Support tc trap such that packets can explicitly be forwarded to slow
path if they match a specific rule.

In the example below, we want packets with src IP equals 7.7.7.8 to be
forwarded to software, in which case it will get to the appropriate
representor net device.

$ tc filter add dev eth1 protocol ip prio 1 root flower skip_sw \
    src_ip 7.7.7.8 action trap
Signed-off-by: NEli Cohen <eli@mellanox.com>
Reviewed-by: NRoi Dayan <roid@nvidia.com>
Reviewed-by: NAriel Levkovich <lariel@nvidia.com>
Reviewed-by: NMaor Dickman <maord@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

Multiple features use metadata matching such as bond vport
in live migration, multi-port RoCE mode, stacked devices;
hence, enable vport metadata matching by default.

Fixes: 1e62e222 ("net/mlx5: E-Switch, Use vport metadata matching only when mandatory")
Signed-off-by: NVu Pham <vuhuong@mellanox.com>
Reviewed-by: NBodong Wang <bodong@nvidia.com>
Reviewed-by: NRoi Dayan <roid@nvidia.com>
Reviewed-by: NParav Pandit <parav@nvidia.com>
Reviewed-by: NMark Bloch <mbloch@nvidia.com>
Reviewed-by: NOz Shlomo <ozsh@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

In merged eswitch configuration, peer miss rule is setup for all
vports. If metadata is enabled, peer miss rule with metadata matching
will be configured instead of source port matching; however, some
vports that have not yet been enabled don't have default_metadata
setup and their default_metadata will be zero.

Hence, setup/cleanup default metadata for all vports when eswitch moves
in/out of offloads mode.

Fixes: 133dcfc5 ("net/mlx5: E-Switch, Alloc and free unique metadata for match")
Signed-off-by: NVu Pham <vuhuong@mellanox.com>
Reviewed-by: NBodong Wang <bodong@nvidia.com>
Reviewed-by: NRoi Dayan <roid@nvidia.com>
Reviewed-by: NMark Bloch <mbloch@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

Uplink vport must have a dedicated metadata with vhca_id
being part of the metadata.

Fixes: 133dcfc5 ("net/mlx5: E-Switch, Alloc and free unique metadata for match")
Signed-off-by: NVu Pham <vuhuong@mellanox.com>
Reviewed-by: NMark Bloch <mbloch@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

Check E-Switch capabilities and enable metadata support flag
before using it to setup other features that need metadata.
Signed-off-by: NVu Pham <vuhuong@mellanox.com>
Reviewed-by: NMark Bloch <mbloch@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

LAG offload can't be enabled if the enslaved PF is not lag master,
which is indicated by HCA capabilities bit. It is cleared if more than
64 VFs are configured for this PF.

Previously, a data structure is created to store lag info, including
PFs to be enslaved, then a handler is registered for netdev notifier.
However, this initialization is skipped if PF is not lag master. So
PF can't handle CHANGEUPPER event from upper bond device. Even worse,
PF is enslaved silently, and LAG offload is not activated.

Fix this by registering netdev notifier for PFs which are not lag
masters. When CHANGEUPPER event is received, and both physical ports
(and only them) on the same NIC are about to be enslaved, a warning is
returned for user to know it.
Signed-off-by: NJianbo Liu <jianbol@mellanox.com>
Reviewed-by: NRaed Salem <raeds@mellanox.com>
Reviewed-by: NRoi Dayan <roid@mellanox.com>
Reviewed-by: NJiri Pirko <jiri@nvidia.com>
Reviewed-by: NRaed Salem <raeds@nvidia.com>
Reviewed-by: NRoi Dayan <roid@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

If bond tx type is not active-backup or hash, LAG offload can't be
enabled. When CHANGEUPPER event is received, and both PFs (and only
them) under the same lag master are about to be enslaved, a warning
is returned for user to know the offload failure, otherwise PFs are
enslaved silently without LAG offload activated.
Signed-off-by: NJianbo Liu <jianbol@mellanox.com>
Reviewed-by: NRaed Salem <raeds@mellanox.com>
Reviewed-by: NRoi Dayan <roid@mellanox.com>
Reviewed-by: NJiri Pirko <jiri@nvidia.com>
Reviewed-by: NRaed Salem <raeds@nvidia.com>
Reviewed-by: NRoi Dayan <roid@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

Change the return value to -ENOENT, to make it more meaningful.
Signed-off-by: NJianbo Liu <jianbol@mellanox.com>
Reviewed-by: NJiri Pirko <jiri@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

Before calling timecounter_cyc2time(), clock->lock must be taken.
Use mlx5_timecounter_cyc2time instead which guarantees a safe access.

Fixes: afc98a0b ("net/mlx5: Update ptp_clock_event foreach PPS event")
Signed-off-by: NEran Ben Elisha <eranbe@mellanox.com>

Holding the clock lock is not required when scheduling a PPS work.
Signed-off-by: NEran Ben Elisha <eranbe@mellanox.com>
Reviewed-by: NMoshe Shemesh <moshe@mellanox.com>

Fix a typo in ptp_clock_info naming: mlx5_p2p -> mlx5_ptp.
Signed-off-by: NEran Ben Elisha <eranbe@mellanox.com>

Clock struct is part of struct mlx5_core_dev. Code was inconsistent, on
some cases used container_of and on another used clock->mdev.

Align code to use container_of amd remove clock->mdev pointer.
While here, fix reverse xmas tree coding style.
Signed-off-by: NEran Ben Elisha <eranbe@mellanox.com>
Reviewed-by: NMoshe Shemesh <moshe@mellanox.com>

This isn't a fall through because it was after a return statement.  The
fall through annotation leads to a Smatch warning:

    drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c:246
    mlx5e_ethtool_get_sset_count() warn: ignoring unreachable code.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: NGustavo A. R. Silva <gustavoars@kernel.org>

Add variable initialization to eliminate the warning
"variable may be used uninitialized".

Fixes: 5f29458b ("net/mlx5e: Support dump callback in TX reporter")
Signed-off-by: NMoshe Tal <moshet@mellanox.com>
Reviewed-by: NAya Levin <ayal@nvidia.com>
Signed-off-by: NSaeed Mahameed <saeedm@nvidia.com>

Clean and rebuild the debugfs info for the queues being swapped.

Fixes: a34e25ab ("ionic: change the descriptor ring length without full reset")
Signed-off-by: NShannon Nelson <snelson@pensando.io>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A DSA master interface has upper network devices, each representing an
Ethernet switch port attached to it. Demultiplexing the source ports and
setting skb->dev accordingly is done through the catch-all ETH_P_XDSA
packet_type handler. Catch-all because DSA vendors have various header
implementations, which can be placed anywhere in the frame: before the
DMAC, before the EtherType, before the FCS, etc. So, the ETH_P_XDSA
handler acts like an rx_handler more than anything.

It is unlikely for the DSA master interface to have any other upper than
the DSA switch interfaces themselves. Only maybe a bridge upper*, but it
is very likely that the DSA master will have no 8021q upper. So
__netif_receive_skb_core() will try to untag the VLAN, despite the fact
that the DSA switch interface might have an 8021q upper. So the skb will
never reach that.

So far, this hasn't been a problem because most of the possible
placements of the DSA switch header mentioned in the first paragraph
will displace the VLAN header when the DSA master receives the frame, so
__netif_receive_skb_core() will not actually execute any VLAN-specific
code for it. This only becomes a problem when the DSA switch header does
not displace the VLAN header (for example with a tail tag).

What the patch does is it bypasses the untagging of the skb when there
is a DSA switch attached to this net device. So, DSA is the only
packet_type handler which requires seeing the VLAN header. Once skb->dev
will be changed, __netif_receive_skb_core() will be invoked again and
untagging, or delivery to an 8021q upper, will happen in the RX of the
DSA switch interface itself.

*see commit 9eb8eff0 ("net: bridge: allow enslaving some DSA master
network devices". This is actually the reason why I prefer keeping DSA
as a packet_type handler of ETH_P_XDSA rather than converting to an
rx_handler. Currently the rx_handler code doesn't support chaining, and
this is a problem because a DSA master might be bridged.
Signed-off-by: NVladimir Oltean <olteanv@gmail.com>
Reviewed-by: NFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Landen Chao says:

====================
net-next: dsa: mt7530: add support for MT7531

This patch series adds support for MT7531.

MT7531 is the next generation of MT7530 which could be found on Mediatek
router platforms such as MT7622 or MT7629.

It is also a 7-ports switch with 5 giga embedded phys, 2 cpu ports, and
the same MAC logic of MT7530. Cpu port 6 only supports SGMII interface.
Cpu port 5 supports either RGMII or SGMII in different HW SKU, but cannot
be muxed to PHY of port 0/4 like mt7530. Due to support for SGMII
interface, pll, and pad setting are different from MT7530.

MT7531 SGMII interface can be configured in following mode:
- 'SGMII AN mode' with in-band negotiation capability
    which is compatible with PHY_INTERFACE_MODE_SGMII.
- 'SGMII force mode' without in-band negotiation
    which is compatible with 10B/8B encoding of
    PHY_INTERFACE_MODE_1000BASEX with fixed full-duplex and fixed pause.
- 2.5 times faster clocked 'SGMII force mode' without in-band negotiation
    which is compatible with 10B/8B encoding of
    PHY_INTERFACE_MODE_2500BASEX with fixed full-duplex and fixed pause.

v4 -> v5
- Add fixed-link node to dsa cpu port in dts file by suggestion of
  Vladimir Oltean.

v3 -> v4
- Adjust the coding style by suggestion of Jakub Kicinski.
  Remove unnecessary jumping label, merge continuous numeric 'switch
  cases' into one line, and keep the variables longest to shortest
  (reverse xmas tree).

v2 -> v3
- Keep the same setup logic of mt7530/mt7621 because these series of
  patches is for adding mt7531 hardware.
- Do not adjust rgmii delay when vendor phy driver presents in order to
  prevent double adjustment by suggestion of Andrew Lunn.
- Remove redundant 'Example 4' from dt-bindings by suggestion of
  Rob Herring.
- Fix typo.

v1 -> v2
- change phylink_validate callback function to support full-duplex
  gigabit only to match hardware capability.
- add description of SGMII interface.
- configure mt7531 cpu port in fastest speed by default.
- parse SGMII control word for in-band negotiation mode.
- configure RGMII delay based on phy.rst.
- Rename the definition in the header file to avoid potential conflicts.
- Add wrapper function for mdio read/write to support both C22 and C45.
- correct fixed-link speed of 2500base-x in dts.
- add MT7531 port mirror setting.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add mt7531 dsa to bananapi-bpi-r64 board for 5 giga Ethernet ports support.
Signed-off-by: NLanden Chao <landen.chao@mediatek.com>
Tested-By: NFrank Wunderlich <frank-w@public-files.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add mt7531 dsa to mt7622-rfb1 board for 5 giga Ethernet ports support.
mt7622 only supports 1 sgmii interface, so either gmac0 or gmac1 can be
configured as sgmii interface. In this patch, change to connect mt7622
gmac0 and mt7531 port6 through sgmii interface.
Signed-off-by: NLanden Chao <landen.chao@mediatek.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add new support for MT7531:

MT7531 is the next generation of MT7530. It is also a 7-ports switch with
5 giga embedded phys, 2 cpu ports, and the same MAC logic of MT7530. Cpu
port 6 only supports SGMII interface. Cpu port 5 supports either RGMII
or SGMII in different HW sku, but cannot be muxed to PHY of port 0/4 like
mt7530. Due to SGMII interface support, pll, and pad setting are different
from MT7530. This patch adds different initial setting, and SGMII phylink
handlers of MT7531.

MT7531 SGMII interface can be configured in following mode:
- 'SGMII AN mode' with in-band negotiation capability
    which is compatible with PHY_INTERFACE_MODE_SGMII.
- 'SGMII force mode' without in-band negotiation
    which is compatible with 10B/8B encoding of
    PHY_INTERFACE_MODE_1000BASEX with fixed full-duplex and fixed pause.
- 2.5 times faster clocked 'SGMII force mode' without in-band negotiation
    which is compatible with 10B/8B encoding of
    PHY_INTERFACE_MODE_2500BASEX with fixed full-duplex and fixed pause.
Signed-off-by: NLanden Chao <landen.chao@mediatek.com>
Signed-off-by: NSean Wang <sean.wang@mediatek.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add devicetree binding to support the compatible mt7531 switch as used
in the MediaTek MT7531 switch.
Signed-off-by: NSean Wang <sean.wang@mediatek.com>
Signed-off-by: NLanden Chao <landen.chao@mediatek.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add a structure holding required operations for each device such as device
initialization, PHY port read or write, a checker whether PHY interface is
supported on a certain port, MAC port setup for either bus pad or a
specific PHY interface.

The patch is done for ready adding a new hardware MT7531, and keep the
same setup logic of existing hardware.
Signed-off-by: NLanden Chao <landen.chao@mediatek.com>
Signed-off-by: NSean Wang <sean.wang@mediatek.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Refine message in Kconfig with fixing typo and an explicit MT7621 support.
Signed-off-by: NLanden Chao <landen.chao@mediatek.com>
Signed-off-by: NSean Wang <sean.wang@mediatek.com>
Reviewed-by: NFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

x25_type_trans only needs to be called before we call netif_rx to pass
the skb to upper layers.

It does not need to be called before lapb_data_received. The LAPB module
does not need the fields that are set by calling it.

In the other two X.25 drivers - lapbether and hdlc_x25. x25_type_trans
is only called before netif_rx and not before lapb_data_received.

Cc: Martin Schiller <ms@dev.tdt.de>
Signed-off-by: NXie He <xie.he.0141@gmail.com>
Acked-by: NMartin Schiller <ms@dev.tdt.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

flush_all_backlogs() may cause deadlock on systems
running processes with FIFO scheduling policy.

The above is critical in -RT scenarios, where user-space
specifically ensure no network activity is scheduled on
the CPU running the mentioned FIFO process, but still get
stuck.

This commit tries to address the problem checking the
backlog status on the remote CPUs before scheduling the
flush operation. If the backlog is empty, we can skip it.

v1 -> v2:
 - explicitly clear flushed cpu mask - Eric
Signed-off-by: NPaolo Abeni <pabeni@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Ido Schimmel says:

====================
mlxsw: Derive SBIB from maximum port speed & MTU

Petr says:

Internal buffer is a part of port headroom used for packets that are
mirrored due to triggers that the Spectrum ASIC considers "egress". Besides
ACL mirroring on port egresss this includes also packets mirrored due to
ECN marking.

This patchset changes the way the internal mirroring buffer is reserved.
Currently the buffer reflects port MTU and speed accurately. In the future,
mlxsw should support dcbnl_setbuffer hook to allow the users to set buffer
sizes by hand. In that case, there might not be enough space for growth of
the internal mirroring buffer due to MTU and speed changes. While vetoing
MTU changes would be merely confusing, port speed changes cannot be vetoed,
and such change would simply lead to issues in packet mirroring.

For these reasons, with these patches the internal mirroring buffer is
derived from maximum MTU and maximum speed achievable on the port.

Patches #1 and #2 introduce a new callback to determine the maximum speed a
given port can achieve.

With patches #3 and #4, the information about, respectively, maximum MTU
and maximum port speed, is kept in struct mlxsw_sp_port.

In patch #5, maximum MTU and maximum speed are used to determine the size
of the internal buffer. MTU update and speed update hooks are dropped,
because they are no longer necessary.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The SBIB register configures the size of an internal buffer that the
Spectrum ASICs use when mirroring traffic on egress. This size should be
taken into account when validating that the port headroom buffers are not
larger than the chip can handle. Up until now this was not done, which is
incidentally not a problem, because the priority group buffers that mlxsw
auto-configures are small enough that the boundary condition could not be
violated.

However when dcbnl_setbuffer is implemented, the user has control over
sizes of PG buffers, and they might overshoot the headroom capacity.
However the size of the SBIB buffer depends on port speed, and that cannot
be vetoed. Therefore SBIB size should be deduced from maximum port speed.

Additionally, once the buffers are configured by hand, the user could get
into an uncomfortable situation where their MTU change requests get vetoed,
because the SBIB does not fit anymore. Therefore derive SBIB size from
maximum permissible MTU as well.

Remove all the code that adjusted the SBIB size whenever speed or MTU
changed.
Signed-off-by: NPetr Machata <petrm@nvidia.com>
Signed-off-by: NIdo Schimmel <idosch@nvidia.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The maximum port speed depends on link modes supported by the port, and for
Ethernet ports is constant. The maximum speed will be handy when setting
SBIB, the internal buffer used for traffic mirroring. Therefore, keep it in
struct mlxsw_sp_port for easy access.
Signed-off-by: NPetr Machata <petrm@nvidia.com>
Signed-off-by: NIdo Schimmel <idosch@nvidia.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The maximum port MTU depends on port type. On Spectrum, mlxsw configures
all ports as Ethernet ports, and the maximum MTU therefore never changes.
Besides checking MTU configuration, maximum MTU will also be handy when
setting SBIB, the internal buffer used for traffic mirroring. Therefore,
keep it in struct mlxsw_sp_port for easy access.
Signed-off-by: NPetr Machata <petrm@nvidia.com>
Signed-off-by: NIdo Schimmel <idosch@nvidia.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The SBIB register configures the size of an internal buffer that the
Spectrum ASICs use when mirroring traffic on egress. This size should be
taken into account when validating that the port headroom buffers are not
larger than the chip can handle. Up until now this was not done, which is
incidentally not a problem, because the priority group buffers that mlxsw
auto-configures are small enough that the boundary condition could not be
violated.

When dcbnl_setbuffer is implemented, the user gets control over sizes of PG
buffers, and they might overshoot the headroom capacity. However the size
of the SBIB buffer depends on port speed, which cannot be vetoed. There is
obviously no way to retroactively push back on requests for overlarge PG
buffers, or reject an overlarge MTU, or cancel losslessness of a certain
PG.

Therefore, instead of taking into account the current speed when
calculating SBIB buffer size, take into account the maximum speed that a
port with given Ethernet protocol capabilities can have.

To that end, add a new ethtool callback, ptys_max_speed, which determines
this maximum speed.
Signed-off-by: NPetr Machata <petrm@nvidia.com>
Signed-off-by: NIdo Schimmel <idosch@nvidia.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In order to allow reusing the logic, extract from
mlxsw_sp_port_get_link_ksettings() the code to obtain Ethernet protocol
attributes, mlxsw_sp_port_ptys_query().
Signed-off-by: NPetr Machata <petrm@nvidia.com>
Signed-off-by: NIdo Schimmel <idosch@nvidia.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Tony Nguyen says:

====================
40GbE Intel Wired LAN Driver Updates 2020-09-14

This series contains updates to i40e driver only.

Li RongQing removes binding affinity mask to a fixed CPU and sets
prefetch of Rx buffer page to occur conditionally.

Björn provides AF_XDP performance improvements by not prefetching HW
descriptors, using 16 byte descriptors, and moving buffer allocation
out of Rx processing loop.

v2: Define prefetch_page_address in a common header for patch 2.
Dropped, previous, patch 5 as it is being reworked to be more
generalized.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

David Howells says:

====================
rxrpc: Fixes for the connection manager rewrite

Here are some fixes for the connection manager rewrite:

 (1) Fix a goto to the wrong place in error handling.

 (2) Fix a missing NULL pointer check.

 (3) The stored allocation error needs to be stored signed.

 (4) Fix a leak of connection bundle when clearing connections due to
     net namespace exit.

 (5) Fix an overget of the bundle when setting up a new client conn.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add NETIF_F_GSO_UDP_TUNNEL and NETIF_F_GSO_UDP_TUNNEL_CSUM features
to support vxlan segmentation and checksum offload. Ipip and ipv6
tunnel packets are regarded as non-tunnel pkt for hw and as for other
type of tunnel pkts, checksum offload is disabled.
Signed-off-by: NLuo bin <luobin9@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixes the following W=1 kernel build warning(s):

drivers/net/ethernet/qlogic/qlcnic/qlcnic_83xx_hw.c:661:6: warning:
 variable 'val' set but not used [-Wunused-but-set-variable]
  661 |  u32 val;
      |      ^~~

After commit 7f966452 ("qlcnic: 83xx memory map and HW access
routines"), variable 'val' is never used in qlcnic_83xx_cam_unlock(), so
removing it to avoid build warning.
Reported-by: NHulk Robot <hulkci@huawei.com>
Signed-off-by: NZhang Changzhong <zhangchangzhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixes the following W=1 kernel build warning(s):

drivers/net/ethernet/marvell/pxa168_eth.c:1190:6: warning:
 variable 'retval' set but not used [-Wunused-but-set-variable]
 1190 |  int retval;
      |      ^~~~~~

Function pxa168_eth_change_mtu() always return zero, so variable 'retval'
is redundant, just remove it.
Reported-by: NHulk Robot <hulkci@huawei.com>
Signed-off-by: NZhang Changzhong <zhangchangzhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixes the following W=1 kernel build warning(s):

drivers/net/ethernet/freescale/fec_ptp.c:523:6: warning:
 variable 'ns' set but not used [-Wunused-but-set-variable]
  523 |  u64 ns;
      |      ^~

After commit 6605b730 ("FEC: Add time stamping code and a PTP
hardware clock"), variable 'ns' is never used in fec_time_keep(),
so removing it to avoid build warning.
Reported-by: NHulk Robot <hulkci@huawei.com>
Signed-off-by: NZhang Changzhong <zhangchangzhong@huawei.com>
Acked-by: NFugang Duan <fugang.duan@nxp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixes the following W=1 kernel build warning(s):

drivers/net/ethernet/dnet.c:510:6: warning:
 variable 'tx_status' set but not used [-Wunused-but-set-variable]
  u32 tx_status, irq_enable;
      ^~~~~~~~~

After commit 47964174 ("dnet: Dave DNET ethernet controller driver
(updated)"), variable 'tx_status' is never used in dnet_start_xmit(),
so removing it to avoid build warning.
Reported-by: NHulk Robot <hulkci@huawei.com>
Signed-off-by: NZhang Changzhong <zhangchangzhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

SOCK_QUEUE_SHRUNK is currently used by TCP as a temporary state
that remembers if some room has been made in the rtx queue
by an incoming ACK packet.

This is later used from tcp_check_space() before
considering to send EPOLLOUT.

Problem is: If we receive SACK packets, and no packet
is removed from RTX queue, we can send fresh packets, thus
moving them from write queue to rtx queue and eventually
empty the write queue.

This stall can happen if TCP_NOTSENT_LOWAT is used.

With this fix, we no longer risk stalling sends while holes
are repaired, and we can fully use socket sndbuf.

This also removes a cache line dirtying for typical RPC
workloads.

Fixes: c9bee3b7 ("tcp: TCP_NOTSENT_LOWAT socket option")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This comment is outdated and no longer reflects the actual implementation
of af_packet.c.

Reasons for the new comment:

1.

In af_packet.c, the function packet_snd first reserves a headroom of
length (dev->hard_header_len + dev->needed_headroom).
Then if the socket is a SOCK_DGRAM socket, it calls dev_hard_header,
which calls dev->header_ops->create, to create the link layer header.
If the socket is a SOCK_RAW socket, it "un-reserves" a headroom of
length (dev->hard_header_len), and checks if the user has provided a
header sized between (dev->min_header_len) and (dev->hard_header_len)
(in dev_validate_header).
This shows the developers of af_packet.c expect hard_header_len to
be consistent with header_ops.

2.

In af_packet.c, the function packet_sendmsg_spkt has a FIXME comment.
That comment states that prepending an LL header internally in a driver
is considered a bug. I believe this bug can be fixed by setting
hard_header_len to 0, making the internal header completely invisible
to af_packet.c (and requesting the headroom in needed_headroom instead).

3.

There is a commit for a WiFi driver:
commit 9454f7a8 ("mwifiex: set needed_headroom, not hard_header_len")
According to the discussion about it at:
  https://patchwork.kernel.org/patch/11407493/
The author tried to set the WiFi driver's hard_header_len to the Ethernet
header length, and request additional header space internally needed by
setting needed_headroom.
This means this usage is already adopted by driver developers.

Cc: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Brian Norris <briannorris@chromium.org>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: NXie He <xie.he.0141@gmail.com>
Acked-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>