OSDN Git Service

tomoyo/tomoyo-test1.git
17 months agoMerge branch 'amd-xgbe-add-support-for-2-5gbe-and-rx-adaptation'
Paolo Abeni [Thu, 2 Feb 2023 14:17:21 +0000 (15:17 +0100)]
Merge branch 'amd-xgbe-add-support-for-2-5gbe-and-rx-adaptation'

Raju Rangoju says:

====================
amd-xgbe: add support for 2.5GbE and rx-adaptation

This patch series adds support for 2.GbE in 10GBaseT mode and
rx-adaptation support for Yellow Carp devices.

1) Support for 2.5GbE:
   Add the necessary changes to the driver to fully recognize and enable
   2.5GbE speed in 10GBaseT mode.

2) Support for rx-adaptation:
   In order to support the 10G backplane mode without Auto-negotiation
   and to support the longer-length DAC cables, it requires PHY to
   perform RX Adaptation sequence as mentioned in the Synopsys databook.
   Add the necessary changes to Yellow Carp devices to ensure seamless
   RX Adaptation for 10G-SFI (LONG DAC), and 10G-KR modes without
   Auto-Negotiation (CL72 not present)
====================

Link: https://lore.kernel.org/r/20230201054932.212700-1-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agoamd-xgbe: add support for rx-adaptation
Raju Rangoju [Wed, 1 Feb 2023 05:49:32 +0000 (11:19 +0530)]
amd-xgbe: add support for rx-adaptation

The existing implementation for non-Autonegotiation 10G speed modes does
not enable RX adaptation in the Driver and FW. The RX Equalization
settings (AFE settings alone) are manually configured and the existing
link-up sequence in the driver does not perform rx adaptation process as
mentioned in the Synopsys databook. There's a customer request for 10G
backplane mode without Auto-negotiation and for the DAC cables of more
significant length that follow the non-Autonegotiation mode. These modes
require PHY to perform RX Adaptation.

The proposed logic adds the necessary changes to Yellow Carp devices to
ensure seamless RX Adaptation for 10G-SFI (LONG DAC) and 10G-KR without
AN (CL72 not present). The RX adaptation core algorithm is executed by
firmware, however, to achieve that a new mailbox sub-command is required
to be sent by the driver.

Co-developed-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agoamd-xgbe: add 2.5GbE support to 10G BaseT mode
Raju Rangoju [Wed, 1 Feb 2023 05:49:31 +0000 (11:19 +0530)]
amd-xgbe: add 2.5GbE support to 10G BaseT mode

Add support to the driver to fully recognize and enable 2.5GbE speed in
10GBaseT mode.

Acked-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agoMerge branch 'net-sched-transition-act_pedit-to-rcu-and-percpu-stats'
Paolo Abeni [Thu, 2 Feb 2023 12:19:06 +0000 (13:19 +0100)]
Merge branch 'net-sched-transition-act_pedit-to-rcu-and-percpu-stats'

Pedro Tammela says:

====================
net/sched: transition act_pedit to rcu and percpu stats

The software pedit action didn't get the same love as some of the
other actions and it's still using spinlocks and shared stats.
Therefore, transition the action to rcu and percpu stats which
improves the action's performance.

We test this change with a very simple packet forwarding setup:

tc filter add dev ens2f0 ingress protocol ip matchall \
   action pedit ex munge eth src set b8:ce:f6:4b:68:35 pipe \
   action pedit ex munge eth dst set ac:1f:6b:e4:ff:93 pipe \
   action mirred egress redirect dev ens2f1
tc filter add dev ens2f1 ingress protocol ip matchall \
   action pedit ex munge eth src set b8:ce:f6:4b:68:34 pipe \
   action pedit ex munge eth dst set ac:1f:6b:e4:ff:92 pipe \
   action mirred egress redirect dev ens2f0

Using TRex with a http-like profile, in our setup with a 25G NIC
and a 26 cores Intel CPU, we observe the following in perf:
   before:
    11.59%  2.30%  [kernel]  [k] tcf_pedit_act
       2.55% tcf_pedit_act
             8.38% _raw_spin_lock
                       6.43% native_queued_spin_lock_slowpath
   after:
    1.46%  1.46%  [kernel]  [k] tcf_pedit_act

tdc results for pedit after the patch:
1..69
ok 1 319a - Add pedit action that mangles IP TTL
ok 2 7e67 - Replace pedit action with invalid goto chain
ok 3 377e - Add pedit action with RAW_OP offset u32
ok 4 a0ca - Add pedit action with RAW_OP offset u32 (INVALID)
ok 5 dd8a - Add pedit action with RAW_OP offset u16 u16
ok 6 53db - Add pedit action with RAW_OP offset u16 (INVALID)
ok 7 5c7e - Add pedit action with RAW_OP offset u8 add value
ok 8 2893 - Add pedit action with RAW_OP offset u8 quad
ok 9 3a07 - Add pedit action with RAW_OP offset u8-u16-u8
ok 10 ab0f - Add pedit action with RAW_OP offset u16-u8-u8
ok 11 9d12 - Add pedit action with RAW_OP offset u32 set u16 clear u8 invert
ok 12 ebfa - Add pedit action with RAW_OP offset overflow u32 (INVALID)
ok 13 f512 - Add pedit action with RAW_OP offset u16 at offmask shift set
ok 14 c2cb - Add pedit action with RAW_OP offset u32 retain value
ok 15 1762 - Add pedit action with RAW_OP offset u8 clear value
ok 16 bcee - Add pedit action with RAW_OP offset u8 retain value
ok 17 e89f - Add pedit action with RAW_OP offset u16 retain value
ok 18 c282 - Add pedit action with RAW_OP offset u32 clear value
ok 19 c422 - Add pedit action with RAW_OP offset u16 invert value
ok 20 d3d3 - Add pedit action with RAW_OP offset u32 invert value
ok 21 57e5 - Add pedit action with RAW_OP offset u8 preserve value
ok 22 99e0 - Add pedit action with RAW_OP offset u16 preserve value
ok 23 1892 - Add pedit action with RAW_OP offset u32 preserve value
ok 24 4b60 - Add pedit action with RAW_OP negative offset u16/u32 set value
ok 25 a5a7 - Add pedit action with LAYERED_OP eth set src
ok 26 86d4 - Add pedit action with LAYERED_OP eth set src & dst
ok 27 f8a9 - Add pedit action with LAYERED_OP eth set dst
ok 28 c715 - Add pedit action with LAYERED_OP eth set src (INVALID)
ok 29 8131 - Add pedit action with LAYERED_OP eth set dst (INVALID)
ok 30 ba22 - Add pedit action with LAYERED_OP eth type set/clear sequence
ok 31 dec4 - Add pedit action with LAYERED_OP eth set type (INVALID)
ok 32 ab06 - Add pedit action with LAYERED_OP eth add type
ok 33 918d - Add pedit action with LAYERED_OP eth invert src
ok 34 a8d4 - Add pedit action with LAYERED_OP eth invert dst
ok 35 ee13 - Add pedit action with LAYERED_OP eth invert type
ok 36 7588 - Add pedit action with LAYERED_OP ip set src
ok 37 0fa7 - Add pedit action with LAYERED_OP ip set dst
ok 38 5810 - Add pedit action with LAYERED_OP ip set src & dst
ok 39 1092 - Add pedit action with LAYERED_OP ip set ihl & dsfield
ok 40 02d8 - Add pedit action with LAYERED_OP ip set ttl & protocol
ok 41 3e2d - Add pedit action with LAYERED_OP ip set ttl (INVALID)
ok 42 31ae - Add pedit action with LAYERED_OP ip ttl clear/set
ok 43 486f - Add pedit action with LAYERED_OP ip set duplicate fields
ok 44 e790 - Add pedit action with LAYERED_OP ip set ce, df, mf, firstfrag, nofrag fields
ok 45 cc8a - Add pedit action with LAYERED_OP ip set tos
ok 46 7a17 - Add pedit action with LAYERED_OP ip set precedence
ok 47 c3b6 - Add pedit action with LAYERED_OP ip add tos
ok 48 43d3 - Add pedit action with LAYERED_OP ip add precedence
ok 49 438e - Add pedit action with LAYERED_OP ip clear tos
ok 50 6b1b - Add pedit action with LAYERED_OP ip clear precedence
ok 51 824a - Add pedit action with LAYERED_OP ip invert tos
ok 52 106f - Add pedit action with LAYERED_OP ip invert precedence
ok 53 6829 - Add pedit action with LAYERED_OP beyond ip set dport & sport
ok 54 afd8 - Add pedit action with LAYERED_OP beyond ip set icmp_type & icmp_code
ok 55 3143 - Add pedit action with LAYERED_OP beyond ip set dport (INVALID)
ok 56 815c - Add pedit action with LAYERED_OP ip6 set src
ok 57 4dae - Add pedit action with LAYERED_OP ip6 set dst
ok 58 fc1f - Add pedit action with LAYERED_OP ip6 set src & dst
ok 59 6d34 - Add pedit action with LAYERED_OP ip6 dst retain value (INVALID)
ok 60 94bb - Add pedit action with LAYERED_OP ip6 traffic_class
ok 61 6f5e - Add pedit action with LAYERED_OP ip6 flow_lbl
ok 62 6795 - Add pedit action with LAYERED_OP ip6 set payload_len, nexthdr, hoplimit
ok 63 1442 - Add pedit action with LAYERED_OP tcp set dport & sport
ok 64 b7ac - Add pedit action with LAYERED_OP tcp sport set (INVALID)
ok 65 cfcc - Add pedit action with LAYERED_OP tcp flags set
ok 66 3bc4 - Add pedit action with LAYERED_OP tcp set dport, sport & flags fields
ok 67 f1c8 - Add pedit action with LAYERED_OP udp set dport & sport
ok 68 d784 - Add pedit action with mixed RAW/LAYERED_OP #1
ok 69 70ca - Add pedit action with mixed RAW/LAYERED_OP #2
====================

Link: https://lore.kernel.org/r/20230131190512.3805897-1-pctammela@mojatatu.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet/sched: simplify tcf_pedit_act
Pedro Tammela [Tue, 31 Jan 2023 19:05:12 +0000 (16:05 -0300)]
net/sched: simplify tcf_pedit_act

Remove the check for a negative number of keys as
this cannot ever happen

Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Pedro Tammela <pctammela@mojatatu.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet/sched: transition act_pedit to rcu and percpu stats
Pedro Tammela [Tue, 31 Jan 2023 19:05:11 +0000 (16:05 -0300)]
net/sched: transition act_pedit to rcu and percpu stats

The software pedit action didn't get the same love as some of the
other actions and it's still using spinlocks and shared stats in the
datapath.
Transition the action to rcu and percpu stats as this improves the
action's performance dramatically on multiple cpu deployments.

Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Pedro Tammela <pctammela@mojatatu.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agoMerge tag 'rxrpc-next-20230131' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Abeni [Thu, 2 Feb 2023 11:47:28 +0000 (12:47 +0100)]
Merge tag 'rxrpc-next-20230131' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
Here's the fifth part of patches in the process of moving rxrpc from doing
a lot of its stuff in softirq context to doing it in an I/O thread in
process context and thereby making it easier to support a larger SACK
table.

The full description is in the description for the first part[1] which is
now upstream.  The second and third parts are also upstream[2].  A subset
of the original fourth part[3] got applied as a fix for a race[4].

The fifth part includes some cleanups:

 (1) Miscellaneous trace header cleanups: fix a trace string, display the
     security index in rx_packet rather than displaying the type twice,
     remove some whitespace to make checkpatch happier and remove some
     excess tabulation.

 (2) Convert ->recvmsg_lock to a spinlock as it's only ever locked
     exclusively.

 (3) Make ->ackr_window and ->ackr_nr_unacked non-atomic as they're only
     used in the I/O thread.

 (4) Don't use call->tx_lock to access ->tx_buffer as that is only accessed
     inside the I/O thread.  sendmsg() loads onto ->tx_sendmsg and the I/O
     thread decants from that to the buffer.

 (5) Remove local->defrag_sem as DATA packets are transmitted serially by
     the I/O thread.

 (6) Remove the service connection bundle is it was only used for its
     channel_lock - which has now gone.

And some more significant changes:

 (7) Add a debugging option to allow a delay to be injected into packet
     reception to help investigate the behaviour over longer links than
     just a few cm.

 (8) Generate occasional PING ACKs to probe for RTT information during a
     receive heavy call.

 (9) Simplify the SACK table maintenance and ACK generation.  Now that both
     parts are done in the same thread, there's no possibility of a race
     and no need to try and be cunning to avoid taking a BH spinlock whilst
     copying the SACK table (which in the future will be up to 2K) and no
     need to rotate the copy to fit the ACK packet table.

(10) Use SKB_CONSUMED when freeing received DATA packets (stop dropwatch
     complaining).

* tag 'rxrpc-next-20230131' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
  rxrpc: Kill service bundle
  rxrpc: Change rx_packet tracepoint to display securityIndex not type twice
  rxrpc: Show consumed and freed packets as non-dropped in dropwatch
  rxrpc: Remove local->defrag_sem
  rxrpc: Don't lock call->tx_lock to access call->tx_buffer
  rxrpc: Simplify ACK handling
  rxrpc: De-atomic call->ackr_window and call->ackr_nr_unacked
  rxrpc: Generate extra pings for RTT during heavy-receive call
  rxrpc: Allow a delay to be injected into packet reception
  rxrpc: Convert call->recvmsg_lock to a spinlock
  rxrpc: Shrink the tabulation in the rxrpc trace header a bit
  rxrpc: Remove whitespace before ')' in trace header
  rxrpc: Fix trace string
====================

Link: https://lore.kernel.org/all/20230131171227.3912130-1-dhowells@redhat.com/
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agoocteontx2-af: Removed unnecessary debug messages.
Sunil Goutham [Wed, 1 Feb 2023 04:03:01 +0000 (09:33 +0530)]
octeontx2-af: Removed unnecessary debug messages.

NPC exact match feature is supported only on one silicon
variant, removed debug messages which print that this
feature is not available on all other silicon variants.

Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230201040301.1034843-1-rkannoth@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agovirtio-net: fix possible unsigned integer overflow
Heng Qi [Tue, 31 Jan 2023 08:50:04 +0000 (16:50 +0800)]
virtio-net: fix possible unsigned integer overflow

When the single-buffer xdp is loaded and after xdp_linearize_page()
is called, *num_buf becomes 0 and (*num_buf - 1) may overflow into
a large integer in virtnet_build_xdp_buff_mrg(), resulting in
unexpected packet dropping.

Fixes: ef75cb51f139 ("virtio-net: build xdp_buff with multi buffers")
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://lore.kernel.org/r/20230131085004.98687-1-hengqi@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonetlink: provide an ability to set default extack message
Leon Romanovsky [Tue, 31 Jan 2023 13:31:57 +0000 (15:31 +0200)]
netlink: provide an ability to set default extack message

In netdev common pattern, extack pointer is forwarded to the drivers
to be filled with error message. However, the caller can easily
overwrite the filled message.

Instead of adding multiple "if (!extack->_msg)" checks before any
NL_SET_ERR_MSG() call, which appears after call to the driver, let's
add new macro to common code.

[1] https://lore.kernel.org/all/Y9Irgrgf3uxOjwUm@unreal
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/6993fac557a40a1973dfa0095107c3d03d40bec1.1675171790.git.leon@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoneighbor: fix proxy_delay usage when it is zero
Brian Haley [Mon, 30 Jan 2023 17:14:28 +0000 (12:14 -0500)]
neighbor: fix proxy_delay usage when it is zero

When set to zero, the neighbor sysctl proxy_delay value
does not cause an immediate reply for ARP/ND requests
as expected, it instead causes a random delay between
[0, U32_MAX). Looking at this comment from
__get_random_u32_below() explains the reason:

/*
 * This function is technically undefined for ceil == 0, and in fact
 * for the non-underscored constant version in the header, we build bug
 * on that. But for the non-constant case, it's convenient to have that
 * evaluate to being a straight call to get_random_u32(), so that
 * get_random_u32_inclusive() can work over its whole range without
 * undefined behavior.
 */

Added helper function that does not call get_random_u32_below()
if proxy_delay is zero and just uses the current value of
jiffies instead, causing pneigh_enqueue() to respond
immediately.

Also added definition of proxy_delay to ip-sysctl.txt since
it was missing.

Signed-off-by: Brian Haley <haleyb.dev@gmail.com>
Link: https://lore.kernel.org/r/20230130171428.367111-1-haleyb.dev@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch 'net-support-ipv4-big-tcp'
Jakub Kicinski [Thu, 2 Feb 2023 04:54:29 +0000 (20:54 -0800)]
Merge branch 'net-support-ipv4-big-tcp'

Xin Long says:

====================
net: support ipv4 big tcp

This is similar to the BIG TCP patchset added by Eric for IPv6:

  https://lwn.net/Articles/895398/

Different from IPv6, IPv4 tot_len is 16-bit long only, and IPv4 header
doesn't have exthdrs(options) for the BIG TCP packets' length. To make
it simple, as David and Paolo suggested, we set IPv4 tot_len to 0 to
indicate this might be a BIG TCP packet and use skb->len as the real
IPv4 total length.

This will work safely, as all BIG TCP packets are GSO/GRO packets and
processed on the same host as they were created; There is no padding
in GSO/GRO packets, and skb->len - network_offset is exactly the IPv4
packet total length; Also, before implementing the feature, all those
places that may get iph tot_len from BIG TCP packets are taken care
with some new APIs:

Patch 1 adds some APIs for iph tot_len setting and getting, which are
used in all these places where IPv4 BIG TCP packets may reach in Patch
2-7, Patch 8 adds a GSO_TCP tp_status for af_packet users, and Patch 9
add new netlink attributes to make IPv4 BIG TCP independent from IPv6
BIG TCP on configuration, and Patch 10 implements this feature.

Note that the similar change as in Patch 2-6 are also needed for IPv6
BIG TCP packets, and will be addressed in another patchset.

The similar performance test is done for IPv4 BIG TCP with 25Gbit NIC
and 1.5K MTU:

No BIG TCP:
for i in {1..10}; do netperf -t TCP_RR -H 192.168.100.1 -- -r80000,80000 -O MIN_LATENCY,P90_LATENCY,P99_LATENCY,THROUGHPUT|tail -1; done
168          322          337          3776.49
143          236          277          4654.67
128          258          288          4772.83
171          229          278          4645.77
175          228          243          4678.93
149          239          279          4599.86
164          234          268          4606.94
155          276          289          4235.82
180          255          268          4418.95
168          241          249          4417.82

Enable BIG TCP:
ip link set dev ens1f0np0 gro_ipv4_max_size 128000 gso_ipv4_max_size 128000
for i in {1..10}; do netperf -t TCP_RR -H 192.168.100.1 -- -r80000,80000 -O MIN_LATENCY,P90_LATENCY,P99_LATENCY,THROUGHPUT|tail -1; done
161          241          252          4821.73
174          205          217          5098.28
167          208          220          5001.43
164          228          249          4883.98
150          233          249          4914.90
180          233          244          4819.66
154          208          219          5004.92
157          209          247          4999.78
160          218          246          4842.31
174          206          217          5080.99

Thanks for the feedback from Eric and David Ahern.
====================

Link: https://lore.kernel.org/r/cover.1674921359.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: add support for ipv4 big tcp
Xin Long [Sat, 28 Jan 2023 15:58:39 +0000 (10:58 -0500)]
net: add support for ipv4 big tcp

Similar to Eric's IPv6 BIG TCP, this patch is to enable IPv4 BIG TCP.

Firstly, allow sk->sk_gso_max_size to be set to a value greater than
GSO_LEGACY_MAX_SIZE by not trimming gso_max_size in sk_trim_gso_size()
for IPv4 TCP sockets.

Then on TX path, set IP header tot_len to 0 when skb->len > IP_MAX_MTU
in __ip_local_out() to allow to send BIG TCP packets, and this implies
that skb->len is the length of a IPv4 packet; On RX path, use skb->len
as the length of the IPv4 packet when the IP header tot_len is 0 and
skb->len > IP_MAX_MTU in ip_rcv_core(). As the API iph_set_totlen() and
skb_ip_totlen() are used in __ip_local_out() and ip_rcv_core(), we only
need to update these APIs.

Also in GRO receive, add the check for ETH_P_IP/IPPROTO_TCP, and allows
the merged packet size >= GRO_LEGACY_MAX_SIZE in skb_gro_receive(). In
GRO complete, set IP header tot_len to 0 when the merged packet size
greater than IP_MAX_MTU in iph_set_totlen() so that it can be processed
on RX path.

Note that by checking skb_is_gso_tcp() in API iph_totlen(), it makes
this implementation safe to use iph->len == 0 indicates IPv4 BIG TCP
packets.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: add gso_ipv4_max_size and gro_ipv4_max_size per device
Xin Long [Sat, 28 Jan 2023 15:58:38 +0000 (10:58 -0500)]
net: add gso_ipv4_max_size and gro_ipv4_max_size per device

This patch introduces gso_ipv4_max_size and gro_ipv4_max_size
per device and adds netlink attributes for them, so that IPV4
BIG TCP can be guarded by a separate tunable in the next patch.

To not break the old application using "gso/gro_max_size" for
IPv4 GSO packets, this patch updates "gso/gro_ipv4_max_size"
in netif_set_gso/gro_max_size() if the new size isn't greater
than GSO_LEGACY_MAX_SIZE, so that nothing will change even if
userspace doesn't realize the new netlink attributes.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agopacket: add TP_STATUS_GSO_TCP for tp_status
Xin Long [Sat, 28 Jan 2023 15:58:37 +0000 (10:58 -0500)]
packet: add TP_STATUS_GSO_TCP for tp_status

Introduce TP_STATUS_GSO_TCP tp_status flag to tell the af_packet user
that this is a TCP GSO packet. When parsing IPv4 BIG TCP packets in
tcpdump/libpcap, it can use tp_len as the IPv4 packet len when this
flag is set, as iph tot_len is set to 0 for IPv4 BIG TCP packets.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoipvlan: use skb_ip_totlen in ipvlan_get_L3_hdr
Xin Long [Sat, 28 Jan 2023 15:58:36 +0000 (10:58 -0500)]
ipvlan: use skb_ip_totlen in ipvlan_get_L3_hdr

ipvlan devices calls netif_inherit_tso_max() to get the tso_max_size/segs
from the lower device, so when lower device supports BIG TCP, the ipvlan
devices support it too. We also should consider its iph tot_len accessing.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agocipso_ipv4: use iph_set_totlen in skbuff_setattr
Xin Long [Sat, 28 Jan 2023 15:58:35 +0000 (10:58 -0500)]
cipso_ipv4: use iph_set_totlen in skbuff_setattr

It may process IPv4 TCP GSO packets in cipso_v4_skbuff_setattr(), so
the iph->tot_len update should use iph_set_totlen().

Note that for these non GSO packets, the new iph tot_len with extra
iph option len added may become greater than 65535, the old process
will cast it and set iph->tot_len to it, which is a bug. In theory,
iph options shouldn't be added for these big packets in here, a fix
may be needed here in the future. For now this patch is only to set
iph->tot_len to 0 when it happens.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonetfilter: use skb_ip_totlen and iph_totlen
Xin Long [Sat, 28 Jan 2023 15:58:34 +0000 (10:58 -0500)]
netfilter: use skb_ip_totlen and iph_totlen

There are also quite some places in netfilter that may process IPv4 TCP
GSO packets, we need to replace them too.

In length_mt(), we have to use u_int32_t/int to accept skb_ip_totlen()
return value, otherwise it may overflow and mismatch. This change will
also help us add selftest for IPv4 BIG TCP in the following patch.

Note that we don't need to replace the one in tcpmss_tg4(), as it will
return if there is data after tcphdr in tcpmss_mangle_packet(). The
same in mangle_contents() in nf_nat_helper.c, it returns false when
skb->len + extra > 65535 in enlarge_skb().

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: sched: use skb_ip_totlen and iph_totlen
Xin Long [Sat, 28 Jan 2023 15:58:33 +0000 (10:58 -0500)]
net: sched: use skb_ip_totlen and iph_totlen

There are 1 action and 1 qdisc that may process IPv4 TCP GSO packets
and access iph->tot_len, replace them with skb_ip_totlen() and
iph_totlen() accordingly.

Note that we don't need to replace the one in tcf_csum_ipv4(), as it
will return for TCP GSO packets in tcf_csum_ipv4_tcp().

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoopenvswitch: use skb_ip_totlen in conntrack
Xin Long [Sat, 28 Jan 2023 15:58:32 +0000 (10:58 -0500)]
openvswitch: use skb_ip_totlen in conntrack

IPv4 GSO packets may get processed in ovs_skb_network_trim(),
and we need to use skb_ip_totlen() to get iph totlen.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Aaron Conole <aconole@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agobridge: use skb_ip_totlen in br netfilter
Xin Long [Sat, 28 Jan 2023 15:58:31 +0000 (10:58 -0500)]
bridge: use skb_ip_totlen in br netfilter

These 3 places in bridge netfilter are called on RX path after GRO
and IPv4 TCP GSO packets may come through, so replace iph tot_len
accessing with skb_ip_totlen() in there.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: add a couple of helpers for iph tot_len
Xin Long [Sat, 28 Jan 2023 15:58:30 +0000 (10:58 -0500)]
net: add a couple of helpers for iph tot_len

This patch adds three APIs to replace the iph->tot_len setting
and getting in all places where IPv4 BIG TCP packets may reach,
they will be used in the following patches.

Note that iph_totlen() will be used when iph is not in linear
data of the skb.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch 'virtio_net-vdpa-update-mac-address-when-it-is-generated-by-virtio-net'
Jakub Kicinski [Thu, 2 Feb 2023 04:53:06 +0000 (20:53 -0800)]
Merge branch 'virtio_net-vdpa-update-mac-address-when-it-is-generated-by-virtio-net'

Laurent Vivier says:

====================
virtio_net: vdpa: update MAC address when it is generated by virtio-net

When the MAC address is not provided by the vdpa device virtio_net
driver assigns a random one without notifying the device.
The consequence, in the case of mlx5_vdpa, is the internal routing
tables of the device are not updated and this can block the
communication between two namespaces.

To fix this problem, use virtnet_send_command(VIRTIO_NET_CTRL_MAC)
to set the address from virtnet_probe() when the MAC address is
not provided by the device.
====================

Link: https://lore.kernel.org/r/20230127204500.51930-1-lvivier@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agovirtio_net: notify MAC address change on device initialization
Laurent Vivier [Fri, 27 Jan 2023 20:45:00 +0000 (21:45 +0100)]
virtio_net: notify MAC address change on device initialization

In virtnet_probe(), if the device doesn't provide a MAC address the
driver assigns a random one.
As we modify the MAC address we need to notify the device to allow it
to update all the related information.

The problem can be seen with vDPA and mlx5_vdpa driver as it doesn't
assign a MAC address by default. The virtio_net device uses a random
MAC address (we can see it with "ip link"), but we can't ping a net
namespace from another one using the virtio-vdpa device because the
new MAC address has not been provided to the hardware:
RX packets are dropped since they don't go through the receive filters,
TX packets go through unaffected.

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agovirtio_net: disable VIRTIO_NET_F_STANDBY if VIRTIO_NET_F_MAC is not set
Laurent Vivier [Fri, 27 Jan 2023 20:44:59 +0000 (21:44 +0100)]
virtio_net: disable VIRTIO_NET_F_STANDBY if VIRTIO_NET_F_MAC is not set

failover relies on the MAC address to pair the primary and the standby
devices:

  "[...] the hypervisor needs to enable VIRTIO_NET_F_STANDBY
   feature on the virtio-net interface and assign the same MAC address
   to both virtio-net and VF interfaces."

  Documentation/networking/net_failover.rst

This patch disables the STANDBY feature if the MAC address is not
provided by the hypervisor.

Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonfp: correct cleanup related to DCB resources
Huayu Chen [Tue, 31 Jan 2023 16:30:33 +0000 (17:30 +0100)]
nfp: correct cleanup related to DCB resources

This patch corrects two oversights relating to releasing resources
and DCB initialisation.

1. If mapping of the dcbcfg_tbl area fails: an error should be
   propagated, allowing partial initialisation (probe) to be unwound.

2. Conversely, if where dcbcfg_tbl is successfully mapped: it should
   be unmapped in nfp_nic_dcb_clean() which is called via various error
   cleanup paths, and shutdown or removal of the PCIE device.

Fixes: 9b7fe8046d74 ("nfp: add DCB IEEE support")
Signed-off-by: Huayu Chen <huayu.chen@corigine.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230131163033.981937-1-simon.horman@corigine.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoipv6: ICMPV6: Use swap() instead of open coding it
Jiapeng Chong [Tue, 31 Jan 2023 06:34:56 +0000 (14:34 +0800)]
ipv6: ICMPV6: Use swap() instead of open coding it

Swap is a function interface that provides exchange function. To avoid
code duplication, we can use swap function.

./net/ipv6/icmp.c:344:25-26: WARNING opportunity for swap().

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3896
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20230131063456.76302-1-jiapeng.chong@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch 'devlink-trivial-names-cleanup'
Jakub Kicinski [Wed, 1 Feb 2023 18:57:03 +0000 (10:57 -0800)]
Merge branch 'devlink-trivial-names-cleanup'

Jiri Pirko says:

====================
devlink: trivial names cleanup

This is a follow-up to Jakub's devlink code split and dump iteration
helper patchset. No functional changes, just couple of renames to makes
things consistent and perhaps easier to follow.
====================

Link: https://lore.kernel.org/r/20230131090613.2131740-1-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agodevlink: rename and reorder instances of struct devlink_cmd
Jiri Pirko [Tue, 31 Jan 2023 09:06:13 +0000 (10:06 +0100)]
devlink: rename and reorder instances of struct devlink_cmd

In order to maintain naming consistency, rename and reorder all usages
of struct struct devlink_cmd in the following way:
1) Remove "gen" and replace it with "cmd" to match the struct name
2) Order devl_cmds[] and the header file to match the order
   of enum devlink_command
3) Move devl_cmd_rate_get among the peers
4) Remove "inst" for DEVLINK_CMD_GET
5) Add "_get" suffix to all to match DEVLINK_CMD_*_GET (only rate had it
   done correctly)

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agodevlink: remove "gen" from struct devlink_gen_cmd name
Jiri Pirko [Tue, 31 Jan 2023 09:06:12 +0000 (10:06 +0100)]
devlink: remove "gen" from struct devlink_gen_cmd name

No need to have "gen" inside name of the structure for devlink commands.
Remove it.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agodevlink: rename devlink_nl_instance_iter_dump() to "dumpit"
Jiri Pirko [Tue, 31 Jan 2023 09:06:11 +0000 (10:06 +0100)]
devlink: rename devlink_nl_instance_iter_dump() to "dumpit"

To have the name of the function consistent with the struct cb name,
rename devlink_nl_instance_iter_dump() to
devlink_nl_instance_iter_dumpit().

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch 'net-ipa-remaining-ipa-v5-0-support'
Jakub Kicinski [Wed, 1 Feb 2023 05:45:53 +0000 (21:45 -0800)]
Merge branch 'net-ipa-remaining-ipa-v5-0-support'

Alex Elder says:

====================
net: ipa: remaining IPA v5.0 support

This series includes almost all remaining IPA code changes required
to support IPA v5.0.  IPA register definitions and configuration
data for IPA v5.0 will be sent later (soon).  Note that the GSI
register definitions still require work.  GSI for IPA v5.0 supports
up to 256 (rather than 32) channels, and this changes the way GSI
register offsets are calculated.  A few GSI register fields also
change.

The first patch in this series increases the number of IPA endpoints
supported by the driver, from 32 to 36.  The next updates the width
of the destination field for the IP_PACKET_INIT immediate command so
it can represent up to 256 endpoints rather than just 32.  The next
adds a few definitions of some IPA registers and fields that are
first available in IPA v5.0.

The next two patches update the code that handles router and filter
table caches.  Previously these were referred to as "hashed" tables,
and the IPv4 and IPv6 tables are now combined into one "unified"
table.  The sixth and seventh patches add support for a new pulse
generator, which allows time periods to be specified with a wider
range of clock resolution.  And the last patch just defines two new
memory regions that were not previously used.
====================

Link: https://lore.kernel.org/r/20230130210158.4126129-1-elder@linaro.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: ipa: define two new memory regions
Alex Elder [Mon, 30 Jan 2023 21:01:58 +0000 (15:01 -0600)]
net: ipa: define two new memory regions

IPA v5.0 uses two memory regions not previously used.  Define them
and treat them as valid only for IPA v5.0.

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: ipa: support a third pulse register
Alex Elder [Mon, 30 Jan 2023 21:01:57 +0000 (15:01 -0600)]
net: ipa: support a third pulse register

The AP has third pulse generator available starting with IPA v5.0.
Redefine ipa_qtime_val() to support that possibility.  Pass the IPA
pointer as an argument so the version can be determined.  And stop
using the sign of the returned tick count to indicate which of two
pulse generators to use.

Instead, have the caller provide the address of a variable that will
hold the selected pulse generator for the Qtime value.  And for
version 5.0, check whether the third pulse generator best represents
the time period.

Add code in ipa_qtime_config() to configure the fourth pulse
generator for IPA v5.0+; in that case configure both the third and
fourth pulse generators to use 10 msec granularity.

Consistently use "ticks" for local variables that represent a tick
count.

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: ipa: greater timer granularity options
Alex Elder [Mon, 30 Jan 2023 21:01:56 +0000 (15:01 -0600)]
net: ipa: greater timer granularity options

Starting with IPA v5.0, the head-of-line blocking timer has more
than two pulse generators available to define timer granularity.
To prepare for that, change the way the field value is encoded
to use ipa_reg_encode() rather than ipa_reg_bit().

The aggregation granularity selection could (in principle) also use
an additional pulse generator starting with IPA v5.0.  Encode the
AGGR_GRAN_SEL field differently to allow that as well.

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: ipa: support zeroing new cache tables
Alex Elder [Mon, 30 Jan 2023 21:01:55 +0000 (15:01 -0600)]
net: ipa: support zeroing new cache tables

IPA v5.0+ separates the configuration of entries in the cached
(previously "hashed") routing and filtering tables into distinct
registers.  Previously a single "filter and router" register updated
entries in both tables at once; now the routing and filter table
caches have separate registers that define their content.

This patch updates the code that zeroes entries in the cached filter
and router tables to support IPA versions including v5.0+.

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: ipa: update table cache flushing
Alex Elder [Mon, 30 Jan 2023 21:01:54 +0000 (15:01 -0600)]
net: ipa: update table cache flushing

Update the code that causes filter and router table caches to be
flushed so that it supports IPA versions 5.0+.  It adds a comment in
ipa_hardware_config_hashing() that explains that cacheing does not
need to be enabled, just as before, because it's enabled by default.
(For the record, the FILT_ROUT_CACHE_CFG register would have been
used if we wanted to explicitly enable these.)

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: ipa: define IPA v5.0+ registers
Alex Elder [Mon, 30 Jan 2023 21:01:53 +0000 (15:01 -0600)]
net: ipa: define IPA v5.0+ registers

Define some new registers that appear starting with IPA v5.0, along
with enumerated types identifying their fields.  Code that uses
these will be added by upcoming patches.

Most of the new registers are related to filter and routing tables,
and in particular, their "hashed" variant.  These tables are better
described as "cached", where a hash value determines which entries
are cached.  From now on, naming related to this functionality will
use "cache" instead of "hash", and that is reflected in these new
register names.  Some registers for managing these caches and their
contents have changed as well.

A few other new field definitions for registers (unrelated to table
caches) are also defined.

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: ipa: extend endpoints in packet init command
Alex Elder [Mon, 30 Jan 2023 21:01:52 +0000 (15:01 -0600)]
net: ipa: extend endpoints in packet init command

The IP_PACKET_INIT immediate command defines the destination
endpoint to which a packet should be sent.  Prior to IPA v5.0, a
5 bit field in that command represents the endpoint, but starting
with IPA v5.0, the field is extended to 8 bits to support more than
32 endpoints.

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: ipa: support more endpoints
Alex Elder [Mon, 30 Jan 2023 21:01:51 +0000 (15:01 -0600)]
net: ipa: support more endpoints

Increase the number of endpoints supported by the driver to 36,
which IPA v5.0 supports.  This makes it impossible to check at build
time whether the supported number is too big to fit within the
(5-bit) PACKET_INIT destination endpoint field.  Instead, convert
the build time check to compare against what fits in 8 bits.

Add a check in ipa_endpoint_config() to also ensure the hardware
reports an endpoint count that's in the expected range.  Just
open-code 32 as the limit (the PACKET_INIT field mask is not
available where we'd want to use it).

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge tag 'mlx5-updates-2023-01-30' of git://git.kernel.org/pub/scm/linux/kernel...
Jakub Kicinski [Wed, 1 Feb 2023 05:35:34 +0000 (21:35 -0800)]
Merge tag 'mlx5-updates-2023-01-30' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2023-01-30

Add fast update encryption key

Jianbo Liu Says:
================

Data encryption keys (DEKs) are the keys used for data encryption and
decryption operations. Starting from version 22.33.0783, firmware is
optimized to accelerate the update of user keys into DEK object in
hardware. The support for bulk allocation and destruction of DEK
objects is added, and the bulk allocated DEKs are uninitialized, as
the bulk creation requires no input key. When offload
encryption/decryption, user gets one object from a bulk, and updates
key by a new "modify DEK" command. This command is the same as create
DEK object, but requires no heavy context memory allocation in
firmware, which consumes most cpu cycles of the create DEK command.

DEKs are cached internally by the NIC, so invalidating internal NIC
caches is required before reusing DEKs. The SYNC_CRYPTO command is
added to support it. DEK object can be reused, the keys in it can be
updated after this command is executed.

This patchset enhances the key creation and destruction flow, to get
use of this new feature. Any user, for example, ktls, ipsec and
macsec, can use it to offload keys. But, only ktls uses it, as others
don't need many keys, and caching two many DEKs in pool is wasteful.

There are two new data struts added:
    a. DEK pool. One pool is created for each key type. The bulks by
the type, are placed in the pool's different bulk lists, according to
the number of available and in_used DEKs in the bulk.
    b. DEK bulk. All DEKs in one bulk allocation are store here. There
are two bitmaps to indicate the state of each DEK.

New APIs are then added. When user need a DEK object,
    a. Fetch one bulk with avail DEKs, from the partial_list or
avail_list, otherwise create new one.
    b. Pick one DEK, and set its need_sync and in_used bits to 1.
Move the bulk to full_list if no more available keys, or put it to
partial_list if the bulk is newly created.
    c. Update DEK object's key with user key, by the "modify DEK"
command.
    d. Return DEK struct to user, then it gets the object id and fills
it into the offload commands.
When user free a DEK,
    a. Set in_use bit to 0. If all need_sync bits are 1 and all in_use
bits of this bulk are 0, move it to sync_list.
    b. If the number of DEKs, which are freed by users, is over the
threshold (128), schedule a workqueue to do the sync process.

For the sync process, the SYNC_CRYPTO command is executed first. Then,
for each bulks in partial_list, full_list and sync_list, reset
need_sync bits of the freed DEK objects. If all need_sync bits in one
bulk are zero, move it to avail_list.

We already supported TIS pool to recycle the TISes. With this series
and TIS pool, TLS CPS performance is improved greatly.
And we tested https on the system:
    CPU: dual AMD EPYC 7763 64-Core processors
    RAM: 512G
    DEV: ConnectX-6 DX, with FW ver 22.33.0838 and TLS_OPTIMISE=true
TLS CPS performance numbers are:
    Before: 11k connections/sec
    After: 101 connections/sec

================

* tag 'mlx5-updates-2023-01-30' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
  net/mlx5e: kTLS, Improve connection rate by using fast update encryption key
  net/mlx5: Keep only one bulk of full available DEKs
  net/mlx5: Add async garbage collector for DEK bulk
  net/mlx5: Reuse DEKs after executing SYNC_CRYPTO command
  net/mlx5: Use bulk allocation for fast update encryption key
  net/mlx5: Add bulk allocation and modify_dek operation
  net/mlx5: Add support SYNC_CRYPTO command
  net/mlx5: Add new APIs for fast update encryption key
  net/mlx5: Refactor the encryption key creation
  net/mlx5: Add const to the key pointer of encryption key creation
  net/mlx5: Prepare for fast crypto key update if hardware supports it
  net/mlx5: Change key type to key purpose
  net/mlx5: Add IFC bits and enums for crypto key
  net/mlx5: Add IFC bits for general obj create param
  net/mlx5: Header file for crypto
====================

Link: https://lore.kernel.org/r/20230131031201.35336-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next...
Jakub Kicinski [Wed, 1 Feb 2023 05:04:25 +0000 (21:04 -0800)]
Merge branch '10GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
Intel Wired LAN: Remove redundant Device Control Error Reporting Enable

Bjorn Helgaas says:

Since f26e58bf6f54 ("PCI/AER: Enable error reporting when AER is native"),
the PCI core sets the Device Control bits that enable error reporting for
PCIe devices.

This series removes redundant calls to pci_enable_pcie_error_reporting()
that do the same thing from several NIC drivers.

There are several more drivers where this should be removed; I started with
just the Intel drivers here.

* '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
  ixgbe: Remove redundant pci_enable_pcie_error_reporting()
  igc: Remove redundant pci_enable_pcie_error_reporting()
  igb: Remove redundant pci_enable_pcie_error_reporting()
  ice: Remove redundant pci_enable_pcie_error_reporting()
  iavf: Remove redundant pci_enable_pcie_error_reporting()
  i40e: Remove redundant pci_enable_pcie_error_reporting()
  fm10k: Remove redundant pci_enable_pcie_error_reporting()
  e1000e: Remove redundant pci_enable_pcie_error_reporting()
====================

Link: https://lore.kernel.org/r/20230130192519.686446-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch 'selftests-mlxsw-convert-to-iproute2-dcb'
Jakub Kicinski [Wed, 1 Feb 2023 05:02:12 +0000 (21:02 -0800)]
Merge branch 'selftests-mlxsw-convert-to-iproute2-dcb'

Petr Machata says:

====================
selftests: mlxsw: Convert to iproute2 dcb

There is a dedicated tool for configuration of DCB in iproute2. Use it
in the selftests instead of lldpad.

Patches #1-#3 convert three tests. Patch #4 drops the now-unnecessary
lldpad helpers.
====================

Link: https://lore.kernel.org/r/cover.1675096231.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoselftests: net: forwarding: lib: Drop lldpad_app_wait_set(), _del()
Petr Machata [Mon, 30 Jan 2023 16:40:04 +0000 (17:40 +0100)]
selftests: net: forwarding: lib: Drop lldpad_app_wait_set(), _del()

The existing users of these helpers have been converted to iproute2 dcb.
Drop the helpers.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoselftests: mlxsw: qos_defprio: Convert from lldptool to dcb
Petr Machata [Mon, 30 Jan 2023 16:40:03 +0000 (17:40 +0100)]
selftests: mlxsw: qos_defprio: Convert from lldptool to dcb

Set up default port priority through the iproute2 dcb tool, which is easier
to understand and manage.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoselftests: mlxsw: qos_dscp_router: Convert from lldptool to dcb
Petr Machata [Mon, 30 Jan 2023 16:40:02 +0000 (17:40 +0100)]
selftests: mlxsw: qos_dscp_router: Convert from lldptool to dcb

Set up DSCP prioritization through the iproute2 dcb tool, which is easier
to understand and manage.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoselftests: mlxsw: qos_dscp_bridge: Convert from lldptool to dcb
Petr Machata [Mon, 30 Jan 2023 16:40:01 +0000 (17:40 +0100)]
selftests: mlxsw: qos_dscp_bridge: Convert from lldptool to dcb

Set up DSCP prioritization through the iproute2 dcb tool, which is easier
to understand and manage.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch 'net-mdio-add-amlogic-gxl-mdio-mux-support'
Jakub Kicinski [Wed, 1 Feb 2023 04:59:09 +0000 (20:59 -0800)]
Merge branch 'net-mdio-add-amlogic-gxl-mdio-mux-support'

Jerome Brunet says:

====================
net: mdio: add amlogic gxl mdio mux support

Add support for the MDIO multiplexer found in the Amlogic GXL SoC family.
This multiplexer allows to choose between the external (SoC pins) MDIO bus,
or the internal one leading to the integrated 10/100M PHY.

This multiplexer has been handled with the mdio-mux-mmioreg generic driver
so far. When it was added, it was thought the logic was handled by a
single register.

It turns out more than a single register need to be properly set.
As long as the device is using the Amlogic vendor bootloader, or upstream
u-boot with net support, it is working fine since the kernel is inheriting
the bootloader settings. Without net support in the bootloader, this glue
comes unset in the kernel and only the external path may operate properly.

With this driver (and the associated change in
arch/arm64/boot/dts/amlogic/meson-gxl.dtsi), the kernel no longer relies
on the bootloader to set things up, fixing the problem.
====================

Link: https://lore.kernel.org/r/20230130151616.375168-1-jbrunet@baylibre.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: mdio: add amlogic gxl mdio mux support
Jerome Brunet [Mon, 30 Jan 2023 15:16:16 +0000 (16:16 +0100)]
net: mdio: add amlogic gxl mdio mux support

Add support for the mdio mux and internal phy glue of the GXL SoC
family

Reported-by: Da Xue <da@lessconfused.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agodt-bindings: net: add amlogic gxl mdio multiplexer
Jerome Brunet [Mon, 30 Jan 2023 15:16:15 +0000 (16:16 +0100)]
dt-bindings: net: add amlogic gxl mdio multiplexer

Add documentation for the MDIO bus multiplexer found on the Amlogic GXL
SoC family

Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agoMerge branch 'tools-ynl-more-docs-and-basic-ethtool-support'
Jakub Kicinski [Wed, 1 Feb 2023 04:36:05 +0000 (20:36 -0800)]
Merge branch 'tools-ynl-more-docs-and-basic-ethtool-support'

Jakub Kicinski says:

====================
tools: ynl: more docs and basic ethtool support

I got discouraged from supporting ethtool in specs, because
generating the user space C code seems a little tricky.
The messages are ID'ed in a "directional" way (to and from
kernel are separate ID "spaces"). There is value, however,
in having the spec and being able to for example use it
in Python.

After paying off some technical debt - add a partial
ethtool spec. Partial because the header for ethtool is almost
a 1000 LoC, so converting in one sitting is tough. But adding
new commands should be trivial now.

Last but not least I add more docs, I realized that I've been
sending a similar "instructions" email to people working on
new families. It's now intro-specs.rst.
====================

Link: https://lore.kernel.org/r/20230131023354.1732677-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: net: use python3 explicitly
Jakub Kicinski [Tue, 31 Jan 2023 02:33:54 +0000 (18:33 -0800)]
tools: net: use python3 explicitly

The scripts require Python 3 and some distros are dropping
Python 2 support.

Reported-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agodocs: netlink: add a starting guide for working with specs
Jakub Kicinski [Tue, 31 Jan 2023 02:33:53 +0000 (18:33 -0800)]
docs: netlink: add a starting guide for working with specs

We have a bit of documentation about the internals of Netlink
and the specs, but really the goal is for most people to not
worry about those. Add a practical guide for beginners who
want to poke at the specs.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonetlink: specs: add partial specification for ethtool
Jakub Kicinski [Tue, 31 Jan 2023 02:33:52 +0000 (18:33 -0800)]
netlink: specs: add partial specification for ethtool

Ethtool is one of the most actively developed families.
With the changes to the CLI it should be possible to use
the YNL based code for easy prototyping and development.
Add a partial family definition. I've tested the string
set and rings. I don't have any MAC Merge implementation
to test with, but I added the definition for it, anyway,
because it's last. New commands can simply be added at
the end without having to worry about manually providing
IDs / values.

Set (with notification support - None is the response,
the data is from the notification):

$ sudo ./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/ethtool.yaml \
    --do rings-set \
    --json '{"header":{"dev-name":"enp0s31f6"}, "rx":129}' \
    --subscribe monitor
None
[{'msg': {'header': {'dev-index': 2, 'dev-name': 'enp0s31f6'},
          'rx': 136,
          'rx-max': 4096,
          'tx': 256,
          'tx-max': 4096,
          'tx-push': 0},
  'name': 'rings-ntf'}]

Do / dump (yes, the kernel requires that even for dump and even
if empty - the "header" nest must be there):

$ ./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/ethtool.yaml \
    --do rings-get \
    --json '{"header":{"dev-index": 2}}'
{'header': {'dev-index': 2, 'dev-name': 'enp0s31f6'},
 'rx': 136,
 'rx-max': 4096,
 'tx': 256,
 'tx-max': 4096,
 'tx-push': 0}

$ ./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/ethtool.yaml \
    --dump rings-get \
    --json '{"header":{}}'
[{'header': {'dev-index': 2, 'dev-name': 'enp0s31f6'},
  'rx': 136,
  'rx-max': 4096,
  'tx': 256,
  'tx-max': 4096,
  'tx-push': 0},
 {'header': {'dev-index': 3, 'dev-name': 'wlp0s20f3'}, 'tx-push': 0},
 {'header': {'dev-index': 19, 'dev-name': 'enp58s0u1u1'},
  'rx': 100,
  'rx-max': 4096,
  'tx-push': 0}]

And error reporting:

$ ./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/ethtool.yaml \
    --dump rings-get \
    --json '{"header":{"flags":5}}'
Netlink error: Invalid argument
nl_len = 68 (52) nl_flags = 0x300 nl_type = 2
error: -22 extack: {'msg': 'reserved bit set',
                         'bad-attr-offs': 24,
 'bad-attr': '.header.flags'}
None

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonetlink: specs: finish up operation enum-models
Jakub Kicinski [Tue, 31 Jan 2023 02:33:51 +0000 (18:33 -0800)]
netlink: specs: finish up operation enum-models

I had a (bright?) idea of introducing the concept of enum-models
to account for all the weird ways families enumerate their messages.
I've never finished it because generating C code for each of them
is pretty daunting. But for languages which can use ID values directly
the support is simple enough, so clean this up a bit.

"unified" model is what I recommend going forward.
"directional" model is what ethtool uses.
"notify-split" is used by the proposed DPLL code, but we can just
make them use "unified", it hasn't been merged :)

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: load jsonschema on demand
Jakub Kicinski [Tue, 31 Jan 2023 02:33:50 +0000 (18:33 -0800)]
tools: ynl: load jsonschema on demand

The CLI script tries to validate jsonschema by default.
It's seems better to validate too many times than too few.
However, when copying the scripts to random servers having
to install jsonschema is tedious. Load jsonschema via
importlib, and let the user opt out.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: use operation names from spec on the CLI
Jakub Kicinski [Tue, 31 Jan 2023 02:33:49 +0000 (18:33 -0800)]
tools: ynl: use operation names from spec on the CLI

When I wrote the first version of the Python code I was quite
excited that we can generate class methods directly from the
spec. Unfortunately we need to use valid identifiers for method
names (specifically no dashes are allowed). Don't reuse those
names on the CLI, it's much more natural to use the operation
names exactly as listed in the spec.

Instead of:
  ./cli --do rings_get
use:
  ./cli --do rings-get

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: support pretty printing bad attribute names
Jakub Kicinski [Tue, 31 Jan 2023 02:33:48 +0000 (18:33 -0800)]
tools: ynl: support pretty printing bad attribute names

One of my favorite features of the Netlink specs is that they
make decoding structured extack a ton easier.
Implement pretty printing bad attribute names in YNL.

For example it will now say:

  'bad-attr': '.header.flags'

rather than the useless:

  'bad-attr-offs': 32

Proof:

  $ ./cli.py --spec ethtool.yaml --do rings_get \
     --json '{"header":{"dev-index":1, "flags":4}}'
  Netlink error: Invalid argument
  nl_len = 68 (52) nl_flags = 0x300 nl_type = 2
error: -22 extack: {'msg': 'reserved bit set',
 'bad-attr': '.header.flags'}

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: support multi-attr
Jakub Kicinski [Tue, 31 Jan 2023 02:33:47 +0000 (18:33 -0800)]
tools: ynl: support multi-attr

Ethtool uses mutli-attr, add the support to YNL.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: support directional enum-model in CLI
Jakub Kicinski [Tue, 31 Jan 2023 02:33:46 +0000 (18:33 -0800)]
tools: ynl: support directional enum-model in CLI

Support families which use different IDs for messages
to and from the kernel.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: add support for types needed by ethtool
Jakub Kicinski [Tue, 31 Jan 2023 02:33:45 +0000 (18:33 -0800)]
tools: ynl: add support for types needed by ethtool

Ethtool needs support for handful of extra types.
It doesn't have the definitions section yet.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: use the common YAML loading and validation code
Jakub Kicinski [Tue, 31 Jan 2023 02:33:44 +0000 (18:33 -0800)]
tools: ynl: use the common YAML loading and validation code

Adapt the common object hierarchy in code gen and CLI.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: add an object hierarchy to represent parsed spec
Jakub Kicinski [Tue, 31 Jan 2023 02:33:43 +0000 (18:33 -0800)]
tools: ynl: add an object hierarchy to represent parsed spec

There's a lot of copy and pasting going on between the "cli"
and code gen when it comes to representing the parsed spec.
Create a library which both can use.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl: move the cli and netlink code around
Jakub Kicinski [Tue, 31 Jan 2023 02:33:42 +0000 (18:33 -0800)]
tools: ynl: move the cli and netlink code around

Move the CLI code out of samples/ and the library part
of it into tools/net/ynl/lib/. This way we can start
sharing some code with the code gen.

Initially I thought that code gen is too C-specific to
share anything but basic stuff like calculating values
for enums can easily be shared.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agotools: ynl-gen: prevent do / dump reordering
Jakub Kicinski [Tue, 31 Jan 2023 02:33:41 +0000 (18:33 -0800)]
tools: ynl-gen: prevent do / dump reordering

An earlier fix tried to address generated code jumping around
one code-gen run to another. Turns out dict()s are already
ordered since Python 3.7, the problem is that we iterate over
operation modes using a set(). Sets are unordered in Python.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agorxrpc: Kill service bundle
David Howells [Wed, 16 Nov 2022 16:42:02 +0000 (16:42 +0000)]
rxrpc: Kill service bundle

Now that the bundle->channel_lock has been eliminated, we don't need the
dummy service bundle anymore.  It's purpose was purely to provide the
channel_lock for service connections.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Change rx_packet tracepoint to display securityIndex not type twice
David Howells [Mon, 5 Dec 2022 17:38:31 +0000 (17:38 +0000)]
rxrpc: Change rx_packet tracepoint to display securityIndex not type twice

Change the rx_packet tracepoint to display the securityIndex from the
packet header instead of displaying the type in numeric form.  There's no
need for the latter, as the display of the type in symbolic form will fall
back automatically to displaying the hex value if no symbol is available.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Show consumed and freed packets as non-dropped in dropwatch
David Howells [Fri, 11 Nov 2022 22:01:04 +0000 (22:01 +0000)]
rxrpc: Show consumed and freed packets as non-dropped in dropwatch

Set a reason when freeing a packet that has been consumed such that
dropwatch doesn't complain that it has been dropped.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Remove local->defrag_sem
David Howells [Mon, 17 Oct 2022 21:52:06 +0000 (22:52 +0100)]
rxrpc: Remove local->defrag_sem

We no longer need local->defrag_sem as all DATA packet transmission is now
done from one thread, so remove it.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Don't lock call->tx_lock to access call->tx_buffer
David Howells [Mon, 17 Oct 2022 21:48:58 +0000 (22:48 +0100)]
rxrpc: Don't lock call->tx_lock to access call->tx_buffer

call->tx_buffer is now only accessed within the I/O thread (->tx_sendmsg is
the way sendmsg passes packets to the I/O thread) so there's no need to
lock around it.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Simplify ACK handling
David Howells [Sun, 16 Oct 2022 07:01:32 +0000 (08:01 +0100)]
rxrpc: Simplify ACK handling

Now that general ACK transmission is done from the same thread as incoming
DATA packet wrangling, there's no possibility that the SACK table will be
being updated by the latter whilst the former is trying to copy it to an
ACK.

This means that we can safely rotate the SACK table whilst updating it
without having to take a lock, rather than keeping all the bits inside it
in fixed place and copying and then rotating it in the transmitter.

Therefore, simplify SACK handing by keeping track of starting point in the
ring and rotate slots down as we consume them.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: De-atomic call->ackr_window and call->ackr_nr_unacked
David Howells [Mon, 17 Oct 2022 10:44:22 +0000 (11:44 +0100)]
rxrpc: De-atomic call->ackr_window and call->ackr_nr_unacked

call->ackr_window doesn't need to be atomic as ACK generation and ACK
transmission are now done in the same thread, so drop the atomic64 handling
and split it into two separate members.

Similarly, call->ackr_nr_unacked doesn't need to be atomic now either.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Generate extra pings for RTT during heavy-receive call
David Howells [Mon, 17 Oct 2022 09:55:41 +0000 (10:55 +0100)]
rxrpc: Generate extra pings for RTT during heavy-receive call

When doing a call that has a single transmitted data packet and a massive
amount of received data packets, we only ping for one RTT sample, which
means we don't get a good reading on it.

Fix this by converting occasional IDLE ACKs into PING ACKs to elicit a
response.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Allow a delay to be injected into packet reception
David Howells [Mon, 17 Oct 2022 07:54:57 +0000 (08:54 +0100)]
rxrpc: Allow a delay to be injected into packet reception

If CONFIG_AF_RXRPC_DEBUG_RX_DELAY=y, then a delay is injected between
packets and errors being received and them being made available to the
processing code, thereby allowing the RTT to be artificially increased.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Convert call->recvmsg_lock to a spinlock
David Howells [Wed, 12 Oct 2022 21:06:52 +0000 (22:06 +0100)]
rxrpc: Convert call->recvmsg_lock to a spinlock

Convert call->recvmsg_lock to a spinlock as it's only ever write-locked.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

17 months agorxrpc: Shrink the tabulation in the rxrpc trace header a bit
David Howells [Tue, 31 Jan 2023 08:39:46 +0000 (08:39 +0000)]
rxrpc: Shrink the tabulation in the rxrpc trace header a bit

Shrink the tabulation in the rxrpc trace header a bit to allow for fields
with long type names that have been removed.

Signed-off-by: David Howells <dhowells@redhat.com>
17 months agorxrpc: Remove whitespace before ')' in trace header
David Howells [Tue, 31 Jan 2023 08:35:16 +0000 (08:35 +0000)]
rxrpc: Remove whitespace before ')' in trace header

Work around checkpatch warnings in the rxrpc trace header by removing
whitespace before ')' on lines defining the trace record struct.

Signed-off-by: David Howells <dhowells@redhat.com>
17 months agonet: ipa: use dev PM wakeirq handling
Caleb Connolly [Fri, 27 Jan 2023 20:27:58 +0000 (20:27 +0000)]
net: ipa: use dev PM wakeirq handling

Replace the enable_irq_wake() call with one to dev_pm_set_wake_irq()
instead. This will let the dev PM framework automatically manage the
the wakeup capability of the ipa IRQ and ensure that userspace requests
to enable/disable wakeup for the IPA via sysfs are respected.

Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
Reviewed-by: Alex Elder <elder@linaro.org>
Link: https://lore.kernel.org/r/20230127202758.2913612-1-caleb.connolly@linaro.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agonet: dsa: microchip: ptp: fix up PTP dependency
Arnd Bergmann [Mon, 30 Jan 2023 13:17:51 +0000 (14:17 +0100)]
net: dsa: microchip: ptp: fix up PTP dependency

When NET_DSA_MICROCHIP_KSZ_COMMON is built-in but PTP is a loadable
module, the ksz_ptp support still causes a link failure:

ld.lld-16: error: undefined symbol: ptp_clock_index
>>> referenced by ksz_ptp.c
>>>               drivers/net/dsa/microchip/ksz_ptp.o:(ksz_get_ts_info) in archive vmlinux.a

This can happen if NET_DSA_MICROCHIP_KSZ8863_SMI is enabled, or
even if none of the KSZ9477_I2C/KSZ_SPI/KSZ8863_SMI ones are active
but only the common module is.

The most straightforward way to address this is to move the
dependency to NET_DSA_MICROCHIP_KSZ_PTP itself, which can now
only be enabled if both PTP_1588_CLOCK support is reachable
from NET_DSA_MICROCHIP_KSZ_COMMON. Alternatively, one could make
NET_DSA_MICROCHIP_KSZ_COMMON a hidden Kconfig symbol and extend the
PTP_1588_CLOCK_OPTIONAL dependency to NET_DSA_MICROCHIP_KSZ8863_SMI as
well, but that is a little more fragile.

Fixes: eac1ea20261e ("net: dsa: microchip: ptp: add the posix clock support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Link: https://lore.kernel.org/r/20230130131808.1084796-1-arnd@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agoDocumentation: networking: correct spelling
Randy Dunlap [Sun, 29 Jan 2023 23:10:48 +0000 (15:10 -0800)]
Documentation: networking: correct spelling

Correct spelling problems for Documentation/networking/ as reported
by codespell.

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Cc: Jiri Pirko <jiri@nvidia.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: netdev@vger.kernel.org
Link: https://lore.kernel.org/r/20230129231053.20863-5-rdunlap@infradead.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agoibmvnic: Toggle between queue types in affinity mapping
Nick Child [Fri, 27 Jan 2023 21:43:58 +0000 (15:43 -0600)]
ibmvnic: Toggle between queue types in affinity mapping

Previously, ibmvnic IRQs were assigned to CPU numbers by assigning all
the IRQs for transmit queues then assigning all the IRQs for receive
queues. With multi-threaded processors, in a heavy RX or TX environment,
physical cores would either be overloaded or underutilized (due to the
IRQ assignment algorithm). This approach is sub-optimal because IRQs for
the same subprocess (RX or TX) would be bound to adjacent CPU numbers,
meaning they were more likely to be contending for the same core.

For example, in a system with 64 CPU's and 32 queues, the IRQs would
be bound to CPU in the following pattern:

IRQ type |  CPU number
-----------------------
TX0  | 0-1
TX1  | 2-3
<etc>
RX0  | 32-33
RX1  | 34-35
<etc>

Observe that in SMT-8, the first 4 tx queues would be sharing the
same core.

A more optimal algorithm would balance the number RX and TX IRQ's across
the physical cores. Therefore, to increase performance, distribute RX and
TX IRQs across cores by alternating between assigning IRQs for RX and TX
queues to CPUs.
With a system with 64 CPUs and 32 queues, this results in the following
pattern:

IRQ type |  CPU number
-----------------------
TX0  | 0-1
RX0  | 2-3
TX1  | 4-5
RX1  | 6-7
<etc>

Observe that in SMT-8, there is equal distribution of RX and TX IRQs
per core. In the above case, each core handles 2 TX and 2 RX IRQ's.

Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Haren Myneni <haren@linux.ibm.com>
Link: https://lore.kernel.org/r/20230127214358.318152-1-nnac123@linux.ibm.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
17 months agoMerge branch 'add-support-for-the-the-vsc7512-internal-copper-phys'
Jakub Kicinski [Tue, 31 Jan 2023 05:07:22 +0000 (21:07 -0800)]
Merge branch 'add-support-for-the-the-vsc7512-internal-copper-phys'

Colin Foster says:

====================
add support for the the vsc7512 internal copper phys

This patch series is a continuation to add support for the VSC7512:
https://patchwork.kernel.org/project/netdevbpf/list/?series=674168&state=*

That series added the framework and initial functionality for the
VSC7512 chip. Several of these patches grew during the initial
development of the framework, which is why v1 will include changelogs.
It was during v9 of that original MFD patch set that these were dropped.

With that out of the way, the VSC7512 is mainly a subset of the VSC7514
chip. The 7512 lacks an internal MIPS processor, but otherwise many of
the register definitions are identical. That is why several of these
patches are simply to expose common resources from
drivers/net/ethernet/mscc/*.

This patch only adds support for the first four ports (swp0-swp3). The
remaining ports require more significant changes to the felix driver,
and will be handled in the future.
====================

Link: https://lore.kernel.org/r/20230127193559.1001051-1-colin.foster@in-advantage.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agomfd: ocelot: add external ocelot switch control
Colin Foster [Fri, 27 Jan 2023 19:35:59 +0000 (11:35 -0800)]
mfd: ocelot: add external ocelot switch control

Utilize the existing ocelot MFD interface to add switch functionality to
the Microsemi VSC7512 chip.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Acked-for-MFD-by: Lee Jones <lee@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: dsa: ocelot: add external ocelot switch control
Colin Foster [Fri, 27 Jan 2023 19:35:58 +0000 (11:35 -0800)]
net: dsa: ocelot: add external ocelot switch control

Add control of an external VSC7512 chip.

Currently the four copper phy ports are fully functional. Communication to
external phys is also functional, but the SGMII / QSGMII interfaces are
currently non-functional.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agodt-bindings: mfd: ocelot: add ethernet-switch hardware support
Colin Foster [Fri, 27 Jan 2023 19:35:57 +0000 (11:35 -0800)]
dt-bindings: mfd: ocelot: add ethernet-switch hardware support

The main purpose of the Ocelot chips are the Ethernet switching
functionalities. Document the support for these features.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agodt-bindings: net: mscc,vsc7514-switch: add dsa binding for the vsc7512
Colin Foster [Fri, 27 Jan 2023 19:35:56 +0000 (11:35 -0800)]
dt-bindings: net: mscc,vsc7514-switch: add dsa binding for the vsc7512

The VSC7511, VSC7512, VSC7513 and VSC7514 all have the ability to be
controlled either internally by a memory-mapped CPU, or externally via
interfaces like SPI and PCIe. The internal CPU of the VSC7511 and 7512
don't have the resources to run Linux, so must be controlled via these
external interfaces in a DSA configuration.

Add mscc,vsc7512-switch compatible string to indicate that the chips are
being controlled externally in a DSA configuration.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agomfd: ocelot: prepend resource size macros to be 32-bit
Colin Foster [Fri, 27 Jan 2023 19:35:55 +0000 (11:35 -0800)]
mfd: ocelot: prepend resource size macros to be 32-bit

The *_RES_SIZE macros are initally <= 0x100. Future resource sizes will be
upwards of 0x200000 in size.

To keep things clean, fully align the RES_SIZE macros to 32-bit to do
nothing more than make the code more consistent.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Acked-for-MFD-by: Lee Jones <lee@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: dsa: felix: add functionality when not all ports are supported
Colin Foster [Fri, 27 Jan 2023 19:35:54 +0000 (11:35 -0800)]
net: dsa: felix: add functionality when not all ports are supported

When the Felix driver would probe the ports and verify functionality, it
would fail if it hit single port mode that wasn't supported by the driver.

The initial case for the VSC7512 driver will have physical ports that
exist, but aren't supported by the driver implementation. Add the
OCELOT_PORT_MODE_NONE macro to handle this scenario, and allow the Felix
driver to continue with all the ports that are currently functional.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: dsa: felix: add support for MFD configurations
Colin Foster [Fri, 27 Jan 2023 19:35:53 +0000 (11:35 -0800)]
net: dsa: felix: add support for MFD configurations

The architecture around the VSC7512 differs from existing felix drivers. In
order to add support for all the chip's features (pinctrl, MDIO, gpio) the
device had to be laid out as a multi-function device (MFD).

One difference between an MFD and a standard platform device is that the
regmaps are allocated to the parent device before the child devices are
probed. As such, there is no need for felix to initialize new regmaps in
these configurations, they can simply be requested from the parent device.

Add support for MFD configurations by performing this request from the
parent device.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: dsa: felix: add configurable device quirks
Colin Foster [Fri, 27 Jan 2023 19:35:52 +0000 (11:35 -0800)]
net: dsa: felix: add configurable device quirks

The define FELIX_MAC_QUIRKS was used directly in the felix.c shared driver.
Other devices (VSC7512 for example) don't require the same quirks, so they
need to be configured on a per-device basis.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: mscc: ocelot: expose vsc7514_regmap definition
Colin Foster [Fri, 27 Jan 2023 19:35:51 +0000 (11:35 -0800)]
net: mscc: ocelot: expose vsc7514_regmap definition

The VSC7514 target regmap is identical for ones shared with similar
hardware, specifically the VSC7512. Share this resource, and change the
name to match the pattern of other exported resources.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: mscc: ocelot: expose ocelot_reset routine
Colin Foster [Fri, 27 Jan 2023 19:35:50 +0000 (11:35 -0800)]
net: mscc: ocelot: expose ocelot_reset routine

Resetting the switch core is the same whether it is done internally or
externally. Move this routine to the ocelot library so it can be used by
other drivers.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: mscc: ocelot: expose vcap_props structure
Colin Foster [Fri, 27 Jan 2023 19:35:49 +0000 (11:35 -0800)]
net: mscc: ocelot: expose vcap_props structure

The vcap_props structure is common to other devices, specifically the
VSC7512 chip that can only be controlled externally. Export this structure
so it doesn't need to be recreated.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: mscc: ocelot: expose regfield definition to be used by other drivers
Colin Foster [Fri, 27 Jan 2023 19:35:48 +0000 (11:35 -0800)]
net: mscc: ocelot: expose regfield definition to be used by other drivers

The ocelot_regfields struct is common between several different chips, some
of which can only be controlled externally. Export this structure so it
doesn't have to be duplicated in these other drivers.

Rename the structure as well, to follow the conventions of other shared
resources.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: mscc: ocelot: expose ocelot wm functions
Colin Foster [Fri, 27 Jan 2023 19:35:47 +0000 (11:35 -0800)]
net: mscc: ocelot: expose ocelot wm functions

Expose ocelot_wm functions so they can be shared with other drivers.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # regression
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: phy: motorcomm: change the phy id of yt8521 and yt8531s to lowercase
Frank Sae [Sat, 28 Jan 2023 06:35:58 +0000 (14:35 +0800)]
net: phy: motorcomm: change the phy id of yt8521 and yt8531s to lowercase

The phy id is usually defined in lower case.

Signed-off-by: Frank Sae <Frank.Sae@motor-comm.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/20230128063558.5850-2-Frank.Sae@motor-comm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: phy: fix the spelling problem of Sentinel
Frank Sae [Sat, 28 Jan 2023 06:35:57 +0000 (14:35 +0800)]
net: phy: fix the spelling problem of Sentinel

CHECK: 'sentinal' may be misspelled - perhaps 'sentinel'?

Signed-off-by: Frank Sae <Frank.Sae@motor-comm.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/20230128063558.5850-1-Frank.Sae@motor-comm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agosh: checksum: add missing linux/uaccess.h include
Jakub Kicinski [Sat, 28 Jan 2023 07:31:08 +0000 (23:31 -0800)]
sh: checksum: add missing linux/uaccess.h include

SuperH does not include uaccess.h, even tho it calls access_ok().

Fixes: 68f4eae781dd ("net: checksum: drop the linux/uaccess.h include")
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Tested-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230128073108.1603095-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet: b44: Remove the unused function __b44_cam_read()
Jiapeng Chong [Sat, 28 Jan 2023 09:04:13 +0000 (17:04 +0800)]
net: b44: Remove the unused function __b44_cam_read()

The function __b44_cam_read() is defined in the b44.c file, but not called
elsewhere, so remove this unused function.

drivers/net/ethernet/broadcom/b44.c:199:20: warning: unused function '__b44_cam_read'.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3858
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230128090413.79824-1-jiapeng.chong@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
17 months agonet/mlx5e: kTLS, Improve connection rate by using fast update encryption key
Jianbo Liu [Mon, 28 Nov 2022 00:55:16 +0000 (00:55 +0000)]
net/mlx5e: kTLS, Improve connection rate by using fast update encryption key

As the fast DEK update is fully implemented, use it for kTLS to get
better performance.
TIS pool was already supported to recycle the TISes. With this series
and TIS pool, TLS CPS is improved by 9x higher, from 11k/s to 101k/s.

Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>