA PDUs.
Sponsored by: Chelsio Communications
D28743 was commited, reverted and then f689cb23b2 landed before it
was recommitted. However, D28743 included an extra length check. Redo
that functionality so we check both the number of entries as well as the
length checks for wacky data.
Sponsored by: Netflix
Reviewed by: gallatin
Differential Revision: https://reviews.freebsd.org/D45763
Issue a warning if we have system interrupt issues. If you get this
warning, then we submitted a request, it timed out without an interrupt
being posted, but when we polled the card's completion, we found
completion events. This indicates that we're missing interrupts, and to
date all the times I've helped people track issues like this down it has
been a system issue, not an NVMe driver isseue.
Sponsored by: Netflix
Reviewed by: gallatin
Differential Revision: https://reviews.freebsd.org/D46031
Optimize timeout code based on three observations.
(1) The tr queues are sorted in order of submission, so the first one
that could time out is the first "real" one on the list.
(2) Timeouts for a given queue are all the same length (well, except
at startup, where timeout doesn't matter, and when you change it
at runtime, where timeouts will still happen eventually and the
difference isn't worth optimizing for).
(3) Calling the ISR races the real ISR and we should avoid that better.
So now, after checking to see if the card is there and working, the
timeout routine scans the pending tracker list until it finds a non-AER
tracker. If the deadline hasn't passed, we return, doing nothing
further. Otherwise, we call poll completions and then process the list
looking for timed out items.
This should move the timeout routine to touching hardware only when it's
really necessary. It thus avoids racing the normal ISR, while still
timig out stuck transactions quickly enough.
There was also some minor code motion to make all of the above flow more
nicely for the reader.
When interrupts aren't working at all, then this will increase latency
somewhat. But when interrupts aren't working at all, there's bigger
problems and we should poll quite often in that case. That will be
handled in future commits.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D46026
When processing an abort completion command, we have to lock. But we
have to lock the qpair of the original transaction (not the abort we're
completing). We do this to avoid races with checking the completion id
to tr mapping array, as well as to manually complete it.
Note: we don't handle the completion status of 'Asked to abort too many
transactions at once.' That will be fixed on subsequent commits. Add a
note to that effect for now since it's a harder problem to solve.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D46025
When we lose a race with the timeout code, shift towards waiting for
that timeout code to complete so we can acquire the lock. This way we
can make sure we're in 'normal' mode before processing I/O
completions. If we're not in 'normal' mode, then we're resetting and we
should avoid completions.
Sponsored by: Netflix
Reviewed by: gallatin
Differential Revision: https://reviews.freebsd.org/D46024
Users may interpret the message as a possible hardware error, but the
issue is in fact unimplemented functionality. Reword the message to
avoid implying it is an error.
Reviewed by: wulf
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D46092
