tcx: Fix splat during dev unregister

During unregister_netdevice_many_notify(), the ordering of our concerned
function calls is like this:

  unregister_netdevice_many_notify
    dev_shutdown
qdisc_put
            clsact_destroy
    tcx_uninstall

The syzbot reproducer triggered a case that the qdisc refcnt is not
zero during dev_shutdown().

tcx_uninstall() will then WARN_ON_ONCE(tcx_entry(entry)->miniq_active)
because the miniq is still active and the entry should not be freed.
The latter assumed that qdisc destruction happens before tcx teardown.

This fix is to avoid tcx_uninstall() doing tcx_entry_free() when the
miniq is still alive and let the clsact_destroy() do the free later, so
that we do not assume any specific ordering for either of them.

If still active, tcx_uninstall() does clear the entry when flushing out
the prog/link. clsact_destroy() will then notice the "!tcx_entry_is_active()"
and then does the tcx_entry_free() eventually.

Fixes: e420bed02507 ("bpf: Add fd-based tcx multi-prog infra with link support")
Reported-by: syzbot+376a289e86a0fd02b9ba@syzkaller.appspotmail.com
Reported-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: syzbot+376a289e86a0fd02b9ba@syzkaller.appspotmail.com
Tested-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/222255fe07cb58f15ee662e7ee78328af5b438e4.1690549248.git.daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: bcmgenet: Remove TX ring full logging

There is no need to spam the kernel log with such an indication, remove
this message.

Signed-off-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230728183945.760531-1-florian.fainelli@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

vsock: Remove unused function declarations

These are never implemented since introduction in
commit d021c344051a ("VSOCK: Introduce VM Sockets")

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://lore.kernel.org/r/20230729122036.32988-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/smc: Remove unused function declarations

commit f9aab6f2ce57 ("net/smc: immediate freeing in smc_lgr_cleanup_early()")
left behind smc_lgr_schedule_free_work_fast() declaration.
And since commit 349d43127dac ("net/smc: fix kernel panic caused by race of smc_sock")
smc_ib_modify_qp_reset() is not used anymore.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wenjia Zhang <wenjia@linux.ibm.com>
Link: https://lore.kernel.org/r/20230729121929.17180-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'connector-proc_filter-test-fixes'

Shuah Khan says:

====================
Connector/proc_filter test fixes

The first patch fixes the LKFT reported compile error, second
one adds .gitignore.
====================

Applying the first 2 patches, third one resent separately.

Link: https://lore.kernel.org/r/cover.1690564372.git.skhan@linuxfoundation.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: connector: Add .gitignore and poupulate it with test

Add gitignore and poupulate it with test name - proc_filter

Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Link: https://lore.kernel.org/r/e3d04cc34e9af07909dc882b50fb1b6f1ce7705b.1690564372.git.skhan@linuxfoundation.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: connector: Fix Makefile to include KHDR_INCLUDES

The test compile fails with following errors. Fix the Makefile
CFLAGS to include KHDR_INCLUDES to pull in uapi defines.

gcc -Wall     proc_filter.c  -o ../tools/testing/selftests/connector/proc_filter
proc_filter.c: In function ‘send_message’:
proc_filter.c:22:33: error: invalid application of ‘sizeof’ to incomplete type ‘struct proc_input’
   22 |                          sizeof(struct proc_input))
      |                                 ^~~~~~
proc_filter.c:42:19: note: in expansion of macro ‘NL_MESSAGE_SIZE’
   42 |         char buff[NL_MESSAGE_SIZE];
      |                   ^~~~~~~~~~~~~~~
proc_filter.c:22:33: error: invalid application of ‘sizeof’ to incomplete type ‘struct proc_input’
   22 |                          sizeof(struct proc_input))
      |                                 ^~~~~~
proc_filter.c:48:34: note: in expansion of macro ‘NL_MESSAGE_SIZE’
   48 |                 hdr->nlmsg_len = NL_MESSAGE_SIZE;
      |                                  ^~~~~~~~~~~~~~~
`

Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Link: https://lore.kernel.org/all/CA+G9fYt=6ysz636XcQ=-KJp7vJcMZ=NjbQBrn77v7vnTcfP2cA@mail.gmail.com/
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Link: https://lore.kernel.org/r/d0055c8cdf18516db8ba9edec99cfc5c08f32a7c.1690564372.git.skhan@linuxfoundation.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

i40e: remove i40e_status

Replace uses of i40e_status to as equivalent as possible error codes.
Remove enum i40e_status as it is no longer needed

Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230728171336.2446156-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: Remove unused function declarations

commit 8a59f9d1e3d4 ("sock: Introduce sk->sk_prot->psock_update_sk_prot()")
left behind tcp_bpf_get_proto() declaration. And tcp_v4_tw_remember_stamp()
function is remove in ccb7c410ddc0 ("timewait_sock: Create and use getpeer op.").
Since commit 686989700cab ("tcp: simplify tcp_mark_skb_lost")
tcp_skb_mark_lost_uncond_verify() declaration is not used anymore.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230729122644.10648-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: Remove unused extern declaration devlink_port_region_destroy()

devlink_port_region_destroy() is never implemented since
commit 544e7c33ec2f ("net: devlink: Add support for port regions").

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230728132113.32888-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: Use sockaddr_storage for getsockopt(SO_PEERNAME).

Commit df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3") started
applying strict rules to standard string functions.

It does not work well with conventional socket code around each protocol-
specific sockaddr_XXX struct, which is cast from sockaddr_storage and has
a bigger size than fortified functions expect.  See these commits:

 commit 06d4c8a80836 ("af_unix: Fix fortify_panic() in unix_bind_bsd().")
 commit ecb4534b6a1c ("af_unix: Terminate sun_path when bind()ing pathname socket.")
 commit a0ade8404c3b ("af_packet: Fix warning of fortified memcpy() in packet_getname().")

We must cast the protocol-specific address back to sockaddr_storage
to call such functions.

However, in the case of getsockaddr(SO_PEERNAME), the rationale is a bit
unclear as the buffer is defined by char[128] which is the same size as
sockaddr_storage.

Let's use sockaddr_storage explicitly.

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: flow_dissector: Use 64bits for used_keys

As 32bits of dissector->used_keys are exhausted,
increase the size to 64bits.

This is base change for ESP/AH flow dissector patch.
Please find patch and discussions at
https://lore.kernel.org/netdev/ZMDNjD46BvZ5zp5I@corigine.com/T/#t

Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
Reviewed-by: Petr Machata <petrm@nvidia.com> # for mlxsw
Tested-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Martin Habets <habetsm.xilinx@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

team: Remove NULL check before dev_{put, hold}

The call netdev_{put, hold} of dev_{put, hold} will check NULL,
so there is no need to check before using dev_{put, hold},
remove it to silence the warning:

./drivers/net/team/team.c:2325:3-10: WARNING: NULL check before dev_{put, hold} functions is not needed.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Closes: https://bugzilla.openanolis.cn/show_bug.cgi?id=5991
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ethernet: mtk_eth_soc: enable nft hw flowtable_offload for MT7988 SoC

Enable hw Packet Process Engine (PPE) for MT7988 SoC.

Tested-by: Daniel Golle <daniel@makrotopia.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://lore.kernel.org/r/5e86341b0220a49620dadc02d77970de5ded9efc.1690441576.git.lorenzo@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ethernet: mtk_eth_soc: enable page_pool support for MT7988 SoC

In order to recycle pages, enable page_pool allocator for MT7988 SoC.

Tested-by: Daniel Golle <daniel@makrotopia.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://lore.kernel.org/r/fd4e8693980e47385a543e7b002eec0b88bd09df.1690440675.git.lorenzo@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: bcmasp: Clean up redundant dev_err_probe()

Referring to platform_get_irq()'s definition, the return value has
already been checked, error message also been printed via
dev_err_probe() if ret < 0. Calling dev_err_probe() one more time
outside platform_get_irq() is obviously redundant.

Removing dev_err_probe() outside platform_get_irq() to clean up
above problem.

Signed-off-by: Chen Jiahao <chenjiahao16@huawei.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Acked-by: Justin Chen <justin.chen@broadcom.com>
Link: https://lore.kernel.org/r/20230727115551.2655840-1-chenjiahao16@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

bonding: 3ad: Remove unused declaration bond_3ad_update_lacp_active()

This is not used since commit 3a755cd8b7c6 ("bonding: add new option lacp_active")

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://lore.kernel.org/r/20230726143816.15280-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'r8152-reduce-control-transfer'

Hayes Wang says:

====================
r8152: reduce control transfer

The two patches are used to reduce the number of control transfer when
access the registers in bulk.
====================

Link: https://lore.kernel.org/r/20230726030808.9093-417-nic_swsd@realtek.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

r8152: set bp in bulk

PLA_BP_0 ~ PLA_BP_15 (0xfc28 ~ 0xfc46) are continuous registers, so we
could combine the control transfers into one control transfer.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Link: https://lore.kernel.org/r/20230726030808.9093-419-nic_swsd@realtek.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

r8152: adjust generic_ocp_write function

Reduce the control transfer if all bytes of first or the last DWORD are
written.

The original method is to split the control transfer into three parts
(the first DWORD, middle continuous data, and the last DWORD). However,
they could be combined if whole bytes of the first DWORD or last DWORD
are written. That is, the first DWORD or the last DWORD could be combined
with the middle continuous data, if the byte_en is 0xff.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Link: https://lore.kernel.org/r/20230726030808.9093-418-nic_swsd@realtek.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ethernet: slicoss: remove redundant increment of pointer data

The pointer data is being incremented but this change to the pointer
is not used afterwards. The increment is redundant and can be removed.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Acked-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Link: https://lore.kernel.org/r/20230726164522.369206-1-colin.i.king@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'in-kernel-support-for-the-tls-alert-protocol'

Chuck Lever says:

====================
In-kernel support for the TLS Alert protocol

IMO the kernel doesn't need user space (ie, tlshd) to handle the TLS
Alert protocol. Instead, a set of small helper functions can be used
to handle sending and receiving TLS Alerts for in-kernel TLS
consumers.
====================

Merged on top of a tag in case it's needed in the NFS tree.

Link: https://lore.kernel.org/r/169047923706.5241.1181144206068116926.stgit@oracle-102.nfsv4bat.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/handshake: Trace events for TLS Alert helpers

Add observability for the new TLS Alert infrastructure.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://lore.kernel.org/r/169047947409.5241.14548832149596892717.stgit@oracle-102.nfsv4bat.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

SUNRPC: Use new helpers to handle TLS Alerts

Use the helpers to parse the level and description fields in
incoming alerts. "Warning" alerts are discarded, and "fatal"
alerts mean the session is no longer valid.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://lore.kernel.org/r/169047944747.5241.1974889594004407123.stgit@oracle-102.nfsv4bat.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/handshake: Add helpers for parsing incoming TLS Alerts

Kernel TLS consumers can replace common TLS Alert parsing code with
these helpers.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://lore.kernel.org/r/169047942074.5241.13791647439480672048.stgit@oracle-102.nfsv4bat.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

SUNRPC: Send TLS Closure alerts before closing a TCP socket

Before closing a TCP connection, the TLS protocol wants peers to
send session close Alert notifications. Add those in both the RPC
client and server.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://lore.kernel.org/r/169047939404.5241.14392506226409865832.stgit@oracle-102.nfsv4bat.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/handshake: Add API for sending TLS Closure alerts

This helper sends an alert only if a TLS session was established.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://lore.kernel.org/r/169047936730.5241.618595693821012638.stgit@oracle-102.nfsv4bat.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/tls: Add TLS Alert definitions

I'm about to add support for kernel handshake API consumers to send
TLS Alerts, so introduce the needed protocol definitions in the new
header tls_prot.h.

This presages support for Closure alerts. Also, support for alerts
is a pre-requite for handling session re-keying, where one peer will
signal the need for a re-key by sending a TLS Alert.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://lore.kernel.org/r/169047934064.5241.8377890858495063518.stgit@oracle-102.nfsv4bat.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/tls: Move TLS protocol elements to a separate header

Kernel TLS consumers will need definitions of various parts of the
TLS protocol, but often do not need the function declarations and
other infrastructure provided in <net/tls.h>.

Break out existing standardized protocol elements into a separate
header, and make room for a few more elements in subsequent patches.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Link: https://lore.kernel.org/r/169047931374.5241.7713175865185969309.stgit@oracle-102.nfsv4bat.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

octeontx2-af: Initialize 'cntr_val' to fix uninitialized symbol error

drivers/net/ethernet/marvell/octeontx2/nic/otx2_tc.c:860
otx2_tc_update_mcam_table_del_req()
error: uninitialized symbol 'cntr_val'.

Fixes: ec87f05402f5 ("octeontx2-af: Install TC filter rules in hardware based on priority")
Signed-off-by: Suman Ghosh <sumang@marvell.com>
Link: https://lore.kernel.org/r/20230727163101.2793453-1-sumang@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'eth-bnxt-fix-a-couple-of-w-1-c-1-warnings'

Jakub Kicinski says:

====================
eth: bnxt: fix a couple of W=1 C=1 warnings

Fix a couple of build warnings.
====================

Link: https://lore.kernel.org/r/20230727190726.1859515-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

eth: bnxt: fix warning for define in struct_group

Fix C=1 warning with sparse 0.6.4:

drivers/net/ethernet/broadcom/bnxt/bnxt.c: note: in included file:
drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.h:30:1: warning: directive in macro's argument list

Don't put defines in a struct_group().

Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://lore.kernel.org/r/20230727190726.1859515-3-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

eth: bnxt: fix one of the W=1 warnings about fortified memcpy()

Fix a W=1 warning with gcc 13.1:

In function ‘fortify_memcpy_chk’,
    inlined from ‘bnxt_hwrm_queue_cos2bw_cfg’ at drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c:133:3:
include/linux/fortify-string.h:592:25: warning: call to ‘__read_overflow2_field’ declared with attribute warning: detected read beyond size of field (2nd parameter); maybe use struct_group()? [-Wattribute-warning]
  592 |                         __read_overflow2_field(q_size_field, size);
      |                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

The field group is already defined and starts at queue_id:

struct bnxt_cos2bw_cfg {
u8 pad[3];
struct_group_attr(cfg, __packed,
u8 queue_id;
__le32 min_bw;

Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://lore.kernel.org/r/20230727190726.1859515-2-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'mlx5-updates-2023-07-24' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2023-07-24

1) Generalize devcom implementation to be independent of number of ports
   or device's GUID.

2) Save memory on command interface statistics.

3) General code cleanups

* tag 'mlx5-updates-2023-07-24' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
  net/mlx5: Give esw_offloads_load/unload_rep() "mlx5_" prefix
  net/mlx5: Make mlx5_eswitch_load/unload_vport() static
  net/mlx5: Make mlx5_esw_offloads_rep_load/unload() static
  net/mlx5: Remove pointless devlink_rate checks
  net/mlx5: Don't check vport->enabled in port ops
  net/mlx5e: Make flow classification filters static
  net/mlx5e: Remove duplicate code for user flow
  net/mlx5: Allocate command stats with xarray
  net/mlx5: split mlx5_cmd_init() to probe and reload routines
  net/mlx5: Remove redundant cmdif revision check
  net/mlx5: Re-organize mlx5_cmd struct
  net/mlx5e: E-Switch, Allow devcom initialization on more vports
  net/mlx5e: E-Switch, Register devcom device with switch id key
  net/mlx5: Devcom, Infrastructure changes
  net/mlx5: Use shared code for checking lag is supported
====================

Link: https://lore.kernel.org/r/20230727183914.69229-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'mlxsw-avoid-non-tracker-helpers-when-holding-and-putting-netdevices'

Petr Machata says:

====================
mlxsw: Avoid non-tracker helpers when holding and putting netdevices

Using the tracking helpers, netdev_hold() and netdev_put(), makes it easier
to debug netdevice refcount imbalances when CONFIG_NET_DEV_REFCNT_TRACKER
is enabled. For example, the following traceback shows the callpath to the
point of an outstanding hold that was never put:

    unregister_netdevice: waiting for swp3 to become free. Usage count = 6
    ref_tracker: eth%d@ffff888123c9a580 has 1/5 users at
mlxsw_sp_switchdev_event+0x6bd/0xcc0 [mlxsw_spectrum]
notifier_call_chain+0xbf/0x3b0
atomic_notifier_call_chain+0x78/0x200
br_switchdev_fdb_notify+0x25f/0x2c0 [bridge]
fdb_notify+0x16a/0x1a0 [bridge]
[...]

In this patchset, get rid of all non-ref-tracking helpers in mlxsw.

- Patch #1 drops two functions that are not used anymore, but contain
  dev_hold() / dev_put() calls.

- Patch #2 avoids taking a reference in one function which is called
  under RTNL.

- The remaining patches convert individual hold/put sites one by one
  from trackerless to tracker-enabled.

Suggested-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/netdev/4c056da27c19d95ffeaba5acf1427ecadfc3f94c.camel@redhat.com/
====================

Link: https://lore.kernel.org/r/cover.1690471774.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: spectrum_router: IPv6 events: Use tracker helpers to hold & put netdevices

Using the tracking helpers makes it easier to debug netdevice refcount
imbalances when CONFIG_NET_DEV_REFCNT_TRACKER is enabled.

Convert dev_hold() / dev_put() to netdev_hold() / netdev_put() in the
router code that deals with IPv6 address events.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/f0af6ad4722b4ca6e598fd4fda8311a3041651ec.1690471775.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: spectrum_router: RIF: Use tracker helpers to hold & put netdevices

Using the tracking helpers makes it easier to debug netdevice refcount
imbalances when CONFIG_NET_DEV_REFCNT_TRACKER is enabled.

Convert dev_hold() / dev_put() to netdev_hold() / netdev_put() in the
router code that deals with RIF allocation.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/8b7701a7b439ac268e4be4040eff99d01e27ae47.1690471775.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: spectrum_router: hw_stats: Use tracker helpers to hold & put netdevices

Using the tracking helpers makes it easier to debug netdevice refcount
imbalances when CONFIG_NET_DEV_REFCNT_TRACKER is enabled.

Convert dev_hold() / dev_put() to netdev_hold() / netdev_put() in the
router code that deals with hw_stats events.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/b972314cfef4f4c24e66e60d13cffa5d606d1bf3.1690471774.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: spectrum_router: FIB: Use tracker helpers to hold & put netdevices

Using the tracking helpers makes it easier to debug netdevice refcount
imbalances when CONFIG_NET_DEV_REFCNT_TRACKER is enabled.

Convert dev_hold() / dev_put() to netdev_hold() / netdev_put() in the
router code that deals with FIB events.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/5221a92e751c40447c55959f622267ccc999ed04.1690471774.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: spectrum_switchdev: Use tracker helpers to hold & put netdevices

Using the tracking helpers makes it easier to debug netdevice refcount
imbalances when CONFIG_NET_DEV_REFCNT_TRACKER is enabled.

Convert dev_hold() / dev_put() to netdev_hold() / netdev_put() in the
switchdev module.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/774c3d7b5b0231f1435df2ec9dd660192e382756.1690471774.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: spectrum_nve: Do not take reference when looking up netdevice

mlxsw_sp_nve_fid_disable() is always called under RTNL. It is therefore
safe to call __dev_get_by_index() to get the netdevice pointer without
bumping the reference count, because we can be sure the netdevice is not
going away. That then obviates the need to put the netdevice later in the
function.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/341d1046f89d8d839d9d00e4a3d58cdc351e9397.1690471774.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: spectrum: Drop unused functions mlxsw_sp_port_lower_dev_hold/_put()

As of commit 151b89f6025a ("mlxsw: spectrum_router: Reuse work neighbor
initialization in work scheduler"), the functions
mlxsw_sp_port_lower_dev_hold() and mlxsw_sp_port_dev_put() have no users.
Drop them.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/d0adcd7cb4ea19416294a0f861100edba84c9f36.1690471774.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: change accept_ra_min_rtr_lft to affect all RA lifetimes

accept_ra_min_rtr_lft only considered the lifetime of the default route
and discarded entire RAs accordingly.

This change renames accept_ra_min_rtr_lft to accept_ra_min_lft, and
applies the value to individual RA sections; in particular, router
lifetime, PIO preferred lifetime, and RIO lifetime. If any of those
lifetimes are lower than the configured value, the specific RA section
is ignored.

In order for the sysctl to be useful to Android, it should really apply
to all lifetimes in the RA, since that is what determines the minimum
frequency at which RAs must be processed by the kernel. Android uses
hardware offloads to drop RAs for a fraction of the minimum of all
lifetimes present in the RA (some networks have very frequent RAs (5s)
with high lifetimes (2h)). Despite this, we have encountered networks
that set the router lifetime to 30s which results in very frequent CPU
wakeups. Instead of disabling IPv6 (and dropping IPv6 ethertype in the
WiFi firmware) entirely on such networks, it seems better to ignore the
misconfigured routers while still processing RAs from other IPv6 routers
on the same network (i.e. to support IoT applications).

The previous implementation dropped the entire RA based on router
lifetime. This turned out to be hard to expand to the other lifetimes
present in the RA in a consistent manner; dropping the entire RA based
on RIO/PIO lifetimes would essentially require parsing the whole thing
twice.

Fixes: 1671bcfd76fd ("net: add sysctl accept_ra_min_rtr_lft")
Cc: Lorenzo Colitti <lorenzo@google.com>
Signed-off-by: Patrick Rohr <prohr@google.com>
Reviewed-by: Maciej Żenczykowski <maze@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20230726230701.919212-1-prohr@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-store-netdevs-in-an-xarray'

Jakub Kicinski says:

====================
net: store netdevs in an xarray

One of more annoying developer experience gaps we have in netlink
is iterating over netdevs. It's painful. Add an xarray to make
it trivial.

v1: https://lore.kernel.org/all/20230722014237.4078962-1-kuba@kernel.org/
====================

Link: https://lore.kernel.org/r/20230726185530.2247698-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: convert some netlink netdev iterators to depend on the xarray

Reap the benefits of easier iteration thanks to the xarray.
Convert just the genetlink ones, those are easier to test.

Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230726185530.2247698-3-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: store netdevs in an xarray

Iterating over the netdev hash table for netlink dumps is hard.
Dumps are done in "chunks" so we need to save the position
after each chunk, so we know where to restart from. Because
netdevs are stored in a hash table we remember which bucket
we were in and how many devices we dumped.

Since we don't hold any locks across the "chunks" - devices may
come and go while we're dumping. If that happens we may miss
a device (if device is deleted from the bucket we were in).
We indicate to user space that this may have happened by setting
NLM_F_DUMP_INTR. User space is supposed to dump again (I think)
if it sees that. Somehow I doubt most user space gets this right..

To illustrate let's look at an example:

               System state:
  start:       # [A, B, C]
  del:  B      # [A, C]

with the hash table we may dump [A, B], missing C completely even
tho it existed both before and after the "del B".

Add an xarray and use it to allocate ifindexes. This way we
can iterate ifindexes in order, without the worry that we'll
skip one. We may still generate a dump of a state which "never
existed", for example for a set of values and sequence of ops:

               System state:
  start:       # [A, B]
  add:  C      # [A, C, B]
  del:  B      # [A, C]

we may generate a dump of [A], if C got an index between A and B.
System has never been in such state. But I'm 90% sure that's perfectly
fine, important part is that we can't _miss_ devices which exist before
and after. User space which wants to mirror kernel's state subscribes
to notifications and does periodic dumps so it will know that C exists
from the notification about its creation or from the next dump
(next dump is _guaranteed_ to include C, if it doesn't get removed).

To avoid any perf regressions keep the hash table for now. Most
net namespaces have very few devices and microbenchmarking 1M lookups
on Skylake I get the following results (not counting loopback
to number of devs):

 #devs | hash |  xa  | delta
    2  | 18.3 | 20.1 | + 9.8%
   16  | 18.3 | 20.1 | + 9.5%
   64  | 18.3 | 26.3 | +43.8%
  128  | 20.4 | 26.3 | +28.6%
  256  | 20.0 | 26.4 | +32.1%
 1024  | 26.6 | 26.7 | + 0.2%
 8192  |541.3 | 33.5 | -93.8%

No surprises since the hash table has 256 entries.
The microbenchmark scans indexes in order, if the pattern is more
random xa starts to win at 512 devices already. But that's a lot
of devices, in practice.

Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230726185530.2247698-2-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'ynl-couple-of-unrelated-fixes'

Stanislav Fomichev says:

====================
ynl: couple of unrelated fixes

- spelling of xdp-features
- s/xdp_zc_max_segs/xdp-zc-max-segs/
- expose xdp-zc-max-segs
- add /* private: */
- regenerate headers
- print xdp_zc_max_segs from sample
====================

Link: https://lore.kernel.org/r/20230727163001.3952878-1-sdf@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ynl: print xdp-zc-max-segs in the sample

Technically we don't have to keep extending the sample, but it
feels useful to run these tools locally to confirm everything
is working.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20230727163001.3952878-5-sdf@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ynl: regenerate all headers

Also add support to pass topdir to ynl-regen.sh (Jakub) and call
it from the makefile to update the UAPI headers.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Co-developed-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20230727163001.3952878-4-sdf@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ynl: mark max/mask as private for kdoc

Simon mentioned in another thread that it makes kdoc happy
and Jakub confirms that commit e27cb89a22ad ("scripts: kernel-doc: support
private / public marking for enums") actually added the needed
support.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20230727163001.3952878-3-sdf@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ynl: expose xdp-zc-max-segs

Also rename it to dashes, to match the rest. And fix unrelated
spelling error while we're at it.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20230727163001.3952878-2-sdf@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch '100GbE' of git://git./linux/kernel/git/tnguy/nex
t-queue

Tony Nguyen says:

====================
ice: Implement support for SRIOV + LAG

Dave Ertman says:

Implement support for SRIOV VF's on interfaces that are in an
aggregate interface.

The first interface added into the aggregate will be flagged as
the primary interface, and this primary interface will be
responsible for managing the VF's resources.  VF's created on the
primary are the only VFs that will be supported on the aggregate.
Only Active-Backup mode will be supported and only aggregates whose
primary interface is in switchdev mode will be supported.

The ice-lag DDP must be loaded to support this feature.

Additional restrictions on what interfaces can be added to the aggregate
and still support SRIOV VFs are:
- interfaces have to all be on the same physical NIC
- all interfaces have to have the same QoS settings
- interfaces have to have the FW LLDP agent disabled
- only the primary interface is to be put into switchdev mode
- no more than two interfaces in the aggregate
---
v2:
- Move NULL check for q_ctx in ice_lag_qbuf_recfg() earlier (patch 6)

v1: https://lore.kernel.org/netdev/20230726182141.3797928-1-anthony.l.nguyen@intel.com/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

IPv6: add extack info for IPv6 address add/delete

Add extack info for IPv6 address add/delete, which would be useful for
users to understand the problem without having to read kernel code.

Suggested-by: Beniamino Galvani <bgalvani@redhat.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'selftest-ptp'

Alex Maftei says:

====================
selftests/ptp: Add support for new timestamp IOCTLs

PTP_SYS_OFFSET_EXTENDED was added in November 2018 in
361800876f80 (" ptp: add PTP_SYS_OFFSET_EXTENDED ioctl")
and PTP_SYS_OFFSET_PRECISE was added in February 2016 in
719f1aa4a671 ("ptp: Add PTP_SYS_OFFSET_PRECISE for driver crosstimestamping")

The PTP selftest code is lacking support for these two IOCTLS.
This short series of patches adds support for them.

Changes in v2:
- Fixed rebase issues (v1 somehow ended up with patch 1 being from the
  first manual split of my changes and patch 2 being from rebase 2 out
  of 3)
- Rebased on top of net-next
====================

Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

selftests/ptp: Add -X option for testing PTP_SYS_OFFSET_PRECISE

The -X option was chosen because X looks like a cross, and the underlying
callback is 'get cross timestamp'.

Signed-off-by: Alex Maftei <alex.maftei@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

selftests/ptp: Add -x option for testing PTP_SYS_OFFSET_EXTENDED

The -x option (where 'x' stands for eXtended) takes an argument which
represents the number of samples to request from the PTP device.
The help message will display the maximum number of samples allowed.
Providing an invalid argument will also display the maximum number of
samples allowed.

Signed-off-by: Alex Maftei <alex.maftei@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'linux-can-next-for-6.6-20230728' of git://git./linux/kernel/git/mkl/linux-can-next

linux-can-next-for-6.6-20230728

Marc Kleine-Budde says:

====================
Hello netdev-team,

this is a pull request of 21 patches for net-next/master.

The 1st patch is by Gerhard Uttenthaler, which adds Gerhard as the
maintainer ems_pci driver.

Peter Seiderer's patch removes a unused function from the peak_usb
driver.

The next 4 patches are by John Watts and add support for the sun4i_can
driver on the Allwinner D1.

Rob Herring's patch corrects the DT includes in various CAN drivers.

Followed by 14 patches from me concerning the gs_usb driver. The first
11 are various cleanups consisting of coding style improvements, error
path printout cleanups, and removal of unneeded
usb_kill_anchored_urbs(). The last 3 convert the driver to use NAPI to
avoid out-of-order reception of CAN frames.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'sfc-siena-next'

Martin Habets says:

====================
sfc: Remove Siena bits from sfc.ko

Last year we split off Siena into it's own driver under directory siena.
This patch series removes the now unused Falcon and Siena code from sfc.ko.
No functional changes are intended.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Remove vfdi.h

It was only used for Siena SRIOV, so nothing includes it any more
in this directory.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Cleanups in io.h

Most of the Falcon locking description does not apply to EF10.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Miscellaneous comment removals

Remove comments that only apply to Falcon and Siena.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Remove struct efx_special_buffer

The attributes index and entries are no longer needed, so use
struct efx_buffer instead.
next_buffer_table was also Siena specific.
Removed some checkpatch warnings.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Filter cleanups for Falcon and Siena

unicast_filter and multicast_hash are no longer needed.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Remove some NIC type indirections that are no longer needed

The special handling for SRIOV reset and FLR is not needed on EF10.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Remove PTP code for Siena

rx_tx_inline is now always true.
The special event list is no longer needed, event handling is always
inline.
Event MCDI_EVENT_CODE_PTP_RX is no longer needed.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Remove EFX_REV_SIENA_A0

The workarounds for bug 8568 and 17213 are no longer needed.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Remove support for siena high priority queue

This also removes TC support code, since that was never supported for EF10.
TC support for EF100 is not handled from efx.c.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Remove siena_nic_data and stats

These are no longer used, and the two  Siena specific functions are
no longer present in sfc.ko.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Remove falcon references

Siena was using functions from the Falcon architecture.
These start with the efx_farch prefix.
Now that both of these are in separate modules the references
are no longer used in the sfc.ko module.

Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'rxfh-custom-rss'

Joe Damato says:

====================
rxfh with custom RSS fixes

Greetings:

Welcome to v2, now via net-next. No functional changes; only style
changes (see the summary below).

While attempting to get the RX flow hash key for a custom RSS context on
my mlx5 NIC, I got an error:

$ sudo ethtool -u eth1 rx-flow-hash tcp4 context 1
Cannot get RX network flow hashing options: Invalid argument

I dug into this a bit and noticed two things:

1. ETHTOOL_GRXFH supports custom RSS contexts, but ETHTOOL_SRXFH does
not. I moved the copy logic out of ETHTOOL_GRXFH and into a helper so
that both ETHTOOL_{G,S}RXFH now call it, which fixes ETHTOOL_SRXFH. This
is patch 1/2.

2. mlx5 defaulted to RSS context 0 for both ETHTOOL_{G,S}RXFH paths. I
have modified the driver to support custom contexts for both paths. It
is now possible to get and set the flow hash key for custom RSS contexts
with mlx5. This is patch 2/2.

See commit messages for more details.

Thanks.

v2:
- Rebased on net-next
- Adjusted arguments of mlx5e_rx_res_rss_get_hash_fields and
  mlx5e_rx_res_rss_set_hash_fields to move rss_idx next to the rss
  argument
- Changed return value of both mlx5e_rx_res_rss_get_hash_fields and
  mlx5e_rx_res_rss_set_hash_fields to -ENOENT when the rss entry is
  NULL
- Changed order of local variables in mlx5e_get_rss_hash_opt and
  mlx5e_set_rss_hash_opt
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

net/mlx5: Fix flowhash key set/get for custom RSS

mlx5 flow hash field retrieval and set only worked on the default
RSS context as of commit f01cc58c18d6 ("net/mlx5e: Support multiple RSS
contexts"), not custom RSS contexts.

For example, before this patch attempting to retrieve the flow hash fields
for RSS context 1 fails:

$ sudo ethtool -u eth1 rx-flow-hash tcp4 context 1
Cannot get RX network flow hashing options: Invalid argument

This patch fixes getting and setting the flow hash fields for contexts
other than the default context.

Getting the flow hash fields for RSS context 1:

$ sudo ethtool -u eth1 rx-flow-hash tcp4 context 1
For RSS context 1:
TCP over IPV4 flows use these fields for computing Hash flow key:
IP DA

Now, setting the flash hash fields to a custom value:

$ sudo ethtool -U eth1 rx-flow-hash tcp4 sdfn context 1

And retrieving them again:

$ sudo ethtool -u eth1 rx-flow-hash tcp4 context 1
For RSS context 1:
TCP over IPV4 flows use these fields for computing Hash flow key:
IP SA
IP DA
L4 bytes 0 & 1 [TCP/UDP src port]
L4 bytes 2 & 3 [TCP/UDP dst port]

Signed-off-by: Joe Damato <jdamato@fastly.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ethtool: Unify ETHTOOL_{G,S}RXFH rxnfc copy

ETHTOOL_GRXFH correctly copies in the full struct ethtool_rxnfc when
FLOW_RSS is set; ETHTOOL_SRXFH needs a similar code path to handle the
FLOW_RSS case so that ethtool can set the flow hash for custom RSS
contexts (if supported by the driver).

The copy code from ETHTOOL_GRXFH has been pulled out in to a helper so
that it can be called in both ETHTOOL_{G,S}RXFH code paths.

Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Joe Damato <jdamato@fastly.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge patch series "can: gs_usb: convert to NAPI"

Marc Kleine-Budde <mkl@pengutronix.de> says:

Traditionally USB drivers used netif_rx to pass the received CAN
frames/skbs to the network stack. In netif_rx() the skbs are queued to
the local CPU. If IRQs are handled in round robin, CAN frames may be
delivered out-of-order to user space.

To support devices without timestamping the TX path of the rx-offload
helper is cleaned up and extended:
- rename rx_offload_get_echo_skb() ->
  can_rx_offload_get_echo_skb_queue_timestamp()
- add can_rx_offload_get_echo_skb_queue_tail()

The last patch converts the gs_usb driver to NAPI with the rx-offload
helper.

Link: https://lore.kernel.org/all/20230718-gs_usb-rx-offload-v2-0-716e542d14d5@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: convert to NAPI/rx-offload to avoid OoO reception

The driver used to pass received CAN frames/skbs to the network stack
with netif_rx(). In netif_rx() the skbs are queued to the local CPU.
If IRQs are handled in round robin, OoO packets may occur.

To avoid out-of-order reception convert the driver from netif_rx() to
NAPI.

For USB devices with timestamping support use the rx-offload helper
can_rx_offload_queue_timestamp() for the RX, and
can_rx_offload_get_echo_skb_queue_timestamp() for the TX path. Devices
without timestamping support use can_rx_offload_queue_tail() for RX,
and can_rx_offload_get_echo_skb_queue_tail() for the TX path.

Link: https://lore.kernel.org/all/559D628C.5020100@hartkopp.net
Link: https://github.com/candle-usb/candleLight_fw/issues/166
Link: https://lore.kernel.org/all/20230718-gs_usb-rx-offload-v2-3-716e542d14d5@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: rx-offload: add can_rx_offload_get_echo_skb_queue_tail()

Add can_rx_offload_get_echo_skb_queue_tail(). This function addds the
echo skb at the end of rx-offload the queue. This is intended for
devices without timestamp support.

Link: https://lore.kernel.org/all/20230718-gs_usb-rx-offload-v2-2-716e542d14d5@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: rx-offload: rename rx_offload_get_echo_skb() -> can_rx_offload_get_echo_skb_queue_timestamp()

Rename the rx_offload_get_echo_skb() function to
can_rx_offload_get_echo_skb_queue_timestamp(), since it inserts the
echo skb into the rx-offload queue sorted by timestamp.

This is a preparation for adding
can_rx_offload_get_echo_skb_queue_tail(), which adds the echo skb to
the end of the queue. This is intended for devices that do not support
timestamps.

Link: https://lore.kernel.org/all/20230718-gs_usb-rx-offload-v2-1-716e542d14d5@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

Merge patch series "can: gs_usb-cleanups: various clenaups"

Marc Kleine-Budde <mkl@pengutronix.de> says:

This is a cleanup series of the gs_usb driver. Align the driver more
to the kernel coding style, make use of locally defined variables,
clean up printouts in various error paths, remove some not needed
usb_kill_anchored_urbs() from the shut down paths.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-0-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_usb_disconnect(): remove not needed usb_kill_anchored_urbs()

In gs_usb_disconnect(), all channels are destroyed first, then all
anchored RX URBs (parent->rx_submitted) are disposed with
usb_kill_anchored_urbs().

The call to usb_kill_anchored_urbs() is not needed, as
gs_destroy_candev() of the last active channel already disposes the RX
URBS.

Remove not needed call to usb_kill_anchored_urbs() from
gs_usb_disconnect().

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-11-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_destroy_candev(): remove not needed usb_kill_anchored_urbs()

In gs_destroy_candev(), the netdev is unregistered first, then all
anchored TX URBs (dev->tx_submitted) are disposed with
usb_kill_anchored_urbs().

The call to usb_kill_anchored_urbs() is not needed, as
unregister_candev() calls gs_can_close(), which already disposes the
TX URBS.

Remove not needed call to usb_kill_anchored_urbs() from
gs_destroy_candev().

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-10-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_can_close(): don't complain about failed device reset during ndo_stop

When the USB device is unplugged, gs_can_close() (which implements the
struct net_device_ops::ndo_stop callback) is called. In this function
an attempt is made to shut down the USB device with a USB control
message. For disconnected devices this will fail and a warning message
is printed.

Silence the driver by removing the printout of the error message if
the reset command fails.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-9-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_can_start_xmit(), gs_can_open(): clean up printouts in error path

Remove unnecessary "out of memory" message from the error path of
gs_can_start_xmit() and gs_can_open().

Convert the printout in case of a failing usb_submit_urb() in
gs_can_open() from numbers to human readable error codes.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-8-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_usb_receive_bulk_callback(): count RX overflow errors also in case of OOM

In case of an RX overflow error from the CAN controller and an OOM
where no skb can be allocated, the error counters are not incremented.

Fix this by first incrementing the error counters and then allocate
the skb.

Fixes: d08e973a77d1 ("can: gs_usb: Added support for the GS_USB CAN devices")
Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-7-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_usb_receive_bulk_callback(): make use of stats

Make use the previously assigned variable stats instead of using
netdev->stats.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-6-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_usb_receive_bulk_callback(): make use of netdev

Make use the previously assigned variable netdev instead of using
dev->netdev.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-5-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: uniformly use "parent" as variable name for struct gs_usb

To ease readability and maintainability uniformly use the variable
name "parent" for the struct gs_usb in the gs_usb driver.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-4-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_usb_set_timestamp(): remove return statements form void function

Remove the return statements from void gs_usb_set_timestamp()
function, as it's not generally useful.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-3-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: gs_usb_probe(): align block comment

Indent block comment so that it aligns the * on each line.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-2-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: gs_usb: remove leading space from goto labels

Remove leading spaces from goto labels in accordance with the kernel
encoding style.

Link: https://lore.kernel.org/all/20230718-gs_usb-cleanups-v1-1-c3b9154ec605@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: Explicitly include correct DT includes, part 2

The DT of_device.h and of_platform.h date back to the separate
of_platform_bus_type before it as merged into the regular platform bus.
As part of that merge prepping Arm DT support 13 years ago, they
"temporarily" include each other. They also include platform_device.h
and of.h. As a result, there's a pretty much random mix of those include
files used throughout the tree. In order to detangle these headers and
replace the implicit includes with struct declarations, users need to
explicitly include the correct includes.

Signed-off-by: Rob Herring <robh@kernel.org>
Link: https://lore.kernel.org/all/20230724211841.805053-1-robh@kernel.org
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

Merge patch series "Add support for Allwinner D1 CAN controllers"

John Watts <contact@jookia.org> says:

This patch series adds support for the Allwinner D1 CAN controllers.
It requires adding a new device tree compatible and driver support to
work around some hardware quirks.

This has been tested on the Mango Pi MQ Dual running a T113 and a
Lichee Panel 86 running a D1.

Changes in v2:
- Re-ordered patches to work with bisecting
- Fixed device tree label underscores
- Fixed email headers
- Link to v1: https://lore.kernel.org/all/20230715112523.2533742-1-contact@jookia.org

Link: https://lore.kernel.org/all/20230721221552.1973203-2-contact@jookia.org
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: peak_usb: remove unused/legacy peak_usb_netif_rx() function

Remove unused/legacy peak_usb_netif_rx() function (not longer used
since commit 28e0a70cede3 ("can: peak_usb: CANFD: store 64-bits hw
timestamps").

Signed-off-by: Peter Seiderer <ps.report@gmx.net>
Link: https://lore.kernel.org/all/20230721180758.26199-1-ps.report@gmx.net
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: sun4i_can: Add support for the Allwinner D1

The controllers present in the D1 are extremely similar to the R40
and require the same reset quirks, but An extra quirk is needed to support
receiving packets.

Signed-off-by: John Watts <contact@jookia.org>
Link: https://lore.kernel.org/all/20230721221552.1973203-6-contact@jookia.org
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

MAINTAINERS: Add myself as maintainer of the ems_pci.c driver

At the suggestion of Marc Kleine-Budde [1], I add myself as maintainer
of the ems_pci.c driver.

[1] https://lore.kernel.org/all/20230720-purplish-quizzical-247024e66671-mkl@pengutronix.de

Signed-off-by: Gerhard Uttenthaler <uttenthaler@ems-wuensche.com>
Link: https://lore.kernel.org/all/20230720144032.28960-1-uttenthaler@ems-wuensche.com
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

can: sun4i_can: Add acceptance register quirk

The Allwinner D1's CAN controllers have the ACPC and ACPM registers
moved down. Compensate for this by adding an offset quirk for the
acceptance registers.

Signed-off-by: John Watts <contact@jookia.org>
Link: https://lore.kernel.org/all/20230721221552.1973203-5-contact@jookia.org
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

riscv: dts: allwinner: d1: Add CAN controller nodes

The Allwinner D1, T113 provide two CAN controllers that are variants
of the R40 controller.

I have tested support for these controllers on two boards:

- A Lichee Panel RV 86 Panel running a D1 chip
- A Mango Pi MQ Dual running a T113-s3 chip

Both of these fully support both CAN controllers.

Signed-off-by: John Watts <contact@jookia.org>
Link: https://lore.kernel.org/all/20230721221552.1973203-4-contact@jookia.org
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

dt-bindings: net: can: Add support for Allwinner D1 CAN controller

The Allwinner D1 has two CAN controllers, both a variant of the R40
controller. Unfortunately the registers for the D1 controllers are
moved around enough to be incompatible and require a new compatible.

Introduce the "allwinner,sun20i-d1-can" compatible to support this.

Signed-off-by: John Watts <contact@jookia.org>
Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Link: https://lore.kernel.org/all/20230721221552.1973203-3-contact@jookia.org
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

net: Explicitly include correct DT includes

The DT of_device.h and of_platform.h date back to the separate
of_platform_bus_type before it as merged into the regular platform bus.
As part of that merge prepping Arm DT support 13 years ago, they
"temporarily" include each other. They also include platform_device.h
and of.h. As a result, there's a pretty much random mix of those include
files used throughout the tree. In order to detangle these headers and
replace the implicit includes with struct declarations, users need to
explicitly include the correct includes.

Acked-by: Alex Elder <elder@linaro.org>
Reviewed-by: Bhupesh Sharma <bhupesh.sharma@linaro.org>
Reviewed-by: Wei Fang <wei.fang@nxp.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230727014944.3972546-1-robh@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Revert "net: stmmac: correct MAC propagation delay"

This reverts commit 20bf98c94146eb6fe62177817cb32f53e72dd2e8.

Richard raised concerns about correctness of the code on previous
generations of the HW.

Fixes: 20bf98c94146 ("net: stmmac: correct MAC propagation delay")
Link: https://lore.kernel.org/all/ZMGIuKVP7BEotbrn@hoboy.vegasvil.org/
Link: https://lore.kernel.org/r/20230726224054.3241127-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-stmmac-increase-clk_ptp_ref-rate'

Andrew Halaney says:

====================
net: stmmac: Increase clk_ptp_ref rate

This series aims to increase the clk_ptp_ref rate to get the best
possible PTP timestamping resolution possible. Some modified disclosure
about my development/testing process from the RFC/RFT v1 follows.

Disclosure: I don't know much about PTP beyond what you can google in an
afternoon, don't have access to documentation about the stmmac IP,
and have only tested that (based on code comments and git commit
history) the programming of the subsecond register (and the clock rate)
makes more sense with these changes. Qualcomm has tested a similar
change offlist, verifying PTP more formally as I understand it.

The last version was an RFC/RFT, but I didn't get a lot of confirmation
that doing patch 3 in that series (essentially setting clk_ptp_ref to
whatever its max value is) for the whole stmmac ecosystem was a safe
idea. So I am erring on the side of caution and doing this for the
Qualcomm platform only. See v1 for an approach that would apply to
all stmmac platform drivers with clk_ptp_ref.

v1: https://lore.kernel.org/netdev/20230711205732.364954-1-ahalaney@redhat.com/
====================

Link: https://lore.kernel.org/r/20230725211853.895832-2-ahalaney@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: dwmac-qcom-ethqos: Use max frequency for clk_ptp_ref

Qualcomm clocks can set their frequency to a variety of levels
generally. Let's use the max for clk_ptp_ref to ensure the best
timestamping resolution possible.

Without this, the default value of the clock is used. For sa8775p-ride
this is 19.2 MHz, far less than the 230.4 MHz possible.

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Link: https://lore.kernel.org/r/20230725211853.895832-4-ahalaney@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>