IEEE802.1Q-2014 supersedes IEEE802.1D-2004. Now Priority Code Point (PCP)
2 is no longer at a lower priority than PCP 0. PCP 1 (Background) is still
at a lower priority than PCP 0 (Best Effort).
Reference:
IEEE802.1Q-2014, Standard for Local and metropolitan area networks
Table I-2 - Traffic type acronyms
Table I-3 - Defining traffic types
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Reviewed-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Philo Lu says:
====================
udp: Add 4-tuple hash for connected sockets
This patchset introduces 4-tuple hash for connected udp sockets, to make
connected udp lookup faster.
Stress test results (with 1 cpu fully used) are shown below, in pps:
(1) _un-connected_ socket as server
[a] w/o hash4: 1,825176
[b] w/ hash4: 1,831750 (+0.36%)
(2) 500 _connected_ sockets as server
[c] w/o hash4: 290860 (only 16% of [a])
[d] w/ hash4: 1,889658 (+3.1% compared with [b])
With hash4, compute_score is skipped when lookup, so [d] is slightly
better than [b].
Patch1: Add a new counter for hslot2 named hash4_cnt, to avoid cache line
miss when lookup.
Patch2: Add hslot/hlist_nulls for 4-tuple hash.
Patch3 and 4: Implement 4-tuple hash for ipv4 and ipv6.
The detailed motivation is described in Patch 3.
The 4-tuple hash increases the size of udp_sock and udp_hslot. Thus add it
with CONFIG_BASE_SMALL, i.e., it's a no op with CONFIG_BASE_SMALL.
Intentionally, the feature is not available for udplite. Though udplite
shares some structs and functions with udp, its connect() keeps unchanged.
So all udplite sockets perform the same as un-connected udp sockets.
Besides, udplite also shares the additional memory consumption in udp_sock
and udptable.
changelogs:
v8 -> v9 (Paolo Abeni):
- Add explanation about udplite in cover letter
- Update tags for co-developers
- Add acked-by tags of Paolo and Willem
v7 -> v8:
- add EXPORT_SYMBOL for ipv6.ko build
v6 -> v7 (Kuniyuki Iwashima):
- export udp_ehashfn to be used by udpv6 rehash
v5 -> v6 (Paolo Abeni):
- move udp_table_hash4_init from patch2 to patch1
- use hlist_nulls for lookup-rehash race
- add test results in commit log
- add more comment, e.g., for rehash4 used in hash4
- add ipv6 support (Patch4), and refactor some functions for better
sharing, without functionality change
v4 -> v5 (Paolo Abeni):
- add CONFIG_BASE_SMALL with which udp hash4 does nothing
v3 -> v4 (Willem de Bruijn):
- fix mistakes in udp_pernet_table_alloc()
RFCv2 -> v3 (Gur Stavi):
- minor fix in udp_hashslot2() and udp_table_init()
- add rcu sync in rehash4()
RFCv1 -> RFCv2:
- add a new struct for hslot2
- remove the sockopt UDP_HASH4 because it has little side effect for
unconnected sockets
- add rehash in connect()
- re-organize the patch into 3 smaller ones
- other minor fix
v8:
https://lore.kernel.org/all/20241108054836.123484-1-lulie@linux.alibaba.com/
v7:
https://lore.kernel.org/all/20241105121225.12513-1-lulie@linux.alibaba.com/
v6:
https://lore.kernel.org/all/20241031124550.20227-1-lulie@linux.alibaba.com/
v5:
https://lore.kernel.org/all/20241018114535.35712-1-lulie@linux.alibaba.com/
v4:
https://lore.kernel.org/all/20241012012918.70888-1-lulie@linux.alibaba.com/
v3:
https://lore.kernel.org/all/20241010090351.79698-1-lulie@linux.alibaba.com/
RFCv2:
https://lore.kernel.org/all/20240924110414.52618-1-lulie@linux.alibaba.com/
RFCv1:
https://lore.kernel.org/all/20240913100941.8565-1-lulie@linux.alibaba.com/
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement ipv6 udp hash4 like that in ipv4. The major difference is that
the hash value should be calculated with udp6_ehashfn(). Besides,
ipv4-mapped ipv6 address is handled before hash() and rehash(). Export
udp_ehashfn because now we use it in udpv6 rehash.
Core procedures of hash/unhash/rehash are same as ipv4, and udpv4 and
udpv6 share the same udptable, so some functions in ipv4 hash4 can also
be shared.
Co-developed-by: Cambda Zhu <cambda@linux.alibaba.com>
Signed-off-by: Cambda Zhu <cambda@linux.alibaba.com>
Co-developed-by: Fred Chen <fred.cc@alibaba-inc.com>
Signed-off-by: Fred Chen <fred.cc@alibaba-inc.com>
Co-developed-by: Yubing Qiu <yubing.qiuyubing@alibaba-inc.com>
Signed-off-by: Yubing Qiu <yubing.qiuyubing@alibaba-inc.com>
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, the udp_table has two hash table, the port hash and portaddr
hash. Usually for UDP servers, all sockets have the same local port and
addr, so they are all on the same hash slot within a reuseport group.
In some applications, UDP servers use connect() to manage clients. In
particular, when firstly receiving from an unseen 4 tuple, a new socket
is created and connect()ed to the remote addr:port, and then the fd is
used exclusively by the client.
Once there are connected sks in a reuseport group, udp has to score all
sks in the same hash2 slot to find the best match. This could be
inefficient with a large number of connections, resulting in high
softirq overhead.
To solve the problem, this patch implement 4-tuple hash for connected
udp sockets. During connect(), hash4 slot is updated, as well as a
corresponding counter, hash4_cnt, in hslot2. In __udp4_lib_lookup(),
hslot4 will be searched firstly if the counter is non-zero. Otherwise,
hslot2 is used like before. Note that only connected sockets enter this
hash4 path, while un-connected ones are not affected.
hlist_nulls is used for hash4, because we probably move to another hslot
wrongly when lookup with concurrent rehash. Then we check nulls at the
list end to see if we should restart lookup. Because udp does not use
SLAB_TYPESAFE_BY_RCU, we don't need to touch sk_refcnt when lookup.
Stress test results (with 1 cpu fully used) are shown below, in pps:
(1) _un-connected_ socket as server
[a] w/o hash4: 1,825176
[b] w/ hash4: 1,831750 (+0.36%)
(2) 500 _connected_ sockets as server
[c] w/o hash4: 290860 (only 16% of [a])
[d] w/ hash4: 1,889658 (+3.1% compared with [b])
With hash4, compute_score is skipped when lookup, so [d] is slightly
better than [b].
Co-developed-by: Cambda Zhu <cambda@linux.alibaba.com>
Signed-off-by: Cambda Zhu <cambda@linux.alibaba.com>
Co-developed-by: Fred Chen <fred.cc@alibaba-inc.com>
Signed-off-by: Fred Chen <fred.cc@alibaba-inc.com>
Co-developed-by: Yubing Qiu <yubing.qiuyubing@alibaba-inc.com>
Signed-off-by: Yubing Qiu <yubing.qiuyubing@alibaba-inc.com>
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Preparing for udp 4-tuple hash (uhash4 for short).
To implement uhash4 without cache line missing when lookup, hslot2 is
used to record the number of hashed sockets in hslot4. Thus adding a new
struct udp_hslot_main with field hash4_cnt, which is used by hash2. The
new struct is used to avoid doubling the size of udp_hslot.
Before uhash4 lookup, firstly checking hash4_cnt to see if there are
hashed sks in hslot4. Because hslot2 is always used in lookup, there is
no cache line miss.
Related helpers are updated, and use the helpers as possible.
uhash4 is implemented in following patches.
Signed-off-by: Philo Lu <lulie@linux.alibaba.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Steffen Klassert says:
====================
ipsec-next-11-15
1) Add support for RFC 9611 per cpu xfrm state handling.
2) Add inbound and outbound xfrm state caches to speed up
state lookups.
3) Convert xfrm to dscp_t. From Guillaume Nault.
4) Fix error handling in build_aevent.
From Everest K.C.
5) Replace strncpy with strscpy_pad in copy_to_user_auth.
From Daniel Yang.
6) Fix an uninitialized symbol during acquire state insertion.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Xuan Zhuo says:
====================
virtio-net: support AF_XDP zero copy (tx)
XDP socket(AF_XDP) is an excellent bypass kernel network framework. The zero
copy feature of xsk (XDP socket) needs to be supported by the driver. The
performance of zero copy is very good. mlx5 and intel ixgbe already support
this feature, This patch set allows virtio-net to support xsk's zerocopy xmit
feature.
At present, we have completed some preparation:
1. vq-reset (virtio spec and kernel code)
2. virtio-core premapped dma
3. virtio-net xdp refactor
So it is time for Virtio-Net to complete the support for the XDP Socket
Zerocopy.
Virtio-net can not increase the queue num at will, so xsk shares the queue with
kernel.
This patch set includes some refactor to the virtio-net to let that to support
AF_XDP.
The current configuration sets the virtqueue (vq) to premapped mode,
implying that all buffers submitted to this queue must be mapped ahead
of time. This presents a challenge for the virtnet send queue (sq): the
virtnet driver would be required to keep track of dma information for vq
size * 17, which can be substantial. However, if the premapped mode were
applied on a per-buffer basis, the complexity would be greatly reduced.
With AF_XDP enabled, AF_XDP buffers would become premapped, while kernel
skb buffers could remain unmapped.
We can distinguish them by sg_page(sg), When sg_page(sg) is NULL, this
indicates that the driver has performed DMA mapping in advance, allowing
the Virtio core to directly utilize sg_dma_address(sg) without
conducting any internal DMA mapping. Additionally, DMA unmap operations
for this buffer will be bypassed.
ENV: Qemu with vhost-user(polling mode).
Host CPU: Intel(R) Xeon(R) Platinum 8163 CPU @ 2.50GHz
testpmd> show port stats all
######################## NIC statistics for port 0 ########################
RX-packets: 19531092064 RX-missed: 0 RX-bytes: 1093741155584
RX-errors: 0
RX-nombuf: 0
TX-packets: 5959955552 TX-errors: 0 TX-bytes: 371030645664
Throughput (since last show)
Rx-pps: 8861574 Rx-bps: 3969985208
Tx-pps: 8861493 Tx-bps: 3969962736
############################################################################
testpmd> show port stats all
######################## NIC statistics for port 0 ########################
RX-packets: 68152727 RX-missed: 0 RX-bytes: 3816552712
RX-errors: 0
RX-nombuf: 0
TX-packets: 68114967 TX-errors: 33216 TX-bytes: 3814438152
Throughput (since last show)
Rx-pps: 6333196 Rx-bps: 2837272088
Tx-pps: 6333227 Tx-bps: 2837285936
############################################################################
But AF_XDP consumes more CPU for tx and rx napi(100% and 86%).
====================
Link: https://patch.msgid.link/20241112012928.102478-1-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Because the af-xdp will introduce a new xmit type, so I refactor the
xmit type mechanism first.
We know both xdp_frame and sk_buff are at least 4 bytes aligned.
For the xdp tx, we do not pass any pointer to virtio core as data,
we just need to pass the len of the packet. So we will push len
to the void pointer. We can make sure the pointer is 4 bytes aligned.
And the data structure of AF_XDP also is at least 4 bytes aligned.
So the last two bits of the pointers are free, we can't use these to
distinguish them.
00 for skb
01 for SKB_ORPHAN
10 for XDP
11 for AF-XDP tx
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241112012928.102478-9-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The current configuration sets the virtqueue (vq) to premapped mode,
implying that all buffers submitted to this queue must be mapped ahead
of time. This presents a challenge for the virtnet send queue (sq): the
virtnet driver would be required to keep track of dma information for vq
size * 17, which can be substantial. However, if the premapped mode were
applied on a per-buffer basis, the complexity would be greatly reduced.
With AF_XDP enabled, AF_XDP buffers would become premapped, while kernel
skb buffers could remain unmapped.
And consider that some sgs are not generated by the virtio driver,
that may be passed from the block stack. So we can not change the
sgs, new APIs are the better way.
So we pass the new argument 'premapped' to indicate the buffers
submitted to virtio are premapped in advance. Additionally,
DMA unmap operations for these buffers will be bypassed.
Suggested-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://patch.msgid.link/20241112012928.102478-5-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2024-11-05 (ice, ixgbe, igc. igb, igbvf, e1000)
For ice:
Mateusz refactors and adds additional SerDes configuration values to be
output.
Przemek refactors processing of DDP and adds support for a flag field in
the DDP's signature segment header.
Joe Damato adds support for persistent NAPI config.
Brett adjusts setting of Tx promiscuous based on unicast/multicast
setting.
Jake moves setting of pf->supported_rxdids to occur directly after DDP
load and changes a small struct to use stack memory.
Frederic Weisbecker adds WQ_UNBOUND flag to the workqueue.
For ixgbe:
Diomidis Spinellis removes a circular dependency.
For igc:
Vitaly removes an unneeded autoneg parameter.
For igb:
Johnny Park fixes a couple of typos.
For igbvf:
Wander Lairson Costa removes an unused spinlock.
For e1000:
Joe Damato adds RTNL lock to some calls where it is expected to be held.
* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
e1000: Hold RTNL when e1000_down can be called
igbvf: remove unused spinlock
igb: Fix 2 typos in comments in igb_main.c
igc: remove autoneg parameter from igc_mac_info
ixgbe: Break include dependency cycle
ice: Unbind the workqueue
ice: use stack variable for virtchnl_supported_rxdids
ice: initialize pf->supported_rxdids immediately after loading DDP
ice: only allow Tx promiscuous for multicast
ice: Add support for persistent NAPI config
ice: support optional flags in signature segment header
ice: refactor "last" segment of DDP pkg
ice: extend dump serdes equalizer values feature
ice: rework of dump serdes equalizer values feature
====================
Link: https://patch.msgid.link/20241113185431.1289708-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Petr Machata says:
====================
net: ndo_fdb_add/del: Have drivers report whether they notified
Currently when FDB entries are added to or deleted from a VXLAN netdevice,
the VXLAN driver emits one notification, including the VXLAN-specific
attributes. The core however always sends a notification as well, a generic
one. Thus two notifications are unnecessarily sent for these operations. A
similar situation comes up with bridge driver, which also emits
notifications on its own.
# ip link add name vx type vxlan id 1000 dstport 4789
# bridge monitor fdb &
[1] 1981693
# bridge fdb add de:ad:be:ef:13:37 dev vx self dst 192.0.2.1
de:ad:be:ef:13:37 dev vx dst 192.0.2.1 self permanent
de:ad:be:ef:13:37 dev vx self permanent
In order to prevent this duplicity, add a parameter, bool *notified, to
ndo_fdb_add and ndo_fdb_del. The flag is primed to false, and if the callee
sends a notification on its own, it sets the flag to true, thus informing
the core that it should not generate another notification.
Patches #1 to #2 are concerned with the above.
In the remaining patches, #3 to #7, add a selftest. This takes place across
several patches. Many of the helpers we would like to use for the test are
in forwarding/lib.sh, whereas net/ is a more suitable place for the test,
so the libraries need to be massaged a bit first.
====================
Link: https://patch.msgid.link/cover.1731589511.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Check that only one notification is produced for various FDB edit
operations.
Regarding the ip_link_add() and ip_link_master() helpers. This pattern of
action plus corresponding defer is bound to come up often, and a dedicated
vocabulary to capture it will be handy. tunnel_create() and vlan_create()
from forwarding/lib.sh are somewhat opaque and perhaps too kitchen-sinky,
so I tried to go in the opposite direction with these ones, and wrapped
only the bare minimum to schedule a corresponding cleanup.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Amit Cohen <amcohen@nvidia.com>
Acked-by: Shuah Khan <skhan@linuxfoundation.org>
Link: https://patch.msgid.link/910c5880ae6d3b558d6889cbdba2be690c2615c6.1731589511.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
For logging to be useful, something has to set RET and retmsg by calling
ret_set_ksft_status(). There is a suite of functions to that end in
forwarding/lib: check_err, check_fail et.al. Move them to net/lib.sh so
that every net test can use them.
Existing lib.sh users might be using these same names for their functions.
However lib.sh is always sourced near the top of the file (checked), and
whatever new definitions will simply override the ones provided by lib.sh.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Amit Cohen <amcohen@nvidia.com>
Acked-by: Shuah Khan <skhan@linuxfoundation.org>
Link: https://patch.msgid.link/f488a00dc85b8e0c1f3c71476b32b21b5189a847.1731589511.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It would be good to use the same mechanism for scheduling and dispatching
general net tests as the many forwarding tests already use. To that end,
move the logging helpers to net/lib.sh so that every net test can use them.
Existing lib.sh users might be using the name themselves. However lib.sh is
always sourced near the top of the file (checked), and whatever new
definition will simply override the one provided by lib.sh.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Amit Cohen <amcohen@nvidia.com>
Acked-by: Shuah Khan <skhan@linuxfoundation.org>
Link: https://patch.msgid.link/a6fc083486493425b2c61185c327845b6ce3233a.1731589511.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Many net selftests invent their own logging helpers. These really should be
in a library sourced by these tests. Currently forwarding/lib.sh has a
suite of perfectly fine logging helpers, but sourcing a forwarding/ library
from a higher-level directory smells of layering violation. In this patch,
move the logging helpers to net/lib.sh so that every net test can use them.
Together with the logging helpers, it's also necessary to move
pause_on_fail(), and EXIT_STATUS and RET.
Existing lib.sh users might be using these same names for their functions
or variables. However lib.sh is always sourced near the top of the
file (checked), and whatever new definitions will simply override the ones
provided by lib.sh.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Amit Cohen <amcohen@nvidia.com>
Acked-by: Shuah Khan <skhan@linuxfoundation.org>
Link: https://patch.msgid.link/edd3785a3bd72ffbe1409300989e993ee50ae98b.1731589511.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently when FDB entries are added to or deleted from a VXLAN netdevice,
the VXLAN driver emits one notification, including the VXLAN-specific
attributes. The core however always sends a notification as well, a generic
one. Thus two notifications are unnecessarily sent for these operations. A
similar situation comes up with bridge driver, which also emits
notifications on its own:
# ip link add name vx type vxlan id 1000 dstport 4789
# bridge monitor fdb &
[1] 1981693
# bridge fdb add de:ad:be:ef:13:37 dev vx self dst 192.0.2.1
de:ad:be:ef:13:37 dev vx dst 192.0.2.1 self permanent
de:ad:be:ef:13:37 dev vx self permanent
In order to prevent this duplicity, add a paremeter to ndo_fdb_add,
bool *notified. The flag is primed to false, and if the callee sends a
notification on its own, it sets it to true, thus informing the core that
it should not generate another notification.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/cbf6ae8195e85cbf922f8058ce4eba770f3b71ed.1731589511.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Breno Leitao says:
====================
net: netpoll: Improve SKB pool management
The netpoll subsystem pre-allocates 32 SKBs in a pool for emergency use
during out-of-memory conditions. However, the current implementation has
several inefficiencies:
* The SKB pool, once allocated, is never freed:
* Resources remain allocated even after netpoll users are removed
* Failed initialization can leave pool populated forever
* The global pool design makes resource tracking difficult
This series addresses these issues through three patches:
Patch 1 ("net: netpoll: Individualize the skb pool"):
- Replace global pool with per-user pools in netpoll struct
Patch 2 ("net: netpoll: flush skb pool during cleanup"):
- Properly free pool resources during netconsole cleanup
These changes improve resource management and make the code more
maintainable. As a side benefit, the improved structure would allow
netpoll to be modularized if desired in the future.
v2: https://lore.kernel.org/20241107-skb_buffers_v2-v2-0-288c6264ba4f@debian.org
v1: https://lore.kernel.org/20241025142025.3558051-1-leitao@debian.org
====================
Link: https://patch.msgid.link/20241114-skb_buffers_v2-v3-0-9be9f52a8b69@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The netpoll subsystem maintains a pool of 32 pre-allocated SKBs per
instance, but these SKBs are not freed when the netpoll user is brought
down. This leads to memory waste as these buffers remain allocated but
unused.
Add skb_pool_flush() to properly clean up these SKBs when netconsole is
terminated, improving memory efficiency.
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20241114-skb_buffers_v2-v3-2-9be9f52a8b69@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The current implementation of the netpoll system uses a global skb
pool, which can lead to inefficient memory usage and
waste when targets are disabled or no longer in use.
This can result in a significant amount of memory being unnecessarily
allocated and retained, potentially causing performance issues and
limiting the availability of resources for other system components.
Modify the netpoll system to assign a skb pool to each target instead of
using a global one.
This approach allows for more fine-grained control over memory
allocation and deallocation, ensuring that resources are only allocated
and retained as needed.
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20241114-skb_buffers_v2-v3-1-9be9f52a8b69@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Current implementation of gettimex64() makes at least 3 PCIe reads to
get current PHC time. It takes at least 2.2us to get this value back to
userspace. At the same time there is cached value of upper bits of PHC
available for packet timestamps already. This patch reuses cached value
to speed up reading of PHC time.
Signed-off-by: Vadim Fedorenko <vadfed@meta.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20241114114820.1411660-1-vadfed@meta.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
