ena: Budget rx descriptors, not packets
We had ENA_RX_BUDGET = 256 in order to allow up to 256 received
packets to be processed before we do other cleanups (handling tx
packets and, critically, refilling the rx buffer ring). Since the
ring holds 1024 buffers by default, this was fine for normal packets:
We refill the ring when it falls below 7/8 full, and even with a large
burst of incoming packets allowing it to fall by another 1/4 before we
consider refilling the ring still leaves it at 7/8 - 1/4 = 5/8 full.
With jumbos, the story is different: A 9k jumbo (as is used by default
within the EC2 network) consumes 3 descriptors, so a single rx cleanup
pass can consume 3/4 of the default-sized rx ring; if the rx buffer
ring wasn't completely full before a packet burst arrives, this puts
us perilously close to running out of rx buffers.
This precise failure mode has been observed on some EC2 instance types
within a Cluster Placement Group, resulting in the nominal 10 Gbps
single-flow throughput between instances dropping to ~100 Mbps as a
[21 lines not shown]
ena: Adjust ena_[rt]x_cleanup to return bool
The ena_[rt]x_cleanup functions are limited internally to a maximum
number of packets; this ensures that TX doesn't starve RX (or vice
versa) and also attempts to ensure that we get a chance to refill
the RX buffer ring before the device runs out of buffers and starts
dropping packets.
Historically these functions have returned the number of packets which
they processed which ena_cleanup compares to their respective budgets
to decide whether to reinvoke them. This is unnecessary complication;
since the precise number of packets processed is never used, adjust
the APIs of those functions to return a bool indicating if they want
to be reinvoked (aka if they hit their limits).
Since ena_tx_cleanup now only uses work_done if diagnostics are
enabled (ena_log_io macros to nothing otherwise) eliminate that
variable and pass its value (ENA_TX_BUDGET - budget) to ena_log_io
directly.
[9 lines not shown]
ena: Report RX overrun errors
Extract rx_overruns from the keep alive descriptor reported by
the device and expose it via sysctl hw stats.
RX overrun errors occur when a packet arrives but there are not
enough free buffers in the RX ring to receive it.
MFC after: 2 weeks
Sponsored by: Amazon, Inc.
Differential Revision: https://reviews.freebsd.org/D56640
(cherry picked from commit e3f4a63af63bea70bc86b6c790b14aa5ee99fcd0)
install.1: Document options incompatible with -s
Reported by: des
Reviewed by: ziaee
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D57409
krpc: Get rid of KRPC_VNET macros
When the krpc was vnet'd, the VNET macros were hidden
behind macros that had the KRPC_ prefix on them.
This was done because, at the time, it was thought
that something other than vnet might be used for this.
That has not happened and probably will not happen,
so this patch replaces these obscuring macros with
the regular vnet ones.
There should be no semantics change caused by
this commit.
(cherry picked from commit d8d9e1cf5e055433565a7e686bc096a8494d2bb8)
nfsd: Get rid of NFSD_VNET macros
When the nfsd was vnet'd, the VNET macros were hidden
behind macros that had the NFSD_ prefix on them.
This was done because, at the time, it was thought
that something other than vnet might be used for this.
That has not happened and probably will not happen,
so this patch replaces these obscuring macros with
the regular vnet ones.
There should be no semantics change caused by
this commit.
(cherry picked from commit 821976facf746ef241e1524b44bb1de0af98fdc0)
nfscl: Fix handling of gssd upcalls for the NFS client
Without this patch, all upcalls to the gssd daemon are
done in vnet0 (outside of any vnet jail). This does
not work well, because a user principal's credential
cache can be within the jail (/tmp/krb5cc_NNN in the
jail's namespace).
This patch modifies the client so that RPCs done
from within vnet jails does an upcall to a gssd
daemon running within the vnet jail. It required
that the cache of uid->credential shorthands in
the rpcsec_gss be vnet'd.
The situation is still less than ideal and sec=krb5[ip]
mounts that are visible within vnet jails is still
not something I would recommend, but it can work ok
with this patch.
[5 lines not shown]
nfs_clrpcops.c: Change initial layout type to flex files
Since the File 4.1 layout is only supported by some
large configurations of Netapp Filers, switch the client
pNFS code to attempting to use Flexible File Format first.
File 4.1 Layout should still work for any servers that do
no support Flexible File Layout.
thunderbolt: Fix INVARIANTS compilation
The problem is that THUNDERBOLT_DEBUG triggers the use of some
variables, but it is independent of INVARIANTS and the variables it uses
were tagged with '__diagused'.
Fix this by using '__maybe_unused' in those places.
Fixes: 183633079178 ("thunderbolt: make code -Wunused clean")
Fixes: 886164895f3f ("amd64: complete thunderbolt KERNCONF integration")
Sponsored by: The FreeBSD Foundation
xinstall: Do not allow -l and -s together
Cannot strip the target if creating a link.
Reviewed by: des
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D57398
xinstall: Add test for -d -s not allowed together
Reviewed by: des
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D57403
i386: Fix build (of 'genassym.o')
i386's genassym.c needs to define some assembly symbols holding the size
of NFS structures to support NFS_ROOT while booting with a nfs_diskless
structure. For this, it needs to include a few NFS headers, which
require definitions from <sys/mount.h> (fhandle_t, vfs_init_t), which
was removed by commit 72ab129799a2 ("x86: remove sys/mount.h from
genassym.c").
Since recently, <sys/mount.h> has been including <sys/vnode.h>, so needs
"vnode_if.h" to have been generated for the compilation of 'genassym.o'
not to fail. Make sure this is the case (for all architectures for
simplicity) by tweaking the rule for 'genassym.o' in
'sys/conf/kern.post.mk', leaving a comment there so that it can be
removed when i386 is dropped (or if the above-mentioned dependency is
broken).
Fixes: 72ab129799a2 ("x86: remove sys/mount.h from genassym.c")
Sponsored by: The FreeBSD Foundation
netpfil: add PF netlink command decoding support
Convert PFNL_CMD values in pf_nl.h from an enum to #define constants,
add a pfnl_cmd table definition for mktable, and implement the
corresponding command decoding helpers in libsysdecode.
This allows mktable to generate PF netlink command lookup tables and
enables symbolic decoding of PF netlink commands.
Reviewed by: kp
Signed-off-by: Ishan Agrawal <iagrawal9990 at gmail.com>
Sponsored by: Google LLC (GSoC 2026)
pfdenied: fix checking root anchor
pfctl doesn't like empty anchors (-a ''), but we can specify the root
anchor as '/' too, so do that instead.
PR: 295324
Tested by: Paweł Krawczyk
MFC after: 1 week
Sponsored by: Rubicon Communications, LLC ("Netgate")
sound: Retire snd_mixer->enuminfo
Instead of caching this when mix_setrecdevs() is called (which many
drivers never call), calculate it when we need it. After all, it is
quite rare that this structure is used by applications.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Pull Request: https://ron-dev.freebsd.org/FreeBSD/src/pulls/18
(cherry picked from commit 5589a7499add5912a88e9424b0aec843e099fb60)
sound: Remove all remaining uses of mixer_get_lock()
These functions are called from sound(4) through MIXER_SET() in
mixer_set(), but nothing mixer-related is used or needs to be locked in
these cases.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Pull Request: https://ron-dev.freebsd.org/FreeBSD/src/pulls/18
(cherry picked from commit 1e72608a3c6a60b76938c5ded2a290d0b9ff6456)
sound: Retire MIXER_SIZE and use correct size in DEFINE_CLASS
I am not sure why this hack was needed, but the size should be
sizeof(struct snd_mixer). Unfortunately, we have to expose snd_mixer,
but it should be better than relying on a hack.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Pull Request: https://ron-dev.freebsd.org/FreeBSD/src/pulls/18
(cherry picked from commit 64ebbb1d74415cb93445cc625da63449f01ac8c5)
snd_uaudio: Stop using mixer_get_lock()
There is no reason to lock snd_mixer->lock here, since the only thing we
are doing with snd_mixer is to call mix_getdevinfo() and mix_get_dev(),
which are one-line getters, which are needed to eventually call
uaudio_get_child_index_by_dev().
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Pull Request: https://ron-dev.freebsd.org/FreeBSD/src/pulls/18
(cherry picked from commit e87654db5a0923856afbaace9bd0975331782044)
sound: Retire mixer_ioctl_channel()
This function never succeeds when it is not called from the same process
that has opened the file descriptor (e.g., mixer(8)). The reason is that
the CHN_FOREACH() loop tries to match the pid of each channel with the
pid of the process performing the ioctl, which will not be the same,
unless it's the same process that both opened the channel and performed
the ioctl.
In the case that the same process opens the channels and performs the
ioctl, however, we still do not need to worry, because mixer_ioctl_cmd()
essentially does the same thing anyway. Additionally, this scenario
should be quite rare, given that most applications do not open both
/dev/dsp* and /dev/mixer*, and in fact, it is actively encouraged by the
official OSSv4 specification not to do that.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Pull Request: https://ron-dev.freebsd.org/FreeBSD/src/pulls/18
[2 lines not shown]
sound: Retire FEEDEQ_DISABLE
We can do this more efficiently by just using the SD_F_EQ* flags. In
fact, the dev.pcm.%d.eq handler will (un)set SD_F_EQ_ENABLED and this is
what we actually test with when choosing to creating the EQ feeder or
not, so setting the state to FEEDEQ_DISABLE does not really an effect in
the first place.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Pull Request: https://ron-dev.freebsd.org/FreeBSD/src/pulls/15
(cherry picked from commit f5d6e5cb5c3e0593d24c90671d72654aa59cdd1a)
