linux-2.6-microblaze.git
2 years agomlxsw: spectrum: Add hash table for IPv6 address mapping
Amit Cohen [Tue, 14 Dec 2021 14:25:44 +0000 (16:25 +0200)]
mlxsw: spectrum: Add hash table for IPv6 address mapping

The device supports forwarding entries such as routes and FDBs that
perform tunnel (e.g., VXLAN, IP-in-IP) encapsulation or decapsulation.
When the underlay is IPv6, these entries do not encode the 128 bit IPv6
address used for encapsulation / decapsulation. Instead, these entries
encode a 24 bit pointer to an array called KVDL where the IPv6 address
is stored.

Currently, only IP-in-IP with IPv6 underlay is supported, but subsequent
patches will add support for VxLAN with IPv6 underlay. To avoid
duplicating the logic required to store and retrieve these IPv6
addresses, introduce a hash table that will store the mapping between
IPv6 addresses and their KVDL index.

Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoMerge tag 'mlx5-updates-2021-12-14' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Wed, 15 Dec 2021 14:46:33 +0000 (14:46 +0000)]
Merge tag 'mlx5-updates-2021-12-14' of git://git./linux/kernel/git/saeed/linux

Saed Mahameed says:

====================
mlx5-updates-2021-12-14

Parsing Infrastructure for TC actions:

The series introduce a TC action infrastructure to help
parsing TC actions in a generic way for both FDB and NIC rules.

To help maintain the parsing code of TC actions, we the parsing code to
action parser per action TC type in separate files, instead of having one
big switch case loop, duplicated between FDB and NIC parsers as before this
patchset.

Each TC flow_action->id is represented by a dedicated mlx5e_tc_act handler
which has callbacks to check if the specific action is offload supported and
to parse the specific action.

We move each case (TC action) handling into the specific handler, which is
responsible for parsing and determining if the action is supported.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next...
David S. Miller [Wed, 15 Dec 2021 10:59:57 +0000 (10:59 +0000)]
Merge branch '100GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
100GbE Intel Wired LAN Driver Updates 2021-12-14

This series contains updates to ice driver only.

Haiyue adds support to query hardware for supported PTYPEs.

Jeff changes PTYPE validation to utilize the capabilities queried from
the hardware instead of maintaining a per DDP support list.

Brett refactors promiscuous functions to provide common and clear
interfaces to call for configuration.

Wojciech modifies DDP package load to simplify determining the final
state of the load.

Tony removes the use of ice_status from the driver. This involves
removing string conversion functions, converting variables and values to
standard errors, and clean up. He also removes an unused define.

Dan Carpenter removes unneeded casts.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: fec: fix system hang during suspend/resume
Joakim Zhang [Tue, 14 Dec 2021 10:50:46 +0000 (18:50 +0800)]
net: fec: fix system hang during suspend/resume

1. During normal suspend (WoL not enabled) process, system has posibility
to hang. The root cause is TXF interrupt coming after clocks disabled,
system hang when accessing registers from interrupt handler. To fix this
issue, disable all interrupts when system suspend.

2. System also has posibility to hang with WoL enabled during suspend,
after entering stop mode, then magic pattern coming after clocks
disabled, system will be waked up, and interrupt handler will be called,
system hang when access registers. To fix this issue, disable wakeup
irq in .suspend(), and enable it in .resume().

Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: ocelot: add support to get port mac from device-tree
Clément Léger [Tue, 14 Dec 2021 09:55:34 +0000 (10:55 +0100)]
net: ocelot: add support to get port mac from device-tree

Add support to get mac from device-tree using of_get_ethdev_address.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agosun4i-emac.c: remove unnecessary branch
Conley Lee [Tue, 14 Dec 2021 09:11:06 +0000 (17:11 +0800)]
sun4i-emac.c: remove unnecessary branch

According to the current implementation of emac_rx, every arrived packet
will be processed in the while loop. So, there is no remain packet last
time. The skb_last field and this branch for dealing with it is
unnecessary.

Signed-off-by: Conley Lee <conleylee@foxmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoethtool: use ethnl_parse_header_dev_put()
Eric Dumazet [Tue, 14 Dec 2021 08:42:30 +0000 (00:42 -0800)]
ethtool: use ethnl_parse_header_dev_put()

It seems I missed that most ethnl_parse_header_dev_get() callers
declare an on-stack struct ethnl_req_info, and that they simply call
dev_put(req_info.dev) when about to return.

Add ethnl_parse_header_dev_put() helper to properly untrack
reference taken by ethnl_parse_header_dev_get().

Fixes: e4b8954074f6 ("netlink: add net device refcount tracker to struct ethnl_req_info")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet/mlx5e: Move goto action checks into tc_action goto post parse op
Roi Dayan [Sun, 15 Aug 2021 10:23:14 +0000 (13:23 +0300)]
net/mlx5e: Move goto action checks into tc_action goto post parse op

Move goto action checks from parse nic/fdb funcs into the tc action
infra goto post parse op.
While moving this part also use NL_SET_ERR_MSG_MOD() instead of
NL_SET_ERR_MSG().

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Move vlan action chunk into tc action vlan post parse op
Roi Dayan [Mon, 23 Aug 2021 08:30:17 +0000 (11:30 +0300)]
net/mlx5e: Move vlan action chunk into tc action vlan post parse op

Move vlan prio tag rewrite handling into tc action infra vlan post parse op.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add post_parse() op to tc action infrastructure
Roi Dayan [Wed, 11 Aug 2021 11:19:56 +0000 (14:19 +0300)]
net/mlx5e: Add post_parse() op to tc action infrastructure

The post_parse() op should be called after the parse op was called
for all actions. It could be an action state is dependent on other
actions. In the new op an action can fail the parse if the state
is not valid anymore.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Move sample attr allocation to tc_action sample parse op
Roi Dayan [Thu, 12 Aug 2021 13:02:58 +0000 (16:02 +0300)]
net/mlx5e: Move sample attr allocation to tc_action sample parse op

There is no reason to wait with the kmalloc to after parsing all
other actions. There could still be a failure later and before
offloading the rule. So alloc the mem when parsing.
The memory is being released on mlx5e_flow_put() which is called
also on error flow.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: TC action parsing loop
Roi Dayan [Sun, 25 Jul 2021 08:02:27 +0000 (11:02 +0300)]
net/mlx5e: TC action parsing loop

Introduce a common function to implement the generic parsing loop.
The same function can be used for parsing NIC and FDB (Switchdev mode) flows.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add redirect ingress to tc action infra
Roi Dayan [Tue, 16 Nov 2021 11:17:56 +0000 (13:17 +0200)]
net/mlx5e: Add redirect ingress to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add sample and ptype to tc_action infra
Roi Dayan [Wed, 27 Oct 2021 06:52:36 +0000 (09:52 +0300)]
net/mlx5e: Add sample and ptype to tc_action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add ct to tc action infra
Roi Dayan [Wed, 21 Jul 2021 12:02:36 +0000 (15:02 +0300)]
net/mlx5e: Add ct to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add mirred/redirect to tc action infra
Roi Dayan [Wed, 21 Jul 2021 07:27:50 +0000 (10:27 +0300)]
net/mlx5e: Add mirred/redirect to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add mpls push/pop to tc action infra
Roi Dayan [Wed, 21 Jul 2021 08:25:35 +0000 (11:25 +0300)]
net/mlx5e: Add mpls push/pop to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add vlan push/pop/mangle to tc action infra
Roi Dayan [Wed, 21 Jul 2021 07:28:55 +0000 (10:28 +0300)]
net/mlx5e: Add vlan push/pop/mangle to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add pedit to tc action infra
Roi Dayan [Thu, 22 Jul 2021 12:26:25 +0000 (15:26 +0300)]
net/mlx5e: Add pedit to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add csum to tc action infra
Roi Dayan [Wed, 21 Jul 2021 07:30:40 +0000 (10:30 +0300)]
net/mlx5e: Add csum to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add tunnel encap/decap to tc action infra
Roi Dayan [Wed, 21 Jul 2021 07:29:37 +0000 (10:29 +0300)]
net/mlx5e: Add tunnel encap/decap to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add goto to tc action infra
Roi Dayan [Wed, 21 Jul 2021 07:26:41 +0000 (10:26 +0300)]
net/mlx5e: Add goto to tc action infra

Add parsing support by implementing struct mlx5e_tc_act
for this action.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agonet/mlx5e: Add tc action infrastructure
Roi Dayan [Mon, 12 Jul 2021 11:13:10 +0000 (14:13 +0300)]
net/mlx5e: Add tc action infrastructure

Add an infrastructure to help parsing tc actions in a generic way.

Supporting an action parser means implementing struct mlx5e_tc_act
for that action.

The infrastructure will give the possibility to be generic when parsing tc
actions, i.e. parse_tc_nic_actions() and parse_tc_fdb_actions().
To parse tc actions a user needs to allocate a parse_state instance
and pass it when iterating over the tc actions parsers.
If a parser doesn't exists then a user can treat it as unsupported.

To add an action parser a user needs to implement two callbacks.
The can_offload() callback to quickly check if an action can be offloaded.
The parse_action() callback to do actual parsing and prepare for offload.

Add implementation for drop, trap, mark and accept action parsers with this
commit to act as examples and implement usage of the new infrastructure for
those actions.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
2 years agoMerge branch 'net-dsa-hellcreek-fix-handling-of-mgmt-protocols'
Jakub Kicinski [Wed, 15 Dec 2021 02:46:31 +0000 (18:46 -0800)]
Merge branch 'net-dsa-hellcreek-fix-handling-of-mgmt-protocols'

Kurt Kanzenbach says:

====================
net: dsa: hellcreek: Fix handling of MGMT protocols

this series fixes some minor issues with regards to management protocols
such as PTP and STP in the hellcreek DSA driver. Configure static FDB
for these protocols. The end result is:

|root@tsn:~# mv88e6xxx_dump --atu
|Using device <platform/ff240000.switch>
|ATU:
|FID  MAC               0123 Age OBT Pass Static Reprio Prio
|   0 01:1b:19:00:00:00 1100   1               X       X    6
|   1 01:00:5e:00:01:81 1100   1               X       X    6
|   2 33:33:00:00:01:81 1100   1               X       X    6
|   3 01:80:c2:00:00:0e 1100   1        X      X       X    6
|   4 01:00:5e:00:00:6b 1100   1        X      X       X    6
|   5 33:33:00:00:00:6b 1100   1        X      X       X    6
|   6 01:80:c2:00:00:00 1100   1        X      X       X    6

Previous version:
 * https://lore.kernel.org/r/20211213101810.121553-1-kurt@linutronix.de/

Changes since v1:
 * Target net-next, as this never worked correctly and is not critical
 * Add STP and PTP over UDP rules
 * Use pass_blocked for PDelay messages only (Richard Cochran)
====================

Link: https://lore.kernel.org/r/20211214134508.57806-1-kurt@linutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: dsa: hellcreek: Add missing PTP via UDP rules
Kurt Kanzenbach [Tue, 14 Dec 2021 13:45:08 +0000 (14:45 +0100)]
net: dsa: hellcreek: Add missing PTP via UDP rules

The switch supports PTP for UDP transport too. Therefore, add the missing static
FDB entries to ensure correct forwarding of these packets.

Fixes: ddd56dfe52c9 ("net: dsa: hellcreek: Add PTP clock support")
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: dsa: hellcreek: Allow PTP P2P measurements on blocked ports
Kurt Kanzenbach [Tue, 14 Dec 2021 13:45:07 +0000 (14:45 +0100)]
net: dsa: hellcreek: Allow PTP P2P measurements on blocked ports

Allow PTP peer delay measurements on blocked ports by STP. In case of topology
changes the PTP stack can directly start with the correct delays.

Fixes: ddd56dfe52c9 ("net: dsa: hellcreek: Add PTP clock support")
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: dsa: hellcreek: Add STP forwarding rule
Kurt Kanzenbach [Tue, 14 Dec 2021 13:45:06 +0000 (14:45 +0100)]
net: dsa: hellcreek: Add STP forwarding rule

Treat STP as management traffic. STP traffic is designated for the CPU port
only. In addition, STP traffic has to pass blocked ports.

Fixes: e4b27ebc780f ("net: dsa: Add DSA driver for Hirschmann Hellcreek switches")
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: dsa: hellcreek: Fix insertion of static FDB entries
Kurt Kanzenbach [Tue, 14 Dec 2021 13:45:05 +0000 (14:45 +0100)]
net: dsa: hellcreek: Fix insertion of static FDB entries

The insertion of static FDB entries ignores the pass_blocked bit. That bit is
evaluated with regards to STP. Add the missing functionality.

Fixes: e4b27ebc780f ("net: dsa: Add DSA driver for Hirschmann Hellcreek switches")
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: dev_replace_track() cleanup
Eric Dumazet [Tue, 14 Dec 2021 15:15:15 +0000 (07:15 -0800)]
net: dev_replace_track() cleanup

Use existing helpers (netdev_tracker_free()
and netdev_tracker_alloc()) to remove ifdefery.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20211214151515.312535-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: linkwatch: be more careful about dev->linkwatch_dev_tracker
Eric Dumazet [Tue, 14 Dec 2021 05:19:55 +0000 (21:19 -0800)]
net: linkwatch: be more careful about dev->linkwatch_dev_tracker

Apparently a concurrent linkwatch_add_event() could
run while we are in __linkwatch_run_queue().

We need to free dev->linkwatch_dev_tracker tracker
under lweventlist_lock protection to avoid this race.

syzbot report:
[   77.935949][ T3661] reference already released.
[   77.941015][ T3661] allocated in:
[   77.944482][ T3661]  linkwatch_fire_event+0x202/0x260
[   77.950318][ T3661]  netif_carrier_on+0x9c/0x100
[   77.955120][ T3661]  __ieee80211_sta_join_ibss+0xc52/0x1590
[   77.960888][ T3661]  ieee80211_sta_create_ibss.cold+0xd2/0x11f
[   77.966908][ T3661]  ieee80211_ibss_work.cold+0x30e/0x60f
[   77.972483][ T3661]  ieee80211_iface_work+0xb70/0xd00
[   77.977715][ T3661]  process_one_work+0x9ac/0x1680
[   77.982671][ T3661]  worker_thread+0x652/0x11c0
[   77.987371][ T3661]  kthread+0x405/0x4f0
[   77.991465][ T3661]  ret_from_fork+0x1f/0x30
[   77.995895][ T3661] freed in:
[   77.999006][ T3661]  linkwatch_do_dev+0x96/0x160
[   78.004014][ T3661]  __linkwatch_run_queue+0x233/0x6a0
[   78.009496][ T3661]  linkwatch_event+0x4a/0x60
[   78.014099][ T3661]  process_one_work+0x9ac/0x1680
[   78.019034][ T3661]  worker_thread+0x652/0x11c0
[   78.023719][ T3661]  kthread+0x405/0x4f0
[   78.027810][ T3661]  ret_from_fork+0x1f/0x30
[   78.042541][ T3661] ------------[ cut here ]------------
[   78.048253][ T3661] WARNING: CPU: 0 PID: 3661 at lib/ref_tracker.c:120 ref_tracker_free.cold+0x110/0x14e
[   78.062364][ T3661] Modules linked in:
[   78.066424][ T3661] CPU: 0 PID: 3661 Comm: kworker/0:5 Not tainted 5.16.0-rc4-next-20211210-syzkaller #0
[   78.076075][ T3661] Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
[   78.090648][ T3661] Workqueue: events linkwatch_event
[   78.095890][ T3661] RIP: 0010:ref_tracker_free.cold+0x110/0x14e
[   78.102191][ T3661] Code: ea 03 48 c1 e0 2a 0f b6 04 02 84 c0 74 04 3c 03 7e 4c 8b 7b 18 e8 6b 54 e9 fa e8 26 4d 57 f8 4c 89 ee 48 89 ef e8 fb 33 36 00 <0f> 0b 41 bd ea ff ff ff e9 bd 60 e9 fa 4c 89 f7 e8 16 45 a2 f8 e9
[   78.127211][ T3661] RSP: 0018:ffffc90002b5fb18 EFLAGS: 00010246
[   78.133684][ T3661] RAX: 0000000000000000 RBX: ffff88807467f700 RCX: 0000000000000000
[   78.141928][ T3661] RDX: 0000000000000001 RSI: 0000000000000001 RDI: 0000000000000001
[   78.150087][ T3661] RBP: ffff888057e105b8 R08: 0000000000000001 R09: ffffffff8ffa1967
[   78.158211][ T3661] R10: 0000000000000001 R11: 0000000000000000 R12: 1ffff9200056bf65
[   78.166204][ T3661] R13: 0000000000000292 R14: ffff88807467f718 R15: 00000000c0e0008c
[   78.174321][ T3661] FS:  0000000000000000(0000) GS:ffff8880b9c00000(0000) knlGS:0000000000000000
[   78.183310][ T3661] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   78.190156][ T3661] CR2: 000000c000208800 CR3: 000000007f7b5000 CR4: 00000000003506f0
[   78.198235][ T3661] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   78.206214][ T3661] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   78.214328][ T3661] Call Trace:
[   78.217679][ T3661]  <TASK>
[   78.220621][ T3661]  ? __sanitizer_cov_trace_const_cmp4+0x1c/0x70
[   78.226981][ T3661]  ? nlmsg_notify+0xbe/0x280
[   78.231607][ T3661]  ? ref_tracker_dir_exit+0x330/0x330
[   78.237654][ T3661]  ? linkwatch_do_dev+0x96/0x160
[   78.242628][ T3661]  ? __linkwatch_run_queue+0x233/0x6a0
[   78.248170][ T3661]  ? linkwatch_event+0x4a/0x60
[   78.252946][ T3661]  ? process_one_work+0x9ac/0x1680
[   78.258136][ T3661]  ? worker_thread+0x853/0x11c0
[   78.263020][ T3661]  ? kthread+0x405/0x4f0
[   78.267905][ T3661]  ? ret_from_fork+0x1f/0x30
[   78.272670][ T3661]  ? netdev_state_change+0xa1/0x130
[   78.278019][ T3661]  ? netdev_exit+0xd0/0xd0
[   78.282466][ T3661]  ? dev_activate+0x420/0xa60
[   78.287261][ T3661]  linkwatch_do_dev+0x96/0x160
[   78.292043][ T3661]  __linkwatch_run_queue+0x233/0x6a0
[   78.297505][ T3661]  ? linkwatch_do_dev+0x160/0x160
[   78.302561][ T3661]  linkwatch_event+0x4a/0x60
[   78.307225][ T3661]  process_one_work+0x9ac/0x1680
[   78.312292][ T3661]  ? pwq_dec_nr_in_flight+0x2a0/0x2a0
[   78.317757][ T3661]  ? rwlock_bug.part.0+0x90/0x90
[   78.322726][ T3661]  ? _raw_spin_lock_irq+0x41/0x50
[   78.327844][ T3661]  worker_thread+0x853/0x11c0
[   78.332543][ T3661]  ? process_one_work+0x1680/0x1680
[   78.338500][ T3661]  kthread+0x405/0x4f0
[   78.342610][ T3661]  ? set_kthread_struct+0x130/0x130

Fixes: 63f13937cbe9 ("net: linkwatch: add net device refcount tracker")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Link: https://lore.kernel.org/r/20211214051955.3569843-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agomptcp: adjust to use netns refcount tracker
Eric Dumazet [Tue, 14 Dec 2021 04:32:08 +0000 (20:32 -0800)]
mptcp: adjust to use netns refcount tracker

MPTCP can change sk_net_refcnt after sock_create_kern() call.

We need to change its corresponding get_net() to avoid
a splat at release time, as in :

refcount_t: decrement hit 0; leaking memory.
WARNING: CPU: 0 PID: 3599 at lib/refcount.c:31 refcount_warn_saturate+0xbf/0x1e0 lib/refcount.c:31
Modules linked in:
CPU: 1 PID: 3599 Comm: syz-fuzzer Not tainted 5.16.0-rc4-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:refcount_warn_saturate+0xbf/0x1e0 lib/refcount.c:31
Code: 1d b1 99 a1 09 31 ff 89 de e8 5d 3a 9c fd 84 db 75 e0 e8 74 36 9c fd 48 c7 c7 60 00 05 8a c6 05 91 99 a1 09 01 e8 cc 4b 27 05 <0f> 0b eb c4 e8 58 36 9c fd 0f b6 1d 80 99 a1 09 31 ff 89 de e8 28
RSP: 0018:ffffc90001f5fab0 EFLAGS: 00010286
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
RDX: ffff888021873a00 RSI: ffffffff815f1e28 RDI: fffff520003ebf48
RBP: 0000000000000004 R08: 0000000000000000 R09: 0000000000000000
R10: ffffffff815ebbce R11: 0000000000000000 R12: 1ffff920003ebf5b
R13: 00000000ffffffef R14: ffffffff8d2fcd94 R15: ffffc90001f5fd10
FS:  000000c00008a090(0000) GS:ffff8880b9d00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f0a5b59e300 CR3: 000000001cbe6000 CR4: 00000000003506e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <TASK>
 __refcount_dec include/linux/refcount.h:344 [inline]
 refcount_dec include/linux/refcount.h:359 [inline]
 ref_tracker_free+0x4fe/0x610 lib/ref_tracker.c:101
 netns_tracker_free include/net/net_namespace.h:327 [inline]
 put_net_track include/net/net_namespace.h:341 [inline]
 __sk_destruct+0x4a6/0x920 net/core/sock.c:2042
 sk_destruct+0xbd/0xe0 net/core/sock.c:2058
 __sk_free+0xef/0x3d0 net/core/sock.c:2069
 sk_free+0x78/0xa0 net/core/sock.c:2080
 sock_put include/net/sock.h:1911 [inline]
 __mptcp_close_ssk+0x435/0x590 net/mptcp/protocol.c:2276
 __mptcp_destroy_sock+0x35f/0x830 net/mptcp/protocol.c:2702
 mptcp_close+0x5f8/0x7f0 net/mptcp/protocol.c:2750
 inet_release+0x12e/0x280 net/ipv4/af_inet.c:428
 inet6_release+0x4c/0x70 net/ipv6/af_inet6.c:476
 __sock_release+0xcd/0x280 net/socket.c:649
 sock_close+0x18/0x20 net/socket.c:1314
 __fput+0x286/0x9f0 fs/file_table.c:280
 task_work_run+0xdd/0x1a0 kernel/task_work.c:164
 tracehook_notify_resume include/linux/tracehook.h:189 [inline]
 exit_to_user_mode_loop kernel/entry/common.c:175 [inline]
 exit_to_user_mode_prepare+0x27e/0x290 kernel/entry/common.c:207
 __syscall_exit_to_user_mode_work kernel/entry/common.c:289 [inline]
 syscall_exit_to_user_mode+0x19/0x60 kernel/entry/common.c:300
 do_syscall_64+0x42/0xb0 arch/x86/entry/common.c:86
 entry_SYSCALL_64_after_hwframe+0x44/0xae

Fixes: ffa84b5ffb37 ("net: add netns refcount tracker to struct sock")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Mat Martineau <mathew.j.martineau@linux.intel.com>
Cc: Florian Westphal <fw@strlen.de>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Link: https://lore.kernel.org/r/20211214043208.3543046-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agoipv6: use GFP_ATOMIC in rt6_probe()
Eric Dumazet [Tue, 14 Dec 2021 02:58:06 +0000 (18:58 -0800)]
ipv6: use GFP_ATOMIC in rt6_probe()

syzbot reminded me that rt6_probe() can run from
atomic contexts.

stack backtrace:

CPU: 1 PID: 7461 Comm: syz-executor.2 Not tainted 5.16.0-rc4-next-20211210-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 <IRQ>
 __dump_stack lib/dump_stack.c:88 [inline]
 dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:106
 print_usage_bug kernel/locking/lockdep.c:203 [inline]
 valid_state kernel/locking/lockdep.c:3945 [inline]
 mark_lock_irq kernel/locking/lockdep.c:4148 [inline]
 mark_lock.cold+0x61/0x8e kernel/locking/lockdep.c:4605
 mark_usage kernel/locking/lockdep.c:4500 [inline]
 __lock_acquire+0x11d5/0x54a0 kernel/locking/lockdep.c:4981
 lock_acquire kernel/locking/lockdep.c:5639 [inline]
 lock_acquire+0x1ab/0x510 kernel/locking/lockdep.c:5604
 __fs_reclaim_acquire mm/page_alloc.c:4550 [inline]
 fs_reclaim_acquire+0x115/0x160 mm/page_alloc.c:4564
 might_alloc include/linux/sched/mm.h:253 [inline]
 slab_pre_alloc_hook mm/slab.h:739 [inline]
 slab_alloc_node mm/slub.c:3145 [inline]
 slab_alloc mm/slub.c:3239 [inline]
 kmem_cache_alloc_trace+0x3b/0x2c0 mm/slub.c:3256
 kmalloc include/linux/slab.h:581 [inline]
 kzalloc include/linux/slab.h:715 [inline]
 ref_tracker_alloc+0xe1/0x430 lib/ref_tracker.c:74
 netdev_tracker_alloc include/linux/netdevice.h:3860 [inline]
 dev_hold_track include/linux/netdevice.h:3877 [inline]
 rt6_probe net/ipv6/route.c:661 [inline]
 find_match.part.0+0xac9/0xd00 net/ipv6/route.c:752
 find_match net/ipv6/route.c:825 [inline]
 __find_rr_leaf+0x17f/0xd20 net/ipv6/route.c:826
 find_rr_leaf net/ipv6/route.c:847 [inline]
 rt6_select net/ipv6/route.c:891 [inline]
 fib6_table_lookup+0x649/0xa20 net/ipv6/route.c:2185
 ip6_pol_route+0x1c5/0x11e0 net/ipv6/route.c:2221
 pol_lookup_func include/net/ip6_fib.h:580 [inline]
 fib6_rule_lookup+0x52a/0x6f0 net/ipv6/fib6_rules.c:120
 ip6_route_output_flags_noref+0x2e2/0x380 net/ipv6/route.c:2629
 ip6_route_output_flags+0x72/0x320 net/ipv6/route.c:2642
 ip6_route_output include/net/ip6_route.h:98 [inline]
 ip6_dst_lookup_tail+0x5ab/0x1620 net/ipv6/ip6_output.c:1070
 ip6_dst_lookup_flow+0x8c/0x1d0 net/ipv6/ip6_output.c:1200
 geneve_get_v6_dst+0x46f/0x9a0 drivers/net/geneve.c:858
 geneve6_xmit_skb drivers/net/geneve.c:991 [inline]
 geneve_xmit+0x520/0x3530 drivers/net/geneve.c:1074
 __netdev_start_xmit include/linux/netdevice.h:4685 [inline]
 netdev_start_xmit include/linux/netdevice.h:4699 [inline]
 xmit_one net/core/dev.c:3473 [inline]
 dev_hard_start_xmit+0x1eb/0x920 net/core/dev.c:3489
 __dev_queue_xmit+0x2983/0x3640 net/core/dev.c:4112
 neigh_resolve_output net/core/neighbour.c:1522 [inline]
 neigh_resolve_output+0x50e/0x820 net/core/neighbour.c:1502
 neigh_output include/net/neighbour.h:541 [inline]
 ip6_finish_output2+0x56e/0x14f0 net/ipv6/ip6_output.c:126
 __ip6_finish_output net/ipv6/ip6_output.c:191 [inline]
 __ip6_finish_output+0x61e/0xe80 net/ipv6/ip6_output.c:170
 ip6_finish_output+0x32/0x200 net/ipv6/ip6_output.c:201
 NF_HOOK_COND include/linux/netfilter.h:296 [inline]
 ip6_output+0x1e4/0x530 net/ipv6/ip6_output.c:224
 dst_output include/net/dst.h:451 [inline]
 NF_HOOK include/linux/netfilter.h:307 [inline]
 ndisc_send_skb+0xa99/0x17f0 net/ipv6/ndisc.c:508
 ndisc_send_rs+0x12e/0x6f0 net/ipv6/ndisc.c:702
 addrconf_rs_timer+0x3f2/0x820 net/ipv6/addrconf.c:3898
 call_timer_fn+0x1a5/0x6b0 kernel/time/timer.c:1421
 expire_timers kernel/time/timer.c:1466 [inline]
 __run_timers.part.0+0x675/0xa20 kernel/time/timer.c:1734
 __run_timers kernel/time/timer.c:1715 [inline]
 run_timer_softirq+0xb3/0x1d0 kernel/time/timer.c:1747
 __do_softirq+0x29b/0x9c2 kernel/softirq.c:558
 invoke_softirq kernel/softirq.c:432 [inline]
 __irq_exit_rcu+0x123/0x180 kernel/softirq.c:637
 irq_exit_rcu+0x5/0x20 kernel/softirq.c:649
 sysvec_apic_timer_interrupt+0x93/0xc0 arch/x86/kernel/apic/apic.c:1097
 </IRQ>

Fixes: fb67510ba9bd ("ipv6: add net device refcount tracker to rt6_probe_deferred()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Link: https://lore.kernel.org/r/20211214025806.3456382-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agoice: Remove unused ICE_FLOW_SEG_HDRS_L2_MASK
Tony Nguyen [Wed, 17 Nov 2021 23:55:05 +0000 (15:55 -0800)]
ice: Remove unused ICE_FLOW_SEG_HDRS_L2_MASK

Remove the unused define ICE_FLOW_SEG_HDRS_L2_MASK.

Reported-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Acked-by: Paul Menzel <pmenzel@molgen.mpg.de>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
2 years agoice: Remove unnecessary casts
Dan Carpenter [Wed, 10 Nov 2021 07:50:23 +0000 (10:50 +0300)]
ice: Remove unnecessary casts

The "bitmap" variable is already an unsigned long so there is no need
for this cast.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
2 years agoice: Propagate error codes
Tony Nguyen [Thu, 7 Oct 2021 23:01:58 +0000 (16:01 -0700)]
ice: Propagate error codes

As all functions now return standard error codes, propagate the values
being returned instead of converting them to generic values.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
2 years agoice: Remove excess error variables
Tony Nguyen [Thu, 7 Oct 2021 23:00:23 +0000 (16:00 -0700)]
ice: Remove excess error variables

ice_status previously had a variable to contain these values where other
error codes had a variable as well. With ice_status now being an int,
there is no need for two variables to hold error values. In cases where
this occurs, remove one of the excess variables and use a single one.
Some initialization of variables are no longer needed and have been
removed.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
2 years agoice: Cleanup after ice_status removal
Tony Nguyen [Thu, 7 Oct 2021 22:59:03 +0000 (15:59 -0700)]
ice: Cleanup after ice_status removal

Clean up code after changing ice_status to int. Rearrange to fix reverse
Christmas tree and pull lines up where applicable.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
2 years agoice: Remove enum ice_status
Tony Nguyen [Thu, 7 Oct 2021 22:58:01 +0000 (15:58 -0700)]
ice: Remove enum ice_status

Replace uses of ice_status to, as equivalent as possible, error codes.
Remove enum ice_status and its helper conversion function as they are no
longer needed.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
2 years agoice: Use int for ice_status
Tony Nguyen [Thu, 7 Oct 2021 22:56:57 +0000 (15:56 -0700)]
ice: Use int for ice_status

To prepare for removal of ice_status, change the variables from
ice_status to int. This eases the transition when values are changed to
return standard int error codes over enum ice_status.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
2 years agoice: Remove string printing for ice_status
Tony Nguyen [Thu, 7 Oct 2021 22:56:02 +0000 (15:56 -0700)]
ice: Remove string printing for ice_status

Remove the ice_stat_str() function which prints the string
representation of the ice_status error code. With upcoming changes
moving away from ice_status, there will be no need for this function.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
2 years agoice: Refactor status flow for DDP load
Wojciech Drewek [Thu, 7 Oct 2021 22:54:37 +0000 (15:54 -0700)]
ice: Refactor status flow for DDP load

Before this change, final state of the DDP pkg load process was
dependent on many variables such as: ice_status, pkg version,
ice_aq_err. The last one had be stored in hw->pkg_dwnld_status.
It was impossible to conclude this state just from ice_status, that's
why logging process of DDP pkg load in the caller was a little bit
complicated.

With this patch new status enum is introduced - ice_ddp_state.
It covers all the possible final states of the loading process.
What's tricky for ice_ddp_state is that not only
ICE_DDP_PKG_SUCCESS(=0) means that load was successful. Actually
three states mean that:
 - ICE_DDP_PKG_SUCCESS
 - ICE_DDP_PKG_SAME_VERSION_ALREADY_LOADED
 - ICE_DDP_PKG_COMPATIBLE_ALREADY_LOADED
ice_is_init_pkg_successful can tell that information.

One ddp_state should not be used outside of ice_init_pkg which is
ICE_DDP_PKG_ALREADY_LOADED. It is more generic, it is used in
ice_dwnld_cfg_bufs to see if pkg is already loaded. At this point
we can't use one of the specific one (SAME_VERSION, COMPATIBLE,
NOT_SUPPORTED) because we don't have information on the package
currently loaded in HW (we are before calling ice_get_pkg_info).

We can get rid of hw->pkg_dwnld_status because we are immediately
mapping aq errors to ice_ddp_state in ice_dwnld_cfg_bufs.

Other errors like ICE_ERR_NO_MEMORY, ICE_ERR_PARAM are mapped the
generic ICE_DDP_PKG_ERR.

Suggested-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Wojciech Drewek <wojciech.drewek@intel.com>
Tested-by: Tony Brelinski <tony.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
2 years agoice: Refactor promiscuous functions
Brett Creeley [Tue, 2 Mar 2021 18:15:34 +0000 (10:15 -0800)]
ice: Refactor promiscuous functions

Some of the promiscuous mode functions take a boolean to indicate
set/clear, which affects readability. Refactor and provide an
interface for the promiscuous mode code with explicit set and clear
promiscuous mode operations.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Konrad Jankowski <konrad0.jankowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
2 years agoice: refactor PTYPE validating
Jeff Guo [Fri, 16 Jul 2021 22:16:43 +0000 (15:16 -0700)]
ice: refactor PTYPE validating

Since the capability of a PTYPE within a specific package could be
negotiated by checking the HW bit map, it means that there's no need
to maintain a different PTYPE list for each type of the package when
parsing PTYPE. So refactor the PTYPE validating mechanism.

Signed-off-by: Jeff Guo <jia.guo@intel.com>
Tested-by: Tony Brelinski <tony.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
2 years agoice: Add package PTYPE enable information
Haiyue Wang [Fri, 16 Jul 2021 22:16:42 +0000 (15:16 -0700)]
ice: Add package PTYPE enable information

Scan the 'Marker Ptype TCAM' section to retrieve the Rx parser PTYPE
enable information from the current package.

Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
Tested-by: Tony Brelinski <tony.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
2 years agoibmvnic: remove unused defines
Dany Madden [Tue, 14 Dec 2021 05:17:48 +0000 (00:17 -0500)]
ibmvnic: remove unused defines

IBMVNIC_STATS_TIMEOUT and IBMVNIC_INIT_FAILED are not used in the driver.
Remove them.

Suggested-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Signed-off-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoibmvnic: Update driver return codes
Dany Madden [Tue, 14 Dec 2021 05:17:47 +0000 (00:17 -0500)]
ibmvnic: Update driver return codes

Update return codes to be more informative.

Signed-off-by: Jacob Root <otis@otisroot.com>
Signed-off-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoRevert "pktgen: use min() to make code cleaner"
David S. Miller [Tue, 14 Dec 2021 12:53:41 +0000 (12:53 +0000)]
Revert "pktgen: use min() to make code cleaner"

This reverts commit 13510fef48a3803d9ee8f044b015dacfb06fe0f5.

Causes build warnings.

Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agopktgen: use min() to make code cleaner
Changcheng Deng [Tue, 14 Dec 2021 11:44:47 +0000 (11:44 +0000)]
pktgen: use min() to make code cleaner

Use min() in order to make code cleaner. Issue found by coccinelle.

Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: Changcheng Deng <deng.changcheng@zte.com.cn>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoMerge branch 'dsa-fixups'
David S. Miller [Tue, 14 Dec 2021 12:45:16 +0000 (12:45 +0000)]
Merge branch 'dsa-fixups'

Vladimir Oltean says:

====================
DSA tagger-owned storage fixups

It seems that the DSA tagger-owned storage changes were insufficiently
tested and do not work in all cases. Specifically, the NXP Bluebox 3
(arch/arm64/boot/dts/freescale/fsl-lx2160a-bluebox3.dts) got broken by
these changes, because
(a) I forgot that DSA_TAG_PROTO_SJA1110 exists and differs from
    DSA_TAG_PROTO_SJA1105
(b) the Bluebox 3 uses a DSA switch tree with 2 switches, and the
    tagger-owned storage patches don't cover that use case well, it
    seems

Therefore, I'm sorry to say that there needs to be an API fixup: tagging
protocol drivers will from now on connect to individual switches from a
tree, rather than to the tree as a whole. This is more robust against
various ordering constraints in the DSA probe and teardown paths, and is
also symmetrical with the connection API exposed to the switch drivers
themselves, which is also per switch.

With these changes, the Bluebox 3 also works fine.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: make tagging protocols connect to individual switches from a tree
Vladimir Oltean [Tue, 14 Dec 2021 01:45:36 +0000 (03:45 +0200)]
net: dsa: make tagging protocols connect to individual switches from a tree

On the NXP Bluebox 3 board which uses a multi-switch setup with sja1105,
the mechanism through which the tagger connects to the switch tree is
broken, due to improper DSA code design. At the time when tag_ops->connect()
is called in dsa_port_parse_cpu(), DSA hasn't finished "touching" all
the ports, so it doesn't know how large the tree is and how many ports
it has. It has just seen the first CPU port by this time. As a result,
this function will call the tagger's ->connect method too early, and the
tagger will connect only to the first switch from the tree.

This could be perhaps addressed a bit more simply by just moving the
tag_ops->connect(dst) call a bit later (for example in dsa_tree_setup),
but there is already a design inconsistency at present: on the switch
side, the notification is on a per-switch basis, but on the tagger side,
it is on a per-tree basis. Furthermore, the persistent storage itself is
per switch (ds->tagger_data). And the tagger connect and disconnect
procedures (at least the ones that exist currently) could see a fair bit
of simplification if they didn't have to iterate through the switches of
a tree.

To fix the issue, this change transforms tag_ops->connect(dst) into
tag_ops->connect(ds) and moves it somewhere where we already iterate
over all switches of a tree. That is in dsa_switch_setup_tag_protocol(),
which is a good placement because we already have there the connection
call to the switch side of things.

As for the dsa_tree_bind_tag_proto() method (called from the code path
that changes the tag protocol), things are a bit more complicated
because we receive the tree as argument, yet when we unwind on errors,
it would be nice to not call tag_ops->disconnect(ds) where we didn't
previously call tag_ops->connect(ds). We didn't have this problem before
because the tag_ops connection operations passed the entire dst before,
and this is more fine grained now. To solve the error rewind case using
the new API, we have to create yet one more cross-chip notifier for
disconnection, and stay connected with the old tag protocol to all the
switches in the tree until we've succeeded to connect with the new one
as well. So if something fails half way, the whole tree is still
connected to the old tagger. But there may still be leaks if the tagger
fails to connect to the 2nd out of 3 switches in a tree: somebody needs
to tell the tagger to disconnect from the first switch. Nothing comes
for free, and this was previously handled privately by the tagging
protocol driver before, but now we need to emit a disconnect cross-chip
notifier for that, because DSA has to take care of the unwind path. We
assume that the tagging protocol has connected to a switch if it has set
ds->tagger_data to something, otherwise we avoid calling its
disconnection method in the error rewind path.

The rest of the changes are in the tagging protocol drivers, and have to
do with the replacement of dst with ds. The iteration is removed and the
error unwind path is simplified, as mentioned above.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: sja1105: fix broken connection with the sja1110 tagger
Vladimir Oltean [Tue, 14 Dec 2021 01:45:35 +0000 (03:45 +0200)]
net: dsa: sja1105: fix broken connection with the sja1110 tagger

The driver was incorrectly converted assuming that "sja1105" is the only
tagger supported by this driver. This results in SJA1110 switches
failing to probe:

sja1105 spi1.0: Unable to connect to tag protocol "sja1110": -EPROTONOSUPPORT
sja1105: probe of spi1.2 failed with error -93

Add DSA_TAG_PROTO_SJA1110 to the list of supported taggers by the
sja1105 driver. The sja1105_tagger_data structure format is common for
the two tagging protocols.

Fixes: c79e84866d2a ("net: dsa: tag_sja1105: convert to tagger-owned data")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: tag_sja1105: fix zeroization of ds->priv on tag proto disconnect
Vladimir Oltean [Tue, 14 Dec 2021 01:45:34 +0000 (03:45 +0200)]
net: dsa: tag_sja1105: fix zeroization of ds->priv on tag proto disconnect

The method was meant to zeroize ds->tagger_data but got the wrong
pointer. Fix this.

Fixes: c79e84866d2a ("net: dsa: tag_sja1105: convert to tagger-owned data")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agobareudp: Add extack support to bareudp_configure()
Guillaume Nault [Mon, 13 Dec 2021 18:17:17 +0000 (19:17 +0100)]
bareudp: Add extack support to bareudp_configure()

Add missing extacks for common configuration errors.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoethtool: fix null-ptr-deref on ref tracker
Jakub Kicinski [Tue, 14 Dec 2021 01:39:02 +0000 (17:39 -0800)]
ethtool: fix null-ptr-deref on ref tracker

dev can be a NULL here, not all requests set require_dev.

Fixes: e4b8954074f6 ("netlink: add net device refcount tracker to struct ethnl_req_info")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dev: Change the order of the arguments for the contended condition.
Sebastian Andrzej Siewior [Mon, 13 Dec 2021 16:29:31 +0000 (17:29 +0100)]
net: dev: Change the order of the arguments for the contended condition.

Change the order of arguments and make qdisc_is_running() appear first.
This is more readable for the general case.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoMerge branch 'hwtstamp_bonding'
David S. Miller [Tue, 14 Dec 2021 12:28:25 +0000 (12:28 +0000)]
Merge branch 'hwtstamp_bonding'

Hangbin Liu says:

====================
net: add new hwtstamp flag HWTSTAMP_FLAG_BONDED_PHC_INDEX

This patchset add a new hwtstamp_config flag HWTSTAMP_FLAG_BONDED_PHC_INDEX.
When user want to get bond active interface's PHC, they need to add this flag
and aware the PHC index may changed.

v3: Use bitwise test to check the flags validation
v2: rename the flag to HWTSTAMP_FLAG_BONDED_PHC_INDEX
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoBonding: force user to add HWTSTAMP_FLAG_BONDED_PHC_INDEX when get/set HWTSTAMP
Hangbin Liu [Fri, 10 Dec 2021 08:59:59 +0000 (16:59 +0800)]
Bonding: force user to add HWTSTAMP_FLAG_BONDED_PHC_INDEX when get/set HWTSTAMP

When there is a failover, the PHC index of bond active interface will be
changed. This may break the user space program if the author didn't aware.

By setting this flag, the user should aware that the PHC index get/set
by syscall is not stable. And the user space is able to deal with it.
Without this flag, the kernel will reject the request forwarding to
bonding.

Reported-by: Jakub Kicinski <kuba@kernel.org>
Fixes: 94dd016ae538 ("bond: pass get_ts_info and SIOC[SG]HWTSTAMP ioctl to active device")
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet_tstamp: add new flag HWTSTAMP_FLAG_BONDED_PHC_INDEX
Hangbin Liu [Fri, 10 Dec 2021 08:59:58 +0000 (16:59 +0800)]
net_tstamp: add new flag HWTSTAMP_FLAG_BONDED_PHC_INDEX

Since commit 94dd016ae538 ("bond: pass get_ts_info and SIOC[SG]HWTSTAMP
ioctl to active device") the user could get bond active interface's
PHC index directly. But when there is a failover, the bond active
interface will change, thus the PHC index is also changed. This may
break the user's program if they did not update the PHC timely.

This patch adds a new hwtstamp_config flag HWTSTAMP_FLAG_BONDED_PHC_INDEX.
When the user wants to get the bond active interface's PHC, they need to
add this flag and be aware the PHC index may be changed.

With the new flag. All flag checks in current drivers are removed. Only
the checking in net_hwtstamp_validate() is kept.

Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: mtk_eth: add COMPILE_TEST support
Lorenzo Bianconi [Mon, 13 Dec 2021 11:23:22 +0000 (12:23 +0100)]
net: mtk_eth: add COMPILE_TEST support

Improve the build testing of mtk_eth drivers by enabling them when
COMPILE_TEST is selected. Moreover COMPILE_TEST will be useful
for the driver development.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dev: Always serialize on Qdisc::busylock in __dev_xmit_skb() on PREEMPT_RT.
Sebastian Andrzej Siewior [Mon, 13 Dec 2021 10:53:29 +0000 (11:53 +0100)]
net: dev: Always serialize on Qdisc::busylock in __dev_xmit_skb() on PREEMPT_RT.

The root-lock is dropped before dev_hard_start_xmit() is invoked and after
setting the __QDISC___STATE_RUNNING bit. If the Qdisc owner is preempted
by another sender/task with a higher priority then this new sender won't
be able to submit packets to the NIC directly instead they will be
enqueued into the Qdisc. The NIC will remain idle until the Qdisc owner
is scheduled again and finishes the job.

By serializing every task on the ->busylock then the task will be
preempted by a sender only after the Qdisc has no owner.

Always serialize on the busylock on PREEMPT_RT.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agomt76: remove variable set but not used
Yang Li [Mon, 13 Dec 2021 09:54:13 +0000 (17:54 +0800)]
mt76: remove variable set but not used

The code that uses variable queued has been removed,
and "mt76_is_usb(dev) ? q->ndesc - q->queued : q->queued"
didn't do anything, so all they should be removed as well.

Eliminate the following clang warnings:
drivers/net/wireless/mediatek/mt76/debugfs.c:77:9: warning: variable
‘queued’ set but not used.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Fixes: 2d8be76c1674 ("mt76: debugfs: improve queue node readability")
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: bonding: debug: avoid printing debug logs when bond is not notifying peers
Suresh Kumar [Mon, 13 Dec 2021 05:47:09 +0000 (11:17 +0530)]
net: bonding: debug: avoid printing debug logs when bond is not notifying peers

Currently "bond_should_notify_peers: slave ..." messages are printed whenever
"bond_should_notify_peers" function is called.

+++
Dec 12 12:33:26 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:26 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:26 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:26 node1 kernel: bond0: (slave enp0s25): Received LACPDU on port 1
Dec 12 12:33:26 node1 kernel: bond0: (slave enp0s25): Rx Machine: Port=1, Last State=6, Curr State=6
Dec 12 12:33:26 node1 kernel: bond0: (slave enp0s25): partner sync=1
Dec 12 12:33:26 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:26 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:26 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
...
Dec 12 12:33:30 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:30 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:30 node1 kernel: bond0: (slave enp4s3): Received LACPDU on port 2
Dec 12 12:33:30 node1 kernel: bond0: (slave enp4s3): Rx Machine: Port=2, Last State=6, Curr State=6
Dec 12 12:33:30 node1 kernel: bond0: (slave enp4s3): partner sync=1
Dec 12 12:33:30 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:30 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
Dec 12 12:33:30 node1 kernel: bond0: bond_should_notify_peers: slave enp0s25
+++

This is confusing and can also clutter up debug logs.
Print logs only when the peer notification happens.

Signed-off-by: Suresh Kumar <suresh2514@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: ocelot: use dma_unmap_addr to get tx buffer dma_addr
Clément Léger [Mon, 13 Dec 2021 08:26:51 +0000 (09:26 +0100)]
net: ocelot: use dma_unmap_addr to get tx buffer dma_addr

dma_addr was declared using DEFINE_DMA_UNMAP_ADDR() which requires to
use dma_unmap_addr() to access it.

Reported-by: kernel test robot <lkp@intel.com>
Fixes: 753a026cfec1 ("net: ocelot: add FDMA support")
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: lan966x: Fix the configuration of the pcs
Horatiu Vultur [Sat, 11 Dec 2021 21:44:20 +0000 (22:44 +0100)]
net: lan966x: Fix the configuration of the pcs

When inserting a SFP that runs at 2.5G, then the Serdes was still
configured to run at 1G. Because the config->speed was 0, and then the
speed of the serdes was not configured at all, it was using the default
value which is 1G. This patch stop calling the serdes function set_speed
and allow the serdes to figure out the speed based on the interface
type.

Fixes: d28d6d2e37d10d ("net: lan966x: add port module support")
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoselftests/net: expand gro with two machine test
Willem de Bruijn [Sat, 11 Dec 2021 20:04:57 +0000 (15:04 -0500)]
selftests/net: expand gro with two machine test

The test is currently run on a single host with private addresses,
either over veth or by setting a nic in loopback mode with macvlan.

Support running between two real devices. Allow overriding addresses.

Also cut timeout to fail faster on error and explicitly log success.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoMerge branch 'mse102x-support'
David S. Miller [Mon, 13 Dec 2021 14:15:41 +0000 (14:15 +0000)]
Merge branch 'mse102x-support'

Stefan Wahren says:

====================
add Vertexcom MSE102x support

This patch series adds support for the Vertexcom MSE102x Homeplug GreenPHY
chips [1]. They can be connected either via RGMII, RMII or SPI to a host CPU.
These patches handles only the last one, with an Ethernet over SPI protocol
driver.

The code has been tested only on Raspberry Pi boards, but should work
on other platforms.

Changes in V3:
- drop IF_PORT_HOMEPLUG again, since it's actually not used

Changes in V2:
- improve lock handling for RX & TX path
- add new patch to introduce IF_PORT_HOMEPLUG as suggested by Andrew Lunn
- address all the comments by Jakub Kicinski, Andrew Lunn, Kernel test robot

[1] - http://www.vertexcom.com/p_homeplug_plc_en.html
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: vertexcom: Add MSE102x SPI support
Stefan Wahren [Sun, 12 Dec 2021 14:50:27 +0000 (15:50 +0100)]
net: vertexcom: Add MSE102x SPI support

This implements an SPI protocol driver for Vertexcom MSE102x
Homeplug GreenPHY chip.

Signed-off-by: Stefan Wahren <stefan.wahren@i2se.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agodt-bindings: net: add Vertexcom MSE102x support
Stefan Wahren [Sun, 12 Dec 2021 14:50:26 +0000 (15:50 +0100)]
dt-bindings: net: add Vertexcom MSE102x support

Add devicetree binding for the Vertexcom MSE102x Homeplug GreenPHY chip
as SPI device.

Signed-off-by: Stefan Wahren <stefan.wahren@i2se.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agodt-bindings: add vendor Vertexcom
Stefan Wahren [Sun, 12 Dec 2021 14:50:25 +0000 (15:50 +0100)]
dt-bindings: add vendor Vertexcom

Add vendor prefix for Vertexcom Technologies, Inc [1].

[1] - http://www.vertexcom.com/

Signed-off-by: Stefan Wahren <stefan.wahren@i2se.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: mvneta: mark as a legacy_pre_march2020 driver
Russell King (Oracle) [Sun, 12 Dec 2021 13:01:21 +0000 (13:01 +0000)]
net: mvneta: mark as a legacy_pre_march2020 driver

mvneta provides mac_an_restart and mac_pcs_get_state methods, so needs
to be marked as a legacy driver. Marek spotted that mvneta had stopped
working in 2500base-X mode - thanks for reporting.

Reported-by: Marek Behún <kabel@kernel.org>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: axienet: mark as a legacy_pre_march2020 driver
Russell King (Oracle) [Sun, 12 Dec 2021 13:01:15 +0000 (13:01 +0000)]
net: axienet: mark as a legacy_pre_march2020 driver

axienet has a PCS, but does not make use of the phylink PCS support.
Mark it was a pre-March 2020 driver.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoisdn: cpai: no need to initialise statics to 0
Jason Wang [Sun, 12 Dec 2021 07:12:04 +0000 (15:12 +0800)]
isdn: cpai: no need to initialise statics to 0

Static variables do not need to be initialised to 0, because compiler
will initialise all uninitialised statics to 0. Thus, remove the
unneeded initializations.

Signed-off-by: Jason Wang <wangborong@cdjrlc.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: ipa: fix IPA v4.5 interconnect data
Alex Elder [Fri, 10 Dec 2021 22:31:23 +0000 (16:31 -0600)]
net: ipa: fix IPA v4.5 interconnect data

Update the definition of the IPA interconnects for IPA v4.5 so
the path between IPA and system memory is represented by a single
"memory" interconnect.

Tested-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoARM: dts: qcom: sdx55: fix IPA interconnect definitions
Alex Elder [Fri, 10 Dec 2021 22:31:22 +0000 (16:31 -0600)]
ARM: dts: qcom: sdx55: fix IPA interconnect definitions

The first two interconnects defined for IPA on the SDX55 SoC are
really two parts of what should be represented as a single path
between IPA and system memory.

Fix this by combining the "memory-a" and "memory-b" interconnects
into a single "memory" interconnect.

Reported-by: David Heidelberg <david@ixit.cz>
Tested-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: Enable max_dgram_qlen unix sysctl to be configurable by non-init user namespaces
Joanne Koong [Fri, 10 Dec 2021 20:40:23 +0000 (12:40 -0800)]
net: Enable max_dgram_qlen unix sysctl to be configurable by non-init user namespaces

This patch enables the "/proc/sys/net/unix/max_dgram_qlen" sysctl to be
exposed to non-init user namespaces. max_dgram_qlen is used as the default
"sk_max_ack_backlog" value for when a unix socket is created.

Currently, when a networking namespace is initialized, its unix sysctls
are exposed only if the user namespace that "owns" it is the init user
namespace. If there is an non-init user namespace that "owns" a networking
namespace (for example, in the case after we call clone() with both
CLONE_NEWUSER and CLONE_NEWNET set), the sysctls are hidden from view
and not configurable.

Exposing the unix sysctl is safe because any changes made to it will be
limited in scope to the networking namespace the non-init user namespace
"owns" and has privileges over (changes won't affect any other net
namespace). There is also no possibility of a non-privileged user namespace
messing up the net namespace sysctls it shares with its parent user namespace.
When a new user namespace is created without unsharing the network namespace
(eg calling clone()  with CLONE_NEWUSER), the new user namespace shares its
parent's network namespace. Write access is protected by the mode set
in the sysctl's ctl_table (and enforced by procfs). Here in the case of
"max_dgram_qlen", 0644 is set; only the user owner has write access.

v1 -> v2:
* Add more detail to commit message, specify the
"/proc/sys/net/unix/max_dgram_qlen" sysctl in commit message.

Signed-off-by: Joanne Koong <joannekoong@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agou64_stats: Disable preemption on 32bit UP+SMP PREEMPT_RT during updates.
Sebastian Andrzej Siewior [Fri, 10 Dec 2021 20:29:59 +0000 (21:29 +0100)]
u64_stats: Disable preemption on 32bit UP+SMP PREEMPT_RT during updates.

On PREEMPT_RT the seqcount_t for synchronisation is required on 32bit
architectures even on UP because the softirq (and the threaded IRQ handler) can
be preempted.

With the seqcount_t for synchronisation, a reader with higher priority can
preempt the writer and then spin endlessly in read_seqcount_begin() while the
writer can't make progress.

To avoid such a lock up on PREEMPT_RT the writer must disable preemption during
the update. There is no need to disable interrupts because no writer is using
this API in hard-IRQ context on PREEMPT_RT.

Disable preemption on 32bit-RT within the u64_stats write section.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoMerge branch 'bareudp-remove-unused'
David S. Miller [Mon, 13 Dec 2021 12:34:10 +0000 (12:34 +0000)]
Merge branch 'bareudp-remove-unused'

Guillaume Nault says:

====================
bareudp: Remove unused code from header file

Stop exporting unused functions and structures in bareudp.h. The only
piece of bareudp.h that is actually used is netif_is_bareudp(). The
rest can be moved to bareudp.c or even dropped entirely.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agobareudp: Move definition of struct bareudp_conf to bareudp.c
Guillaume Nault [Fri, 10 Dec 2021 19:56:39 +0000 (20:56 +0100)]
bareudp: Move definition of struct bareudp_conf to bareudp.c

This structure is used only in bareudp.c.

While there, adjust include files: we need netdevice.h, not skbuff.h.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agobareudp: Remove bareudp_dev_create()
Guillaume Nault [Fri, 10 Dec 2021 19:56:36 +0000 (20:56 +0100)]
bareudp: Remove bareudp_dev_create()

There's no user for this function.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agotipc: discard MSG_CRYPTO msgs when key_exchange_enabled is not set
Xin Long [Fri, 10 Dec 2021 18:50:40 +0000 (13:50 -0500)]
tipc: discard MSG_CRYPTO msgs when key_exchange_enabled is not set

When key_exchange is disabled, there is no reason to accept MSG_CRYPTO
msgs if it doesn't send MSG_CRYPTO msgs.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: stmmac: bump tc when get underflow error from DMA descriptor
Xiaoliang Yang [Wed, 8 Dec 2021 10:06:51 +0000 (18:06 +0800)]
net: stmmac: bump tc when get underflow error from DMA descriptor

In DMA threshold mode, frame underflow errors may sometimes occur when
the TC(threshold control) value is not enough. The TC value need to be
bumped up in this case.

There is no underflow interrupt bit on DMA_CH(#i)_Status of dwmac4, so
the DMA threshold cannot be bumped up in stmmac_dma_interrupt(). The
i.mx8mp board observed an underflow error while running NFS boot, the
NFS rootfs could not be mounted.

The underflow error can be got from the DMA descriptor TDES3 on dwmac4.
This patch bump up tc value once underflow error is got from TDES3.

Signed-off-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoMerge branch 'dsa-tagger-storage'
David S. Miller [Sun, 12 Dec 2021 12:51:34 +0000 (12:51 +0000)]
Merge branch 'dsa-tagger-storage'

Vladimir Oltean says:

====================
Replace DSA dp->priv with tagger-owned storage

Ansuel's recent work on qca8k register access over Ethernet:
https://patchwork.kernel.org/project/netdevbpf/cover/20211207145942.7444-1-ansuelsmth@gmail.com/
has triggered me to do something which I should've done for a longer
time:
https://patchwork.kernel.org/project/netdevbpf/patch/20211109095013.27829-7-martin.kaistra@linutronix.de/#24585521
which is to replace dp->priv with something that has less caveats.

The dp->priv was introduced when sja1105 needed to hold stateful
information in the tagging protocol driver. In that design, dp->priv
held memory allocated by the switch driver, because the tagging protocol
driver design was 100% stateless.

Some years have passed and others have started to feel the need for
stateful information kept by the tagger, as well as passing data back
and forth between the tagging protocol driver and the switch driver.
This isn't possible cleanly in DSA due to a circular dependency which
leads to broken module autoloading:
https://lore.kernel.org/netdev/20210908220834.d7gmtnwrorhharna@skbuf/

This patchset introduces a framework that resembles something normal,
which allows data to be passed from the tagging protocol driver (things
like switch management packets, which aren't intended for the network
stack) to the switch driver, while the tagging protocol still remains
more or less stateless. The overall design of the framework was
discussed with Ansuel too and it appears to be flexible enough to cover
the "register access over Ethernet" use case. Additionally, the existing
uses of dp->priv, which have mainly to do with PTP timestamping, have
also been migrated.

Changes in v2:
Fix transient build breakage in patch 5/11 due to a missing parenthesis,
https://patchwork.hopto.org/static/nipa/592567/12665213/build_clang/
and another transient build warning in patch 4/11 that for some reason
doesn't appear in my W=1 C=1 build.
https://patchwork.hopto.org/static/nipa/592567/12665209/build_clang/stderr
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: remove dp->priv
Vladimir Oltean [Thu, 9 Dec 2021 23:34:47 +0000 (01:34 +0200)]
net: dsa: remove dp->priv

All current in-tree uses of dp->priv have been replaced with
ds->tagger_data, which provides for a safer API especially when the
connection isn't the regular 1:1 link between one switch driver and one
tagging protocol driver, but could be either one switch to many taggers,
or many switches to one tagger.

Therefore, we can remove this unused pointer.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: tag_sja1105: split sja1105_tagger_data into private and public sections
Vladimir Oltean [Thu, 9 Dec 2021 23:34:46 +0000 (01:34 +0200)]
net: dsa: tag_sja1105: split sja1105_tagger_data into private and public sections

The sja1105 driver messes with the tagging protocol's state when PTP RX
timestamping is enabled/disabled. This is fundamentally necessary
because the tagger needs to know what to do when it receives a PTP
packet. If RX timestamping is enabled, then a metadata follow-up frame
is expected, and this holds the (partial) timestamp. So the tagger plays
hide-and-seek with the network stack until it also gets the metadata
frame, and then presents a single packet, the timestamped PTP packet.
But when RX timestamping isn't enabled, there is no metadata frame
expected, so the hide-and-seek game must be turned off and the packet
must be delivered right away to the network stack.

Considering this, we create a pseudo isolation by devising two tagger
methods callable by the switch: one to get the RX timestamping state,
and one to set it. Since we can't export symbols between the tagger and
the switch driver, these methods are exposed through function pointers.

After this change, the public portion of the sja1105_tagger_data
contains only function pointers.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoRevert "net: dsa: move sja1110_process_meta_tstamp inside the tagging protocol driver"
Vladimir Oltean [Thu, 9 Dec 2021 23:34:45 +0000 (01:34 +0200)]
Revert "net: dsa: move sja1110_process_meta_tstamp inside the tagging protocol driver"

This reverts commit 6d709cadfde68dbd12bef12fcced6222226dcb06.

The above change was done to avoid calling symbols exported by the
switch driver from the tagging protocol driver.

With the tagger-owned storage model, we have a new option on our hands,
and that is for the switch driver to provide a data consumer handler in
the form of a function pointer inside the ->connect_tag_protocol()
method. Having a function pointer avoids the problems of the exported
symbols approach.

By creating a handler for metadata frames holding TX timestamps on
SJA1110, we are able to eliminate an skb queue from the tagger data, and
replace it with a simple, and stateless, function pointer. This skb
queue is now handled exclusively by sja1105_ptp.c, which makes the code
easier to follow, as it used to be before the reverted patch.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: tag_sja1105: convert to tagger-owned data
Vladimir Oltean [Thu, 9 Dec 2021 23:34:44 +0000 (01:34 +0200)]
net: dsa: tag_sja1105: convert to tagger-owned data

Currently, struct sja1105_tagger_data is a part of struct
sja1105_private, and is used by the sja1105 driver to populate dp->priv.

With the movement towards tagger-owned storage, the sja1105 driver
should not be the owner of this memory.

This change implements the connection between the sja1105 switch driver
and its tagging protocol, which means that sja1105_tagger_data no longer
stays in dp->priv but in ds->tagger_data, and that the sja1105 driver
now only populates the sja1105_port_deferred_xmit callback pointer.
The kthread worker is now the responsibility of the tagger.

The sja1105 driver also alters the tagger's state some more, especially
with regard to the PTP RX timestamping state. This will be fixed up a
bit in further changes.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: sja1105: move ts_id from sja1105_tagger_data
Vladimir Oltean [Thu, 9 Dec 2021 23:34:43 +0000 (01:34 +0200)]
net: dsa: sja1105: move ts_id from sja1105_tagger_data

The TX timestamp ID is incremented by the SJA1110 PTP timestamping
callback (->port_tx_timestamp) for every packet, when cloning it.
It isn't used by the tagger at all, even though it sits inside the
struct sja1105_tagger_data.

Also, serialization to this structure is currently done through
tagger_data->meta_lock, which is a cheap hack because the meta_lock
isn't used for anything else on SJA1110 (sja1105_rcv_meta_state_machine
isn't called).

This change moves ts_id from sja1105_tagger_data to sja1105_private and
introduces a dedicated spinlock for it, also in sja1105_private.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: sja1105: make dp->priv point directly to sja1105_tagger_data
Vladimir Oltean [Thu, 9 Dec 2021 23:34:42 +0000 (01:34 +0200)]
net: dsa: sja1105: make dp->priv point directly to sja1105_tagger_data

The design of the sja1105 tagger dp->priv is that each port has a
separate struct sja1105_port, and the sp->data pointer points to a
common struct sja1105_tagger_data.

We have removed all per-port members accessible by the tagger, and now
only struct sja1105_tagger_data remains. Make dp->priv point directly to
this.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: sja1105: remove hwts_tx_en from tagger data
Vladimir Oltean [Thu, 9 Dec 2021 23:34:41 +0000 (01:34 +0200)]
net: dsa: sja1105: remove hwts_tx_en from tagger data

This tagger property is in fact not used at all by the tagger, only by
the switch driver. Therefore it makes sense to be moved to
sja1105_private.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: sja1105: bring deferred xmit implementation in line with ocelot-8021q
Vladimir Oltean [Thu, 9 Dec 2021 23:34:40 +0000 (01:34 +0200)]
net: dsa: sja1105: bring deferred xmit implementation in line with ocelot-8021q

When the ocelot-8021q driver was converted to deferred xmit as part of
commit 8d5f7954b7c8 ("net: dsa: felix: break at first CPU port during
init and teardown"), the deferred implementation was deliberately made
subtly different from what sja1105 has.

The implementation differences lied on the following observations:

- There might be a race between these two lines in tag_sja1105.c:

       skb_queue_tail(&sp->xmit_queue, skb_get(skb));
       kthread_queue_work(sp->xmit_worker, &sp->xmit_work);

  and the skb dequeue logic in sja1105_port_deferred_xmit(). For
  example, the xmit_work might be already queued, however the work item
  has just finished walking through the skb queue. Because we don't
  check the return code from kthread_queue_work, we don't do anything if
  the work item is already queued.

  However, nobody will take that skb and send it, at least until the
  next timestampable skb is sent. This creates additional (and
  avoidable) TX timestamping latency.

  To close that race, what the ocelot-8021q driver does is it doesn't
  keep a single work item per port, and a skb timestamping queue, but
  rather dynamically allocates a work item per packet.

- It is also unnecessary to have more than one kthread that does the
  work. So delete the per-port kthread allocations and replace them with
  a single kthread which is global to the switch.

This change brings the two implementations in line by applying those
observations to the sja1105 driver as well.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: sja1105: let deferred packets time out when sent to ports going down
Vladimir Oltean [Thu, 9 Dec 2021 23:34:39 +0000 (01:34 +0200)]
net: dsa: sja1105: let deferred packets time out when sent to ports going down

This code is not necessary and complicates the conversion of this driver
to tagger-owned memory. If there is a PTP packet that is sent
concurrently with the port getting disabled, the deferred xmit mechanism
is robust enough to time out when it sees that it hasn't been delivered,
and recovers.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: tag_ocelot: convert to tagger-owned data
Vladimir Oltean [Thu, 9 Dec 2021 23:34:38 +0000 (01:34 +0200)]
net: dsa: tag_ocelot: convert to tagger-owned data

The felix driver makes very light use of dp->priv, and the tagger is
effectively stateless. dp->priv is practically only needed to set up a
callback to perform deferred xmit of PTP and STP packets using the
ocelot-8021q tagging protocol (the main ocelot tagging protocol makes no
use of dp->priv, although this driver sets up dp->priv irrespective of
actual tagging protocol in use).

struct felix_port (what used to be pointed to by dp->priv) is removed
and replaced with a two-sided structure. The public side of this
structure, visible to the switch driver, is ocelot_8021q_tagger_data.
The private side is ocelot_8021q_tagger_private, and the latter
structure physically encapsulates the former. The public half of the
tagger data structure can be accessed through a helper of the same name
(ocelot_8021q_tagger_data) which also sanity-checks the protocol
currently in use by the switch. The public/private split was requested
by Andrew Lunn.

Suggested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: introduce tagger-owned storage for private and shared data
Vladimir Oltean [Thu, 9 Dec 2021 23:34:37 +0000 (01:34 +0200)]
net: dsa: introduce tagger-owned storage for private and shared data

Ansuel is working on register access over Ethernet for the qca8k switch
family. This requires the qca8k tagging protocol driver to receive
frames which aren't intended for the network stack, but instead for the
qca8k switch driver itself.

The dp->priv is currently the prevailing method for passing data back
and forth between the tagging protocol driver and the switch driver.
However, this method is riddled with caveats.

The DSA design allows in principle for any switch driver to return any
protocol it desires in ->get_tag_protocol(). The dsa_loop driver can be
modified to do just that. But in the current design, the memory behind
dp->priv has to be allocated by the switch driver, so if the tagging
protocol is paired to an unexpected switch driver, we may end up in NULL
pointer dereferences inside the kernel, or worse (a switch driver may
allocate dp->priv according to the expectations of a different tagger).

The latter possibility is even more plausible considering that DSA
switches can dynamically change tagging protocols in certain cases
(dsa <-> edsa, ocelot <-> ocelot-8021q), and the current design lends
itself to mistakes that are all too easy to make.

This patch proposes that the tagging protocol driver should manage its
own memory, instead of relying on the switch driver to do so.
After analyzing the different in-tree needs, it can be observed that the
required tagger storage is per switch, therefore a ds->tagger_data
pointer is introduced. In principle, per-port storage could also be
introduced, although there is no need for it at the moment. Future
changes will replace the current usage of dp->priv with ds->tagger_data.

We define a "binding" event between the DSA switch tree and the tagging
protocol. During this binding event, the tagging protocol's ->connect()
method is called first, and this may allocate some memory for each
switch of the tree. Then a cross-chip notifier is emitted for the
switches within that tree, and they are given the opportunity to fix up
the tagger's memory (for example, they might set up some function
pointers that represent virtual methods for consuming packets).
Because the memory is owned by the tagger, there exists a ->disconnect()
method for the tagger (which is the place to free the resources), but
there doesn't exist a ->disconnect() method for the switch driver.
This is part of the design. The switch driver should make minimal use of
the public part of the tagger data, and only after type-checking it
using the supplied "proto" argument.

In the code there are in fact two binding events, one is the initial
event in dsa_switch_setup_tag_protocol(). At this stage, the cross chip
notifier chains aren't initialized, so we call each switch's connect()
method by hand. Then there is dsa_tree_bind_tag_proto() during
dsa_tree_change_tag_proto(), and here we have an old protocol and a new
one. We first connect to the new one before disconnecting from the old
one, to simplify error handling a bit and to ensure we remain in a valid
state at all times.

Co-developed-by: Ansuel Smith <ansuelsmth@gmail.com>
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: dsa: mv88e6xxx: Add tx fwd offload PVT on intermediate devices
Tobias Waldekranz [Thu, 9 Dec 2021 22:24:24 +0000 (23:24 +0100)]
net: dsa: mv88e6xxx: Add tx fwd offload PVT on intermediate devices

In a typical mv88e6xxx switch tree like this:

  CPU
   |    .----.
.--0--. | .--0--.
| sw0 | | | sw1 |
'-1-2-' | '-1-2-'
    '---'

If sw1p{1,2} are added to a bridge that sw0p1 is not a part of, sw0
still needs to add a crosschip PVT entry for the virtual DSA device
assigned to represent the bridge.

Fixes: ce5df6894a57 ("net: dsa: mv88e6xxx: map virtual bridges with forwarding offload in the PVT")
Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agonet: Enable neighbor sysctls that is save for userns root
xu xin [Wed, 8 Dec 2021 08:58:44 +0000 (08:58 +0000)]
net: Enable neighbor sysctls that is save for userns root

Inside netns owned by non-init userns, sysctls about ARP/neighbor is
currently not visible and configurable.

For the attributes these sysctls correspond to, any modifications make
effects on the performance of networking(ARP, especilly) only in the
scope of netns, which does not affect other netns.

Actually, some tools via netlink can modify these attribute. iproute2 is
an example. see as follows:

$ unshare -ur -n
$ cat /proc/sys/net/ipv4/neigh/lo/retrans_time
cat: can't open '/proc/sys/net/ipv4/neigh/lo/retrans_time': No such file
or directory
$ ip ntable show dev lo
inet arp_cache
    dev lo
    refcnt 1 reachable 19494 base_reachable 30000 retrans 1000
    gc_stale 60000 delay_probe 5000 queue 101
    app_probes 0 ucast_probes 3 mcast_probes 3
    anycast_delay 1000 proxy_delay 800 proxy_queue 64 locktime 1000

inet6 ndisc_cache
    dev lo
    refcnt 1 reachable 42394 base_reachable 30000 retrans 1000
    gc_stale 60000 delay_probe 5000 queue 101
    app_probes 0 ucast_probes 3 mcast_probes 3
    anycast_delay 1000 proxy_delay 800 proxy_queue 64 locktime 0
$ ip ntable change name arp_cache dev <if> retrans 2000
inet arp_cache
    dev lo
    refcnt 1 reachable 22917 base_reachable 30000 retrans 2000
    gc_stale 60000 delay_probe 5000 queue 101
    app_probes 0 ucast_probes 3 mcast_probes 3
    anycast_delay 1000 proxy_delay 800 proxy_queue 64 locktime 1000

inet6 ndisc_cache
    dev lo
    refcnt 1 reachable 35524 base_reachable 30000 retrans 1000
    gc_stale 60000 delay_probe 5000 queue 101
    app_probes 0 ucast_probes 3 mcast_probes 3
    anycast_delay 1000 proxy_delay 800 proxy_queue 64 locktime 0

Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: xu xin <xu.xin16@zte.com.cn>
Acked-by: Joanne Koong <joannekoong@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 years agoMerge branch 'add-fdma-support-on-ocelot-switch-driver'
Jakub Kicinski [Sat, 11 Dec 2021 04:56:59 +0000 (20:56 -0800)]
Merge branch 'add-fdma-support-on-ocelot-switch-driver'

Clément Léger says:

====================
Add FDMA support on ocelot switch driver

This series adds support for the Frame DMA present on the VSC7514
switch. The FDMA is able to extract and inject packets on the various
ethernet interfaces present on the switch.
====================

Link: https://lore.kernel.org/r/20211209154911.3152830-1-clement.leger@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: add FDMA support
Clément Léger [Thu, 9 Dec 2021 15:49:11 +0000 (16:49 +0100)]
net: ocelot: add FDMA support

Ethernet frames can be extracted or injected autonomously to or from
the device’s DDR3/DDR3L memory and/or PCIe memory space. Linked list
data structures in memory are used for injecting or extracting Ethernet
frames. The FDMA generates interrupts when frame extraction or
injection is done and when the linked lists need updating.

The FDMA is shared between all the ethernet ports of the switch and
uses a linked list of descriptors (DCB) to inject and extract packets.
Before adding descriptors, the FDMA channels must be stopped. It would
be inefficient to do that each time a descriptor would be added so the
channels are restarted only once they stopped.

Both channels uses ring-like structure to feed the DCBs to the FDMA.
head and tail are never touched by hardware and are completely handled
by the driver. On top of that, page recycling has been added and is
mostly taken from gianfar driver.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Co-developed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: add support for ndo_change_mtu
Clément Léger [Thu, 9 Dec 2021 15:49:10 +0000 (16:49 +0100)]
net: ocelot: add support for ndo_change_mtu

This commit adds support for changing MTU for the ocelot register based
interface. For ocelot, JUMBO frame size can be set up to 25000 bytes
but has been set to 9000 which is a saner value and allows for maximum
gain of performance with FDMA.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: add and export ocelot_ptp_rx_timestamp()
Clément Léger [Thu, 9 Dec 2021 15:49:09 +0000 (16:49 +0100)]
net: ocelot: add and export ocelot_ptp_rx_timestamp()

In order to support PTP in FDMA, PTP handling code is needed. Since
this is the same as for register-based extraction, export it with
a new ocelot_ptp_rx_timestamp() function.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2 years agonet: ocelot: export ocelot_ifh_port_set() to setup IFH
Clément Léger [Thu, 9 Dec 2021 15:49:08 +0000 (16:49 +0100)]
net: ocelot: export ocelot_ifh_port_set() to setup IFH

FDMA will need this code to prepare the injection frame header when
sending SKBs. Move this code into ocelot_ifh_port_set() and add
conditional IFH setting for vlan and rew op if they are not set.

Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>