platform/kernel/linux-rpi.git
19 months agonet/packet: convert po->tp_loss to an atomic flag
Eric Dumazet [Thu, 16 Mar 2023 01:10:11 +0000 (01:10 +0000)]
net/packet: convert po->tp_loss to an atomic flag

tp_loss can be read locklessly.

Convert it to an atomic flag to avoid races.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet/packet: convert po->tp_tx_has_off to an atomic flag
Eric Dumazet [Thu, 16 Mar 2023 01:10:10 +0000 (01:10 +0000)]
net/packet: convert po->tp_tx_has_off to an atomic flag

This is to use existing space in po->flags, and reclaim
the storage used by the non atomic bit fields.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet/packet: annotate accesses to po->tp_tstamp
Eric Dumazet [Thu, 16 Mar 2023 01:10:09 +0000 (01:10 +0000)]
net/packet: annotate accesses to po->tp_tstamp

tp_tstamp is read locklessly.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet/packet: convert po->auxdata to an atomic flag
Eric Dumazet [Thu, 16 Mar 2023 01:10:08 +0000 (01:10 +0000)]
net/packet: convert po->auxdata to an atomic flag

po->auxdata can be read while another thread
is changing its value, potentially raising KCSAN splat.

Convert it to PACKET_SOCK_AUXDATA flag.

Fixes: 8dc419447415 ("[PACKET]: Add optional checksum computation for recvmsg")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet/packet: convert po->origdev to an atomic flag
Eric Dumazet [Thu, 16 Mar 2023 01:10:07 +0000 (01:10 +0000)]
net/packet: convert po->origdev to an atomic flag

syzbot/KCAN reported that po->origdev can be read
while another thread is changing its value.

We can avoid this splat by converting this field
to an actual bit.

Following patches will convert remaining 1bit fields.

Fixes: 80feaacb8a64 ("[AF_PACKET]: Add option to return orig_dev to userspace.")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet/packet: annotate accesses to po->xmit
Eric Dumazet [Thu, 16 Mar 2023 01:10:06 +0000 (01:10 +0000)]
net/packet: annotate accesses to po->xmit

po->xmit can be set from setsockopt(PACKET_QDISC_BYPASS),
while read locklessly.

Use READ_ONCE()/WRITE_ONCE() to avoid potential load/store
tearing issues.

Fixes: d346a3fae3ff ("packet: introduce PACKET_QDISC_BYPASS socket option")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoMerge branch 'gve-xdp-support'
David S. Miller [Fri, 17 Mar 2023 08:29:21 +0000 (08:29 +0000)]
Merge branch 'gve-xdp-support'

Praveen Kaligineedi says:

====================
gve: Add XDP support for GQI-QPL format

Adding support for XDP DROP, PASS, TX, REDIRECT for GQI QPL format.
Add AF_XDP zero-copy support.

When an XDP program is installed, dedicated TX queues are created to
handle XDP traffic. The user needs to ensure that the number of
configured TX queues is equal to the number of configured RX queues; and
the number of TX/RX queues is less than or equal to half the maximum
number of TX/RX queues.

The XDP traffic from AF_XDP sockets and from other NICs (arriving via
XDP_REDIRECT) will also egress through the dedicated XDP TX queues.

Although these changes support AF_XDP socket in zero-copy mode, there is
still a copy happening within the driver between XSK buffer pool and QPL
bounce buffers in GQI-QPL format.

The following example demonstrates how the XDP packets are mapped to
TX queues:

Example configuration:
Max RX queues : 2N, Max TX queues : 2N
Configured RX queues : N, Configured TX queues : N

TX queue mapping:
TX queues with queue id 0,...,N-1 will handle traffic from the stack.
TX queues with queue id N,...,2N-1 will handle XDP traffic.

For the XDP packets transmitted using XDP_TX action:
<Egress TX queue id> = N + <Ingress RX queue id>

For the XDP packets that arrive from other NICs via XDP_REDIRECT action:
<Egress TX queue id> = N + ( smp_processor_id % N )

For AF_XDP zero-copy mode:
<Egress TX queue id> = N + <AF_XDP TX queue id>

Changes in v2:
- Removed gve_close/gve_open when adding XDP dedicated queues. Instead
we add and register additional TX queues when the XDP program is
installed. If the allocation/registration fails we return error and do
not install the XDP program. Added a new patch to enable adding TX queues
without gve_close/gve_open
- Removed xdp tx spin lock from this patch. It is needed for XDP_REDIRECT
support as both XDP_REDIRECT and XDP_TX traffic share the dedicated XDP
queues. Moved the code to add xdp tx spinlock to the subsequent patch
that adds XDP_REDIRECT support.
- Added netdev_err when the user tries to set rx/tx queues to the values
not supported when XDP is enabled.
- Removed rcu annotation for xdp_prog. We disable the napi prior to
adding/removing the xdp_prog and reenable it after the program has
been installed for all the queues.
- Ring the tx doorbell once for napi instead of every XDP TX packet.
- Added a new helper function for freeing the FIFO buffer
- Unregister xdp rxq for all the queues when the registration
fails during XDP program installation
- Register xsk rxq only when XSK buff pool is enabled
- Removed code accessing internal xsk_buff_pool fields
- Removed sleep driven code when disabling XSK buff pool. Disable
napi and re-enable it after disabling XSK pool.
- Make sure that we clean up dma mappings on XSK pool disable
- Use napi_if_scheduled_mark_missed to avoid unnecessary napi move
to the CPU calling ndo_xsk_wakeup()

Changes in v3:
- Padding bytes are used if the XDP TX packet headers do not
fit at tail of TX FIFO. Taking these padding bytes into account
while checking if enough space is available in TX FIFO.

Changes in v4:
- Turn on the carrier based on the link status synchronously rather
than asynchronously when XDP is installed/uninstalled
- Set the supported flags in net_device.xdp_features
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agogve: Add AF_XDP zero-copy support for GQI-QPL format
Praveen Kaligineedi [Wed, 15 Mar 2023 23:33:12 +0000 (16:33 -0700)]
gve: Add AF_XDP zero-copy support for GQI-QPL format

Adding AF_XDP zero-copy support.

Note: Although these changes support AF_XDP socket in zero-copy
mode, there is still a copy happening within the driver between
XSK buffer pool and QPL bounce buffers in GQI-QPL format.
In GQI-QPL queue format, the driver needs to allocate a fixed size
memory, the size specified by vNIC device, for RX/TX and register this
memory as a bounce buffer with the vNIC device when a queue is
created. The number of pages in the bounce buffer is limited and the
pages need to be made available to the vNIC by copying the RX data out
to prevent head-of-line blocking. Therefore, we cannot pass the XSK
buffer pool to the vNIC.

The number of copies on RX path from the bounce buffer to XSK buffer is 2
for AF_XDP copy mode (bounce buffer -> allocated page frag -> XSK buffer)
and 1 for AF_XDP zero-copy mode (bounce buffer -> XSK buffer).

This patch contains the following changes:
1) Enable and disable XSK buffer pool
2) Copy XDP packets from QPL bounce buffers to XSK buffer on rx
3) Copy XDP packets from XSK buffer to QPL bounce buffers and
   ring the doorbell as part of XDP TX napi poll
4) ndo_xsk_wakeup callback support

Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Reviewed-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agogve: Add XDP REDIRECT support for GQI-QPL format
Praveen Kaligineedi [Wed, 15 Mar 2023 23:33:11 +0000 (16:33 -0700)]
gve: Add XDP REDIRECT support for GQI-QPL format

This patch contains the following changes:
1) Support for XDP REDIRECT action on rx
2) ndo_xdp_xmit callback support

In GQI-QPL queue format, the driver needs to allocate a fixed size
memory, the size specified by vNIC device, for RX/TX and register this
memory as a bounce buffer with the vNIC device when a queue is created.
The number of pages in the bounce buffer is limited and the pages need to
be made available to the vNIC by copying the RX data out to prevent
head-of-line blocking. The XDP_REDIRECT packets are therefore immediately
copied to a newly allocated page.

Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Reviewed-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agogve: Add XDP DROP and TX support for GQI-QPL format
Praveen Kaligineedi [Wed, 15 Mar 2023 23:33:10 +0000 (16:33 -0700)]
gve: Add XDP DROP and TX support for GQI-QPL format

Add support for XDP PASS, DROP and TX actions.

This patch contains the following changes:
1) Support installing/uninstalling XDP program
2) Add dedicated XDP TX queues
3) Add support for XDP DROP action
4) Add support for XDP TX action

Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Reviewed-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agogve: Changes to add new TX queues
Praveen Kaligineedi [Wed, 15 Mar 2023 23:33:09 +0000 (16:33 -0700)]
gve: Changes to add new TX queues

Changes to enable adding and removing TX queues without calling
gve_close() and gve_open().

Made the following changes:
1) priv->tx, priv->rx and priv->qpls arrays are allocated based on
   max tx queues and max rx queues
2) Changed gve_adminq_create_tx_queues(), gve_adminq_destroy_tx_queues(),
gve_tx_alloc_rings() and gve_tx_free_rings() functions to add/remove a
subset of TX queues rather than all the TX queues.

Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Reviewed-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agogve: XDP support GQI-QPL: helper function changes
Praveen Kaligineedi [Wed, 15 Mar 2023 23:33:08 +0000 (16:33 -0700)]
gve: XDP support GQI-QPL: helper function changes

This patch adds/modifies helper functions needed to add XDP
support.

Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Reviewed-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoMerge branch 'net-sk_err-lockless-annotate'
David S. Miller [Fri, 17 Mar 2023 08:25:06 +0000 (08:25 +0000)]
Merge branch 'net-sk_err-lockless-annotate'

Eric Dumazet says:

====================
net: annotate lockless accesses to sk_err[_soft]

This patch series is inspired by yet another syzbot report.

Most poll() handlers are lockless and read sk->sk_err
while other cpus can change it.

Add READ_ONCE/WRITE_ONCE() to major/usual offenders.

More to come later.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoaf_unix: annotate lockless accesses to sk->sk_err
Eric Dumazet [Wed, 15 Mar 2023 20:57:46 +0000 (20:57 +0000)]
af_unix: annotate lockless accesses to sk->sk_err

unix_poll() and unix_dgram_poll() read sk->sk_err
without any lock held.

Add relevant READ_ONCE()/WRITE_ONCE() annotations.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agomptcp: annotate lockless accesses to sk->sk_err
Eric Dumazet [Wed, 15 Mar 2023 20:57:45 +0000 (20:57 +0000)]
mptcp: annotate lockless accesses to sk->sk_err

mptcp_poll() reads sk->sk_err without socket lock held/owned.

Add READ_ONCE() and WRITE_ONCE() to avoid load/store tearing.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agotcp: annotate lockless access to sk->sk_err
Eric Dumazet [Wed, 15 Mar 2023 20:57:44 +0000 (20:57 +0000)]
tcp: annotate lockless access to sk->sk_err

tcp_poll() reads sk->sk_err without socket lock held/owned.

We should used READ_ONCE() here, and update writers
to use WRITE_ONCE().

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: annotate lockless accesses to sk->sk_err_soft
Eric Dumazet [Wed, 15 Mar 2023 20:57:43 +0000 (20:57 +0000)]
net: annotate lockless accesses to sk->sk_err_soft

This field can be read/written without lock synchronization.

tcp and dccp have been handled in different patches.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agodccp: annotate lockless accesses to sk->sk_err_soft
Eric Dumazet [Wed, 15 Mar 2023 20:57:42 +0000 (20:57 +0000)]
dccp: annotate lockless accesses to sk->sk_err_soft

This field can be read/written without lock synchronization.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agotcp: annotate lockless accesses to sk->sk_err_soft
Eric Dumazet [Wed, 15 Mar 2023 20:57:41 +0000 (20:57 +0000)]
tcp: annotate lockless accesses to sk->sk_err_soft

This field can be read/written without lock synchronization.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agovlan: partially enable SIOCSHWTSTAMP in container
Vadim Fedorenko [Wed, 15 Mar 2023 15:33:02 +0000 (08:33 -0700)]
vlan: partially enable SIOCSHWTSTAMP in container

Setting timestamp filter was explicitly disabled on vlan devices in
containers because it might affect other processes on the host. But it's
absolutely legit in case when real device is in the same namespace.

Fixes: 873017af7784 ("vlan: disable SIOCSHWTSTAMP in container")
Signed-off-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoMerge branch 'pcs_get_state-fixes'
David S. Miller [Fri, 17 Mar 2023 08:19:03 +0000 (08:19 +0000)]
Merge branch 'pcs_get_state-fixes'

Russell King (Oracle) says:

====================
Minor fixes for pcs_get_state() implementations

This series contains a number fixes for minor issues with some
pcs_get_state() implementations, particualrly for the phylink
state->an_enabled member. As they are minor, I'm suggesting we
queue them in net-next as there is follow-on work for these, and
there is no urgency for them to be in -rc.

Just like phylib, state->advertising's Autoneg bit is a copy of
state->an_enabled, and thus it is my intention to remove
state->an_enabled from phylink to simplify things.

This series gets rid of state->an_enabled assignments or
reporting that should never have been there.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: pcs: lynx: don't print an_enabled in pcs_get_state()
Russell King (Oracle) [Wed, 15 Mar 2023 14:46:49 +0000 (14:46 +0000)]
net: pcs: lynx: don't print an_enabled in pcs_get_state()

an_enabled will be going away, and in any case, pcs_get_state() should
not be updating this member. Remove the print.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Steen Hegelund <Steen.Hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: pcs: xpcs: remove double-read of link state when using AN
Russell King (Oracle) [Wed, 15 Mar 2023 14:46:43 +0000 (14:46 +0000)]
net: pcs: xpcs: remove double-read of link state when using AN

Phylink does not want the current state of the link when reading the
PCS link state - it wants the latched state. Don't double-read the
MII status register. Phylink will re-read as necessary to capture
transient link-down events as of dbae3388ea9c ("net: phylink: Force
retrigger in case of latched link-fail indicator").

The above referenced commit is a dependency for this change, and thus
this change should not be backported to any kernel that does not
contain the above referenced commit.

Fixes: fcb26bd2b6ca ("net: phy: Add Synopsys DesignWare XPCS MDIO module")
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoMerge branch 'vxlan-MDB-support'
David S. Miller [Fri, 17 Mar 2023 08:05:50 +0000 (08:05 +0000)]
Merge branch 'vxlan-MDB-support'

Ido Schimmel says:

====================
vxlan: Add MDB support

tl;dr
=====

This patchset implements MDB support in the VXLAN driver, allowing it to
selectively forward IP multicast traffic to VTEPs with interested
receivers instead of flooding it to all the VTEPs as BUM. The motivating
use case is intra and inter subnet multicast forwarding using EVPN
[1][2], which means that MDB entries are only installed by the user
space control plane and no snooping is implemented, thereby avoiding a
lot of unnecessary complexity in the kernel.

Background
==========

Both the bridge and VXLAN drivers have an FDB that allows them to
forward Ethernet frames based on their destination MAC addresses and
VLAN/VNI. These FDBs are managed using the same PF_BRIDGE/RTM_*NEIGH
netlink messages and bridge(8) utility.

However, only the bridge driver has an MDB that allows it to selectively
forward IP multicast packets to bridge ports with interested receivers
behind them, based on (S, G) and (*, G) MDB entries. When these packets
reach the VXLAN driver they are flooded using the "all-zeros" FDB entry
(00:00:00:00:00:00). The entry either includes the list of all the VTEPs
in the tenant domain (when ingress replication is used) or the multicast
address of the BUM tunnel (when P2MP tunnels are used), to which all the
VTEPs join.

Networks that make heavy use of multicast in the overlay can benefit
from a solution that allows them to selectively forward IP multicast
traffic only to VTEPs with interested receivers. Such a solution is
described in the next section.

Motivation
==========

RFC 7432 [3] defines a "MAC/IP Advertisement route" (type 2) [4] that
allows VTEPs in the EVPN network to advertise and learn reachability
information for unicast MAC addresses. Traffic destined to a unicast MAC
address can therefore be selectively forwarded to a single VTEP behind
which the MAC is located.

The same is not true for IP multicast traffic. Such traffic is simply
flooded as BUM to all VTEPs in the broadcast domain (BD) / subnet,
regardless if a VTEP has interested receivers for the multicast stream
or not. This is especially problematic for overlay networks that make
heavy use of multicast.

The issue is addressed by RFC 9251 [1] that defines a "Selective
Multicast Ethernet Tag Route" (type 6) [5] which allows VTEPs in the
EVPN network to advertise multicast streams that they are interested in.
This is done by having each VTEP suppress IGMP/MLD packets from being
transmitted to the NVE network and instead communicate the information
over BGP to other VTEPs.

The draft in [2] further extends RFC 9251 with procedures to allow
efficient forwarding of IP multicast traffic not only in a given subnet,
but also between different subnets in a tenant domain.

The required changes in the bridge driver to support the above were
already merged in merge commit 8150f0cfb24f ("Merge branch
'bridge-mcast-extensions-for-evpn'"). However, full support entails MDB
support in the VXLAN driver so that it will be able to selectively
forward IP multicast traffic only to VTEPs with interested receivers.
The implementation of this MDB is described in the next section.

Implementation
==============

The user interface is extended to allow user space to specify the
destination VTEP(s) and related parameters. Example usage:

 # bridge mdb add dev vxlan0 port vxlan0 grp 239.1.1.1 permanent dst 198.51.100.1
 # bridge mdb add dev vxlan0 port vxlan0 grp 239.1.1.1 permanent dst 192.0.2.1

 $ bridge -d -s mdb show
 dev vxlan0 port vxlan0 grp 239.1.1.1 permanent filter_mode exclude proto static dst 192.0.2.1    0.00
 dev vxlan0 port vxlan0 grp 239.1.1.1 permanent filter_mode exclude proto static dst 198.51.100.1    0.00

Since the MDB is fully managed by user space and since snooping is not
implemented, only permanent entries can be installed and temporary
entries are rejected by the kernel.

The netlink interface is extended with a few new attributes in the
RTM_NEWMDB / RTM_DELMDB request messages:

[ struct nlmsghdr ]
[ struct br_port_msg ]
[ MDBA_SET_ENTRY ]
struct br_mdb_entry
[ MDBA_SET_ENTRY_ATTRS ]
[ MDBE_ATTR_SOURCE ]
struct in_addr / struct in6_addr
[ MDBE_ATTR_SRC_LIST ]
[ MDBE_SRC_LIST_ENTRY ]
[ MDBE_SRCATTR_ADDRESS ]
struct in_addr / struct in6_addr
[ ...]
[ MDBE_ATTR_GROUP_MODE ]
u8
[ MDBE_ATTR_RTPORT ]
u8
[ MDBE_ATTR_DST ] // new
struct in_addr / struct in6_addr
[ MDBE_ATTR_DST_PORT ] // new
u16
[ MDBE_ATTR_VNI ] // new
u32
[ MDBE_ATTR_IFINDEX ] // new
s32
[ MDBE_ATTR_SRC_VNI ] // new
u32

RTM_NEWMDB / RTM_DELMDB responses and notifications are extended with
corresponding attributes.

One MDB entry that can be installed in the VXLAN MDB, but not in the
bridge MDB is the catchall entry (0.0.0.0 / ::). It is used to transmit
unregistered multicast traffic that is not link-local and is especially
useful when inter-subnet multicast forwarding is required. See patch #12
for a detailed explanation and motivation. It is similar to the
"all-zeros" FDB entry that can be installed in the VXLAN FDB, but not
the bridge FDB.

"added_by_star_ex" entries
--------------------------

The bridge driver automatically installs (S, G) MDB port group entries
marked as "added_by_star_ex" whenever it detects that an (S, G) entry
can prevent traffic from being forwarded via a port associated with an
EXCLUDE (*, G) entry. The bridge will add the port to the port group of
the (S, G) entry, thereby creating a new port group entry. The
complexity associated with these entries is not trivial, but it needs to
reside in the bridge driver because it automatically installs MDB
entries in response to snooped IGMP / MLD packets.

The same in not true for the VXLAN MDB which is entirely managed by user
space who is fully capable of forming the correct replication lists on
its own. In addition, the complexity associated with the
"added_by_star_ex" entries in the VXLAN driver is higher compared to the
bridge: Whenever a remote VTEP is added to the catchall entry, it needs
to be added to all the existing MDB entries, as such a remote requested
all the multicast traffic to be forwarded to it. Similarly, whenever an
(*, G) or (S, G) entry is added, all the remotes associated with the
catchall entry need to be added to it.

Given the above, this patchset does not implement support for such
entries.  One argument against this decision can be that in the future
someone might want to populate the VXLAN MDB in response to decapsulated
IGMP / MLD packets and not according to EVPN routes. Regardless of my
doubts regarding this possibility, it can be implemented using a new
VXLAN device knob that will also enable the "added_by_star_ex"
functionality.

Testing
=======

Tested using existing VXLAN and MDB selftests under "net/" and
"net/forwarding/". Added a dedicated selftest in the last patch.

Patchset overview
=================

Patches #1-#3 are small preparations in the bridge driver. I plan to
submit them separately together with an MDB dump test case.

Patches #4-#6 are additional preparations centered around the extraction
of the MDB netlink handlers from the bridge driver to the common
rtnetlink code. This allows reusing the existing MDB netlink messages
for the configuration of the VXLAN MDB.

Patches #7-#9 include more small preparations in the common rtnetlink
code and the VXLAN driver.

Patch #10 implements the MDB control path in the VXLAN driver, which
will allow user space to create, delete, replace and dump MDB entries.

Patches #11-#12 implement the MDB data path in the VXLAN driver,
allowing it to selectively forward IP multicast traffic according to the
matched MDB entry.

Patch #13 finally enables MDB support in the VXLAN driver.

iproute2 patches can be found here [6].

Note that in order to fully support the specifications in [1] and [2],
additional functionality is required from the data path. However, it can
be achieved using existing kernel interfaces which is why it is not
described here.

Changelog
=========

Since v1 [7]:

Patch #9: Use htons() in 'case' instead of ntohs() in 'switch'.

Since RFC [8]:

Patch #3: Use NL_ASSERT_DUMP_CTX_FITS().
Patch #3: memset the entire context when moving to the next device.
Patch #3: Reset sequence counters when moving to the next device.
Patch #3: Use NL_SET_ERR_MSG_ATTR() in rtnl_validate_mdb_entry().
Patch #7: Remove restrictions regarding mixing of multicast and unicast
remote destination IPs in an MDB entry. While such configuration does
not make sense to me, it is no forbidden by the VXLAN FDB code and does
not crash the kernel.
Patch #7: Fix check regarding all-zeros MDB entry and source.
Patch #11: New patch.

[1] https://datatracker.ietf.org/doc/html/rfc9251
[2] https://datatracker.ietf.org/doc/html/draft-ietf-bess-evpn-irb-mcast
[3] https://datatracker.ietf.org/doc/html/rfc7432
[4] https://datatracker.ietf.org/doc/html/rfc7432#section-7.2
[5] https://datatracker.ietf.org/doc/html/rfc9251#section-9.1
[6] https://github.com/idosch/iproute2/commits/submit/mdb_vxlan_rfc_v1
[7] https://lore.kernel.org/netdev/20230313145349.3557231-1-idosch@nvidia.com/
[8] https://lore.kernel.org/netdev/20230204170801.3897900-1-idosch@nvidia.com/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoselftests: net: Add VXLAN MDB test
Ido Schimmel [Wed, 15 Mar 2023 13:11:55 +0000 (15:11 +0200)]
selftests: net: Add VXLAN MDB test

Add test cases for VXLAN MDB, testing the control and data paths. Two
different sets of namespaces (i.e., ns{1,2}_v4 and ns{1,2}_v6) are used
in order to test VXLAN MDB with both IPv4 and IPv6 underlays,
respectively.

Example truncated output:

 # ./test_vxlan_mdb.sh
 [...]
 Tests passed: 620
 Tests failed:   0

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agovxlan: Enable MDB support
Ido Schimmel [Wed, 15 Mar 2023 13:11:54 +0000 (15:11 +0200)]
vxlan: Enable MDB support

Now that the VXLAN MDB control and data paths are in place we can expose
the VXLAN MDB functionality to user space.

Set the VXLAN MDB net device operations to the appropriate functions,
thereby allowing the rtnetlink code to reach the VXLAN driver.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agovxlan: Add MDB data path support
Ido Schimmel [Wed, 15 Mar 2023 13:11:53 +0000 (15:11 +0200)]
vxlan: Add MDB data path support

Integrate MDB support into the Tx path of the VXLAN driver, allowing it
to selectively forward IP multicast traffic according to the matched MDB
entry.

If MDB entries are configured (i.e., 'VXLAN_F_MDB' is set) and the
packet is an IP multicast packet, perform up to three different lookups
according to the following priority:

1. For an (S, G) entry, using {Source VNI, Source IP, Destination IP}.
2. For a (*, G) entry, using {Source VNI, Destination IP}.
3. For the catchall MDB entry (0.0.0.0 or ::), using the source VNI.

The catchall MDB entry is similar to the catchall FDB entry
(00:00:00:00:00:00) that is currently used to transmit BUM (broadcast,
unknown unicast and multicast) traffic. However, unlike the catchall FDB
entry, this entry is only used to transmit unregistered IP multicast
traffic that is not link-local. Therefore, when configured, the catchall
FDB entry will only transmit BULL (broadcast, unknown unicast,
link-local multicast) traffic.

The catchall MDB entry is useful in deployments where inter-subnet
multicast forwarding is used and not all the VTEPs in a tenant domain
are members in all the broadcast domains. In such deployments it is
advantageous to transmit BULL (broadcast, unknown unicast and link-local
multicast) and unregistered IP multicast traffic on different tunnels.
If the same tunnel was used, a VTEP only interested in IP multicast
traffic would also pull all the BULL traffic and drop it as it is not a
member in the originating broadcast domain [1].

If the packet did not match an MDB entry (or if the packet is not an IP
multicast packet), return it to the Tx path, allowing it to be forwarded
according to the FDB.

If the packet did match an MDB entry, forward it to the associated
remote VTEPs. However, if the entry is a (*, G) entry and the associated
remote is in INCLUDE mode, then skip over it as the source IP is not in
its source list (otherwise the packet would have matched on an (S, G)
entry). Similarly, if the associated remote is marked as BLOCKED (can
only be set on (S, G) entries), then skip over it as well as the remote
is in EXCLUDE mode and the source IP is in its source list.

[1] https://datatracker.ietf.org/doc/html/draft-ietf-bess-evpn-irb-mcast#section-2.6

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agovxlan: mdb: Add an internal flag to indicate MDB usage
Ido Schimmel [Wed, 15 Mar 2023 13:11:52 +0000 (15:11 +0200)]
vxlan: mdb: Add an internal flag to indicate MDB usage

Add an internal flag to indicate whether MDB entries are configured or
not. Set the flag after installing the first MDB entry and clear it
before deleting the last one.

The flag will be consulted by the data path which will only perform an
MDB lookup if the flag is set, thereby keeping the MDB overhead to a
minimum when the MDB is not used.

Another option would have been to use a static key, but it is global and
not per-device, unlike the current approach.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agovxlan: mdb: Add MDB control path support
Ido Schimmel [Wed, 15 Mar 2023 13:11:51 +0000 (15:11 +0200)]
vxlan: mdb: Add MDB control path support

Implement MDB control path support, enabling the creation, deletion,
replacement and dumping of MDB entries in a similar fashion to the
bridge driver. Unlike the bridge driver, each entry stores a list of
remote VTEPs to which matched packets need to be replicated to and not a
list of bridge ports.

The motivating use case is the installation of MDB entries by a user
space control plane in response to received EVPN routes. As such, only
allow permanent MDB entries to be installed and do not implement
snooping functionality, avoiding a lot of unnecessary complexity.

Since entries can only be modified by user space under RTNL, use RTNL as
the write lock. Use RCU to ensure that MDB entries and remotes are not
freed while being accessed from the data path during transmission.

In terms of uAPI, reuse the existing MDB netlink interface, but add a
few new attributes to request and response messages:

* IP address of the destination VXLAN tunnel endpoint where the
  multicast receivers reside.

* UDP destination port number to use to connect to the remote VXLAN
  tunnel endpoint.

* VXLAN VNI Network Identifier to use to connect to the remote VXLAN
  tunnel endpoint. Required when Ingress Replication (IR) is used and
  the remote VTEP is not a member of originating broadcast domain
  (VLAN/VNI) [1].

* Source VNI Network Identifier the MDB entry belongs to. Used only when
  the VXLAN device is in external mode.

* Interface index of the outgoing interface to reach the remote VXLAN
  tunnel endpoint. This is required when the underlay destination IP is
  multicast (P2MP), as the multicast routing tables are not consulted.

All the new attributes are added under the 'MDBA_SET_ENTRY_ATTRS' nest
which is strictly validated by the bridge driver, thereby automatically
rejecting the new attributes.

[1] https://datatracker.ietf.org/doc/html/draft-ietf-bess-evpn-irb-mcast#section-3.2.2

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agovxlan: Expose vxlan_xmit_one()
Ido Schimmel [Wed, 15 Mar 2023 13:11:50 +0000 (15:11 +0200)]
vxlan: Expose vxlan_xmit_one()

Given a packet and a remote destination, the function will take care of
encapsulating the packet and transmitting it to the destination.

Expose it so that it could be used in subsequent patches by the MDB code
to transmit a packet to the remote destination(s) stored in the MDB
entry.

It will allow us to keep the MDB code self-contained, not exposing its
data structures to the rest of the VXLAN driver.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agovxlan: Move address helpers to private headers
Ido Schimmel [Wed, 15 Mar 2023 13:11:49 +0000 (15:11 +0200)]
vxlan: Move address helpers to private headers

Move the helpers out of the core C file to the private header so that
they could be used by the upcoming MDB code.

While at it, constify the second argument of vxlan_nla_get_addr().

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agortnetlink: bridge: mcast: Relax group address validation in common code
Ido Schimmel [Wed, 15 Mar 2023 13:11:48 +0000 (15:11 +0200)]
rtnetlink: bridge: mcast: Relax group address validation in common code

In the upcoming VXLAN MDB implementation, the 0.0.0.0 and :: MDB entries
will act as catchall entries for unregistered IP multicast traffic in a
similar fashion to the 00:00:00:00:00:00 VXLAN FDB entry that is used to
transmit BUM traffic.

In deployments where inter-subnet multicast forwarding is used, not all
the VTEPs in a tenant domain are members in all the broadcast domains.
It is therefore advantageous to transmit BULL (broadcast, unknown
unicast and link-local multicast) and unregistered IP multicast traffic
on different tunnels. If the same tunnel was used, a VTEP only
interested in IP multicast traffic would also pull all the BULL traffic
and drop it as it is not a member in the originating broadcast domain
[1].

Prepare for this change by allowing the 0.0.0.0 group address in the
common rtnetlink MDB code and forbid it in the bridge driver. A similar
change is not needed for IPv6 because the common code only validates
that the group address is not the all-nodes address.

[1] https://datatracker.ietf.org/doc/html/draft-ietf-bess-evpn-irb-mcast#section-2.6

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agortnetlink: bridge: mcast: Move MDB handlers out of bridge driver
Ido Schimmel [Wed, 15 Mar 2023 13:11:47 +0000 (15:11 +0200)]
rtnetlink: bridge: mcast: Move MDB handlers out of bridge driver

Currently, the bridge driver registers handlers for MDB netlink
messages, making it impossible for other drivers to implement MDB
support.

As a preparation for VXLAN MDB support, move the MDB handlers out of the
bridge driver to the core rtnetlink code. The rtnetlink code will call
into individual drivers by invoking their previously added MDB net
device operations.

Note that while the diffstat is large, the change is mechanical. It
moves code out of the bridge driver to rtnetlink code. Also note that a
similar change was made in 2012 with commit 77162022ab26 ("net: add
generic PF_BRIDGE:RTM_ FDB hooks") that moved FDB handlers out of the
bridge driver to the core rtnetlink code.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agobridge: mcast: Implement MDB net device operations
Ido Schimmel [Wed, 15 Mar 2023 13:11:46 +0000 (15:11 +0200)]
bridge: mcast: Implement MDB net device operations

Implement the previously added MDB net device operations in the bridge
driver so that they could be invoked by core rtnetlink code in the next
patch.

The operations are identical to the existing br_mdb_{dump,add,del}
functions. The '_new' suffix will be removed in the next patch. The
functions are re-implemented in this patch to make the conversion in the
next patch easier to review.

Add dummy implementations when 'CONFIG_BRIDGE_IGMP_SNOOPING' is
disabled, so that an error will be returned to user space when it is
trying to add or delete an MDB entry. This is consistent with existing
behavior where the bridge driver does not even register rtnetlink
handlers for RTM_{NEW,DEL,GET}MDB messages when this Kconfig option is
disabled.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: Add MDB net device operations
Ido Schimmel [Wed, 15 Mar 2023 13:11:45 +0000 (15:11 +0200)]
net: Add MDB net device operations

Add MDB net device operations that will be invoked by rtnetlink code in
response to received RTM_{NEW,DEL,GET}MDB messages. Subsequent patches
will implement these operations in the bridge and VXLAN drivers.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoMerge branch 'J784S4-CPSW9G-bindings'
David S. Miller [Fri, 17 Mar 2023 08:01:35 +0000 (08:01 +0000)]
Merge branch 'J784S4-CPSW9G-bindings'

Siddharth Vadapalli says:

====================
Add J784S4 CPSW9G NET Bindings

This series cleans up the bindings by reordering the compatibles, followed
by adding the bindings for CPSW9G instance of CPSW Ethernet Switch on TI's
J784S4 SoC.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agodt-bindings: net: ti: k3-am654-cpsw-nuss: Add J784S4 CPSW9G support
Siddharth Vadapalli [Wed, 15 Mar 2023 07:59:48 +0000 (13:29 +0530)]
dt-bindings: net: ti: k3-am654-cpsw-nuss: Add J784S4 CPSW9G support

Update bindings for TI K3 J784S4 SoC which contains 9 ports (8 external
ports) CPSW9G module and add compatible for it.

Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agodt-bindings: net: ti: k3-am654-cpsw-nuss: Fix compatible order
Siddharth Vadapalli [Wed, 15 Mar 2023 07:59:47 +0000 (13:29 +0530)]
dt-bindings: net: ti: k3-am654-cpsw-nuss: Fix compatible order

Reorder compatibles to follow alphanumeric order.

Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: mana: Add new MANA VF performance counters for easier troubleshooting
Shradha Gupta [Wed, 15 Mar 2023 11:55:13 +0000 (04:55 -0700)]
net: mana: Add new MANA VF performance counters for easier troubleshooting

Extended performance counter stats in 'ethtool -S <interface>' output
for MANA VF to facilitate troubleshooting.

Tested-on: Ubuntu22
Signed-off-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: wangxun: Implement the ndo change mtu interface
Mengyuan Lou [Wed, 15 Mar 2023 07:43:04 +0000 (15:43 +0800)]
net: wangxun: Implement the ndo change mtu interface

Add ngbe and txgbe ndo_change_mtu support.

Signed-off-by: Mengyuan Lou <mengyuanlou@net-swift.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: dsa: realtek: rtl8365mb: add change_mtu
Luiz Angelo Daros de Luca [Wed, 15 Mar 2023 03:49:22 +0000 (00:49 -0300)]
net: dsa: realtek: rtl8365mb: add change_mtu

The rtl8365mb was using a fixed MTU size of 1536, which was probably
inspired by the rtl8366rb's initial frame size. However, unlike that
family, the rtl8365mb family can specify the max frame size in bytes,
rather than in fixed steps.

DSA calls change_mtu for the CPU port once the max MTU value among the
ports changes. As the max frame size is defined globally, the switch
is configured only when the call affects the CPU port.

The available specifications do not directly define the max supported
frame size, but it mentions a 16k limit. This driver will use the 0x3FFF
limit as it is used in the vendor API code. However, the switch sets the
max frame size to 16368 bytes (0x3FF0) after it resets.

change_mtu uses MTU size, or ethernet payload size, while the switch
works with frame size. The frame size is calculated considering the
ethernet header (14 bytes), a possible 802.1Q tag (4 bytes), the payload
size (MTU), and the Ethernet FCS (4 bytes). The CPU tag (8 bytes) is
consumed before the switch enforces the limit.

During setup, the driver will use the default 1500-byte MTU of DSA to
set the maximum frame size. The current sum will be
VLAN_ETH_HLEN+1500+ETH_FCS_LEN, which results in 1522 bytes.  Although
it is lower than the previous initial value of 1536 bytes, the driver
will increase the frame size for a larger MTU. However, if something
requires more space without increasing the MTU, such as QinQ, we would
need to add the extra length to the rtl8365mb_port_change_mtu() formula.

MTU was tested up to 2018 (with 802.1Q) as that is as far as mt7620
(where rtl8367s is stacked) can go. The register was manually
manipulated byte-by-byte to ensure the MTU to frame size conversion was
correct. For frames without 802.1Q tag, the frame size limit will be 4
bytes over the required size.

There is a jumbo register, enabled by default at 6k frame size.
However, the jumbo settings do not seem to limit nor expand the maximum
tested MTU (2018), even when jumbo is disabled. More tests are needed
with a device that can handle larger frames.

Signed-off-by: Luiz Angelo Daros de Luca <luizluca@gmail.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoMerge branch 'add-ptp-support-for-sama7g5'
Jakub Kicinski [Fri, 17 Mar 2023 00:21:52 +0000 (17:21 -0700)]
Merge branch 'add-ptp-support-for-sama7g5'

Durai Manickam says:

====================
Add PTP support for sama7g5

This patch series is intended to add PTP capability to the GEM and
EMAC for sama7g5.
====================

Link: https://lore.kernel.org/r/20230315095053.53969-1-durai.manickamkr@microchip.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: macb: Add PTP support to EMAC for sama7g5
Durai Manickam KR [Wed, 15 Mar 2023 09:50:53 +0000 (15:20 +0530)]
net: macb: Add PTP support to EMAC for sama7g5

Add PTP capability to the Ethernet MAC.

Signed-off-by: Durai Manickam KR <durai.manickamkr@microchip.com>
Reviewed-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: macb: Add PTP support to GEM for sama7g5
Durai Manickam KR [Wed, 15 Mar 2023 09:50:52 +0000 (15:20 +0530)]
net: macb: Add PTP support to GEM for sama7g5

Add PTP capability to the Gigabit Ethernet MAC.

Signed-off-by: Durai Manickam KR <durai.manickamkr@microchip.com>
Reviewed-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: dsa: hellcreek: Get rid of custom led_init_default_state_get()
Andy Shevchenko [Tue, 14 Mar 2023 18:18:24 +0000 (20:18 +0200)]
net: dsa: hellcreek: Get rid of custom led_init_default_state_get()

LED core provides a helper to parse default state from firmware node.
Use it instead of custom implementation.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Link: https://lore.kernel.org/r/20230314181824.56881-1-andriy.shevchenko@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonfc: mrvl: Use of_property_read_bool() for boolean properties
Rob Herring [Tue, 14 Mar 2023 20:13:09 +0000 (15:13 -0500)]
nfc: mrvl: Use of_property_read_bool() for boolean properties

It is preferred to use typed property access functions (i.e.
of_property_read_<type> functions) rather than low-level
of_get_property/of_find_property functions for reading properties.
Convert reading boolean properties to of_property_read_bool().

Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonfc: mrvl: Move platform_data struct into driver
Rob Herring [Tue, 14 Mar 2023 20:13:08 +0000 (15:13 -0500)]
nfc: mrvl: Move platform_data struct into driver

There are no users of nfcmrvl platform_data struct outside of the
driver and none will be added, so move it into the driver.

Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: phy: mxl-gpy: enhance delay time required by loopback disable function
Xu Liang [Tue, 14 Mar 2023 16:30:23 +0000 (00:30 +0800)]
net: phy: mxl-gpy: enhance delay time required by loopback disable function

GPY2xx devices need 3 seconds to fully switch out of loopback mode
before it can safely re-enter loopback mode. Implement timeout mechanism
to guarantee 3 seconds waited before re-enter loopback mode.

Signed-off-by: Xu Liang <lxu@maxlinear.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: phy: micrel: Fix spelling mistake "minimim" -> "minimum"
Colin Ian King [Tue, 14 Mar 2023 08:23:15 +0000 (08:23 +0000)]
net: phy: micrel: Fix spelling mistake "minimim" -> "minimum"

There is a spelling mistake in a pr_warn_ratelimited message. Fix it.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Link: https://lore.kernel.org/r/20230314082315.26532-1-colin.i.king@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoMerge branch 'nfp-flower-add-support-for-multi-zone-conntrack'
Jakub Kicinski [Thu, 16 Mar 2023 05:16:26 +0000 (22:16 -0700)]
Merge branch 'nfp-flower-add-support-for-multi-zone-conntrack'

Louis Peens says:

====================
nfp: flower: add support for multi-zone conntrack

This series add changes to support offload of connection tracking across
multiple zones. Previously the driver only supported offloading of a
single goto_chain, spanning a single zone. This was implemented by
merging a pre_ct rule, post_ct rule and the nft rule. This series
provides updates to let the original post_ct rule act as the new pre_ct
rule for a next set of merges if it contains another goto and
conntrack action. In pseudo-tc rule format this adds support for:

    ingress chain 0 proto ip flower
        action ct zone 1 pipe action goto 1

    ingress chain 1 proto ip flower ct_state +tr+new ct_zone 1
        action ct_clear pipe action ct zone 2 pipe action goto 2
    ingress chain 1 proto ip flower ct_state +tr+est ct_zone 1
        action ct_clear pipe action ct zone 2 pipe action goto 2

    ingress chain 2 proto ip flower ct_state +tr+new ct_zone 2
        action mirred egress redirect dev ...
    ingress chain 2 proto ip flower ct_state +tr+est ct_zone 2
        action mirred egress redirect dev ...

This can continue for up to a maximum of 4 zone recirculations.

The first few patches are some smaller preparation patches while the
last one introduces the functionality.
====================

Link: https://lore.kernel.org/r/20230314063610.10544-1-louis.peens@corigine.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonfp: flower: offload tc flows of multiple conntrack zones
Wentao Jia [Tue, 14 Mar 2023 06:36:10 +0000 (08:36 +0200)]
nfp: flower: offload tc flows of multiple conntrack zones

If goto_chain action present in the post ct flow rule, merge flow rules
in this ct-zone, create a new pre_ct entry as the pre ct flow rule of
next ct-zone, but do not offload merged flow rules to firmware. Repeat
the process in the next ct-zone until no goto_chain action present in
the post ct flow rule in a certain ct-zone, merged all the flow rules.
Offload to firmware finally.

Signed-off-by: Wentao Jia <wentao.jia@corigine.com>
Acked-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonfp: flower: prepare for parameterisation of number of offload rules
Wentao Jia [Tue, 14 Mar 2023 06:36:09 +0000 (08:36 +0200)]
nfp: flower: prepare for parameterisation of number of offload rules

The fixed number of offload flow rule is only supported scenario of one
ct zone, in the scenario of multiple ct zones, dynamic number and more
number of offload flow rules are required. In order to support scenario
of multiple ct zones, parameter num_rules is added for to offload flow
rules

Signed-off-by: Wentao Jia <wentao.jia@corigine.com>
Acked-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonfp: flower: add goto_chain_index for ct entry
Wentao Jia [Tue, 14 Mar 2023 06:36:08 +0000 (08:36 +0200)]
nfp: flower: add goto_chain_index for ct entry

The chain_index has different means in pre ct entry and post ct entry.
In pre ct entry, it means chain index, but in post ct entry, it means
goto chain index, it is confused.

chain_index and goto_chain_index may be present in one flow rule, It
cannot be distinguished by one field chain_index, both chain_index
and goto_chain_index are required in the follow-up patch to support
multiple ct zones

Another field goto_chain_index is added to record the goto chain index.
If no goto action in post ct entry, goto_chain_index is 0.

Signed-off-by: Wentao Jia <wentao.jia@corigine.com>
Acked-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonfp: flower: refactor function "is_post_ct_flow"
Wentao Jia [Tue, 14 Mar 2023 06:36:07 +0000 (08:36 +0200)]
nfp: flower: refactor function "is_post_ct_flow"

'ct_clear' action only or no ct action is supported for 'post_ct_flow'.
But in scenario of multiple ct zones, one non 'ct_clear' ct action or
more ct actions, including 'ct_clear action', may be present in one flow
rule. If ct state match key is 'ct_established', the flow rule is still
expected to be classified as 'post_ct_flow'. Check ct status first in
function "is_post_ct_flow" to achieve this.

Signed-off-by: Wentao Jia <wentao.jia@corigine.com>
Acked-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonfp: flower: refactor function "is_pre_ct_flow"
Wentao Jia [Tue, 14 Mar 2023 06:36:06 +0000 (08:36 +0200)]
nfp: flower: refactor function "is_pre_ct_flow"

In the scenario of multiple ct zones, ct state key match and ct action
is present in one flow rule, the flow rule is classified to post_ct_flow
in design.

There is no ct state key match for pre ct flow, the judging condition
is added to function "is_pre_ct_flow".

Chain_index is another field for judging which flows are pre ct flow
If chain_index not 0, the flow is not pre ct flow.

Signed-off-by: Wentao Jia <wentao.jia@corigine.com>
Acked-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonfp: flower: add get_flow_act_ct() for ct action
Wentao Jia [Tue, 14 Mar 2023 06:36:05 +0000 (08:36 +0200)]
nfp: flower: add get_flow_act_ct() for ct action

CT action is a special case different from other actions, CT clear action
is not required when get ct action, but this case is not considered.
If CT clear action in the flow rule, skip the CT clear action when get ct
action, return the first ct action that is not a CT clear action

Signed-off-by: Wentao Jia <wentao.jia@corigine.com>
Acked-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoMerge mlx5 updates 2023-03-13
Jakub Kicinski [Thu, 16 Mar 2023 05:09:45 +0000 (22:09 -0700)]
Merge mlx5 updates 2023-03-13

Saeed Mahameed says:

====================
mlx5-updates-2023-03-13

1) Trivial cleanup patches
2) By Sandipan Patra: Implement thermal zone to report NIC temperature
3) Adham Faris, Improves devlink health diagnostics for netdev objects
4) From Maor, Enable TC offload for egress and engress MACVLAN over bond
5) From Gal, add devlink hairpin queues parameters to replace debugfs
   as was discussed in [1]:
[1] https://lore.kernel.org/all/20230111194608.7f15b9a1@kernel.org/
====================

Link: https://lore.kernel.org/all/20230314054234.267365-1-saeed@kernel.org/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Enable TC offload for egress MACVLAN over bond
Maor Dickman [Tue, 14 Mar 2023 05:42:34 +0000 (22:42 -0700)]
net/mlx5e: Enable TC offload for egress MACVLAN over bond

Support offloading of TC rules that mirror/redirect egress traffic to a
MACVLAN device, which is attached to bond device which master mlx5 devices.

Signed-off-by: Maor Dickman <maord@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-16-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Enable TC offload for ingress MACVLAN over bond
Maor Dickman [Tue, 14 Mar 2023 05:42:33 +0000 (22:42 -0700)]
net/mlx5e: Enable TC offload for ingress MACVLAN over bond

Support offloading of TC rules that filter ingress traffic from a MACVLAN
device, which is attached to bond device.

Signed-off-by: Maor Dickman <maord@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-15-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: TC, Extract indr setup block checks to function
Maor Dickman [Tue, 14 Mar 2023 05:42:32 +0000 (22:42 -0700)]
net/mlx5e: TC, Extract indr setup block checks to function

In preparation for next patch which will add new check
if device block can be setup, extract all existing checks
to function to make it more readable and maintainable.

Signed-off-by: Maor Dickman <maord@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-14-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Add more information to hairpin table dump
Gal Pressman [Tue, 14 Mar 2023 05:42:31 +0000 (22:42 -0700)]
net/mlx5e: Add more information to hairpin table dump

Print the number of hairpin queues and size as part of the hairpin table
dump.

Signed-off-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-13-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Add devlink hairpin queues parameters
Gal Pressman [Tue, 14 Mar 2023 05:42:30 +0000 (22:42 -0700)]
net/mlx5e: Add devlink hairpin queues parameters

We refer to a TC NIC rule that involves forwarding as "hairpin".
Hairpin queues are mlx5 hardware specific implementation for hardware
forwarding of such packets.

Per the discussion in [1], move the hairpin queues control (number and
size) from debugfs to devlink.

Expose two devlink params:
- hairpin_num_queues: control the number of hairpin queues
- hairpin_queue_size: control the size (in packets) of the hairpin queues

[1] https://lore.kernel.org/all/20230111194608.7f15b9a1@kernel.org/

Signed-off-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-12-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5: Move needed PTYS functions to core layer
Gal Pressman [Tue, 14 Mar 2023 05:42:29 +0000 (22:42 -0700)]
net/mlx5: Move needed PTYS functions to core layer

Downstream patches require devlink params to access the PTYS register,
move the needed functions from mlx5e to the core layer.

Signed-off-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-11-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Add XSK RQ state flag for RQ devlink health diagnostics
Adham Faris [Tue, 14 Mar 2023 05:42:28 +0000 (22:42 -0700)]
net/mlx5e: Add XSK RQ state flag for RQ devlink health diagnostics

Currently RQ health diagnostics doesn't inform the user whether an RQ
is an XSK RQ or not.

Address this, by adding XSK state flag to RQ SW state enum in core/en.h.
XSK will be '1' if current RQ is an XSK RQ, and it will be '0' if it's
not.

In this example below, it can be seen that XSK field value is '1' since
xdpsock program have been attached to channel 0 before issuing the
devlink query command:

$ devlink health diagnose auxiliary/mlx5_core.eth.0/65535 reporter rx

Output:
=======================================================================
 Common config:
    RQ:
      type: 2 stride size: 4096 size: 16 ts_format: FRC
      CQ:
        stride size: 64 size: 1024
  RQs:
      channel ix: 0 rqn: 4236 HW state: 1 WQE counter: 15 posted WQEs: 15 cc: 15
        SW State:
          enabled: 1 recovering: 0 am: 1 no_csum_complete: 1 csum_full: 0 mini_cqe_hw_stridx: 1 shampo: 0 mini_cqe_enhanced: 0 xsk: 1
      CQ:
        cqn: 1085 HW status: 0 ci: 0 size: 1024
      EQ:
        eqn: 7 irqn: 32 vecidx: 0 ci: 5 size: 2048
      ICOSQ:
        sqn: 4229 HW state: 1 cc: 158 pc: 158 WQE size: 2048
        CQ:
          cqn: 1080 cc: 1 size: 2048

Signed-off-by: Adham Faris <afaris@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-10-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Expose SQ SW state as part of SQ health diagnostics
Adham Faris [Tue, 14 Mar 2023 05:42:27 +0000 (22:42 -0700)]
net/mlx5e: Expose SQ SW state as part of SQ health diagnostics

Add SQ SW state textual representation to devlink health diagnostics
for tx reporter.

SQ SW state can be retrieved by issuing the devlink command below:

$ devlink health diagnose auxiliary/mlx5_core.eth.0/65535 reporter tx

Output
=======================================================================
 Common Config:
    SQ:
      stride size: 64 size: 1024 ts_format: FRC
      CQ:
        stride size: 64 size: 1024
  SQs:
      channel ix: 0 tc: 0 txq ix: 0 sqn: 4170 HW state: 1 stopped: false cc: 0 pc: 0
        SW State:
          enabled: 1 mpwqe: 1 recovering: 0 ipsec: 0 am: 1 vlan_need_l2_inline: 1 pending_xsk_tx: 0 pending_tls_rx_resync: 0 xdp_multibuf: 0
      CQ:
        cqn: 1031 HW status: 0 ci: 0 size: 1024
      EQ:
        eqn: 7 irqn: 32 vecidx: 0 ci: 2 size: 2048

Signed-off-by: Adham Faris <afaris@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-9-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Stringify RQ SW state in RQ devlink health diagnostics
Adham Faris [Tue, 14 Mar 2023 05:42:26 +0000 (22:42 -0700)]
net/mlx5e: Stringify RQ SW state in RQ devlink health diagnostics

One of the parameters that is retrieved/printed as a response to
devlink health diagnostics for rx reporter is the RQ SW state.

It's printed as a bitmap decimal number. Printing it as bitmap is
problematic and non informative.

In addition User can't count on SW state without accessing the kernel
sources (mlx5e rq state enum in en.h).

This patch prints RQ SW state in a textual representation, as a key:
value pairs, where disabled rq states will appear as '0' and enabled
ones will appear as '1'.

See below the generated output for rx health diagnostics devlink
command:

$ devlink health diagnose auxiliary/mlx5_core.eth.0/65535 reporter rx

Before:
=======================================================================
 Common config:
    RQ:
      type: 2 stride size: 2048 size: 8 ts_format: FRC
      CQ:
        stride size: 64 size: 1024
  RQs:
      channel ix: 0 rqn: 4172 HW state: 1 SW state: 37 WQE counter: 7 posted WQEs: 7 cc: 7
      CQ:
        cqn: 1033 HW status: 0 ci: 0 size: 1024
      EQ:
        eqn: 7 irqn: 32 vecidx: 0 ci: 2 size: 2048
      ICOSQ:
        sqn: 4169 HW state: 1 cc: 74 pc: 74 WQE size: 128
        CQ:
          cqn: 1030 cc: 1 size: 128
      channel ix: 1 ...
        .
        .

After:
=======================================================================
 Common config:
    RQ:
      type: 2 stride size: 2048 size: 8 ts_format: FRC
      CQ:
        stride size: 64 size: 1024
  RQs:
      channel ix: 0 rqn: 4172 HW state: 1 WQE counter: 7 posted WQEs: 7 cc: 7
        SW State:
          enabled: 1 recovering: 0 am: 1 no_csum_complete: 0 csum_full: 0 mini_cqe_hw_stridx: 1 shampo: 0 mini_cqe_enhanced: 0
      CQ:
        cqn: 1033 HW status: 0 ci: 0 size: 1024
      EQ:
        eqn: 7 irqn: 32 vecidx: 0 ci: 2 size: 2048
      ICOSQ:
        sqn: 4169 HW state: 1 cc: 74 pc: 74 WQE size: 128
        CQ:
          cqn: 1030 cc: 1 size: 128
       channel: ix: 1 ...
        .
        .

Signed-off-by: Adham Faris <afaris@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-8-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Rename RQ/SQ adaptive moderation state flag
Adham Faris [Tue, 14 Mar 2023 05:42:25 +0000 (22:42 -0700)]
net/mlx5e: Rename RQ/SQ adaptive moderation state flag

Dynamic interrupt moderation RQ and SQ feature represented by
MLX5E_RQ_STATE_AM and MLX5E_SQ_STATE_AM enums respectively, is not
consistent with the feature naming in the driver, and with the formal
feature and library names.

Hence, change MLX5E_RQ_STATE_AM and MLX5E_SQ_STATE_AM enum type names in
core/en.h to MLX5E_RQ_STATE_DIM and MLX5E_SQ_STATE_DIM respectively.

Signed-off-by: Adham Faris <afaris@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-7-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5e: Utilize the entire fifo
Rahul Rameshbabu [Tue, 14 Mar 2023 05:42:24 +0000 (22:42 -0700)]
net/mlx5e: Utilize the entire fifo

Previous check was comparing against the fifo mask. The mask is size of the
fifo (power of two) minus one, so a less than or equal comparator should be
used for checking if the fifo has room for the SKB.

Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-6-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5: Implement thermal zone
Sandipan Patra [Tue, 14 Mar 2023 05:42:23 +0000 (22:42 -0700)]
net/mlx5: Implement thermal zone

Implement thermal zone support for mlx5 based HW. The NIC
uses temperature sensor provided by ASIC to report current temperature
to thermal core.

Signed-off-by: Sandipan Patra <spatra@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-5-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5: Add comment to mlx5_devlink_params_register()
Jiri Pirko [Tue, 14 Mar 2023 05:42:22 +0000 (22:42 -0700)]
net/mlx5: Add comment to mlx5_devlink_params_register()

Add comment to mlx5_devlink_params_register() functions so it is clear
that only driver init params should be registered here.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-4-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5: Stop waiting for PCI up if teardown was triggered
Moshe Shemesh [Tue, 14 Mar 2023 05:42:21 +0000 (22:42 -0700)]
net/mlx5: Stop waiting for PCI up if teardown was triggered

If driver teardown is called while PCI is turned off, there is a race
between health recovery and teardown. If health recovery already started
it will wait 60 sec trying to see if PCI gets back and it can recover,
but actually there is no need to wait anymore once teardown was called.

Use the MLX5_BREAK_FW_WAIT flag which is set on driver teardown to break
waiting for PCI up.

Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-3-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet/mlx5: remove redundant clear_bit
Moshe Shemesh [Tue, 14 Mar 2023 05:42:20 +0000 (22:42 -0700)]
net/mlx5: remove redundant clear_bit

When shutdown or remove callbacks are called the driver sets the flag
MLX5_BREAK_FW_WAIT, to stop waiting for FW as teardown was called. There
is no need to clear the bit as once shutdown or remove were called as
there is no way back, the driver is going down. Furthermore, if not
cleared the flag can be used also in other loops where we may wait while
teardown was already called.

Use test_bit() instead of test_and_clear_bit() as there is no need to
clear the flag.

Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230314054234.267365-2-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: phy: micrel: drop superfluous use of temp variable
Wolfram Sang [Tue, 14 Mar 2023 12:49:27 +0000 (13:49 +0100)]
net: phy: micrel: drop superfluous use of temp variable

'temp' was used before commit c0c99d0cd107 ("net: phy: micrel: remove
the use of .ack_interrupt()") refactored the code. Now, we can simplify
it a little.

Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Link: https://lore.kernel.org/r/20230314124928.44948-1-wsa+renesas@sang-engineering.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agonet: phy: update obsolete comment about PHY_STARTING
Wolfram Sang [Tue, 14 Mar 2023 12:48:56 +0000 (13:48 +0100)]
net: phy: update obsolete comment about PHY_STARTING

Commit 899a3cbbf77a ("net: phy: remove states PHY_STARTING and
PHY_PENDING") missed to update a comment in phy_probe. Remove
superfluous "Description:" prefix while we are here.

Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Reviewed-by: Heiner Kallweit <hkallweit1@gmail.com>
Link: https://lore.kernel.org/r/20230314124856.44878-1-wsa+renesas@sang-engineering.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next...
Jakub Kicinski [Thu, 16 Mar 2023 04:25:24 +0000 (21:25 -0700)]
Merge branch '100GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
ice: refactor mailbox overflow detection

Jake Keller says:

The primary motivation of this series is to cleanup and refactor the mailbox
overflow detection logic such that it will work with Scalable IOV. In
addition a few other minor cleanups are done while I was working on the
code in the area.

First, the mailbox overflow functions in ice_vf_mbx.c are refactored to
store the data per-VF as an embedded structure in struct ice_vf, rather than
stored separately as a fixed-size array which only works with Single Root
IOV. This reduces the overall memory footprint when only a handful of VFs
are used.

The overflow detection functions are also cleaned up to reduce the need for
multiple separate calls to determine when to report a VF as potentially
malicious.

Finally, the ice_is_malicious_vf function is cleaned up and moved into
ice_virtchnl.c since it is not Single Root IOV specific, and thus does not
belong in ice_sriov.c

I could probably have done this in fewer patches, but I split pieces out to
hopefully aid in reviewing the overall sequence of changes. This does cause
some additional thrash as it results in intermediate versions of the
refactor, but I think its worth it for making each step easier to
understand.

* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
  ice: call ice_is_malicious_vf() from ice_vc_process_vf_msg()
  ice: move ice_is_malicious_vf() to ice_virtchnl.c
  ice: print message if ice_mbx_vf_state_handler returns an error
  ice: pass mbxdata to ice_is_malicious_vf()
  ice: remove unnecessary &array[0] and just use array
  ice: always report VF overflowing mailbox even without PF VSI
  ice: declare ice_vc_process_vf_msg in ice_virtchnl.h
  ice: initialize mailbox snapshot earlier in PF init
  ice: merge ice_mbx_report_malvf with ice_mbx_vf_state_handler
  ice: remove ice_mbx_deinit_snapshot
  ice: move VF overflow message count into struct ice_mbx_vf_info
  ice: track malicious VFs in new ice_mbx_vf_info structure
  ice: convert ice_mbx_clear_malvf to void and use WARN
  ice: re-order ice_mbx_reset_snapshot function
====================

Link: https://lore.kernel.org/r/20230313182123.483057-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoptp: ines: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sun, 12 Mar 2023 13:26:37 +0000 (14:26 +0100)]
ptp: ines: drop of_match_ptr for ID table

The driver can match only via the DT table so the table should be always
used and the of_match_ptr does not have any sense (this also allows ACPI
matching via PRP0001, even though it might not be relevant here).  This
also fixes !CONFIG_OF error:

  drivers/ptp/ptp_ines.c:783:34: error: ‘ines_ptp_ctrl_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Link: https://lore.kernel.org/r/20230312132637.352755-1-krzysztof.kozlowski@linaro.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
19 months agoscm: fix MSG_CTRUNC setting condition for SO_PASSSEC
Alexander Mikhalitsyn [Mon, 13 Mar 2023 11:32:11 +0000 (12:32 +0100)]
scm: fix MSG_CTRUNC setting condition for SO_PASSSEC

Currently, kernel would set MSG_CTRUNC flag if msg_control buffer
wasn't provided and SO_PASSCRED was set or if there was pending SCM_RIGHTS.

For some reason we have no corresponding check for SO_PASSSEC.

In the recvmsg(2) doc we have:
       MSG_CTRUNC
              indicates that some control data was discarded due to lack
              of space in the buffer for ancillary data.

So, we need to set MSG_CTRUNC flag for all types of SCM.

This change can break applications those don't check MSG_CTRUNC flag.

Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: Leon Romanovsky <leon@kernel.org>
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
v2:
- commit message was rewritten according to Eric's suggestion
Acked-by: Paul Moore <paul@paul-moore.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoMerge branch 'net-smc-updates'
David S. Miller [Wed, 15 Mar 2023 08:18:35 +0000 (08:18 +0000)]
Merge branch 'net-smc-updates'

Wenjia Zhang says:

====================
smc: Updates 2023-03-01

The 1st patch is to make implements later do not need to adhere to a
specific SEID format. The 2nd patch does some cleanup.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet/ism: Remove extra include
Stefan Raspl [Mon, 13 Mar 2023 10:10:32 +0000 (11:10 +0100)]
net/ism: Remove extra include

Signed-off-by: Stefan Raspl <raspl@linux.ibm.com>
Signed-off-by: Wenjia Zhang <wenjia@linux.ibm.com>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet/smc: Introduce explicit check for v2 support
Stefan Raspl [Mon, 13 Mar 2023 10:10:31 +0000 (11:10 +0100)]
net/smc: Introduce explicit check for v2 support

Previously, v2 support was derived from a very specific format of the SEID
as part of the SMC-D codebase. Make this part of the SMC-D device API, so
implementers do not need to adhere to a specific SEID format.

Signed-off-by: Stefan Raspl <raspl@linux.ibm.com>
Reviewed-and-tested-by: Jan Karcher <jaka@linux.ibm.com>
Reviewed-by: Wenjia Zhang <wenjia@linux.ibm.com>
Signed-off-by: Wenjia Zhang <wenjia@linux.ibm.com>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: geneve: set IFF_POINTOPOINT with IFLA_GENEVE_INNER_PROTO_INHERIT
Josef Miegl [Sun, 12 Mar 2023 16:45:57 +0000 (17:45 +0100)]
net: geneve: set IFF_POINTOPOINT with IFLA_GENEVE_INNER_PROTO_INHERIT

The GENEVE tunnel used with IFLA_GENEVE_INNER_PROTO_INHERIT is
point-to-point, so set IFF_POINTOPOINT to reflect that.

Signed-off-by: Josef Miegl <josef@miegl.cz>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: ieee802154: ca8210: drop owner from driver
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:33:03 +0000 (18:33 +0100)]
net: ieee802154: ca8210: drop owner from driver

Core already sets owner in spi_driver.

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Acked-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: ieee802154: adf7242: drop owner from driver
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:33:02 +0000 (18:33 +0100)]
net: ieee802154: adf7242: drop owner from driver

Core already sets owner in spi_driver.

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Acked-by: Michael Hennerich <michael.hennerich@analog.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: ieee802154: ca8210: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:33:01 +0000 (18:33 +0100)]
net: ieee802154: ca8210: drop of_match_ptr for ID table

The driver can match only via the DT table so the table should be always
used and the of_match_ptr does not have any sense (this also allows ACPI
matching via PRP0001, even though it might not be relevant here).

  drivers/net/ieee802154/ca8210.c:3174:34: error: ‘ca8210_of_ids’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Acked-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: ieee802154: at86rf230: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:33:00 +0000 (18:33 +0100)]
net: ieee802154: at86rf230: drop of_match_ptr for ID table

The driver will match mostly by DT table (even thought there is regular
ID table) so there is little benefit in of_match_ptr (this also allows
ACPI matching via PRP0001, even though it might not be relevant here).

  drivers/net/ieee802154/at86rf230.c:1644:34: error: ‘at86rf230_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Acked-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: ieee802154: mcr20a: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:32:59 +0000 (18:32 +0100)]
net: ieee802154: mcr20a: drop of_match_ptr for ID table

The driver will match mostly by DT table (even thought there is regular
ID table) so there is little benefit in of_match_ptr (this also allows
ACPI matching via PRP0001, even though it might not be relevant here).

  drivers/net/ieee802154/mcr20a.c:1340:34: error: ‘mcr20a_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Acked-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: ieee802154: adf7242: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:32:58 +0000 (18:32 +0100)]
net: ieee802154: adf7242: drop of_match_ptr for ID table

The driver will match mostly by DT table (even thought there is regular
ID table) so there is little benefit in of_match_ptr (this also allows
ACPI matching via PRP0001, even though it might not be relevant here).

  drivers/net/ieee802154/adf7242.c:1322:34: error: ‘adf7242_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Acked-by: Michael Hennerich <michael.hennerich@analog.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: phy: ks8995: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:32:57 +0000 (18:32 +0100)]
net: phy: ks8995: drop of_match_ptr for ID table

The driver will match mostly by DT table (even thought there is regular
ID table) so there is little benefit in of_match_ptr (this also allows
ACPI matching via PRP0001, even though it might not be relevant here).

  drivers/net/phy/spi_ks8995.c:156:34: error: ‘ks8895_spi_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: dsa: ocelot: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:32:56 +0000 (18:32 +0100)]
net: dsa: ocelot: drop of_match_ptr for ID table

The driver can match only via the DT table so the table should be always
used and the of_match_ptr does not have any sense (this also allows ACPI
matching via PRP0001, even though it might not be relevant here).

  drivers/net/dsa/ocelot/ocelot_ext.c:143:34: error: ‘ocelot_ext_switch_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Acked-by: Colin Foster <colin.foster@in-advantage.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: dsa: ksz9477: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:32:55 +0000 (18:32 +0100)]
net: dsa: ksz9477: drop of_match_ptr for ID table

The driver will match mostly by DT table (even thought there is regular
ID table) so there is little benefit in of_match_ptr (this also allows
ACPI matching via PRP0001, even though it might not be relevant here).

  drivers/net/dsa/microchip/ksz9477_i2c.c:84:34: error: ‘ksz9477_dt_ids’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: dsa: seville_vsc9953: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:32:54 +0000 (18:32 +0100)]
net: dsa: seville_vsc9953: drop of_match_ptr for ID table

The driver can match only via the DT table so the table should be always
used and the of_match_ptr does not have any sense (this also allows ACPI
matching via PRP0001, even though it might not be relevant here).

  drivers/net/dsa/ocelot/seville_vsc9953.c:1070:34: error: ‘seville_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: dsa: lan9303: drop of_match_ptr for ID table
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:32:53 +0000 (18:32 +0100)]
net: dsa: lan9303: drop of_match_ptr for ID table

The driver will match mostly or only by DT table (even thought there is
regular ID table) so there is little benefit in of_match_ptr (this also
allows ACPI matching via PRP0001, even though it might not be relevant
here).

  drivers/net/dsa/lan9303_i2c.c:97:34: error: ‘lan9303_i2c_of_match’ defined but not used [-Werror=unused-const-variable=]
  drivers/net/dsa/lan9303_mdio.c:157:34: error: ‘lan9303_mdio_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: dsa: lantiq_gswip: mark OF related data as maybe unused
Krzysztof Kozlowski [Sat, 11 Mar 2023 17:32:52 +0000 (18:32 +0100)]
net: dsa: lantiq_gswip: mark OF related data as maybe unused

The driver can be compile tested with !CONFIG_OF making certain data
unused:

  drivers/net/dsa/lantiq_gswip.c:1888:34: error: ‘xway_gphy_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonfc: trf7970a: mark OF related data as maybe unused
Krzysztof Kozlowski [Sat, 11 Mar 2023 11:13:28 +0000 (12:13 +0100)]
nfc: trf7970a: mark OF related data as maybe unused

The driver can be compile tested with !CONFIG_OF making certain data
unused:

  drivers/nfc/trf7970a.c:2232:34: error: ‘trf7970a_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Mark Greer <mgreer@animalcreek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: ni: drop of_match_ptr for ID table
Krzysztof Kozlowski [Fri, 10 Mar 2023 21:46:32 +0000 (22:46 +0100)]
net: ni: drop of_match_ptr for ID table

The driver can match only via the DT table so the table should be always
used and the of_match_ptr does not have any sense (this also allows ACPI
matching via PRP0001, even though it is not relevant here).

  drivers/net/ethernet/ni/nixge.c:1253:34: error: ‘nixge_dt_ids’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: samsung: sxgbe: drop of_match_ptr for ID table
Krzysztof Kozlowski [Fri, 10 Mar 2023 21:46:31 +0000 (22:46 +0100)]
net: samsung: sxgbe: drop of_match_ptr for ID table

The driver can match only via the DT table so the table should be always
used and the of_match_ptr does not have any sense (this also allows ACPI
matching via PRP0001, even though it is not relevant here).

  drivers/net/ethernet/samsung/sxgbe/sxgbe_platform.c:220:34: error: ‘sxgbe_dt_ids’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: marvell: pxa168_eth: drop of_match_ptr for ID table
Krzysztof Kozlowski [Fri, 10 Mar 2023 21:46:30 +0000 (22:46 +0100)]
net: marvell: pxa168_eth: drop of_match_ptr for ID table

The driver can match only via the DT table so the table should be always
used and the of_match_ptr does not have any sense (this also allows ACPI
matching via PRP0001, even though it is not relevant here).

  drivers/net/ethernet/marvell/pxa168_eth.c:1575:34: error: ‘pxa168_eth_of_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: stmmac: generic: drop of_match_ptr for ID table
Krzysztof Kozlowski [Fri, 10 Mar 2023 21:46:29 +0000 (22:46 +0100)]
net: stmmac: generic: drop of_match_ptr for ID table

The driver can match only via the DT table so the table should be always
used and the of_match_ptr does not have any sense (this also allows ACPI
matching via PRP0001, even though it is not relevant here).

  drivers/net/ethernet/stmicro/stmmac/dwmac-generic.c:72:34: error: ‘dwmac_generic_match’ defined but not used [-Werror=unused-const-variable=]

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agonet: stmmac: qcom: drop of_match_ptr for ID table
Krzysztof Kozlowski [Fri, 10 Mar 2023 21:46:28 +0000 (22:46 +0100)]
net: stmmac: qcom: drop of_match_ptr for ID table

The driver is specific to ARCH_QCOM which depends on OF thus the driver
is OF-only.  Its of_device_id table is built unconditionally, thus
of_match_ptr() for ID table does not make sense.

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 months agoMerge branch 'dsa-microchip-tc-ets'
David S. Miller [Wed, 15 Mar 2023 08:00:35 +0000 (08:00 +0000)]
Merge branch 'dsa-microchip-tc-ets'

Oleksij Rempel says:

====================
net: dsa: microchip: tc-ets support

changes v3:
- add tc_ets_supported to match supported devices
- dynamically regenerated default TC to queue map.
- add Acked-by to the first patch

changes v2:
- run egress limit configuration on all queue separately. Otherwise
  configuration may not apply correctly.
====================

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