platform/kernel/linux-starfive.git
19 months agoMerge branch 'mptcp-fixes-for-6-3'
Jakub Kicinski [Sat, 11 Mar 2023 05:42:58 +0000 (21:42 -0800)]
Merge branch 'mptcp-fixes-for-6-3'

Matthieu Baerts says:

====================
mptcp: fixes for 6.3

Patch 1 fixes a possible deadlock in subflow_error_report() reported by
lockdep. The report was in fact a false positive but the modification
makes sense and silences lockdep to allow syzkaller to find real issues.
The regression has been introduced in v5.12.

Patch 2 is a refactoring needed to be able to fix the two next issues.
It improves the situation and can be backported up to v6.0.

Patches 3 and 4 fix UaF reported by KASAN. It fixes issues potentially
visible since v5.7 and v5.19 but only reproducible until recently
(v6.0). These two patches depend on patch 2/7.

Patch 5 fixes the order of the printed values: expected vs seen values.
The regression has been introduced recently: v6.3-rc1.

Patch 6 adds missing ro_after_init flags. A previous patch added them
for other functions but these two have been missed. This previous patch
has been backported to stable versions (up to v5.12) so probably better
to do the same here.

Patch 7 fixes tcp_set_state() being called twice in a row since v5.10.

Patch 8 fixes another lockdep false positive issue but this time in
MPTCP PM code. Same here, some modifications in the code has been made
to silence this issue and help finding real ones later. This issue can
be seen since v6.2.

v1: https://lore.kernel.org/r/20230227-upstream-net-20230227-mptcp-fixes-v1-0-070e30ae4a8e@tessares.net
====================

Link: https://lore.kernel.org/r/20230227-upstream-net-20230227-mptcp-fixes-v2-0-47c2e95eada9@tessares.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agomptcp: fix lockdep false positive in mptcp_pm_nl_create_listen_socket()
Paolo Abeni [Thu, 9 Mar 2023 14:50:04 +0000 (15:50 +0100)]
mptcp: fix lockdep false positive in mptcp_pm_nl_create_listen_socket()

Christoph reports a lockdep splat in the mptcp_subflow_create_socket()
error path, when such function is invoked by
mptcp_pm_nl_create_listen_socket().

Such code path acquires two separates, nested socket lock, with the
internal lock operation lacking the "nested" annotation. Adding that
in sock_release() for mptcp's sake only could be confusing.

Instead just add a new lockclass to the in-kernel msk socket,
re-initializing the lockdep infra after the socket creation.

Fixes: ad2171009d96 ("mptcp: fix locking for in-kernel listener creation")
Cc: stable@vger.kernel.org
Reported-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/354
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agomptcp: avoid setting TCP_CLOSE state twice
Matthieu Baerts [Thu, 9 Mar 2023 14:50:03 +0000 (15:50 +0100)]
mptcp: avoid setting TCP_CLOSE state twice

tcp_set_state() is called from tcp_done() already.

There is then no need to first set the state to TCP_CLOSE, then call
tcp_done().

Fixes: d582484726c4 ("mptcp: fix fallback for MP_JOIN subflows")
Cc: stable@vger.kernel.org
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/362
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agomptcp: add ro_after_init for tcp{,v6}_prot_override
Geliang Tang [Thu, 9 Mar 2023 14:50:02 +0000 (15:50 +0100)]
mptcp: add ro_after_init for tcp{,v6}_prot_override

Add __ro_after_init labels for the variables tcp_prot_override and
tcpv6_prot_override, just like other variables adjacent to them, to
indicate that they are initialised from the init hooks and no writes
occur afterwards.

Fixes: b19bc2945b40 ("mptcp: implement delegated actions")
Cc: stable@vger.kernel.org
Fixes: 51fa7f8ebf0e ("mptcp: mark ops structures as ro_after_init")
Signed-off-by: Geliang Tang <geliang.tang@suse.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoselftests: mptcp: userspace pm: fix printed values
Matthieu Baerts [Thu, 9 Mar 2023 14:50:01 +0000 (15:50 +0100)]
selftests: mptcp: userspace pm: fix printed values

In case of errors, the printed message had the expected and the seen
value inverted.

This patch simply correct the order: first the expected value, then the
one that has been seen.

Fixes: 10d4273411be ("selftests: mptcp: userspace: print error details if any")
Cc: stable@vger.kernel.org
Acked-by: Geliang Tang <geliang.tang@suse.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agomptcp: fix UaF in listener shutdown
Paolo Abeni [Thu, 9 Mar 2023 14:50:00 +0000 (15:50 +0100)]
mptcp: fix UaF in listener shutdown

As reported by Christoph after having refactored the passive
socket initialization, the mptcp listener shutdown path is prone
to an UaF issue.

  BUG: KASAN: use-after-free in _raw_spin_lock_bh+0x73/0xe0
  Write of size 4 at addr ffff88810cb23098 by task syz-executor731/1266

  CPU: 1 PID: 1266 Comm: syz-executor731 Not tainted 6.2.0-rc59af4eaa31c1f6c00c8f1e448ed99a45c66340dd5 #6
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
  Call Trace:
   <TASK>
   dump_stack_lvl+0x6e/0x91
   print_report+0x16a/0x46f
   kasan_report+0xad/0x130
   kasan_check_range+0x14a/0x1a0
   _raw_spin_lock_bh+0x73/0xe0
   subflow_error_report+0x6d/0x110
   sk_error_report+0x3b/0x190
   tcp_disconnect+0x138c/0x1aa0
   inet_child_forget+0x6f/0x2e0
   inet_csk_listen_stop+0x209/0x1060
   __mptcp_close_ssk+0x52d/0x610
   mptcp_destroy_common+0x165/0x640
   mptcp_destroy+0x13/0x80
   __mptcp_destroy_sock+0xe7/0x270
   __mptcp_close+0x70e/0x9b0
   mptcp_close+0x2b/0x150
   inet_release+0xe9/0x1f0
   __sock_release+0xd2/0x280
   sock_close+0x15/0x20
   __fput+0x252/0xa20
   task_work_run+0x169/0x250
   exit_to_user_mode_prepare+0x113/0x120
   syscall_exit_to_user_mode+0x1d/0x40
   do_syscall_64+0x48/0x90
   entry_SYSCALL_64_after_hwframe+0x72/0xdc

The msk grace period can legitly expire in between the last
reference count dropped in mptcp_subflow_queue_clean() and
the later eventual access in inet_csk_listen_stop()

After the previous patch we don't need anymore special-casing
msk listener socket cleanup: the mptcp worker will process each
of the unaccepted msk sockets.

Just drop the now unnecessary code.

Please note this commit depends on the two parent ones:

  mptcp: refactor passive socket initialization
  mptcp: use the workqueue to destroy unaccepted sockets

Fixes: 6aeed9045071 ("mptcp: fix race on unaccepted mptcp sockets")
Cc: stable@vger.kernel.org
Reported-and-tested-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/346
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agomptcp: use the workqueue to destroy unaccepted sockets
Paolo Abeni [Thu, 9 Mar 2023 14:49:59 +0000 (15:49 +0100)]
mptcp: use the workqueue to destroy unaccepted sockets

Christoph reported a UaF at token lookup time after having
refactored the passive socket initialization part:

  BUG: KASAN: use-after-free in __token_bucket_busy+0x253/0x260
  Read of size 4 at addr ffff88810698d5b0 by task syz-executor653/3198

  CPU: 1 PID: 3198 Comm: syz-executor653 Not tainted 6.2.0-rc59af4eaa31c1f6c00c8f1e448ed99a45c66340dd5 #6
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
  Call Trace:
   <TASK>
   dump_stack_lvl+0x6e/0x91
   print_report+0x16a/0x46f
   kasan_report+0xad/0x130
   __token_bucket_busy+0x253/0x260
   mptcp_token_new_connect+0x13d/0x490
   mptcp_connect+0x4ed/0x860
   __inet_stream_connect+0x80e/0xd90
   tcp_sendmsg_fastopen+0x3ce/0x710
   mptcp_sendmsg+0xff1/0x1a20
   inet_sendmsg+0x11d/0x140
   __sys_sendto+0x405/0x490
   __x64_sys_sendto+0xdc/0x1b0
   do_syscall_64+0x3b/0x90
   entry_SYSCALL_64_after_hwframe+0x72/0xdc

We need to properly clean-up all the paired MPTCP-level
resources and be sure to release the msk last, even when
the unaccepted subflow is destroyed by the TCP internals
via inet_child_forget().

We can re-use the existing MPTCP_WORK_CLOSE_SUBFLOW infra,
explicitly checking that for the critical scenario: the
closed subflow is the MPC one, the msk is not accepted and
eventually going through full cleanup.

With such change, __mptcp_destroy_sock() is always called
on msk sockets, even on accepted ones. We don't need anymore
to transiently drop one sk reference at msk clone time.

Please note this commit depends on the parent one:

  mptcp: refactor passive socket initialization

Fixes: 58b09919626b ("mptcp: create msk early")
Cc: stable@vger.kernel.org
Reported-and-tested-by: Christoph Paasch <cpaasch@apple.com>
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/347
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agomptcp: refactor passive socket initialization
Paolo Abeni [Thu, 9 Mar 2023 14:49:58 +0000 (15:49 +0100)]
mptcp: refactor passive socket initialization

After commit 30e51b923e43 ("mptcp: fix unreleased socket in accept queue")
unaccepted msk sockets go throu complete shutdown, we don't need anymore
to delay inserting the first subflow into the subflow lists.

The reference counting deserve some extra care, as __mptcp_close() is
unaware of the request socket linkage to the first subflow.

Please note that this is more a refactoring than a fix but because this
modification is needed to include other corrections, see the following
commits. Then a Fixes tag has been added here to help the stable team.

Fixes: 30e51b923e43 ("mptcp: fix unreleased socket in accept queue")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agomptcp: fix possible deadlock in subflow_error_report
Paolo Abeni [Thu, 9 Mar 2023 14:49:57 +0000 (15:49 +0100)]
mptcp: fix possible deadlock in subflow_error_report

Christoph reported a possible deadlock while the TCP stack
destroys an unaccepted subflow due to an incoming reset: the
MPTCP socket error path tries to acquire the msk-level socket
lock while TCP still owns the listener socket accept queue
spinlock, and the reverse dependency already exists in the
TCP stack.

Note that the above is actually a lockdep false positive, as
the chain involves two separate sockets. A different per-socket
lockdep key will address the issue, but such a change will be
quite invasive.

Instead, we can simply stop earlier the socket error handling
for orphaned or unaccepted subflows, breaking the critical
lockdep chain. Error handling in such a scenario is a no-op.

Reported-and-tested-by: Christoph Paasch <cpaasch@apple.com>
Fixes: 15cc10453398 ("mptcp: deliver ssk errors to msk")
Cc: stable@vger.kernel.org
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/355
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoMerge branch 'update-xdp_features-flag-according-to-nic-re-configuration'
Jakub Kicinski [Sat, 11 Mar 2023 05:33:49 +0000 (21:33 -0800)]
Merge branch 'update-xdp_features-flag-according-to-nic-re-configuration'

Lorenzo Bianconi says:

====================
update xdp_features flag according to NIC re-configuration

Changes since v1:
- rebase on top of net tree
- remove NETDEV_XDP_ACT_NDO_XMIT_SG support in mlx5e driver
- always enable NETDEV_XDP_ACT_NDO_XMIT support in mlx5e driver
====================

Link: https://lore.kernel.org/r/cover.1678364612.git.lorenzo@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agomvpp2: take care of xdp_features when reconfiguring queues
Matteo Croce [Thu, 9 Mar 2023 12:25:32 +0000 (13:25 +0100)]
mvpp2: take care of xdp_features when reconfiguring queues

XDP is supported only if enough queues are present, so when reconfiguring
the queues set xdp_features accordingly.

Fixes: 66c0e13ad236 ("drivers: net: turn on XDP features")
Suggested-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Matteo Croce <teknoraver@meta.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: take into account device reconfiguration for xdp_features flag
Lorenzo Bianconi [Thu, 9 Mar 2023 12:25:31 +0000 (13:25 +0100)]
net/mlx5e: take into account device reconfiguration for xdp_features flag

Take into account LRO and GRO configuration setting device xdp_features
flag. Consider channel rq_wq_type enabling rx scatter-gatter support in
xdp_features flag and disable NETDEV_XDP_ACT_NDO_XMIT_SG since it is not
supported yet by the driver.
Moreover always enable NETDEV_XDP_ACT_NDO_XMIT as the ndo_xdp_xmit
callback does not require to load a dummy xdp program on the NIC.

Fixes: 66c0e13ad236 ("drivers: net: turn on XDP features")
Co-developed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoveth: take into account device reconfiguration for xdp_features flag
Lorenzo Bianconi [Thu, 9 Mar 2023 12:25:30 +0000 (13:25 +0100)]
veth: take into account device reconfiguration for xdp_features flag

Take into account tx/rx queues reconfiguration setting device
xdp_features flag. Moreover consider NETIF_F_GRO flag in order to enable
ndo_xdp_xmit callback.

Fixes: 66c0e13ad236 ("drivers: net: turn on XDP features")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: ena: take into account xdp_features setting tx/rx queues
Lorenzo Bianconi [Thu, 9 Mar 2023 12:25:29 +0000 (13:25 +0100)]
net: ena: take into account xdp_features setting tx/rx queues

ena nic allows xdp just if enough hw queues are available for XDP.
Take into account queues configuration setting xdp_features.

Fixes: 66c0e13ad236 ("drivers: net: turn on XDP features")
Reviewed-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: thunderx: take into account xdp_features setting tx/rx queues
Lorenzo Bianconi [Thu, 9 Mar 2023 12:25:28 +0000 (13:25 +0100)]
net: thunderx: take into account xdp_features setting tx/rx queues

thunderx nic allows xdp just if enough hw queues are available for XDP.
Take into account queues configuration setting xdp_features.

Fixes: 66c0e13ad236 ("drivers: net: turn on XDP features")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoxdp: add xdp_set_features_flag utility routine
Lorenzo Bianconi [Thu, 9 Mar 2023 12:25:27 +0000 (13:25 +0100)]
xdp: add xdp_set_features_flag utility routine

Introduce xdp_set_features_flag utility routine in order to update
dynamically xdp_features according to the dynamic hw configuration via
ethtool (e.g. changing number of hw rx/tx queues).
Add xdp_clear_features_flag() in order to clear all xdp_feature flag.

Reviewed-by: Shay Agroskin <shayagr@amazon.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agotools: ynl: fix get_mask utility routine
Lorenzo Bianconi [Thu, 9 Mar 2023 12:25:26 +0000 (13:25 +0100)]
tools: ynl: fix get_mask utility routine

Fix get_mask utility routine in order to take into account possible gaps
in the elements list.

Fixes: be5bea1cc0bf ("net: add basic C code generators for Netlink")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agotools: ynl: fix render-max for flags definition
Lorenzo Bianconi [Thu, 9 Mar 2023 12:25:25 +0000 (13:25 +0100)]
tools: ynl: fix render-max for flags definition

Properly manage render-max property for flags definition type
introducing mask value and setting it to (last_element << 1) - 1
instead of adding max value set to last_element + 1

Fixes: be5bea1cc0bf ("net: add basic C code generators for Netlink")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoi40e: Fix kernel crash during reboot when adapter is in recovery mode
Ivan Vecera [Thu, 9 Mar 2023 18:45:09 +0000 (10:45 -0800)]
i40e: Fix kernel crash during reboot when adapter is in recovery mode

If the driver detects during probe that firmware is in recovery
mode then i40e_init_recovery_mode() is called and the rest of
probe function is skipped including pci_set_drvdata(). Subsequent
i40e_shutdown() called during shutdown/reboot dereferences NULL
pointer as pci_get_drvdata() returns NULL.

To fix call pci_set_drvdata() also during entering to recovery mode.

Reproducer:
1) Lets have i40e NIC with firmware in recovery mode
2) Run reboot

Result:
[  139.084698] i40e: Intel(R) Ethernet Connection XL710 Network Driver
[  139.090959] i40e: Copyright (c) 2013 - 2019 Intel Corporation.
[  139.108438] i40e 0000:02:00.0: Firmware recovery mode detected. Limiting functionality.
[  139.116439] i40e 0000:02:00.0: Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.
[  139.129499] i40e 0000:02:00.0: fw 8.3.64775 api 1.13 nvm 8.30 0x8000b78d 1.3106.0 [8086:1583] [15d9:084a]
[  139.215932] i40e 0000:02:00.0 enp2s0f0: renamed from eth0
[  139.223292] i40e 0000:02:00.1: Firmware recovery mode detected. Limiting functionality.
[  139.231292] i40e 0000:02:00.1: Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.
[  139.244406] i40e 0000:02:00.1: fw 8.3.64775 api 1.13 nvm 8.30 0x8000b78d 1.3106.0 [8086:1583] [15d9:084a]
[  139.329209] i40e 0000:02:00.1 enp2s0f1: renamed from eth0
...
[  156.311376] BUG: kernel NULL pointer dereference, address: 00000000000006c2
[  156.318330] #PF: supervisor write access in kernel mode
[  156.323546] #PF: error_code(0x0002) - not-present page
[  156.328679] PGD 0 P4D 0
[  156.331210] Oops: 0002 [#1] PREEMPT SMP NOPTI
[  156.335567] CPU: 26 PID: 15119 Comm: reboot Tainted: G            E      6.2.0+ #1
[  156.343126] Hardware name: Abacus electric, s.r.o. - servis@abacus.cz Super Server/H12SSW-iN, BIOS 2.4 04/13/2022
[  156.353369] RIP: 0010:i40e_shutdown+0x15/0x130 [i40e]
[  156.358430] Code: c1 fc ff ff 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 f3 0f 1e fa 0f 1f 44 00 00 55 48 89 fd 53 48 8b 9f 48 01 00 00 <f0> 80 8b c2 06 00 00 04 f0 80 8b c0 06 00 00 08 48 8d bb 08 08 00
[  156.377168] RSP: 0018:ffffb223c8447d90 EFLAGS: 00010282
[  156.382384] RAX: ffffffffc073ee70 RBX: 0000000000000000 RCX: 0000000000000001
[  156.389510] RDX: 0000000080000001 RSI: 0000000000000246 RDI: ffff95db49988000
[  156.396634] RBP: ffff95db49988000 R08: ffffffffffffffff R09: ffffffff8bd17d40
[  156.403759] R10: 0000000000000001 R11: ffffffff8a5e3d28 R12: ffff95db49988000
[  156.410882] R13: ffffffff89a6fe17 R14: ffff95db49988150 R15: 0000000000000000
[  156.418007] FS:  00007fe7c0cc3980(0000) GS:ffff95ea8ee80000(0000) knlGS:0000000000000000
[  156.426083] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  156.431819] CR2: 00000000000006c2 CR3: 00000003092fc005 CR4: 0000000000770ee0
[  156.438944] PKRU: 55555554
[  156.441647] Call Trace:
[  156.444096]  <TASK>
[  156.446199]  pci_device_shutdown+0x38/0x60
[  156.450297]  device_shutdown+0x163/0x210
[  156.454215]  kernel_restart+0x12/0x70
[  156.457872]  __do_sys_reboot+0x1ab/0x230
[  156.461789]  ? vfs_writev+0xa6/0x1a0
[  156.465362]  ? __pfx_file_free_rcu+0x10/0x10
[  156.469635]  ? __call_rcu_common.constprop.85+0x109/0x5a0
[  156.475034]  do_syscall_64+0x3e/0x90
[  156.478611]  entry_SYSCALL_64_after_hwframe+0x72/0xdc
[  156.483658] RIP: 0033:0x7fe7bff37ab7

Fixes: 4ff0ee1af016 ("i40e: Introduce recovery mode support")
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Tested-by: Arpana Arland <arpanax.arland@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20230309184509.984639-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoipvlan: Make skb->skb_iif track skb->dev for l3s mode
Jianguo Wu [Thu, 9 Mar 2023 02:03:36 +0000 (10:03 +0800)]
ipvlan: Make skb->skb_iif track skb->dev for l3s mode

For l3s mode, skb->dev is set to ipvlan interface in ipvlan_nf_input():
  skb->dev = addr->master->dev
but, skb->skb_iif remain unchanged, this will cause socket lookup failed
if a target socket is bound to a interface, like the following example:

  ip link add ipvlan0 link eth0 type ipvlan mode l3s
  ip addr add dev ipvlan0 192.168.124.111/24
  ip link set ipvlan0 up

  ping -c 1 -I ipvlan0 8.8.8.8
  100% packet loss

This is because there is no match sk in __raw_v4_lookup() as sk->sk_bound_dev_if != dif(skb->skb_iif).
Fix this by make skb->skb_iif track skb->dev in ipvlan_nf_input().

Fixes: c675e06a98a4 ("ipvlan: decouple l3s mode dependencies from other modes")
Signed-off-by: Jianguo Wu <wujianguo@chinatelecom.cn>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/29865b1f-6db7-c07a-de89-949d3721ea30@163.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: phy: nxp-c45-tja11xx: fix MII_BASIC_CONFIG_REV bit
Radu Pirea (OSS) [Thu, 9 Mar 2023 10:01:11 +0000 (12:01 +0200)]
net: phy: nxp-c45-tja11xx: fix MII_BASIC_CONFIG_REV bit

According to the TJA1103 user manual, the bit for the reversed role in MII
or RMII modes is bit 4.

Cc: <stable@vger.kernel.org> # 5.15+
Fixes: b050f2f15e04 ("phy: nxp-c45: add driver for tja1103")
Signed-off-by: Radu Pirea (OSS) <radu-nicolae.pirea@oss.nxp.com>
Link: https://lore.kernel.org/r/20230309100111.1246214-1-radu-nicolae.pirea@oss.nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoMerge tag 'wireless-2023-03-10' of git://git.kernel.org/pub/scm/linux/kernel/git...
Jakub Kicinski [Sat, 11 Mar 2023 02:19:55 +0000 (18:19 -0800)]
Merge tag 'wireless-2023-03-10' of git://git./linux/kernel/git/wireless/wireless

Johannes Berg says:

====================
Just a few fixes:

 * MLO connection socket ownership didn't work
 * basic rates validation was missing (reported by
   by a private syzbot instances)
 * puncturing bitmap netlink policy was completely broken
 * properly check chandef for NULL channel, it can be
   pointing to a chandef that's still uninitialized

* tag 'wireless-2023-03-10' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless:
  wifi: cfg80211: fix MLO connection ownership
  wifi: mac80211: check basic rates validity
  wifi: nl80211: fix puncturing bitmap policy
  wifi: nl80211: fix NULL-ptr deref in offchan check
====================

Link: https://lore.kernel.org/r/20230310114647.35422-1-johannes@sipsolutions.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoMAINTAINERS: make my email address consistent
Jiri Pirko [Thu, 9 Mar 2023 11:49:11 +0000 (12:49 +0100)]
MAINTAINERS: make my email address consistent

Use jiri@resnulli.us in all MAINTAINERS entries and fixup .mailmap
so all other addresses point to that one.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Link: https://lore.kernel.org/r/20230309114911.923460-1-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoMerge branch 'add-checking-sq-is-full-inside-xdp-xmit'
Jakub Kicinski [Sat, 11 Mar 2023 00:46:02 +0000 (16:46 -0800)]
Merge branch 'add-checking-sq-is-full-inside-xdp-xmit'

Xuan Zhuo says:

====================
add checking sq is full inside xdp xmit

If the queue of xdp xmit is not an independent queue, then when the xdp
xmit used all the desc, the xmit from the __dev_queue_xmit() may encounter
the following error.

net ens4: Unexpected TXQ (0) queue failure: -28

This patch adds a check whether sq is full in XDP Xmit.
====================

Link: https://lore.kernel.org/r/20230308024935.91686-1-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agovirtio_net: add checking sq is full inside xdp xmit
Xuan Zhuo [Wed, 8 Mar 2023 02:49:35 +0000 (10:49 +0800)]
virtio_net: add checking sq is full inside xdp xmit

If the queue of xdp xmit is not an independent queue, then when the xdp
xmit used all the desc, the xmit from the __dev_queue_xmit() may encounter
the following error.

net ens4: Unexpected TXQ (0) queue failure: -28

This patch adds a check whether sq is full in xdp xmit.

Fixes: 56434a01b12e ("virtio_net: add XDP_TX support")
Reported-by: Yichun Zhang <yichun@openresty.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agovirtio_net: separate the logic of checking whether sq is full
Xuan Zhuo [Wed, 8 Mar 2023 02:49:34 +0000 (10:49 +0800)]
virtio_net: separate the logic of checking whether sq is full

Separate the logic of checking whether sq is full. The subsequent patch
will reuse this func.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agovirtio_net: reorder some funcs
Xuan Zhuo [Wed, 8 Mar 2023 02:49:33 +0000 (10:49 +0800)]
virtio_net: reorder some funcs

The purpose of this is to facilitate the subsequent addition of new
functions without introducing a separate declaration.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonfc: pn533: initialize struct pn533_out_arg properly
Fedor Pchelkin [Thu, 9 Mar 2023 16:50:50 +0000 (19:50 +0300)]
nfc: pn533: initialize struct pn533_out_arg properly

struct pn533_out_arg used as a temporary context for out_urb is not
initialized properly. Its uninitialized 'phy' field can be dereferenced in
error cases inside pn533_out_complete() callback function. It causes the
following failure:

general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] PREEMPT SMP KASAN
KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
CPU: 1 PID: 0 Comm: swapper/1 Not tainted 6.2.0-rc3-next-20230110-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 10/26/2022
RIP: 0010:pn533_out_complete.cold+0x15/0x44 drivers/nfc/pn533/usb.c:441
Call Trace:
 <IRQ>
 __usb_hcd_giveback_urb+0x2b6/0x5c0 drivers/usb/core/hcd.c:1671
 usb_hcd_giveback_urb+0x384/0x430 drivers/usb/core/hcd.c:1754
 dummy_timer+0x1203/0x32d0 drivers/usb/gadget/udc/dummy_hcd.c:1988
 call_timer_fn+0x1da/0x800 kernel/time/timer.c:1700
 expire_timers+0x234/0x330 kernel/time/timer.c:1751
 __run_timers kernel/time/timer.c:2022 [inline]
 __run_timers kernel/time/timer.c:1995 [inline]
 run_timer_softirq+0x326/0x910 kernel/time/timer.c:2035
 __do_softirq+0x1fb/0xaf6 kernel/softirq.c:571
 invoke_softirq kernel/softirq.c:445 [inline]
 __irq_exit_rcu+0x123/0x180 kernel/softirq.c:650
 irq_exit_rcu+0x9/0x20 kernel/softirq.c:662
 sysvec_apic_timer_interrupt+0x97/0xc0 arch/x86/kernel/apic/apic.c:1107

Initialize the field with the pn533_usb_phy currently used.

Found by Linux Verification Center (linuxtesting.org) with Syzkaller.

Fixes: 9dab880d675b ("nfc: pn533: Wait for out_urb's completion in pn533_usb_send_frame()")
Reported-by: syzbot+1e608ba4217c96d1952f@syzkaller.appspotmail.com
Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230309165050.207390-1-pchelkin@ispras.ru
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agowifi: cfg80211: fix MLO connection ownership
Johannes Berg [Wed, 1 Mar 2023 10:09:33 +0000 (12:09 +0200)]
wifi: cfg80211: fix MLO connection ownership

When disconnecting from an MLO connection we need the AP
MLD address, not an arbitrary BSSID. Fix the code to do
that.

Fixes: 9ecff10e82a5 ("wifi: nl80211: refactor BSS lookup in nl80211_associate()")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Link: https://lore.kernel.org/r/20230301115906.4c1b3b18980e.I008f070c7f3b8e8bde9278101ef9e40706a82902@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
19 months agowifi: mac80211: check basic rates validity
Johannes Berg [Fri, 24 Feb 2023 09:52:19 +0000 (10:52 +0100)]
wifi: mac80211: check basic rates validity

When userspace sets basic rates, it might send us some rates
list that's empty or consists of invalid values only. We're
currently ignoring invalid values and then may end up with a
rates bitmap that's empty, which later results in a warning.

Reject the call if there were no valid rates.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
19 months agowifi: nl80211: fix puncturing bitmap policy
Johannes Berg [Fri, 24 Feb 2023 12:36:57 +0000 (13:36 +0100)]
wifi: nl80211: fix puncturing bitmap policy

This was meant to be a u32, and while applying the patch
I tried to use policy validation for it. However, not only
did I copy/paste it to u8 instead of u32, but also used
the policy range erroneously. Fix both of these issues.

Fixes: d7c1a9a0ed18 ("wifi: nl80211: validate and configure puncturing bitmap")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
19 months agowifi: nl80211: fix NULL-ptr deref in offchan check
Johannes Berg [Wed, 1 Mar 2023 10:09:29 +0000 (12:09 +0200)]
wifi: nl80211: fix NULL-ptr deref in offchan check

If, e.g. in AP mode, the link was already created by userspace
but not activated yet, it has a chandef but the chandef isn't
valid and has no channel. Check for this and ignore this link.

Fixes: 7b0a0e3c3a88 ("wifi: cfg80211: do some rework towards MLO link APIs")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Link: https://lore.kernel.org/r/20230301115906.71bd4803fbb9.Iee39c0f6c2d3a59a8227674dc55d52e38b1090cf@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
19 months agotcp: tcp_make_synack() can be called from process context
Breno Leitao [Wed, 8 Mar 2023 19:07:45 +0000 (11:07 -0800)]
tcp: tcp_make_synack() can be called from process context

tcp_rtx_synack() now could be called in process context as explained in
0a375c822497 ("tcp: tcp_rtx_synack() can be called from process
context").

tcp_rtx_synack() might call tcp_make_synack(), which will touch per-CPU
variables with preemption enabled. This causes the following BUG:

    BUG: using __this_cpu_add() in preemptible [00000000] code: ThriftIO1/5464
    caller is tcp_make_synack+0x841/0xac0
    Call Trace:
     <TASK>
     dump_stack_lvl+0x10d/0x1a0
     check_preemption_disabled+0x104/0x110
     tcp_make_synack+0x841/0xac0
     tcp_v6_send_synack+0x5c/0x450
     tcp_rtx_synack+0xeb/0x1f0
     inet_rtx_syn_ack+0x34/0x60
     tcp_check_req+0x3af/0x9e0
     tcp_rcv_state_process+0x59b/0x2030
     tcp_v6_do_rcv+0x5f5/0x700
     release_sock+0x3a/0xf0
     tcp_sendmsg+0x33/0x40
     ____sys_sendmsg+0x2f2/0x490
     __sys_sendmsg+0x184/0x230
     do_syscall_64+0x3d/0x90

Avoid calling __TCP_INC_STATS() with will touch per-cpu variables. Use
TCP_INC_STATS() which is safe to be called from context switch.

Fixes: 8336886f786f ("tcp: TCP Fast Open Server - support TFO listeners")
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230308190745.780221-1-leitao@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoMerge tag 'net-6.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 9 Mar 2023 18:56:58 +0000 (10:56 -0800)]
Merge tag 'net-6.3-rc2' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Including fixes from netfilter and bpf.

  Current release - regressions:

   - core: avoid skb end_offset change in __skb_unclone_keeptruesize()

   - sched:
      - act_connmark: handle errno on tcf_idr_check_alloc
      - flower: fix fl_change() error recovery path

   - ieee802154: prevent user from crashing the host

  Current release - new code bugs:

   - eth: bnxt_en: fix the double free during device removal

   - tools: ynl:
      - fix enum-as-flags in the generic CLI
      - fully inherit attrs in subsets
      - re-license uniformly under GPL-2.0 or BSD-3-clause

  Previous releases - regressions:

   - core: use indirect calls helpers for sk_exit_memory_pressure()

   - tls:
      - fix return value for async crypto
      - avoid hanging tasks on the tx_lock

   - eth: ice: copy last block omitted in ice_get_module_eeprom()

  Previous releases - always broken:

   - core: avoid double iput when sock_alloc_file fails

   - af_unix: fix struct pid leaks in OOB support

   - tls:
      - fix possible race condition
      - fix device-offloaded sendpage straddling records

   - bpf:
      - sockmap: fix an infinite loop error
      - test_run: fix &xdp_frame misplacement for LIVE_FRAMES
      - fix resolving BTF_KIND_VAR after ARRAY, STRUCT, UNION, PTR

   - netfilter: tproxy: fix deadlock due to missing BH disable

   - phylib: get rid of unnecessary locking

   - eth: bgmac: fix *initial* chip reset to support BCM5358

   - eth: nfp: fix csum for ipsec offload

   - eth: mtk_eth_soc: fix RX data corruption issue

  Misc:

   - usb: qmi_wwan: add telit 0x1080 composition"

* tag 'net-6.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (64 commits)
  tools: ynl: fix enum-as-flags in the generic CLI
  tools: ynl: move the enum classes to shared code
  net: avoid double iput when sock_alloc_file fails
  af_unix: fix struct pid leaks in OOB support
  eth: fealnx: bring back this old driver
  net: dsa: mt7530: permit port 5 to work without port 6 on MT7621 SoC
  net: microchip: sparx5: fix deletion of existing DSCP mappings
  octeontx2-af: Unlock contexts in the queue context cache in case of fault detection
  net/smc: fix fallback failed while sendmsg with fastopen
  ynl: re-license uniformly under GPL-2.0 OR BSD-3-Clause
  mailmap: update entries for Stephen Hemminger
  mailmap: add entry for Maxim Mikityanskiy
  nfc: change order inside nfc_se_io error path
  ethernet: ice: avoid gcc-9 integer overflow warning
  ice: don't ignore return codes in VSI related code
  ice: Fix DSCP PFC TLV creation
  net: usb: qmi_wwan: add Telit 0x1080 composition
  net: usb: cdc_mbim: avoid altsetting toggling for Telit FE990
  netfilter: conntrack: adopt safer max chain length
  net: tls: fix device-offloaded sendpage straddling records
  ...

19 months agoMerge tag 'for-linus-2023030901' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 9 Mar 2023 18:17:23 +0000 (10:17 -0800)]
Merge tag 'for-linus-2023030901' of git://git./linux/kernel/git/hid/hid

Pull HID fixes from Benjamin Tissoires:

 - fix potential out of bound write of zeroes in HID core with a
   specially crafted uhid device (Lee Jones)

 - fix potential use-after-free in work function in intel-ish-hid (Reka
   Norman)

 - selftests config fixes (Benjamin Tissoires)

 - few device small fixes and support

* tag 'for-linus-2023030901' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid:
  HID: intel-ish-hid: ipc: Fix potential use-after-free in work function
  HID: logitech-hidpp: Add support for Logitech MX Master 3S mouse
  HID: cp2112: Fix driver not registering GPIO IRQ chip as threaded
  selftest: hid: fix hid_bpf not set in config
  HID: uhid: Over-ride the default maximum data buffer value with our own
  HID: core: Provide new max_buffer_size attribute to over-ride the default

19 months agoMerge tag 'm68k-for-v6.3-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git/geert...
Linus Torvalds [Thu, 9 Mar 2023 18:08:46 +0000 (10:08 -0800)]
Merge tag 'm68k-for-v6.3-tag2' of git://git./linux/kernel/git/geert/linux-m68k

Pull m68k fixes from Geert Uytterhoeven:

 - Fix systems with memory at end of 32-bit address space

 - Fix initrd on systems where memory does not start at address zero

 - Fix 68030 handling of bus errors for addresses in exception tables

* tag 'm68k-for-v6.3-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k:
  m68k: Only force 030 bus error if PC not in exception table
  m68k: mm: Move initrd phys_to_virt handling after paging_init()
  m68k: mm: Fix systems with memory at end of 32-bit address space

19 months agosh: sanitize the flags on sigreturn
Al Viro [Mon, 6 Mar 2023 01:20:30 +0000 (01:20 +0000)]
sh: sanitize the flags on sigreturn

We fetch %SR value from sigframe; it might have been modified by signal
handler, so we can't trust it with any bits that are not modifiable in
user mode.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Cc: Rich Felker <dalias@libc.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
19 months agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net...
Paolo Abeni [Thu, 9 Mar 2023 10:45:08 +0000 (11:45 +0100)]
Merge branch '100GbE' of git://git./linux/kernel/git/tnguy/net-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2023-03-07 (ice)

This series contains updates to ice driver only.

Dave removes masking from pfcena field as it was incorrectly preventing
valid traffic classes from being enabled.

Michal resolves various smatch issues such as not propagating error
codes and returning 0 explicitly.

Arnd Bergmann resolves gcc-9 warning for integer overflow.

* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
  ethernet: ice: avoid gcc-9 integer overflow warning
  ice: don't ignore return codes in VSI related code
  ice: Fix DSCP PFC TLV creation
====================

Link: https://lore.kernel.org/r/20230307220714.3997294-1-anthony.l.nguyen@intel.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
19 months agoMerge branch 'tools-ynl-fix-enum-as-flags-in-the-generic-cli'
Jakub Kicinski [Thu, 9 Mar 2023 07:28:23 +0000 (23:28 -0800)]
Merge branch 'tools-ynl-fix-enum-as-flags-in-the-generic-cli'

Jakub Kicinski says:

====================
tools: ynl: fix enum-as-flags in the generic CLI

The CLI needs to use proper classes when looking at Enum definitions
rather than interpreting the YAML spec ad-hoc, because we have more
than on format of the definition supported.
====================

Link: https://lore.kernel.org/r/20230308003923.445268-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agotools: ynl: fix enum-as-flags in the generic CLI
Jakub Kicinski [Wed, 8 Mar 2023 00:39:23 +0000 (16:39 -0800)]
tools: ynl: fix enum-as-flags in the generic CLI

Lorenzo points out that the generic CLI is broken for the netdev
family. When I added the support for documentation of enums
(and sparse enums) the client script was not updated.
It expects the values in enum to be a list of names,
now it can also be a dict (YAML object).

Reported-by: Lorenzo Bianconi <lorenzo@kernel.org>
Fixes: e4b48ed460d3 ("tools: ynl: add a completely generic client")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agotools: ynl: move the enum classes to shared code
Jakub Kicinski [Wed, 8 Mar 2023 00:39:22 +0000 (16:39 -0800)]
tools: ynl: move the enum classes to shared code

Move bulk of the EnumSet and EnumEntry code to shared
code for reuse by cli.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: avoid double iput when sock_alloc_file fails
Thadeu Lima de Souza Cascardo [Tue, 7 Mar 2023 17:37:07 +0000 (14:37 -0300)]
net: avoid double iput when sock_alloc_file fails

When sock_alloc_file fails to allocate a file, it will call sock_release.
__sys_socket_file should then not call sock_release again, otherwise there
will be a double free.

[   89.319884] ------------[ cut here ]------------
[   89.320286] kernel BUG at fs/inode.c:1764!
[   89.320656] invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
[   89.321051] CPU: 7 PID: 125 Comm: iou-sqp-124 Not tainted 6.2.0+ #361
[   89.321535] RIP: 0010:iput+0x1ff/0x240
[   89.321808] Code: d1 83 e1 03 48 83 f9 02 75 09 48 81 fa 00 10 00 00 77 05 83 e2 01 75 1f 4c 89 ef e8 fb d2 ba 00 e9 80 fe ff ff c3 cc cc cc cc <0f> 0b 0f 0b e9 d0 fe ff ff 0f 0b eb 8d 49 8d b4 24 08 01 00 00 48
[   89.322760] RSP: 0018:ffffbdd60068bd50 EFLAGS: 00010202
[   89.323036] RAX: 0000000000000000 RBX: ffff9d7ad3cacac0 RCX: 0000000000001107
[   89.323412] RDX: 000000000003af00 RSI: 0000000000000000 RDI: ffff9d7ad3cacb40
[   89.323785] RBP: ffffbdd60068bd68 R08: ffffffffffffffff R09: ffffffffab606438
[   89.324157] R10: ffffffffacb3dfa0 R11: 6465686361657256 R12: ffff9d7ad3cacb40
[   89.324529] R13: 0000000080000001 R14: 0000000080000001 R15: 0000000000000002
[   89.324904] FS:  00007f7b28516740(0000) GS:ffff9d7aeb1c0000(0000) knlGS:0000000000000000
[   89.325328] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   89.325629] CR2: 00007f0af52e96c0 CR3: 0000000002a02006 CR4: 0000000000770ee0
[   89.326004] PKRU: 55555554
[   89.326161] Call Trace:
[   89.326298]  <TASK>
[   89.326419]  __sock_release+0xb5/0xc0
[   89.326632]  __sys_socket_file+0xb2/0xd0
[   89.326844]  io_socket+0x88/0x100
[   89.327039]  ? io_issue_sqe+0x6a/0x430
[   89.327258]  io_issue_sqe+0x67/0x430
[   89.327450]  io_submit_sqes+0x1fe/0x670
[   89.327661]  io_sq_thread+0x2e6/0x530
[   89.327859]  ? __pfx_autoremove_wake_function+0x10/0x10
[   89.328145]  ? __pfx_io_sq_thread+0x10/0x10
[   89.328367]  ret_from_fork+0x29/0x50
[   89.328576] RIP: 0033:0x0
[   89.328732] Code: Unable to access opcode bytes at 0xffffffffffffffd6.
[   89.329073] RSP: 002b:0000000000000000 EFLAGS: 00000202 ORIG_RAX: 00000000000001a9
[   89.329477] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 00007f7b28637a3d
[   89.329845] RDX: 00007fff4e4318a8 RSI: 00007fff4e4318b0 RDI: 0000000000000400
[   89.330216] RBP: 00007fff4e431830 R08: 00007fff4e431711 R09: 00007fff4e4318b0
[   89.330584] R10: 0000000000000000 R11: 0000000000000202 R12: 00007fff4e441b38
[   89.330950] R13: 0000563835e3e725 R14: 0000563835e40d10 R15: 00007f7b28784040
[   89.331318]  </TASK>
[   89.331441] Modules linked in:
[   89.331617] ---[ end trace 0000000000000000 ]---

Fixes: da214a475f8b ("net: add __sys_socket_file()")
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230307173707.468744-1-cascardo@canonical.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoaf_unix: fix struct pid leaks in OOB support
Eric Dumazet [Tue, 7 Mar 2023 16:45:30 +0000 (16:45 +0000)]
af_unix: fix struct pid leaks in OOB support

syzbot reported struct pid leak [1].

Issue is that queue_oob() calls maybe_add_creds() which potentially
holds a reference on a pid.

But skb->destructor is not set (either directly or by calling
unix_scm_to_skb())

This means that subsequent kfree_skb() or consume_skb() would leak
this reference.

In this fix, I chose to fully support scm even for the OOB message.

[1]
BUG: memory leak
unreferenced object 0xffff8881053e7f80 (size 128):
comm "syz-executor242", pid 5066, jiffies 4294946079 (age 13.220s)
hex dump (first 32 bytes):
01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace:
[<ffffffff812ae26a>] alloc_pid+0x6a/0x560 kernel/pid.c:180
[<ffffffff812718df>] copy_process+0x169f/0x26c0 kernel/fork.c:2285
[<ffffffff81272b37>] kernel_clone+0xf7/0x610 kernel/fork.c:2684
[<ffffffff812730cc>] __do_sys_clone+0x7c/0xb0 kernel/fork.c:2825
[<ffffffff849ad699>] do_syscall_x64 arch/x86/entry/common.c:50 [inline]
[<ffffffff849ad699>] do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
[<ffffffff84a0008b>] entry_SYSCALL_64_after_hwframe+0x63/0xcd

Fixes: 314001f0bf92 ("af_unix: Add OOB support")
Reported-by: syzbot+7699d9e5635c10253a27@syzkaller.appspotmail.com
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Rao Shoaib <rao.shoaib@oracle.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230307164530.771896-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoeth: fealnx: bring back this old driver
Jakub Kicinski [Tue, 7 Mar 2023 17:19:30 +0000 (09:19 -0800)]
eth: fealnx: bring back this old driver

This reverts commit d5e2d038dbece821f1af57acbeded3aa9a1832c1.

We have a report of this chip being used on a

  SURECOM EP-320X-S 100/10M Ethernet PCI Adapter

which could still have been purchased in some parts
of the world 3 years ago.

Cc: stable@vger.kernel.org
Link: https://bugzilla.kernel.org/show_bug.cgi?id=217151
Fixes: d5e2d038dbec ("eth: fealnx: delete the driver for Myson MTD-800")
Link: https://lore.kernel.org/r/20230307171930.4008454-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: dsa: mt7530: permit port 5 to work without port 6 on MT7621 SoC
Vladimir Oltean [Tue, 7 Mar 2023 15:54:11 +0000 (17:54 +0200)]
net: dsa: mt7530: permit port 5 to work without port 6 on MT7621 SoC

The MT7530 switch from the MT7621 SoC has 2 ports which can be set up as
internal: port 5 and 6. Arınç reports that the GMAC1 attached to port 5
receives corrupted frames, unless port 6 (attached to GMAC0) has been
brought up by the driver. This is true regardless of whether port 5 is
used as a user port or as a CPU port (carrying DSA tags).

Offline debugging (blind for me) which began in the linked thread showed
experimentally that the configuration done by the driver for port 6
contains a step which is needed by port 5 as well - the write to
CORE_GSWPLL_GRP2 (note that I've no idea as to what it does, apart from
the comment "Set core clock into 500Mhz"). Prints put by Arınç show that
the reset value of CORE_GSWPLL_GRP2 is RG_GSWPLL_POSDIV_500M(1) |
RG_GSWPLL_FBKDIV_500M(40) (0x128), both on the MCM MT7530 from the
MT7621 SoC, as well as on the standalone MT7530 from MT7623NI Bananapi
BPI-R2. Apparently, port 5 on the standalone MT7530 can work under both
values of the register, while on the MT7621 SoC it cannot.

The call path that triggers the register write is:

mt753x_phylink_mac_config() for port 6
-> mt753x_pad_setup()
   -> mt7530_pad_clk_setup()

so this fully explains the behavior noticed by Arınç, that bringing port
6 up is necessary.

The simplest fix for the problem is to extract the register writes which
are needed for both port 5 and 6 into a common mt7530_pll_setup()
function, which is called at mt7530_setup() time, immediately after
switch reset. We can argue that this mirrors the code layout introduced
in mt7531_setup() by commit 42bc4fafe359 ("net: mt7531: only do PLL once
after the reset"), in that the PLL setup has the exact same positioning,
and further work to consolidate the separate setup() functions is not
hindered.

Testing confirms that:

- the slight reordering of writes to MT7530_P6ECR and to
  CORE_GSWPLL_GRP1 / CORE_GSWPLL_GRP2 introduced by this change does not
  appear to cause problems for the operation of port 6 on MT7621 and on
  MT7623 (where port 5 also always worked)

- packets sent through port 5 are not corrupted anymore, regardless of
  whether port 6 is enabled by phylink or not (or even present in the
  device tree)

My algorithm for determining the Fixes: tag is as follows. Testing shows
that some logic from mt7530_pad_clk_setup() is needed even for port 5.
Prior to commit ca366d6c889b ("net: dsa: mt7530: Convert to PHYLINK
API"), a call did exist for all phy_is_pseudo_fixed_link() ports - so
port 5 included. That commit replaced it with a temporary "Port 5 is not
supported!" comment, and the following commit 38f790a80560 ("net: dsa:
mt7530: Add support for port 5") replaced that comment with a
configuration procedure in mt7530_setup_port5() which was insufficient
for port 5 to work. I'm laying the blame on the patch that claimed
support for port 5, although one would have also needed the change from
commit c3b8e07909db ("net: dsa: mt7530: setup core clock even in TRGMII
mode") for the write to be performed completely independently from port
6's configuration.

Thanks go to Arınç for describing the problem, for debugging and for
testing.

Reported-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Link: https://lore.kernel.org/netdev/f297c2c4-6e7c-57ac-2394-f6025d309b9d@arinc9.com/
Fixes: 38f790a80560 ("net: dsa: mt7530: Add support for port 5")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230307155411.868573-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoMerge tag 'fs_for_v6.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/jack...
Linus Torvalds [Wed, 8 Mar 2023 20:02:09 +0000 (12:02 -0800)]
Merge tag 'fs_for_v6.3-rc2' of git://git./linux/kernel/git/jack/linux-fs

Pull udf fixes from Jan Kara:
 "Fix bugs in UDF caused by the big pile of changes that went in during
  the merge window"

* tag 'fs_for_v6.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs:
  udf: Warn if block mapping is done for in-ICB files
  udf: Fix reading of in-ICB files
  udf: Fix lost writes in udf_adinicb_writepage()

20 months agoMerge tag 'platform-drivers-x86-v6.3-2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 8 Mar 2023 19:56:45 +0000 (11:56 -0800)]
Merge tag 'platform-drivers-x86-v6.3-2' of git://git./linux/kernel/git/pdx86/platform-drivers-x86

Pull x86 platform driver fixes from Hans de Goede:
 "A small set of assorted bug and build/warning fixes"

* tag 'platform-drivers-x86-v6.3-2' of git://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86:
  platform: mellanox: mlx-platform: Initialize shift variable to 0
  platform/x86: int3472: Add GPIOs to Surface Go 3 Board data
  platform/x86: ISST: Fix kernel documentation warnings
  platform: x86: MLX_PLATFORM: select REGMAP instead of depending on it
  platform: mellanox: select REGMAP instead of depending on it
  platform/x86/intel/tpmi: Fix double free reported by Smatch
  platform/x86: ISST: Increase range of valid mail box commands
  platform/x86: dell-ddv: Fix temperature scaling
  platform/x86: dell-ddv: Fix cache invalidation on resume
  platform/x86/amd: pmc: remove CONFIG_SUSPEND checks

20 months agox86/resctl: fix scheduler confusion with 'current'
Linus Torvalds [Tue, 7 Mar 2023 21:06:29 +0000 (13:06 -0800)]
x86/resctl: fix scheduler confusion with 'current'

The implementation of 'current' on x86 is very intentionally special: it
is a very common thing to look up, and it uses 'this_cpu_read_stable()'
to get the current thread pointer efficiently from per-cpu storage.

And the keyword in there is 'stable': the current thread pointer never
changes as far as a single thread is concerned.  Even if when a thread
is preempted, or moved to another CPU, or even across an explicit call
'schedule()' that thread will still have the same value for 'current'.

It is, after all, the kernel base pointer to thread-local storage.
That's why it's stable to begin with, but it's also why it's important
enough that we have that special 'this_cpu_read_stable()' access for it.

So this is all done very intentionally to allow the compiler to treat
'current' as a value that never visibly changes, so that the compiler
can do CSE and combine multiple different 'current' accesses into one.

However, there is obviously one very special situation when the
currently running thread does actually change: inside the scheduler
itself.

So the scheduler code paths are special, and do not have a 'current'
thread at all.  Instead there are _two_ threads: the previous and the
next thread - typically called 'prev' and 'next' (or prev_p/next_p)
internally.

So this is all actually quite straightforward and simple, and not all
that complicated.

Except for when you then have special code that is run in scheduler
context, that code then has to be aware that 'current' isn't really a
valid thing.  Did you mean 'prev'? Did you mean 'next'?

In fact, even if then look at the code, and you use 'current' after the
new value has been assigned to the percpu variable, we have explicitly
told the compiler that 'current' is magical and always stable.  So the
compiler is quite free to use an older (or newer) value of 'current',
and the actual assignment to the percpu storage is not relevant even if
it might look that way.

Which is exactly what happened in the resctl code, that blithely used
'current' in '__resctrl_sched_in()' when it really wanted the new
process state (as implied by the name: we're scheduling 'into' that new
resctl state).  And clang would end up just using the old thread pointer
value at least in some configurations.

This could have happened with gcc too, and purely depends on random
compiler details.  Clang just seems to have been more aggressive about
moving the read of the per-cpu current_task pointer around.

The fix is trivial: just make the resctl code adhere to the scheduler
rules of using the prev/next thread pointer explicitly, instead of using
'current' in a situation where it just wasn't valid.

That same code is then also used outside of the scheduler context (when
a thread resctl state is explicitly changed), and then we will just pass
in 'current' as that pointer, of course.  There is no ambiguity in that
case.

The fix may be trivial, but noticing and figuring out what went wrong
was not.  The credit for that goes to Stephane Eranian.

Reported-by: Stephane Eranian <eranian@google.com>
Link: https://lore.kernel.org/lkml/20230303231133.1486085-1-eranian@google.com/
Link: https://lore.kernel.org/lkml/alpine.LFD.2.01.0908011214330.3304@localhost.localdomain/
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Tested-by: Stephane Eranian <eranian@google.com>
Tested-by: Babu Moger <babu.moger@amd.com>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
20 months agonet: microchip: sparx5: fix deletion of existing DSCP mappings
Daniel Machon [Tue, 7 Mar 2023 11:21:03 +0000 (12:21 +0100)]
net: microchip: sparx5: fix deletion of existing DSCP mappings

Fix deletion of existing DSCP mappings in the APP table.

Adding and deleting DSCP entries are replicated per-port, since the
mapping table is global for all ports in the chip. Whenever a mapping
for a DSCP value already exists, the old mapping is deleted first.
However, it is only deleted for the specified port. Fix this by calling
sparx5_dcb_ieee_delapp() instead of dcb_ieee_delapp() as it ought to be.

Reproduce:

// Map and remap DSCP value 63
$ dcb app add dev eth0 dscp-prio 63:1
$ dcb app add dev eth0 dscp-prio 63:2

$ dcb app show dev eth0 dscp-prio
dscp-prio 63:2

$ dcb app show dev eth1 dscp-prio
dscp-prio 63:1 63:2 <-- 63:1 should not be there

Fixes: 8dcf69a64118 ("net: microchip: sparx5: add support for offloading dscp table")
Signed-off-by: Daniel Machon <daniel.machon@microchip.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 months agoocteontx2-af: Unlock contexts in the queue context cache in case of fault detection
Suman Ghosh [Tue, 7 Mar 2023 10:49:08 +0000 (16:19 +0530)]
octeontx2-af: Unlock contexts in the queue context cache in case of fault detection

NDC caches contexts of frequently used queue's (Rx and Tx queues)
contexts. Due to a HW errata when NDC detects fault/poision while
accessing contexts it could go into an illegal state where a cache
line could get locked forever. To makesure all cache lines in NDC
are available for optimum performance upon fault/lockerror/posion
errors scan through all cache lines in NDC and clear the lock bit.

Fixes: 4a3581cd5995 ("octeontx2-af: NPA AQ instruction enqueue support")
Signed-off-by: Suman Ghosh <sumang@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: Sai Krishna <saikrishnag@marvell.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 months agonet/smc: fix fallback failed while sendmsg with fastopen
D. Wythe [Tue, 7 Mar 2023 03:23:46 +0000 (11:23 +0800)]
net/smc: fix fallback failed while sendmsg with fastopen

Before determining whether the msg has unsupported options, it has been
prematurely terminated by the wrong status check.

For the application, the general usages of MSG_FASTOPEN likes

fd = socket(...)
/* rather than connect */
sendto(fd, data, len, MSG_FASTOPEN)

Hence, We need to check the flag before state check, because the sock
state here is always SMC_INIT when applications tries MSG_FASTOPEN.
Once we found unsupported options, fallback it to TCP.

Fixes: ee9dfbef02d1 ("net/smc: handle sockopts forcing fallback")
Signed-off-by: D. Wythe <alibuda@linux.alibaba.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
v2 -> v1: Optimize code style
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 months agoynl: re-license uniformly under GPL-2.0 OR BSD-3-Clause
Jakub Kicinski [Mon, 6 Mar 2023 20:04:57 +0000 (12:04 -0800)]
ynl: re-license uniformly under GPL-2.0 OR BSD-3-Clause

I was intending to make all the Netlink Spec code BSD-3-Clause
to ease the adoption but it appears that:
 - I fumbled the uAPI and used "GPL WITH uAPI note" there
 - it gives people pause as they expect GPL in the kernel
As suggested by Chuck re-license under dual. This gives us benefit
of full BSD freedom while fulfilling the broad "kernel is under GPL"
expectations.

Link: https://lore.kernel.org/all/20230304120108.05dd44c5@kernel.org/
Link: https://lore.kernel.org/r/20230306200457.3903854-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agomailmap: update entries for Stephen Hemminger
Stephen Hemminger [Mon, 6 Mar 2023 19:44:05 +0000 (11:44 -0800)]
mailmap: update entries for Stephen Hemminger

Map all my old email addresses to current address.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Link: https://lore.kernel.org/r/20230306194405.108236-1-stephen@networkplumber.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agomailmap: add entry for Maxim Mikityanskiy
Jakub Kicinski [Mon, 6 Mar 2023 19:20:18 +0000 (11:20 -0800)]
mailmap: add entry for Maxim Mikityanskiy

Map Maxim's old corporate addresses to his personal one.

Link: https://lore.kernel.org/r/20230306192018.3894988-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agonfc: change order inside nfc_se_io error path
Fedor Pchelkin [Mon, 6 Mar 2023 21:26:50 +0000 (00:26 +0300)]
nfc: change order inside nfc_se_io error path

cb_context should be freed on the error path in nfc_se_io as stated by
commit 25ff6f8a5a3b ("nfc: fix memory leak of se_io context in
nfc_genl_se_io").

Make the error path in nfc_se_io unwind everything in reverse order, i.e.
free the cb_context after unlocking the device.

Suggested-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Link: https://lore.kernel.org/r/20230306212650.230322-1-pchelkin@ispras.ru
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agoethernet: ice: avoid gcc-9 integer overflow warning
Arnd Bergmann [Tue, 14 Feb 2023 15:25:36 +0000 (16:25 +0100)]
ethernet: ice: avoid gcc-9 integer overflow warning

With older compilers like gcc-9, the calculation of the vlan
priority field causes a false-positive warning from the byteswap:

In file included from drivers/net/ethernet/intel/ice/ice_tc_lib.c:4:
drivers/net/ethernet/intel/ice/ice_tc_lib.c: In function 'ice_parse_cls_flower':
include/uapi/linux/swab.h:15:15: error: integer overflow in expression '(int)(short unsigned int)((int)match.key-><U67c8>.<U6698>.vlan_priority << 13) & 57344 & 255' of type 'int' results in '0' [-Werror=overflow]
   15 |  (((__u16)(x) & (__u16)0x00ffU) << 8) |   \
      |   ~~~~~~~~~~~~^~~~~~~~~~~~~~~~~
include/uapi/linux/swab.h:106:2: note: in expansion of macro '___constant_swab16'
  106 |  ___constant_swab16(x) :   \
      |  ^~~~~~~~~~~~~~~~~~
include/uapi/linux/byteorder/little_endian.h:42:43: note: in expansion of macro '__swab16'
   42 | #define __cpu_to_be16(x) ((__force __be16)__swab16((x)))
      |                                           ^~~~~~~~
include/linux/byteorder/generic.h:96:21: note: in expansion of macro '__cpu_to_be16'
   96 | #define cpu_to_be16 __cpu_to_be16
      |                     ^~~~~~~~~~~~~
drivers/net/ethernet/intel/ice/ice_tc_lib.c:1458:5: note: in expansion of macro 'cpu_to_be16'
 1458 |     cpu_to_be16((match.key->vlan_priority <<
      |     ^~~~~~~~~~~

After a change to be16_encode_bits(), the code becomes more
readable to both people and compilers, which avoids the warning.

Fixes: 34800178b302 ("ice: Add support for VLAN priority filters in switchdev")
Suggested-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Tested-by: Sujai Buvaneswaran <sujai.buvaneswaran@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
20 months agoice: don't ignore return codes in VSI related code
Michal Swiatkowski [Mon, 13 Feb 2023 11:27:33 +0000 (12:27 +0100)]
ice: don't ignore return codes in VSI related code

There were few smatch warnings reported by Dan:
- ice_vsi_cfg_xdp_txqs can return 0 instead of ret, which is cleaner
- return values in ice_vsi_cfg_def were ignored
- in ice_vsi_rebuild return value was ignored in case rebuild failed,
  it was a never reached code, however, rewrite it for clarity.
- ice_vsi_cfg_tc can return 0 instead of ret

Fixes: 6624e780a577 ("ice: split ice_vsi_setup into smaller functions")
Reported-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Michal Swiatkowski <michal.swiatkowski@linux.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>
20 months agoice: Fix DSCP PFC TLV creation
Dave Ertman [Fri, 27 Jan 2023 13:24:10 +0000 (14:24 +0100)]
ice: Fix DSCP PFC TLV creation

When creating the TLV to send to the FW for configuring DSCP mode PFC,the
PFCENABLE field was being masked with a 4 bit mask (0xF), but this is an 8
bit bitmask for enabled classes for PFC.  This means that traffic classes
4-7 could not be enabled for PFC.

Remove the mask completely, as it is not necessary, as we are assigning 8
bits to an 8 bit field.

Fixes: 2a87bd73e50d ("ice: Add DSCP support")
Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Signed-off-by: Karen Ostrowska <karen.ostrowska@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>
20 months agocpumask: be more careful with 'cpumask_setall()'
Linus Torvalds [Tue, 7 Mar 2023 20:16:18 +0000 (12:16 -0800)]
cpumask: be more careful with 'cpumask_setall()'

Commit 596ff4a09b89 ("cpumask: re-introduce constant-sized cpumask
optimizations") changed cpumask_setall() to use "bitmap_set()" instead
of "bitmap_fill()", because bitmap_fill() would explicitly set all the
bits of a constant sized small bitmap, and that's exactly what we don't
want: we want to only set bits up to 'nr_cpu_ids', which is what
"bitmap_set()" does.

However, Yury correctly points out that while "bitmap_set()" does indeed
only set bits up to the required bitmap size, it doesn't _clear_ bits
above that size, so the upper bits would still not have well-defined
values.

Now, none of this should really matter, since any bits set past
'nr_cpu_ids' should always be ignored in the first place.  Yes, the bit
scanning functions might return them as a result, but since users should
always consider the ">= nr_cpu_ids" condition to mean "no more bits",
that shouldn't have any actual effect (see previous commit 8ca09d5fa354
"cpumask: fix incorrect cpumask scanning result checks").

But let's just do it right, the way the code was _intended_ to work.  We
have had enough lazy code that works but bites us in the *rse later
(again, see previous commit) that there's no reason to not just do this
properly.

It turns out that "bitmap_fill()" gets this all right for the complex
case, and really only fails for the inlined optimized case that just
fills the whole word.  And while we could just fix bitmap_fill() to use
the proper last word mask, there's two issues with that:

 - the cpumask case wants to do the _optimization_ based on "NR_CPUS is
   a small constant", but then wants to do the actual bit _fill_ based
   on "nr_cpu_ids" that isn't necessarily that same constant

 - we have lots of non-cpumask users of bitmap_fill(), and while they
   hopefully don't care, and probably would want the proper semantics
   anyway ("only set bits up to the limit"), I do not want the cpumask
   changes to impact other parts

So this ends up just doing the single-word optimization by hand in the
cpumask code.  If our cpumask is fundamentally limited to a single word,
just do the proper "fill in that word" exactly.  And if it's the more
complex multi-word case, then the generic bitmap_fill() will DTRT.

This is all an example of how our bitmap function optimizations really
are somewhat broken.  They conflate the "this is size of the bitmap"
optimizations with the actual bit(s) we want to set.

In many cases we really want to have the two be separate things:
sometimes we base our optimizations on the size of the whole bitmap ("I
know this whole bitmap fits in a single word, so I'll just use
single-word accesses"), and sometimes we base them on the bit we are
looking at ("this is just acting on bits that are in the first word, so
I'll use single-word accesses").

Notice how the end result of the two optimizations are the same, but the
way we get to them are quite different.

And all our cpumask optimization games are really about that fundamental
distinction, and we'd often really want to pass in both the "this is the
bit I'm working on" (which _can_ be a small constant but might be
variable), and "I know it's in this range even if it's variable" (based
on CONFIG_NR_CPUS).

So this cpumask_setall() implementation just makes that explicit.  It
checks the "I statically know the size is small" using the known static
size of the cpumask (which is what that 'small_cpumask_bits' is all
about), but then sets the actual bits using the exact number of cpus we
have (ie 'nr_cpumask_bits')

Of course, in a perfect world, the compiler would have done all the
range analysis (possibly with help from us just telling it that
"this value is always in this range"), and would do all of this for us.
But that is not the world we live in.

While we dream of that perfect world, this does that manual logic to
make it all work out.  And this was a very long explanation for a small
code change that shouldn't even matter.

Reported-by: Yury Norov <yury.norov@gmail.com>
Link: https://lore.kernel.org/lkml/ZAV9nGG9e1%2FrV+L%2F@yury-laptop/
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
20 months agonet: usb: qmi_wwan: add Telit 0x1080 composition
Enrico Sau [Mon, 6 Mar 2023 12:05:28 +0000 (13:05 +0100)]
net: usb: qmi_wwan: add Telit 0x1080 composition

Add the following Telit FE990 composition:

0x1080: tty, adb, rmnet, tty, tty, tty, tty

Signed-off-by: Enrico Sau <enrico.sau@gmail.com>
Link: https://lore.kernel.org/r/20230306120528.198842-1-enrico.sau@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agonet: usb: cdc_mbim: avoid altsetting toggling for Telit FE990
Enrico Sau [Mon, 6 Mar 2023 11:59:33 +0000 (12:59 +0100)]
net: usb: cdc_mbim: avoid altsetting toggling for Telit FE990

Add quirk CDC_MBIM_FLAG_AVOID_ALTSETTING_TOGGLE for Telit FE990
0x1081 composition in order to avoid bind error.

Signed-off-by: Enrico Sau <enrico.sau@gmail.com>
Link: https://lore.kernel.org/r/20230306115933.198259-1-enrico.sau@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agoMerge branch 'main' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf
Paolo Abeni [Tue, 7 Mar 2023 12:52:03 +0000 (13:52 +0100)]
Merge branch 'main' of git://git./linux/kernel/git/netfilter/nf

Pablo Neira Ayuso says:

====================
Netfilter fixes for net

The following patchset contains Netfilter fixes for net:

1) Restore ctnetlink zero mark in events and dump, from Ivan Delalande.

2) Fix deadlock due to missing disabled bh in tproxy, from Florian Westphal.

3) Safer maximum chain load in conntrack, from Eric Dumazet.

* 'main' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
  netfilter: conntrack: adopt safer max chain length
  netfilter: tproxy: fix deadlock due to missing BH disable
  netfilter: ctnetlink: revert to dumping mark regardless of event type
====================

Link: https://lore.kernel.org/r/20230307100424.2037-1-pablo@netfilter.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
20 months agoplatform: mellanox: mlx-platform: Initialize shift variable to 0
Hans de Goede [Tue, 7 Mar 2023 10:58:42 +0000 (11:58 +0100)]
platform: mellanox: mlx-platform: Initialize shift variable to 0

Initialize shift variable in mlxplat_mlxcpld_verify_bus_topology()
to 0 to avoid the following compile error:

drivers/platform/x86/mlx-platform.c:6013
 mlxplat_mlxcpld_verify_bus_topology() error: uninitialized symbol 'shift'.

Fixes: 50b823fdd357 ("platform: mellanox: mlx-platform: Move bus shift assignment out of the loop")
Cc: Vadim Pasternak <vadimp@nvidia.com>
Cc: Michael Shych <michaelsh@nvidia.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Link: https://lore.kernel.org/r/20230307105842.286118-1-hdegoede@redhat.com
20 months agoplatform/x86: int3472: Add GPIOs to Surface Go 3 Board data
Daniel Scally [Thu, 2 Mar 2023 10:26:11 +0000 (10:26 +0000)]
platform/x86: int3472: Add GPIOs to Surface Go 3 Board data

Add the INT347E GPIO lookup table to the board data for the Surface
Go 3. This is necessary to allow the ov7251 IR camera to probe
properly on that platform.

Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Link: https://lore.kernel.org/r/20230302102611.314341-1-dan.scally@ideasonboard.com
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
20 months agoplatform/x86: ISST: Fix kernel documentation warnings
Srinivas Pandruvada [Sat, 11 Feb 2023 06:32:46 +0000 (22:32 -0800)]
platform/x86: ISST: Fix kernel documentation warnings

Fix warning displayed for "make W=1" for kernel documentation.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Link: https://lore.kernel.org/r/20230211063257.311746-2-srinivas.pandruvada@linux.intel.com
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
20 months agoplatform: x86: MLX_PLATFORM: select REGMAP instead of depending on it
Randy Dunlap [Sun, 26 Feb 2023 05:39:51 +0000 (21:39 -0800)]
platform: x86: MLX_PLATFORM: select REGMAP instead of depending on it

REGMAP is a hidden (not user visible) symbol. Users cannot set it
directly thru "make *config", so drivers should select it instead of
depending on it if they need it.

Consistently using "select" or "depends on" can also help reduce
Kconfig circular dependency issues.

Therefore, change the use of "depends on REGMAP" to "select REGMAP".

Fixes: ef0f62264b2a ("platform/x86: mlx-platform: Add physical bus number auto detection")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Vadim Pasternak <vadimp@mellanox.com>
Cc: Darren Hart <dvhart@infradead.org>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Mark Gross <markgross@kernel.org>
Cc: platform-driver-x86@vger.kernel.org
Link: https://lore.kernel.org/r/20230226053953.4681-7-rdunlap@infradead.org
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
20 months agoplatform: mellanox: select REGMAP instead of depending on it
Randy Dunlap [Sun, 26 Feb 2023 05:39:50 +0000 (21:39 -0800)]
platform: mellanox: select REGMAP instead of depending on it

REGMAP is a hidden (not user visible) symbol. Users cannot set it
directly thru "make *config", so drivers should select it instead of
depending on it if they need it.

Consistently using "select" or "depends on" can also help reduce
Kconfig circular dependency issues.

Therefore, change the use of "depends on REGMAP" to "select REGMAP".

For NVSW_SN2201, select REGMAP_I2C instead of depending on it.

Fixes: c6acad68eb2d ("platform/mellanox: mlxreg-hotplug: Modify to use a regmap interface")
Fixes: 5ec4a8ace06c ("platform/mellanox: Introduce support for Mellanox register access driver")
Fixes: 62f9529b8d5c ("platform/mellanox: mlxreg-lc: Add initial support for Nvidia line card devices")
Fixes: 662f24826f95 ("platform/mellanox: Add support for new SN2201 system")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Darren Hart <dvhart@infradead.org>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Michael Shych <michaelsh@nvidia.com>
Cc: Mark Gross <markgross@kernel.org>
Cc: Vadim Pasternak <vadimp@nvidia.com>
Cc: platform-driver-x86@vger.kernel.org
Link: https://lore.kernel.org/r/20230226053953.4681-6-rdunlap@infradead.org
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
20 months agoplatform/x86/intel/tpmi: Fix double free reported by Smatch
Srinivas Pandruvada [Mon, 27 Feb 2023 14:06:14 +0000 (06:06 -0800)]
platform/x86/intel/tpmi: Fix double free reported by Smatch

Fix warning:
drivers/platform/x86/intel/tpmi.c:253 tpmi_create_device()
warn: 'feature_vsec_dev' was already freed.

If there is some error, feature_vsec_dev memory is freed as part
of resource managed call intel_vsec_add_aux(). So, additional
kfree() call is not required.

Reordered res allocation and feature_vsec_dev, so that on error
only res is freed.

Reported-by: Dan Carpenter <error27@gmail.com>
Link: https://lore.kernel.org/platform-driver-x86/Y%2FxYR7WGiPayZu%2FR@kili/T/#u
Fixes: 47731fd2865f ("platform/x86/intel: Intel TPMI enumeration driver")
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Link: https://lore.kernel.org/r/20230227140614.2913474-1-srinivas.pandruvada@linux.intel.com
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
20 months agoplatform/x86: ISST: Increase range of valid mail box commands
Srinivas Pandruvada [Mon, 27 Feb 2023 05:35:04 +0000 (21:35 -0800)]
platform/x86: ISST: Increase range of valid mail box commands

A new command CONFIG_TDP_GET_RATIO_INFO is added, with sub command type
of 0x0C. The previous range of valid sub commands was from 0x00 to 0x0B.
Change the valid range from 0x00 to 0x0C.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Link: https://lore.kernel.org/r/20230227053504.2734214-1-srinivas.pandruvada@linux.intel.com
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
20 months agoplatform/x86: dell-ddv: Fix temperature scaling
Armin Wolf [Sat, 18 Feb 2023 11:53:18 +0000 (12:53 +0100)]
platform/x86: dell-ddv: Fix temperature scaling

After using the built-in UEFI hardware diagnostics to compare
the measured battery temperature, i noticed that the temperature
is actually expressed in tenth degree kelvin, similar to the
SBS-Data standard. For example, a value of 2992 is displayed as
26 degrees celsius.
Fix the scaling so that the correct values are being displayed.

Tested on a Dell Inspiron 3505.

Fixes: a77272c16041 ("platform/x86: dell: Add new dell-wmi-ddv driver")
Signed-off-by: Armin Wolf <W_Armin@gmx.de>
Link: https://lore.kernel.org/r/20230218115318.20662-2-W_Armin@gmx.de
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
20 months agoplatform/x86: dell-ddv: Fix cache invalidation on resume
Armin Wolf [Sat, 18 Feb 2023 11:53:17 +0000 (12:53 +0100)]
platform/x86: dell-ddv: Fix cache invalidation on resume

If one or both sensor buffers could not be initialized, either
due to missing hardware support or due to some error during probing,
the resume handler will encounter undefined behaviour when
attempting to lock buffers then protected by an uninitialized or
destroyed mutex.
Fix this by introducing a "active" flag which is set during probe,
and only invalidate buffers which where flaged as "active".

Tested on a Dell Inspiron 3505.

Fixes: 3b7eeff93d29 ("platform/x86: dell-ddv: Add hwmon support")
Signed-off-by: Armin Wolf <W_Armin@gmx.de>
Link: https://lore.kernel.org/r/20230218115318.20662-1-W_Armin@gmx.de
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
20 months agoplatform/x86/amd: pmc: remove CONFIG_SUSPEND checks
Arnd Bergmann [Tue, 14 Feb 2023 15:25:07 +0000 (16:25 +0100)]
platform/x86/amd: pmc: remove CONFIG_SUSPEND checks

The amd_pmc_write_stb() function was previously hidden in an
ifdef to avoid a warning when CONFIG_SUSPEND is disabled, but
now there is an additional caller:

drivers/platform/x86/amd/pmc.c: In function 'amd_pmc_stb_debugfs_open_v2':
drivers/platform/x86/amd/pmc.c:256:8: error: implicit declaration of function 'amd_pmc_write_stb'; did you mean 'amd_pmc_read_stb'? [-Werror=implicit-function-declaration]
  256 |  ret = amd_pmc_write_stb(dev, AMD_PMC_STB_DUMMY_PC);
      |        ^~~~~~~~~~~~~~~~~
      |        amd_pmc_read_stb

There is now an easier way to handle this using DEFINE_SIMPLE_DEV_PM_OPS()
to replace all the #ifdefs, letting gcc drop any of the unused functions
silently.

Fixes: b0d4bb973539 ("platform/x86/amd: pmc: Write dummy postcode into the STB DRAM")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Link: https://lore.kernel.org/r/20230214152512.806188-1-arnd@kernel.org
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
20 months agonetfilter: conntrack: adopt safer max chain length
Eric Dumazet [Tue, 7 Mar 2023 05:22:54 +0000 (05:22 +0000)]
netfilter: conntrack: adopt safer max chain length

Customers using GKE 1.25 and 1.26 are facing conntrack issues
root caused to commit c9c3b6811f74 ("netfilter: conntrack: make
max chain length random").

Even if we assume Uniform Hashing, a bucket often reachs 8 chained
items while the load factor of the hash table is smaller than 0.5

With a limit of 16, we reach load factors of 3.
With a limit of 32, we reach load factors of 11.
With a limit of 40, we reach load factors of 15.
With a limit of 50, we reach load factors of 24.

This patch changes MIN_CHAINLEN to 50, to minimize risks.

Ideally, we could in the future add a cushion based on expected
load factor (2 * nf_conntrack_max / nf_conntrack_buckets),
because some setups might expect unusual values.

Fixes: c9c3b6811f74 ("netfilter: conntrack: make max chain length random")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
20 months agoMerge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
Jakub Kicinski [Tue, 7 Mar 2023 04:28:00 +0000 (20:28 -0800)]
Merge tag 'for-netdev' of https://git./linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2023-03-06

We've added 8 non-merge commits during the last 7 day(s) which contain
a total of 9 files changed, 64 insertions(+), 18 deletions(-).

The main changes are:

1) Fix BTF resolver for DATASEC sections when a VAR points at a modifier,
   that is, keep resolving such instances instead of bailing out,
   from Lorenz Bauer.

2) Fix BPF test framework with regards to xdp_frame info misplacement
   in the "live packet" code, from Alexander Lobakin.

3) Fix an infinite loop in BPF sockmap code for TCP/UDP/AF_UNIX,
   from Liu Jian.

4) Fix a build error for riscv BPF JIT under PERF_EVENTS=n,
   from Randy Dunlap.

5) Several BPF doc fixes with either broken links or external instead
   of internal doc links, from Bagas Sanjaya.

* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf:
  selftests/bpf: check that modifier resolves after pointer
  btf: fix resolving BTF_KIND_VAR after ARRAY, STRUCT, UNION, PTR
  bpf, test_run: fix &xdp_frame misplacement for LIVE_FRAMES
  bpf, doc: Link to submitting-patches.rst for general patch submission info
  bpf, doc: Do not link to docs.kernel.org for kselftest link
  bpf, sockmap: Fix an infinite loop error when len is 0 in tcp_bpf_recvmsg_parser()
  riscv, bpf: Fix patch_text implicit declaration
  bpf, docs: Fix link to BTF doc
====================

Link: https://lore.kernel.org/r/20230306215944.11981-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agonet: tls: fix device-offloaded sendpage straddling records
Jakub Kicinski [Sat, 4 Mar 2023 19:26:10 +0000 (11:26 -0800)]
net: tls: fix device-offloaded sendpage straddling records

Adrien reports that incorrect data is transmitted when a single
page straddles multiple records. We would transmit the same
data in all iterations of the loop.

Reported-by: Adrien Moulin <amoulin@corp.free.fr>
Link: https://lore.kernel.org/all/61481278.42813558.1677845235112.JavaMail.zimbra@corp.free.fr
Fixes: c1318b39c7d3 ("tls: Add opt-in zerocopy mode of sendfile()")
Tested-by: Adrien Moulin <amoulin@corp.free.fr>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Acked-by: Maxim Mikityanskiy <maxtram95@gmail.com>
Link: https://lore.kernel.org/r/20230304192610.3818098-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agonet: ethernet: mtk_eth_soc: fix RX data corruption issue
Daniel Golle [Sat, 4 Mar 2023 13:43:20 +0000 (13:43 +0000)]
net: ethernet: mtk_eth_soc: fix RX data corruption issue

Fix data corruption issue with SerDes connected PHYs operating at 1.25
Gbps speed where we could previously observe about 30% packet loss while
the bad packet counter was increasing.

As almost all boards with MediaTek MT7622 or MT7986 use either the MT7531
switch IC operating at 3.125Gbps SerDes rate or single-port PHYs using
rate-adaptation to 2500Base-X mode, this issue only got exposed now when
we started trying to use SFP modules operating with 1.25 Gbps with the
BananaPi R3 board.

The fix is to set bit 12 which disables the RX FIFO clear function when
setting up MAC MCR, MediaTek SDK did the same change stating:
"If without this patch, kernel might receive invalid packets that are
corrupted by GMAC."[1]

[1]: https://git01.mediatek.com/plugins/gitiles/openwrt/feeds/mtk-openwrt-feeds/+/d8a2975939a12686c4a95c40db21efdc3f821f63

Fixes: 42c03844e93d ("net-next: mediatek: add support for MediaTek MT7622 SoC")
Tested-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Link: https://lore.kernel.org/r/138da2735f92c8b6f8578ec2e5a794ee515b665f.1677937317.git.daniel@makrotopia.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
20 months agonet: phy: smsc: fix link up detection in forced irq mode
Heiner Kallweit [Sat, 4 Mar 2023 10:52:44 +0000 (11:52 +0100)]
net: phy: smsc: fix link up detection in forced irq mode

Currently link up can't be detected in forced mode if polling
isn't used. Only link up interrupt source we have is aneg
complete which isn't applicable in forced mode. Therefore we
have to use energy-on as link up indicator.

Fixes: 7365494550f6 ("net: phy: smsc: skip ENERGYON interrupt if disabled")
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 months agocpumask: fix incorrect cpumask scanning result checks
Linus Torvalds [Mon, 6 Mar 2023 20:15:13 +0000 (12:15 -0800)]
cpumask: fix incorrect cpumask scanning result checks

It turns out that commit 596ff4a09b89 ("cpumask: re-introduce
constant-sized cpumask optimizations") exposed a number of cases of
drivers not checking the result of "cpumask_next()" and friends
correctly.

The documented correct check for "no more cpus in the cpumask" is to
check for the result being equal or larger than the number of possible
CPU ids, exactly _because_ we've always done those constant-sized
cpumask scans using a widened type before.  So the return value of a
cpumask scan should be checked with

if (cpu >= nr_cpu_ids)
...

because the cpumask scan did not necessarily stop exactly *at* that
maximum CPU id.

But a few cases ended up instead using checks like

if (cpu == nr_cpumask_bits)
...

which used that internal "widened" number of bits.  And that used to
work pretty much by accident (ok, in this case "by accident" is simply
because it matched the historical internal implementation of the cpumask
scanning, so it was more of a "intentionally using implementation
details rather than an accident").

But the extended constant-sized optimizations then did that internal
implementation differently, and now that code that did things wrong but
matched the old implementation no longer worked at all.

Which then causes subsequent odd problems due to using what ends up
being an invalid CPU ID.

Most of these cases require either unusual hardware or special uses to
hit, but the random.c one triggers quite easily.

All you really need is to have a sufficiently small CONFIG_NR_CPUS value
for the bit scanning optimization to be triggered, but not enough CPUs
to then actually fill that widened cpumask.  At that point, the cpumask
scanning will return the NR_CPUS constant, which is _not_ the same as
nr_cpumask_bits.

This just does the mindless fix with

   sed -i 's/== nr_cpumask_bits/>= nr_cpu_ids/'

to fix the incorrect uses.

The ones in the SCSI lpfc driver in particular could probably be fixed
more cleanly by just removing that repeated pattern entirely, but I am
not emptionally invested enough in that driver to care.

Reported-and-tested-by: Guenter Roeck <linux@roeck-us.net>
Link: https://lore.kernel.org/lkml/481b19b5-83a0-4793-b4fd-194ad7b978c3@roeck-us.net/
Reported-and-tested-by: Geert Uytterhoeven <geert+renesas@glider.be>
Link: https://lore.kernel.org/lkml/CAMuHMdUKo_Sf7TjKzcNDa8Ve+6QrK+P8nSQrSQ=6LTRmcBKNww@mail.gmail.com/
Reported-by: Vernon Yang <vernon2gm@gmail.com>
Link: https://lore.kernel.org/lkml/20230306160651.2016767-1-vernon2gm@gmail.com/
Cc: Yury Norov <yury.norov@gmail.com>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
20 months agoMerge branch 'fix resolving VAR after DATASEC'
Martin KaFai Lau [Mon, 6 Mar 2023 19:44:14 +0000 (11:44 -0800)]
Merge branch 'fix resolving VAR after DATASEC'

Lorenz Bauer says:

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

See the first patch for a detailed explanation.

v2:
- Move RESOLVE_TBD assignment out of the loop (Martin)
====================

Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
20 months agoselftests/bpf: check that modifier resolves after pointer
Lorenz Bauer [Mon, 6 Mar 2023 11:21:38 +0000 (11:21 +0000)]
selftests/bpf: check that modifier resolves after pointer

Add a regression test that ensures that a VAR pointing at a
modifier which follows a PTR (or STRUCT or ARRAY) is resolved
correctly by the datasec validator.

Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/r/20230306112138.155352-3-lmb@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
20 months agobtf: fix resolving BTF_KIND_VAR after ARRAY, STRUCT, UNION, PTR
Lorenz Bauer [Mon, 6 Mar 2023 11:21:37 +0000 (11:21 +0000)]
btf: fix resolving BTF_KIND_VAR after ARRAY, STRUCT, UNION, PTR

btf_datasec_resolve contains a bug that causes the following BTF
to fail loading:

    [1] DATASEC a size=2 vlen=2
        type_id=4 offset=0 size=1
        type_id=7 offset=1 size=1
    [2] INT (anon) size=1 bits_offset=0 nr_bits=8 encoding=(none)
    [3] PTR (anon) type_id=2
    [4] VAR a type_id=3 linkage=0
    [5] INT (anon) size=1 bits_offset=0 nr_bits=8 encoding=(none)
    [6] TYPEDEF td type_id=5
    [7] VAR b type_id=6 linkage=0

This error message is printed during btf_check_all_types:

    [1] DATASEC a size=2 vlen=2
        type_id=7 offset=1 size=1 Invalid type

By tracing btf_*_resolve we can pinpoint the problem:

    btf_datasec_resolve(depth: 1, type_id: 1, mode: RESOLVE_TBD) = 0
        btf_var_resolve(depth: 2, type_id: 4, mode: RESOLVE_TBD) = 0
            btf_ptr_resolve(depth: 3, type_id: 3, mode: RESOLVE_PTR) = 0
        btf_var_resolve(depth: 2, type_id: 4, mode: RESOLVE_PTR) = 0
    btf_datasec_resolve(depth: 1, type_id: 1, mode: RESOLVE_PTR) = -22

The last invocation of btf_datasec_resolve should invoke btf_var_resolve
by means of env_stack_push, instead it returns EINVAL. The reason is that
env_stack_push is never executed for the second VAR.

    if (!env_type_is_resolve_sink(env, var_type) &&
        !env_type_is_resolved(env, var_type_id)) {
        env_stack_set_next_member(env, i + 1);
        return env_stack_push(env, var_type, var_type_id);
    }

env_type_is_resolve_sink() changes its behaviour based on resolve_mode.
For RESOLVE_PTR, we can simplify the if condition to the following:

    (btf_type_is_modifier() || btf_type_is_ptr) && !env_type_is_resolved()

Since we're dealing with a VAR the clause evaluates to false. This is
not sufficient to trigger the bug however. The log output and EINVAL
are only generated if btf_type_id_size() fails.

    if (!btf_type_id_size(btf, &type_id, &type_size)) {
        btf_verifier_log_vsi(env, v->t, vsi, "Invalid type");
        return -EINVAL;
    }

Most types are sized, so for example a VAR referring to an INT is not a
problem. The bug is only triggered if a VAR points at a modifier. Since
we skipped btf_var_resolve that modifier was also never resolved, which
means that btf_resolved_type_id returns 0 aka VOID for the modifier.
This in turn causes btf_type_id_size to return NULL, triggering EINVAL.

To summarise, the following conditions are necessary:

- VAR pointing at PTR, STRUCT, UNION or ARRAY
- Followed by a VAR pointing at TYPEDEF, VOLATILE, CONST, RESTRICT or
  TYPE_TAG

The fix is to reset resolve_mode to RESOLVE_TBD before attempting to
resolve a VAR from a DATASEC.

Fixes: 1dc92851849c ("bpf: kernel side support for BTF Var and DataSec")
Signed-off-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/r/20230306112138.155352-2-lmb@isovalent.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
20 months agobpf, test_run: fix &xdp_frame misplacement for LIVE_FRAMES
Alexander Lobakin [Fri, 24 Feb 2023 16:36:07 +0000 (17:36 +0100)]
bpf, test_run: fix &xdp_frame misplacement for LIVE_FRAMES

&xdp_buff and &xdp_frame are bound in a way that

xdp_buff->data_hard_start == xdp_frame

It's always the case and e.g. xdp_convert_buff_to_frame() relies on
this.
IOW, the following:

for (u32 i = 0; i < 0xdead; i++) {
xdpf = xdp_convert_buff_to_frame(&xdp);
xdp_convert_frame_to_buff(xdpf, &xdp);
}

shouldn't ever modify @xdpf's contents or the pointer itself.
However, "live packet" code wrongly treats &xdp_frame as part of its
context placed *before* the data_hard_start. With such flow,
data_hard_start is sizeof(*xdpf) off to the right and no longer points
to the XDP frame.

Instead of replacing `sizeof(ctx)` with `offsetof(ctx, xdpf)` in several
places and praying that there are no more miscalcs left somewhere in the
code, unionize ::frm with ::data in a flex array, so that both starts
pointing to the actual data_hard_start and the XDP frame actually starts
being a part of it, i.e. a part of the headroom, not the context.
A nice side effect is that the maximum frame size for this mode gets
increased by 40 bytes, as xdp_buff::frame_sz includes everything from
data_hard_start (-> includes xdpf already) to the end of XDP/skb shared
info.
Also update %MAX_PKT_SIZE accordingly in the selftests code. Leave it
hardcoded for 64 bit && 4k pages, it can be made more flexible later on.

Minor: align `&head->data` with how `head->frm` is assigned for
consistency.
Minor #2: rename 'frm' to 'frame' in &xdp_page_head while at it for
clarity.

(was found while testing XDP traffic generator on ice, which calls
 xdp_convert_frame_to_buff() for each XDP frame)

Fixes: b530e9e1063e ("bpf: Add "live packet" mode for XDP in BPF_PROG_RUN")
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Link: https://lore.kernel.org/r/20230224163607.2994755-1-aleksander.lobakin@intel.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
20 months agocpumask: Fix typo nr_cpumask_size --> nr_cpumask_bits
Andy Shevchenko [Mon, 6 Mar 2023 15:22:04 +0000 (17:22 +0200)]
cpumask: Fix typo nr_cpumask_size --> nr_cpumask_bits

The never used nr_cpumask_size is just a typo, hence use existing
redefinition that's called nr_cpumask_bits.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
20 months agobpf, doc: Link to submitting-patches.rst for general patch submission info
Bagas Sanjaya [Tue, 28 Feb 2023 07:45:23 +0000 (14:45 +0700)]
bpf, doc: Link to submitting-patches.rst for general patch submission info

The link for patch submission information in general refers to index
page for "Working with the kernel development community" section of
kernel docs, whereas the link should have been
Documentation/process/submitting-patches.rst instead.

Fix it by replacing the index target with the appropriate doc.

Fixes: 542228384888f5 ("bpf, doc: convert bpf_devel_QA.rst to use RST formatting")
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230228074523.11493-3-bagasdotme@gmail.com
20 months agobpf, doc: Do not link to docs.kernel.org for kselftest link
Bagas Sanjaya [Tue, 28 Feb 2023 07:45:22 +0000 (14:45 +0700)]
bpf, doc: Do not link to docs. for kselftest link

The question on how to run BPF selftests have a reference link to kernel
selftest documentation (Documentation/dev-tools/kselftest.rst). However,
it uses external link to the documentation at kernel.org/docs (aka
docs.kernel.org) instead, which requires Internet access.

Fix this and replace the link with internal linking, by using :doc: directive
while keeping the anchor text.

Fixes: b7a27c3aafa252 ("bpf, doc: howto use/run the BPF selftests")
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230228074523.11493-2-bagasdotme@gmail.com
20 months agoudf: Warn if block mapping is done for in-ICB files
Jan Kara [Tue, 28 Feb 2023 11:11:38 +0000 (12:11 +0100)]
udf: Warn if block mapping is done for in-ICB files

Now that address space operations are merge dfor in-ICB and normal
files, it is more likely some code mistakenly tries to map blocks for
in-ICB files. WARN and return error instead of silently returning
garbage.

Signed-off-by: Jan Kara <jack@suse.cz>
20 months agoudf: Fix reading of in-ICB files
Jan Kara [Tue, 28 Feb 2023 11:00:25 +0000 (12:00 +0100)]
udf: Fix reading of in-ICB files

After merging address space operations of normal and in-ICB files,
readahead could get called for in-ICB files which resulted in
udf_get_block() being called for these files. udf_get_block() is not
prepared to be called for in-ICB files and ends up returning garbage
results as it interprets file data as extent list. Fix the problem by
skipping readahead for in-ICB files.

Fixes: 37a8a39f7ad3 ("udf: Switch to single address_space_operations")
Signed-off-by: Jan Kara <jack@suse.cz>
20 months agoudf: Fix lost writes in udf_adinicb_writepage()
Jan Kara [Mon, 27 Feb 2023 18:32:19 +0000 (19:32 +0100)]
udf: Fix lost writes in udf_adinicb_writepage()

The patch converting udf_adinicb_writepage() to avoid manually kmapping
the page used memcpy_to_page() however that copies in the wrong
direction (effectively overwriting file data with the old contents).
What we should be using is memcpy_from_page() to copy data from the page
into the inode and then mark inode dirty to store the data.

Fixes: 5cfc45321a6d ("udf: Convert udf_adinicb_writepage() to memcpy_to_page()")
Signed-off-by: Jan Kara <jack@suse.cz>
20 months agom68k: Only force 030 bus error if PC not in exception table
Michael Schmitz [Wed, 1 Mar 2023 02:11:07 +0000 (15:11 +1300)]
m68k: Only force 030 bus error if PC not in exception table

__get_kernel_nofault() does copy data in supervisor mode when
forcing a task backtrace log through /proc/sysrq_trigger.
This is expected cause a bus error exception on e.g. NULL
pointer dereferencing when logging a kernel task has no
workqueue associated. This bus error ought to be ignored.

Our 030 bus error handler is ill equipped to deal with this:

Whenever ssw indicates a kernel mode access on a data fault,
we don't even attempt to handle the fault and instead always
send a SEGV signal (or panic). As a result, the check
for exception handling at the fault PC (buried in
send_sig_fault() which gets called from do_page_fault()
eventually) is never used.

In contrast, both 040 and 060 access error handlers do not
care whether a fault happened on supervisor mode access,
and will call do_page_fault() on those, ultimately honoring
the exception table.

Add a check in bus_error030 to call do_page_fault() in case
we do have an entry for the fault PC in our exception table.

I had attempted a fix for this earlier in 2019 that did rely
on testing pagefault_disabled() (see link below) to achieve
the same thing, but this patch should be more generic.

Tested on 030 Atari Falcon.

Reported-by: Eero Tamminen <oak@helsinkinet.fi>
Link: https://lore.kernel.org/r/alpine.LNX.2.21.1904091023540.25@nippy.intranet
Link: https://lore.kernel.org/r/63130691-1984-c423-c1f2-73bfd8d3dcd3@gmail.com
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Link: https://lore.kernel.org/r/20230301021107.26307-1-schmitzmic@gmail.com
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
20 months agom68k: mm: Move initrd phys_to_virt handling after paging_init()
Geert Uytterhoeven [Mon, 27 Feb 2023 20:14:13 +0000 (21:14 +0100)]
m68k: mm: Move initrd phys_to_virt handling after paging_init()

When booting with an initial ramdisk on platforms where physical memory
does not start at address zero (e.g. on Amiga):

    initrd: 0ef0602c - 0f800000
    Zone ranges:
      DMA      [mem 0x0000000008000000-0x000000f7ffffffff]
      Normal   empty
    Movable zone start for each node
    Early memory node ranges
      node   0: [mem 0x0000000008000000-0x000000000f7fffff]
    Initmem setup node 0 [mem 0x0000000008000000-0x000000000f7fffff]
    Unable to handle kernel access at virtual address (ptrval)
    Oops: 00000000
    Modules linked in:
    PC: [<00201d3c>] memcmp+0x28/0x56

As phys_to_virt() relies on m68k_memoffset and module_fixup(), it must
not be called before paging_init().  Hence postpone the phys_to_virt
handling for the initial ramdisk until after calling paging_init().

While at it, reduce #ifdef clutter by using IS_ENABLED() instead.

Fixes: 376e3fdecb0dcae2 ("m68k: Enable memtest functionality")
Reported-by: Stephen Walsh <vk3heg@vk3heg.net>
Link: https://lists.debian.org/debian-68k/2022/09/msg00007.html
Reported-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Link: https://lore.kernel.org/r/4f45f05f377bf3f5baf88dbd5c3c8aeac59d94f0.camel@physik.fu-berlin.de
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Finn Thain <fthain@linux-m68k.org>
Link: https://lore.kernel.org/r/dff216da09ab7a60217c3fc2147e671ae07d636f.1677528627.git.geert@linux-m68k.org
20 months agom68k: mm: Fix systems with memory at end of 32-bit address space
Kars de Jong [Thu, 23 Feb 2023 11:23:49 +0000 (12:23 +0100)]
m68k: mm: Fix systems with memory at end of 32-bit address space

The calculation of end addresses of memory chunks overflowed to 0 when
a memory chunk is located at the end of 32-bit address space.
This is the case for the HP300 architecture.

Link: https://lore.kernel.org/linux-m68k/CACz-3rhUo5pgNwdWHaPWmz+30Qo9xCg70wNxdf7o5x-6tXq8QQ@mail.gmail.com/
Signed-off-by: Kars de Jong <jongk@linux-m68k.org>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Link: https://lore.kernel.org/r/20230223112349.26675-1-jongk@linux-m68k.org
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
20 months agonetfilter: tproxy: fix deadlock due to missing BH disable
Florian Westphal [Fri, 3 Mar 2023 09:58:56 +0000 (10:58 +0100)]
netfilter: tproxy: fix deadlock due to missing BH disable

The xtables packet traverser performs an unconditional local_bh_disable(),
but the nf_tables evaluation loop does not.

Functions that are called from either xtables or nftables must assume
that they can be called in process context.

inet_twsk_deschedule_put() assumes that no softirq interrupt can occur.
If tproxy is used from nf_tables its possible that we'll deadlock
trying to aquire a lock already held in process context.

Add a small helper that takes care of this and use it.

Link: https://lore.kernel.org/netfilter-devel/401bd6ed-314a-a196-1cdc-e13c720cc8f2@balasys.hu/
Fixes: 4ed8eb6570a4 ("netfilter: nf_tables: Add native tproxy support")
Reported-and-tested-by: Major Dávid <major.david@balasys.hu>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
20 months agonetfilter: ctnetlink: revert to dumping mark regardless of event type
Ivan Delalande [Fri, 3 Mar 2023 01:48:31 +0000 (17:48 -0800)]
netfilter: ctnetlink: revert to dumping mark regardless of event type

It seems that change was unintentional, we have userspace code that
needs the mark while listening for events like REPLY, DESTROY, etc.
Also include 0-marks in requested dumps, as they were before that fix.

Fixes: 1feeae071507 ("netfilter: ctnetlink: fix compilation warning after data race fixes in ct mark")
Signed-off-by: Ivan Delalande <colona@arista.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
20 months agobnxt_en: Fix the double free during device removal
Selvin Xavier [Sat, 4 Mar 2023 02:43:58 +0000 (18:43 -0800)]
bnxt_en: Fix the double free during device removal

Following warning reported by KASAN during driver unload

==================================================================
BUG: KASAN: double-free in bnxt_remove_one+0x103/0x200 [bnxt_en]
Free of addr ffff88814e8dd4c0 by task rmmod/17469
CPU: 47 PID: 17469 Comm: rmmod Kdump: loaded Tainted: G S                 6.2.0-rc7+ #2
Hardware name: Dell Inc. PowerEdge R740/01YM03, BIOS 2.3.10 08/15/2019
Call Trace:
 <TASK>
 dump_stack_lvl+0x33/0x46
 print_report+0x17b/0x4b3
 ? __call_rcu_common.constprop.79+0x27e/0x8c0
 ? __pfx_free_object_rcu+0x10/0x10
 ? __virt_addr_valid+0xe3/0x160
 ? bnxt_remove_one+0x103/0x200 [bnxt_en]
 kasan_report_invalid_free+0x64/0xd0
 ? bnxt_remove_one+0x103/0x200 [bnxt_en]
 ? bnxt_remove_one+0x103/0x200 [bnxt_en]
 __kasan_slab_free+0x179/0x1c0
 ? bnxt_remove_one+0x103/0x200 [bnxt_en]
 __kmem_cache_free+0x194/0x350
 bnxt_remove_one+0x103/0x200 [bnxt_en]
 pci_device_remove+0x62/0x110
 device_release_driver_internal+0xf6/0x1c0
 driver_detach+0x76/0xe0
 bus_remove_driver+0x89/0x160
 pci_unregister_driver+0x26/0x110
 ? strncpy_from_user+0x188/0x1c0
 bnxt_exit+0xc/0x24 [bnxt_en]
 __x64_sys_delete_module+0x21f/0x390
 ? __pfx___x64_sys_delete_module+0x10/0x10
 ? __pfx_mem_cgroup_handle_over_high+0x10/0x10
 ? _raw_spin_lock+0x87/0xe0
 ? __pfx__raw_spin_lock+0x10/0x10
 ? __audit_syscall_entry+0x185/0x210
 ? ktime_get_coarse_real_ts64+0x51/0x80
 ? syscall_trace_enter.isra.18+0x126/0x1a0
 do_syscall_64+0x37/0x90
 entry_SYSCALL_64_after_hwframe+0x72/0xdc
RIP: 0033:0x7effcb6fd71b
Code: 73 01 c3 48 8b 0d 6d 17 2c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa b8 b0 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 3d 17 2c 00 f7 d8 64 89 01 48
RSP: 002b:00007ffeada270b8 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0
RAX: ffffffffffffffda RBX: 00005623660e0750 RCX: 00007effcb6fd71b
RDX: 000000000000000a RSI: 0000000000000800 RDI: 00005623660e07b8
RBP: 0000000000000000 R08: 00007ffeada26031 R09: 0000000000000000
R10: 00007effcb771280 R11: 0000000000000206 R12: 00007ffeada272e0
R13: 00007ffeada28bc4 R14: 00005623660e02a0 R15: 00005623660e0750
 </TASK>

Auxiliary device structures are freed in bnxt_aux_dev_release. So avoid
calling kfree from bnxt_remove_one.

Also, set bp->edev to NULL before freeing the auxilary private structure.

Fixes: d80d88b0dfff ("bnxt_en: Add auxiliary driver support")
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 months agobnxt_en: Avoid order-5 memory allocation for TPA data
Michael Chan [Sat, 4 Mar 2023 02:43:57 +0000 (18:43 -0800)]
bnxt_en: Avoid order-5 memory allocation for TPA data

The driver needs to keep track of all the possible concurrent TPA (GRO/LRO)
completions on the aggregation ring.  On P5 chips, the maximum number
of concurrent TPA is 256 and the amount of memory we allocate is order-5
on systems using 4K pages.  Memory allocation failure has been reported:

NetworkManager: page allocation failure: order:5, mode:0x40dc0(GFP_KERNEL|__GFP_COMP|__GFP_ZERO), nodemask=(null),cpuset=/,mems_allowed=0-1
CPU: 15 PID: 2995 Comm: NetworkManager Kdump: loaded Not tainted 5.10.156 #1
Hardware name: Dell Inc. PowerEdge R660/0M1CC5, BIOS 0.2.25 08/12/2022
Call Trace:
 dump_stack+0x57/0x6e
 warn_alloc.cold.120+0x7b/0xdd
 ? _cond_resched+0x15/0x30
 ? __alloc_pages_direct_compact+0x15f/0x170
 __alloc_pages_slowpath.constprop.108+0xc58/0xc70
 __alloc_pages_nodemask+0x2d0/0x300
 kmalloc_order+0x24/0xe0
 kmalloc_order_trace+0x19/0x80
 bnxt_alloc_mem+0x1150/0x15c0 [bnxt_en]
 ? bnxt_get_func_stat_ctxs+0x13/0x60 [bnxt_en]
 __bnxt_open_nic+0x12e/0x780 [bnxt_en]
 bnxt_open+0x10b/0x240 [bnxt_en]
 __dev_open+0xe9/0x180
 __dev_change_flags+0x1af/0x220
 dev_change_flags+0x21/0x60
 do_setlink+0x35c/0x1100

Instead of allocating this big chunk of memory and dividing it up for the
concurrent TPA instances, allocate each small chunk separately for each
TPA instance.  This will reduce it to order-0 allocations.

Fixes: 79632e9ba386 ("bnxt_en: Expand bnxt_tpa_info struct to support 57500 chips.")
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Damodharam Ammepalli <damodharam.ammepalli@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 months agonet: phylib: get rid of unnecessary locking
Russell King (Oracle) [Fri, 3 Mar 2023 16:37:54 +0000 (16:37 +0000)]
net: phylib: get rid of unnecessary locking

The locking in phy_probe() and phy_remove() does very little to prevent
any races with e.g. phy_attach_direct(), but instead causes lockdep ABBA
warnings. Remove it.

======================================================
WARNING: possible circular locking dependency detected
6.2.0-dirty #1108 Tainted: G        W   E
------------------------------------------------------
ip/415 is trying to acquire lock:
ffff5c268f81ef50 (&dev->lock){+.+.}-{3:3}, at: phy_attach_direct+0x17c/0x3a0 [libphy]

but task is already holding lock:
ffffaef6496cb518 (rtnl_mutex){+.+.}-{3:3}, at: rtnetlink_rcv_msg+0x154/0x560

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (rtnl_mutex){+.+.}-{3:3}:
       __lock_acquire+0x35c/0x6c0
       lock_acquire.part.0+0xcc/0x220
       lock_acquire+0x68/0x84
       __mutex_lock+0x8c/0x414
       mutex_lock_nested+0x34/0x40
       rtnl_lock+0x24/0x30
       sfp_bus_add_upstream+0x34/0x150
       phy_sfp_probe+0x4c/0x94 [libphy]
       mv3310_probe+0x148/0x184 [marvell10g]
       phy_probe+0x8c/0x200 [libphy]
       call_driver_probe+0xbc/0x15c
       really_probe+0xc0/0x320
       __driver_probe_device+0x84/0x120
       driver_probe_device+0x44/0x120
       __device_attach_driver+0xc4/0x160
       bus_for_each_drv+0x80/0xe0
       __device_attach+0xb0/0x1f0
       device_initial_probe+0x1c/0x2c
       bus_probe_device+0xa4/0xb0
       device_add+0x360/0x53c
       phy_device_register+0x60/0xa4 [libphy]
       fwnode_mdiobus_phy_device_register+0xc0/0x190 [fwnode_mdio]
       fwnode_mdiobus_register_phy+0x160/0xd80 [fwnode_mdio]
       of_mdiobus_register+0x140/0x340 [of_mdio]
       orion_mdio_probe+0x298/0x3c0 [mvmdio]
       platform_probe+0x70/0xe0
       call_driver_probe+0x34/0x15c
       really_probe+0xc0/0x320
       __driver_probe_device+0x84/0x120
       driver_probe_device+0x44/0x120
       __driver_attach+0x104/0x210
       bus_for_each_dev+0x78/0xdc
       driver_attach+0x2c/0x3c
       bus_add_driver+0x184/0x240
       driver_register+0x80/0x13c
       __platform_driver_register+0x30/0x3c
       xt_compat_calc_jump+0x28/0xa4 [x_tables]
       do_one_initcall+0x50/0x1b0
       do_init_module+0x50/0x1fc
       load_module+0x684/0x744
       __do_sys_finit_module+0xc4/0x140
       __arm64_sys_finit_module+0x28/0x34
       invoke_syscall+0x50/0x120
       el0_svc_common.constprop.0+0x6c/0x1b0
       do_el0_svc+0x34/0x44
       el0_svc+0x48/0xf0
       el0t_64_sync_handler+0xb8/0xc0
       el0t_64_sync+0x1a0/0x1a4

-> #0 (&dev->lock){+.+.}-{3:3}:
       check_prev_add+0xb4/0xc80
       validate_chain+0x414/0x47c
       __lock_acquire+0x35c/0x6c0
       lock_acquire.part.0+0xcc/0x220
       lock_acquire+0x68/0x84
       __mutex_lock+0x8c/0x414
       mutex_lock_nested+0x34/0x40
       phy_attach_direct+0x17c/0x3a0 [libphy]
       phylink_fwnode_phy_connect.part.0+0x70/0xe4 [phylink]
       phylink_fwnode_phy_connect+0x48/0x60 [phylink]
       mvpp2_open+0xec/0x2e0 [mvpp2]
       __dev_open+0x104/0x214
       __dev_change_flags+0x1d4/0x254
       dev_change_flags+0x2c/0x7c
       do_setlink+0x254/0xa50
       __rtnl_newlink+0x430/0x514
       rtnl_newlink+0x58/0x8c
       rtnetlink_rcv_msg+0x17c/0x560
       netlink_rcv_skb+0x64/0x150
       rtnetlink_rcv+0x20/0x30
       netlink_unicast+0x1d4/0x2b4
       netlink_sendmsg+0x1a4/0x400
       ____sys_sendmsg+0x228/0x290
       ___sys_sendmsg+0x88/0xec
       __sys_sendmsg+0x70/0xd0
       __arm64_sys_sendmsg+0x2c/0x40
       invoke_syscall+0x50/0x120
       el0_svc_common.constprop.0+0x6c/0x1b0
       do_el0_svc+0x34/0x44
       el0_svc+0x48/0xf0
       el0t_64_sync_handler+0xb8/0xc0
       el0t_64_sync+0x1a0/0x1a4

other info that might help us debug this:

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(rtnl_mutex);
                               lock(&dev->lock);
                               lock(rtnl_mutex);
  lock(&dev->lock);

 *** DEADLOCK ***

Fixes: 298e54fa810e ("net: phy: add core phylib sfp support")
Reported-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 months agonet: stmmac: add to set device wake up flag when stmmac init phy
Rongguang Wei [Thu, 2 Mar 2023 06:21:43 +0000 (14:21 +0800)]
net: stmmac: add to set device wake up flag when stmmac init phy

When MAC is not support PMT, driver will check PHY's WoL capability
and set device wakeup capability in stmmac_init_phy(). We can enable
the WoL through ethtool, the driver would enable the device wake up
flag. Now the device_may_wakeup() return true.

But if there is a way which enable the PHY's WoL capability derectly,
like in BIOS. The driver would not know the enable thing and would not
set the device wake up flag. The phy_suspend may failed like this:

[   32.409063] PM: dpm_run_callback(): mdio_bus_phy_suspend+0x0/0x50 returns -16
[   32.409065] PM: Device stmmac-1:00 failed to suspend: error -16
[   32.409067] PM: Some devices failed to suspend, or early wake event detected

Add to set the device wakeup enable flag according to the get_wol
function result in PHY can fix the error in this scene.

v2: add a Fixes tag.

Fixes: 1d8e5b0f3f2c ("net: stmmac: Support WOL with phy")
Signed-off-by: Rongguang Wei <weirongguang@kylinos.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
20 months agoLinux 6.3-rc1
Linus Torvalds [Sun, 5 Mar 2023 22:52:03 +0000 (14:52 -0800)]
Linux 6.3-rc1

20 months agocpumask: re-introduce constant-sized cpumask optimizations
Linus Torvalds [Sat, 4 Mar 2023 21:35:43 +0000 (13:35 -0800)]
cpumask: re-introduce constant-sized cpumask optimizations

Commit aa47a7c215e7 ("lib/cpumask: deprecate nr_cpumask_bits") resulted
in the cpumask operations potentially becoming hugely less efficient,
because suddenly the cpumask was always considered to be variable-sized.

The optimization was then later added back in a limited form by commit
6f9c07be9d02 ("lib/cpumask: add FORCE_NR_CPUS config option"), but that
FORCE_NR_CPUS option is not useful in a generic kernel and more of a
special case for embedded situations with fixed hardware.

Instead, just re-introduce the optimization, with some changes.

Instead of depending on CPUMASK_OFFSTACK being false, and then always
using the full constant cpumask width, this introduces three different
cpumask "sizes":

 - the exact size (nr_cpumask_bits) remains identical to nr_cpu_ids.

   This is used for situations where we should use the exact size.

 - the "small" size (small_cpumask_bits) is the NR_CPUS constant if it
   fits in a single word and the bitmap operations thus end up able
   to trigger the "small_const_nbits()" optimizations.

   This is used for the operations that have optimized single-word
   cases that get inlined, notably the bit find and scanning functions.

 - the "large" size (large_cpumask_bits) is the NR_CPUS constant if it
   is an sufficiently small constant that makes simple "copy" and
   "clear" operations more efficient.

   This is arbitrarily set at four words or less.

As a an example of this situation, without this fixed size optimization,
cpumask_clear() will generate code like

        movl    nr_cpu_ids(%rip), %edx
        addq    $63, %rdx
        shrq    $3, %rdx
        andl    $-8, %edx
        callq   memset@PLT

on x86-64, because it would calculate the "exact" number of longwords
that need to be cleared.

In contrast, with this patch, using a MAX_CPU of 64 (which is quite a
reasonable value to use), the above becomes a single

movq $0,cpumask

instruction instead, because instead of caring to figure out exactly how
many CPU's the system has, it just knows that the cpumask will be a
single word and can just clear it all.

Note that this does end up tightening the rules a bit from the original
version in another way: operations that set bits in the cpumask are now
limited to the actual nr_cpu_ids limit, whereas we used to do the
nr_cpumask_bits thing almost everywhere in the cpumask code.

But if you just clear bits, or scan for bits, we can use the simpler
compile-time constants.

In the process, remove 'cpumask_complement()' and 'for_each_cpu_not()'
which were not useful, and which fundamentally have to be limited to
'nr_cpu_ids'.  Better remove them now than have somebody introduce use
of them later.

Of course, on x86-64 with MAXSMP there is no sane small compile-time
constant for the cpumask sizes, and we end up using the actual CPU bits,
and will generate the above kind of horrors regardless.  Please don't
use MAXSMP unless you really expect to have machines with thousands of
cores.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
20 months agoMerge tag 'v6.3-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Sun, 5 Mar 2023 19:32:30 +0000 (11:32 -0800)]
Merge tag 'v6.3-p2' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto fix from Herbert Xu:
 "Fix a regression in the caam driver"

* tag 'v6.3-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: caam - Fix edesc/iv ordering mixup