Currently, when a VLAN entry is added multiple times in a row to a
bridge port, nbp_vlan_add() calls br_switchdev_port_vlan_add() each
time, even if the VLAN already exists and nothing about it has changed:

bridge vlan add dev lan12 vid 100 master static

Similarly, when a VLAN is added multiple times in a row to a bridge,
br_vlan_add_existing() doesn't filter at all the calls to
br_switchdev_port_vlan_add():

bridge vlan add dev br0 vid 100 self

This behavior makes driver-level accounting of VLANs impossible, since
it is enough for a single deletion event to remove a VLAN, but the
addition event can be emitted an unlimited number of times.

The cause for this can be identified as follows: we rely on
__vlan_add_flags() to retroactively tell us whether it has changed
anything about the VLAN flags or VLAN group pvid. So we'd first have to
call __vlan_add_flags() before calling br_switchdev_port_vlan_add(), in
order to have access to the "bool *changed" information. But we don't
want to change the event ordering, because we'd have to revert the
struct net_bridge_vlan changes we've made if switchdev returns an error.

So to solve this, we need another function that tells us whether any
change is going to occur in the VLAN or VLAN group, _prior_ to calling
__vlan_add_flags().

Split __vlan_add_flags() into a precommit and a commit stage, and rename
it to __vlan_flags_update(). The precommit stage,
__vlan_flags_would_change(), will determine whether there is any reason
to notify switchdev due to a change of flags (note: the BRENTRY flag
transition from false to true is treated separately: as a new switchdev
entry, because we skipped notifying the master VLAN when it wasn't a
brentry yet, and therefore not as a change of flags).

With this lookahead/precommit function in place, we can avoid notifying
switchdev if nothing changed for the VLAN and VLAN group.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently there is a very subtle aspect to the behavior of
__vlan_add_flags(): it changes the struct net_bridge_vlan flags and
pvid, yet it returns true ("changed") even if none of those changed,
just a transition of br_vlan_is_brentry(v) took place from false to
true.

This can be seen in br_vlan_add_existing(), however we do not actually
rely on this subtle behavior, since the "if" condition that checks that
the vlan wasn't a brentry before had a useless (until now) assignment:

	*changed = true;

Make things more obvious by actually making __vlan_add_flags() do what's
written on the box, and be more specific about what is actually written
on the box. This is needed because further transformations will be done
to __vlan_add_flags().
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When a VLAN is added to a bridge port and it doesn't exist on the bridge
device yet, it gets created for the multicast context, but it is
'hidden', since it doesn't have the BRENTRY flag yet:

ip link add br0 type bridge && ip link set swp0 master br0
bridge vlan add dev swp0 vid 100 # the master VLAN 100 gets created
bridge vlan add dev br0 vid 100 self # that VLAN becomes brentry just now

All switchdev drivers ignore switchdev notifiers for VLAN entries which
have the BRENTRY unset, and for good reason: these are merely private
data structures used by the bridge driver. So we might just as well not
notify those at all.

Cleanup in the switchdev drivers that check for the BRENTRY flag is now
possible, and will be handled separately, since those checks just became
dead code.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When a VLAN is added to a bridge port, a master VLAN gets created on the
bridge for context, but it doesn't have the BRENTRY flag.

Then, when the same VLAN is added to the bridge itself, that enters
through the br_vlan_add_existing() code path and gains the BRENTRY flag,
thus it becomes "existing".

It seems natural to check for this condition early, because the current
code flow is to notify switchdev of the addition of a VLAN that isn't a
brentry, just to delete it immediately afterwards.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The commit
a5886ef4 ("gve: Introduce per netdev `enum gve_queue_format`")
introduces three queue format type, only GVE_GQI_QPL_FORMAT queue has
page list. So it should use the queue page list number to detect the
zero size queue page list. Correct the design logic.

Using the 'queue_format == GVE_GQI_RDA_FORMAT' may lead to request zero
sized memory allocation, like if the queue format is GVE_DQO_RDA_FORMAT.

The kernel memory subsystem will return ZERO_SIZE_PTR, which is not NULL
address, so the driver can run successfully. Also the code still checks
the queue page list number firstly, then accesses the allocated memory,
so zero number queue page list allocation will not lead to access fault.
Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
Reviewed-by: Bailey Forrest <bcf@google.com>
Link: https://lore.kernel.org/r/20220215051751.260866-1-haiyue.wang@intel.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

There are spelling mistakes in debug messages. Fix them.
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fix the following coccicheck warnings:
./drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c:1199:42-47: WARNING:
conversion to bool not needed here
./drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c:1218:54-59: WARNING:
conversion to bool not needed here
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If the following call path returns an error from switchdev:

nbp_vlan_flush
-> __vlan_del
   -> __vlan_vid_del
      -> br_switchdev_port_vlan_del
-> __vlan_group_free
   -> WARN_ON(!list_empty(&vg->vlan_list));

then the deletion of the net_bridge_vlan is silently halted, which will
trigger the WARN_ON from __vlan_group_free().

The WARN_ON is rather unhelpful, because nothing about the source of the
error is printed. Add a print to catch errors from __vlan_del.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

hso_create_device() is only called from function that already use
GFP_KERNEL. And all the callers are called from the probe function.

So there is no need here to explicitly require a GFP_ATOMIC when
allocating memory.

Use GFP_KERNEL instead.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch fixes the checkpatch.pl warning:

ERROR: code indent should use tabs where possible #3453: FILE: drivers/net/virtio_net.c:3453: ret = register_virtio_driver(&virtio_net_driver);$

Uneccessary newline was also removed making line 3453 now 3452.
Signed-off-by: Michael Catanzaro <mcatanzaro.kernel@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Saeed Mahameed says:

====================
mlx5-updates-2022-02-14

mlx5 TX routines improvements

1) From Aya and Tariq, first 3 patches, Use the Max size of the TX descriptor
as advertised by the device and not the fixed value of 16 that the driver
always assumed, this is not a bug fix as all existing devices have Max value
larger than 16, but the series is necessary for future proofing the driver.

2) TX Synchronization improvements from Maxim, last 12 patches

Maxim Mikityanskiy Says:
=======================
mlx5e: Synchronize ndo_select_queue with configuration changes

The kernel can call ndo_select_queue at any time, and there is no direct
way to block it. The implementation of ndo_select_queue in mlx5e expects
the parameters to be consistent and may crash (invalid pointer, division
by zero) if they aren't.

There were attempts to partially fix some of the most frequent crashes,
see commit 846d6da1 ("net/mlx5e: Fix division by 0 in
mlx5e_select_queue") and commit 84c8a874 ("net/mlx5e: Fix division
by 0 in mlx5e_select_queue for representors"). However, they don't
address the issue completely.

This series introduces the proper synchronization mechanism between
mlx5e configuration and TX data path:

1. txq2sq updates are synchronized properly with ndo_start_xmit
   (mlx5e_xmit). The TX queue is stopped when it configuration is being
   updated, and memory barriers ensure the changes are visible before
   restarting.

2. The set of parameters needed for mlx5e_select_queue is reduced, and
   synchronization using RCU is implemented. This way, changes are
   atomic, and the state in mlx5e_select_queue is always consistent.

3. A few optimizations are applied to the new implementation of
   mlx5e_select_queue.

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

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

Check all booleans for special queues at once, when deciding whether to
go to the fast path in mlx5e_select_queue. Pack them into bitfields to
have some room for extensibility.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

To improve the performance of the modulo operation (%), it's replaced by
a subtracting the divisor in a loop. The modulo is used to fix up an
out-of-bounds value that might be returned by netdev_pick_tx or to
convert the queue number to the channel number when num_tcs > 1. Both
situations are unlikely, because XPS is configured not to pick higher
queues (qid >= num_channels) by default, so under normal circumstances
the flow won't go inside the loop, and it will be faster than %.

num_tcs == 8 adds at most 7 iterations to the loop. PTP adds at most 1
iteration to the loop. HTB would add at most 256 iterations (when
num_channels == 1), so there is an additional boundary check in the HTB
flow, which falls back to % if more than 7 iterations are expected.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

This commit optimizes mlx5e_select_queue for HTB and PTP cases by
short-cutting some checks, without sacrificing performance of the common
non-HTB non-PTP flow.

1. The HTB flow uses the fact that num_tcs == 1 to drop these checks
(it's not possible to attach both mqprio and htb as the root qdisc).
It's also enough to calculate `txq_ix % num_channels` only once, instead
of twice.

2. The PTP flow drops the check for HTB and the second calculation of
`txq_ix % num_channels`.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

trust_state can be written while mlx5e_select_queue() is reading it. To
avoid inconsistencies, use READ_ONCE and WRITE_ONCE for access and
updates, and touch the variable only once per operation.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

Both mlx5e_select_queue and mlx5e_select_ptpsq contain the same logic to
get user priority of a packet, according to the current trust state
settings. This commit moves this repeating code to its own function.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

Start using the select queue parameters introduced in the previous
commit to have proper synchronization with changing the configuration
(such as number of channels and queues). It ensures that the state that
mlx5e_select_queue() sees is always consistent and stays the same while
the function is running. Also it allows mlx5e_select_queue to stop using
data structures that weren't synchronized properly: txq2sq,
channel_tc2realtxq, port_ptp_tc2realtxq. The last two are removed
completely, as they were used only in mlx5e_select_queue.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

This commit moves mlx5e_select_queue and all stuff related to
ndo_select_queue to en/selq.c to put all stuff working with selq into a
separate file.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

ndo_select_queue can be called at any time, and there is no way to stop
the kernel from calling it to synchronize with configuration changes
(real_num_tx_queues, num_tc). This commit introduces an internal way in
mlx5e to sync mlx5e_select_queue() with these changes. The configuration
needed by this function is stored in a struct mlx5e_selq_params, which
is modified and accessed in an atomic way using RCU methods. The whole
ndo_select_queue is called under an RCU lock, providing the necessary
guarantees.

The parameters stored in the new struct mlx5e_selq_params should only be
used from inside mlx5e_select_queue. It's the minimal set of parameters
needed for mlx5e_select_queue to do its job efficiently, derived from
parameters stored elsewhere. That means that when the configuration
change, mlx5e_selq_params may need to be updated. In such cases, the
mlx5e_selq_prepare/mlx5e_selq_apply API should be used.

struct mlx5e_selq contains two slots for the params: active and standby.
mlx5e_selq_prepare updates the standby slot, and mlx5e_selq_apply swaps
the slots in a safe atomic way using the RCU API. It integrates well
with the open/activate stages of the configuration change flow.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

This commit makes necessary changes to guarantee that txq2sq remains
stable while mlx5e_xmit is running. Proper synchronization is added for
HTB TX queues.

All updates to txq2sq are performed while the corresponding queue is
disabled (i.e. mlx5e_xmit doesn't run on that queue). smp_wmb after each
change guarantees that mlx5e_xmit can see the updated value after the
queue is enabled. Comments explaining this mechanism are added to
mlx5e_xmit.

When an HTB SQ can be deleted (after deleting an HTB node), synchronize
with RCU to wait for mlx5e_select_queue to finish and stop selecting
that queue, before we re-enable it to avoid TX timeout watchdog alarms.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

mlx5e_build_txq_maps updates txq2sq while TX queues are stopped. Add a
barrier to ensure that these changes are visible before the queues are
started and mlx5e_xmit reads from txq2sq.

This commit handles regular TX queues. Synchronization between HTB TX
queues and mlx5e_xmit is handled in the following commit.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

Normally, the queues are disabled when the channels are deactivated, and
enabled when the channels are activated. However, on register, the
channels are not active, but the queues are enabled by default. This
change fixes it, preventing mlx5e_xmit from running when the channels
are deactivated in the beginning.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

mlx5e_activate_priv_channels() and mlx5e_deactivate_priv_channels()
start and stop all netdev TX queues. This commit removes the unneeded
call to netif_tx_stop_all_queues and adds explanatory comments why these
operations are needed.

netif_tx_disable() does the same thing that netif_tx_stop_all_queues(),
but taking the TX lock, thus guaranteeing that ndo_start_xmit is not
running after return. That means that the netif_tx_stop_all_queues()
call is not really necessary.

The comments are improved: the TX watchdog timeout explanation is moved
to the start stage where it really belongs (it used to be in both
places, but was lost during some old refactoring) and rephrased in more
details; the explanation for stopping all TX queues is added.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

Calculate maximal count of MPW WQEBBs on SQ's creation and store it
there. Remove MLX5E_TX_MPW_MAX_NUM_DS and MLX5E_TX_MPW_MAX_WQEBBS.
Update mlx5e_tx_mpwqe_is_full() and mlx5e_xdp_mpqwe_is_full() .
Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

Prior to this patch the maximal value for max WQEBBs (WQE Basic Blocks,
where WQE is a Work Queue Element) on the TX side was assumed to be 16
(fixed value). All firmware versions till today comply to this. In order
to be more flexible and resilient, read from FW the corresponding:
max_wqe_sz_sq. This value describes the maximum WQE size given in bytes,
thus max WQEBBs is given by the division in WQEBB's byte size. The
driver uses the top between 16 and the division result. This ensures
synchronization between driver and firmware and avoids unexpected
behavior. Store this value on the different SQs (Send Queues) for easy
access.
Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

Remove tstamp pointer in mlx5e_txqsq as it's no longer used after
commit 7c39afb3 ("net/mlx5: PTP code migration to driver core section").
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Aya Levin <ayal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

These chips have 8 built-in FE PHYs and 3 SERDES interfaces that can
run at 1G. With the blamed commit, the built-in PHYs could no longer
be connected to, using an MII PHY interface mode.

Create a separate .phylink_get_caps callback for these chips, which
takes the FE/GE split into consideration.

Fixes: 2ee84cfe ("net: dsa: mv88e6xxx: convert to phylink_generic_validate()")
Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/20220213185154.3262207-1-tobias@waldekranz.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

There is a spelling mistake in an error message. Fix it.
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support of rule offloading added to the non-zero index chain,
which was previously forbidden. Also, goto action is offloaded
allowing to jump for processing of desired chain.

Note that only implicit chain 0 is bound to the device port(s) for
processing. The rest of chains have to be jumped by actions.
Signed-off-by: Volodymyr Mytnyk <vmytnyk@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

M Chetan Kumar says:

====================
net: wwan: debugfs dev reference not dropped

This patch series contains WWAN subsystem & IOSM Driver changes to
drop dev reference obtained as part of wwan debugfs dir entry retrieval.

PATCH1: A new debugfs interface is introduced in wwan subsystem so
that wwan driver can drop the obtained dev reference post debugfs use.

PATCH2: IOSM Driver uses new debugfs interface to drop dev reference.

Please refer to commit messages for details.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Post debugfs use call wwan_put_debugfs_dir()to drop
debugfs dev reference.
Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

WWAN driver call's wwan_get_debugfs_dir() to obtain
WWAN debugfs dir entry. As part of this procedure it
returns a reference to a found device.

Since there is no debugfs interface available at WWAN
subsystem, it is not possible to drop dev reference post
debugfs use. This leads to side effects like post wwan
driver load and reload the wwan instance gets increment
from wwanX to wwanX+1.

A new debugfs interface is added in wwan subsystem so that
wwan driver can drop the obtained dev reference post debugfs
use.

void wwan_put_debugfs_dir(struct dentry *dir)
Signed-off-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Luiz Angelo Daros de Luca says:

====================
net: dsa: realtek: realtek-mdio: reset before setup

This patch series cleans the realtek-smi reset code and copy that to the
realtek-mdio.

v1-v2)
- do not run reset code block if GPIO is missing. It was printing "RESET
  deasserted" even when there is no GPIO configured.
- reset switch after dsa_unregister_switch()
- demote reset messages to debug

v2-v3)
- do not assert the reset on gpiod_get. Do it explicitly aferwards.
- split the commit into two (one for each module)
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Some devices, like the switch in Banana Pi BPI R64 only starts to answer
after a HW reset. It is the same reset code from realtek-smi.
Reported-by: Frank Wunderlich <frank-w@public-files.de>
Signed-off-by: Luiz Angelo Daros de Luca <luizluca@gmail.com>
Tested-by: Frank Wunderlich <frank-w@public-files.de>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Acked-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When reset GPIO was missing, the driver was still printing an info
message and still trying to assert the reset. Although gpiod_set_value()
will silently ignore calls with NULL gpio_desc, it is better to make it
clear the driver might allow gpio_desc to be NULL.

The initial value for the reset pin was changed to GPIOD_OUT_LOW,
followed by a gpiod_set_value() asserting the reset. This way, it will
be easier to spot if and where the reset really happens.

A new "asserted RESET" message was added just after the reset is
asserted, similar to the existing "deasserted RESET" message. Both
messages were demoted to dbg. The code comment is not needed anymore.
Signed-off-by: Luiz Angelo Daros de Luca <luizluca@gmail.com>
Acked-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Whenever rt6_uncached_list_flush_dev() swaps rt->rt6_idev
to the blackhole device, parts of IPv6 stack might still need
to increment one SNMP counter.

Root cause, patch from Ido, changelog from Eric :)

This bug suggests that we need to audit rt->rt6_idev usages
and make sure they are properly using RCU protection.

Fixes: e5f80fcf ("ipv6: give an IPv6 dev to blackhole_netdev")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The macro was missed while renaming realtek-smi.h to realtek.h.

Fixes: f5f11907 (net: dsa: realtek: rename realtek_smi to)
Signed-off-by: Luiz Angelo Daros de Luca <luizluca@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Acked-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Sebastian Andrzej Siewior says:

====================
net: dev: PREEMPT_RT fixups.

this series removes or replaces preempt_disable() and local_irq_save()
sections which are problematic on PREEMPT_RT.
Patch 2 makes netif_rx() work from any context after I found suggestions
for it in an old thread. Should that work, then the context-specific
variants could be removed.

v2…v3:
   - #2
     - Export __netif_rx() so it can be used by everyone.
     - Add a lockdep assert to check for interrupt context.
     - Update the kernel doc and mention that the skb is posted to
       backlog NAPI.
     - Use __netif_rx() also in drivers/net/*.c.
     - Added Toke''s review tag and kept Eric's desptite the changes
       made.

v1…v2:
  - #1 and #2
    - merge patch 1 und 2 from the series (as per Toke).
    - updated patch description and corrected the first commit number (as
      per Eric).
   - #2
     - Provide netif_rx() as in v1 and additionally __netif_rx() without
       local_bh disable()+enable() for the loopback driver. __netif_rx() is
       not exported (loopback is built-in only) so it won't be used
       drivers. If this doesn't work then we can still export/ define a
       wrapper as Eric suggested.
     - Added a comment that netif_rx() considered legacy.
   - #3
     - Moved ____napi_schedule() into rps_ipi_queued() and
       renamed it napi_schedule_rps().
   https://lore.kernel.org/all/20220204201259.1095226-1-bigeasy@linutronix.de/

v1:
   https://lore.kernel.org/all/20220202122848.647635-1-bigeasy@linutronix.de
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Disabling interrupts and in the RPS case locking input_pkt_queue is
split into local_irq_disable() and optional spin_lock().

This breaks on PREEMPT_RT because the spinlock_t typed lock can not be
acquired with disabled interrupts.
The sections in which the lock is acquired is usually short in a sense that it
is not causing long und unbounded latiencies. One exception is the
skb_flow_limit() invocation which may invoke a BPF program (and may
require sleeping locks).

By moving local_irq_disable() + spin_lock() into rps_lock(), we can keep
interrupts disabled on !PREEMPT_RT and enabled on PREEMPT_RT kernels.
Without RPS on a PREEMPT_RT kernel, the needed synchronisation happens
as part of local_bh_disable() on the local CPU.
____napi_schedule() is only invoked if sd is from the local CPU. Replace
it with __napi_schedule_irqoff() which already disables interrupts on
PREEMPT_RT as needed. Move this call to rps_ipi_queued() and rename the
function to napi_schedule_rps as suggested by Jakub.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Dave suggested a while ago (eleven years by now) "Let's make netif_rx()
work in all contexts and get rid of netif_rx_ni()". Eric agreed and
pointed out that modern devices should use netif_receive_skb() to avoid
the overhead.
In the meantime someone added another variant, netif_rx_any_context(),
which behaves as suggested.

netif_rx() must be invoked with disabled bottom halves to ensure that
pending softirqs, which were raised within the function, are handled.
netif_rx_ni() can be invoked only from process context (bottom halves
must be enabled) because the function handles pending softirqs without
checking if bottom halves were disabled or not.
netif_rx_any_context() invokes on the former functions by checking
in_interrupts().

netif_rx() could be taught to handle both cases (disabled and enabled
bottom halves) by simply disabling bottom halves while invoking
netif_rx_internal(). The local_bh_enable() invocation will then invoke
pending softirqs only if the BH-disable counter drops to zero.

Eric is concerned about the overhead of BH-disable+enable especially in
regard to the loopback driver. As critical as this driver is, it will
receive a shortcut to avoid the additional overhead which is not needed.

Add a local_bh_disable() section in netif_rx() to ensure softirqs are
handled if needed.
Provide __netif_rx() which does not disable BH and has a lockdep assert
to ensure that interrupts are disabled. Use this shortcut in the
loopback driver and in drivers/net/*.c.
Make netif_rx_ni() and netif_rx_any_context() invoke netif_rx() so they
can be removed once they are no more users left.

Link: https://lkml.kernel.org/r/20100415.020246.218622820.davem@davemloft.netSigned-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>