linux-2.6-microblaze.git
3 years agotools, bpftool: Print correct error message when failing to load BTF
Tobias Klauser [Mon, 25 May 2020 13:54:21 +0000 (15:54 +0200)]
tools, bpftool: Print correct error message when failing to load BTF

btf__parse_raw and btf__parse_elf return negative error numbers wrapped
in an ERR_PTR, so the extracted value needs to be negated before passing
them to strerror which expects a positive error number.

Before:
  Error: failed to load BTF from .../vmlinux: Unknown error -2

After:
  Error: failed to load BTF from .../vmlinux: No such file or directory

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200525135421.4154-1-tklauser@distanz.ch
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
3 years agotools, bpftool: Make capability check account for new BPF caps
Quentin Monnet [Sat, 23 May 2020 01:02:47 +0000 (02:02 +0100)]
tools, bpftool: Make capability check account for new BPF caps

Following the introduction of CAP_BPF, and the switch from CAP_SYS_ADMIN
to other capabilities for various BPF features, update the capability
checks (and potentially, drops) in bpftool for feature probes. Because
bpftool and/or the system might not know of CAP_BPF yet, some caution is
necessary:

- If compiled and run on a system with CAP_BPF, check CAP_BPF,
  CAP_SYS_ADMIN, CAP_PERFMON, CAP_NET_ADMIN.

- Guard against CAP_BPF being undefined, to allow compiling bpftool from
  latest sources on older systems. If the system where feature probes
  are run does not know of CAP_BPF, stop checking after CAP_SYS_ADMIN,
  as this should be the only capability required for all the BPF
  probing.

- If compiled from latest sources on a system without CAP_BPF, but later
  executed on a newer system with CAP_BPF knowledge, then we only test
  CAP_SYS_ADMIN. Some probes may fail if the bpftool process has
  CAP_SYS_ADMIN but misses the other capabilities. The alternative would
  be to redefine the value for CAP_BPF in bpftool, but this does not
  look clean, and the case sounds relatively rare anyway.

Note that libcap offers a cap_to_name() function to retrieve the name of
a given capability (e.g. "cap_sys_admin"). We do not use it because
deriving the names from the macros looks simpler than using
cap_to_name() (doing a strdup() on the string) + cap_free() + handling
the case of failed allocations, when we just want to use the name of the
capability in an error message.

The checks when compiling without libcap (i.e. root versus non-root) are
unchanged.

v2:
- Do not allocate cap_list dynamically.
- Drop BPF-related capabilities when running with "unprivileged", even
  if we didn't have the full set in the first place (in v1, we would
  skip dropping them in that case).
- Keep track of what capabilities we have, print the names of the
  missing ones for privileged probing.
- Attempt to drop only the capabilities we actually have.
- Rename a couple variables.

Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200523010247.20654-1-quentin@isovalent.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
3 years agotools, bpftool: Clean subcommand help messages
Quentin Monnet [Sat, 23 May 2020 01:07:51 +0000 (02:07 +0100)]
tools, bpftool: Clean subcommand help messages

This is a clean-up for the formatting of the do_help functions for
bpftool's subcommands. The following fixes are included:

- Do not use argv[-2] for "iter" help message, as the help is shown by
  default if no "iter" action is selected, resulting in messages looking
  like "./bpftool bpftool pin...".

- Do not print unused HELP_SPEC_PROGRAM in help message for "bpftool
  link".

- Andrii used argument indexing to avoid having multiple occurrences of
  bin_name and argv[-2] in the fprintf() for the help message, for
  "bpftool gen" and "bpftool link". Let's reuse this for all other help
  functions. We can remove up to thirty arguments for the "bpftool map"
  help message.

- Harmonise all functions, e.g. use ending quotes-comma on a separate
  line.

Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200523010751.23465-1-quentin@isovalent.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
3 years agonet: dsa: sja1105: suppress -Wmissing-prototypes in sja1105_vl.c
Vladimir Oltean [Sun, 31 May 2020 18:25:51 +0000 (21:25 +0300)]
net: dsa: sja1105: suppress -Wmissing-prototypes in sja1105_vl.c

Newer C compilers are complaining about the fact that there are no
function prototypes in sja1105_vl.c for the non-static functions.
Give them what they want.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Mon, 1 Jun 2020 19:09:08 +0000 (12:09 -0700)]
Merge branch '100GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
100GbE Intel Wired LAN Driver Updates 2020-05-31

This series contains updates to the ice driver only.

Brett modifies the driver to allow users to clear a VF's
administratively set MAC address on the PF.  Fixes the driver to
recognize an existing VLAN tag when DMAC/SMAC is enabled in a packet.
Fixes an issue, so that VF's are reset after any VF port VLAN
modifications are made on the PF.  Made sure the register QRXFLXP_CNTXT
is cleared before writing a new value to ensure the previous value is
not passed forward.  Updates the PF to allow the VF to request a reset
as soon as it has been initialized.  Fixes an issue to ensure when a VSI
is created, it uses the current coalesce value, not the default value.

Paul allows untrusted VF's to add 16 filters.

Dan increases the timeout needed after a PFR to allow ample time for
package download.

Chinh adjust the define value for the number of PHY speeds we currently
support.  Changes the driver to ignore EMODE error when configuring the
PHY.

Jesse fixes an issue which was preventing a user from configuring the
interface before bringing it up.

Henry fixes the logic for adding back perfect flows after flow director
filter does a deletion.

Bruce fixes line wrappings to make it more consistent.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovxlan: fix dereference of nexthop group in nexthop update path
Roopa Prabhu [Sun, 31 May 2020 05:17:20 +0000 (22:17 -0700)]
vxlan: fix dereference of nexthop group in nexthop update path

fix dereference of nexthop group in fdb nexthop group
update validation path.

Fixes: 1274e1cc4226 ("vxlan: ecmp support for mac fdb entries")
Reported-by: Ido Schimmel <idosch@idosch.org>
Suggested-by: Ido Schimmel <idosch@idosch.org>
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoswitch cmsghdr_from_user_compat_to_kern() to copy_from_user()
Al Viro [Sun, 31 May 2020 01:06:55 +0000 (02:06 +0100)]
switch cmsghdr_from_user_compat_to_kern() to copy_from_user()

no point getting compat_cmsghdr field-by-field

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'dpaa2-eth-add-PFC-support'
David S. Miller [Mon, 1 Jun 2020 19:04:32 +0000 (12:04 -0700)]
Merge branch 'dpaa2-eth-add-PFC-support'

Ioana Ciornei says:

====================
dpaa2-eth: add PFC support

This patch set adds support for Priority Flow Control in DPAA2 Ethernet
devices.

The first patch make the necessary changes so that multiple
traffic classes are configured. The dequeue priority
of the maximum 8 traffic classes is configured to be equal.
The second patch adds a static distribution to said traffic
classes based on the VLAN PCP field. In the future, this could be
extended through the .setapp() DCB callback for dynamic configuration.

Also, add support for the congestion group taildrop mechanism that
allows us to control the number of frames that can accumulate on a group
of Rx frame queues belonging to the same traffic class.

The basic subset of the DCB ops is implemented so that the user can
query the number of PFC capable traffic classes, their state and
reconfigure them if necessary.

Changes in v3:
 - add patches 6-7 which add the PFC functionality
 - patch 2/7: revert to explicitly cast mask to u16 * to not get into
   sparse warnings
Changes in v4:
 - really fix the sparse warnings in 2/7
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: Keep congestion group taildrop enabled when PFC on
Ioana Ciornei [Sat, 30 May 2020 21:08:14 +0000 (00:08 +0300)]
dpaa2-eth: Keep congestion group taildrop enabled when PFC on

Leave congestion group taildrop enabled for all traffic classes
when PFC is enabled. Notification threshold is low enough such
that it will be hit first and this also ensures that FQs on
traffic classes which are not PFC enabled won't drain the buffer
pool.

FQ taildrop threshold is kept disabled as long as any form of
flow control is on. Since FQ taildrop works with bytes, not number
of frames, we can't guarantee it will not interfere with the
congestion notification mechanism for all frame sizes.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: Add PFC support through DCB ops
Ioana Ciornei [Sat, 30 May 2020 21:08:13 +0000 (00:08 +0300)]
dpaa2-eth: Add PFC support through DCB ops

Add support in dpaa2-eth for PFC (Priority Flow Control)
through the DCB ops.

Instruct the hardware to respond to received PFC frames.
Current firmware doesn't allow us to selectively enable PFC
on the Rx side for some priorities only, so we will react to
all incoming PFC frames (and stop transmitting on the traffic
classes specified in the frame).

Also, configure the hardware to generate PFC frames based on Rx
congestion notifications. When a certain number of frames accumulate in
the ingress queues corresponding to a traffic class, priority flow
control frames are generated for that TC.

The number of PFC traffic classes available can be queried through
lldptool. Also, which of those traffic classes have PFC enabled is also
controlled through the same dcbnl_rtnl_ops callbacks.

Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: Update FQ taildrop threshold and buffer pool count
Ioana Radulescu [Sat, 30 May 2020 21:08:12 +0000 (00:08 +0300)]
dpaa2-eth: Update FQ taildrop threshold and buffer pool count

Now that we have congestion group taildrop configured at all
times, we can afford to increase the frame queue taildrop
threshold; this will ensure a better response when receiving
bursts of large-sized frames.

Also decouple the buffer pool count from the Rx FQ taildrop
threshold, as above change would increase it too much. Instead,
keep the old count as a hardcoded value.

With the new limits, we try to ensure that:
* we allow enough leeway for large frame bursts (by buffering
enough of them in queues to avoid heavy dropping in case of
bursty traffic, but when overall ingress bandwidth is manageable)
* allow pending frames to be evenly spread between ingress FQs,
regardless of frame size
* avoid dropping frames due to the buffer pool being empty; this
is not a bad behaviour per se, but system overall response is
more linear and predictable when frames are dropped at frame
queue/group level.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: Add congestion group taildrop
Ioana Radulescu [Sat, 30 May 2020 21:08:11 +0000 (00:08 +0300)]
dpaa2-eth: Add congestion group taildrop

The increase in number of ingress frame queues means we now risk
depleting the buffer pool before the FQ taildrop kicks in.

Congestion group taildrop allows us to control the number of frames that
can accumulate on a group of Rx frame queues belonging to the same
traffic class.  This setting coexists with the frame queue based
taildrop: whichever limit gets hit first triggers the frame drop.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: Add helper functions
Ioana Radulescu [Sat, 30 May 2020 21:08:10 +0000 (00:08 +0300)]
dpaa2-eth: Add helper functions

Add convenient helper functions that determines whether Rx/Tx pause
frames are enabled based on link state flags received from firmware.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: Distribute ingress frames based on VLAN prio
Ioana Radulescu [Sat, 30 May 2020 21:08:09 +0000 (00:08 +0300)]
dpaa2-eth: Distribute ingress frames based on VLAN prio

Configure static ingress classification based on VLAN PCP field.
If the DPNI doesn't have enough traffic classes to accommodate all
priority levels, the lowest ones end up on TC 0 (default on miss).

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: Add support for Rx traffic classes
Ioana Radulescu [Sat, 30 May 2020 21:08:08 +0000 (00:08 +0300)]
dpaa2-eth: Add support for Rx traffic classes

The firmware reserves for each DPNI a number of RX frame queues
equal to the number of configured flows x number of configured
traffic classes.

Current driver configuration directs all incoming traffic to
FQs corresponding to TC0, leaving all other priority levels unused.

Start adding support for multiple ingress traffic classes, by
configuring the FQs associated with all priority levels, not just
TC0. All settings that are per-TC, such as those related to
hashing and flow steering, are also updated.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: broadcom: don't export RDB/legacy access methods
Michael Walle [Sat, 30 May 2020 20:34:04 +0000 (22:34 +0200)]
net: phy: broadcom: don't export RDB/legacy access methods

Don't export __bcm_phy_enable_rdb_access() and
__bcm_phy_enable_legacy_access() functions. They aren't used outside this
module and it was forgotten to provide a prototype for these functions.
Just make them static for now.

Fixes: 11ecf8c55b91 ("net: phy: broadcom: add cable test support")
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agotun: correct header offsets in napi frags mode
Willem de Bruijn [Sat, 30 May 2020 19:41:31 +0000 (15:41 -0400)]
tun: correct header offsets in napi frags mode

Tun in IFF_NAPI_FRAGS mode calls napi_gro_frags. Unlike netif_rx and
netif_gro_receive, this expects skb->data to point to the mac layer.

But skb_probe_transport_header, __skb_get_hash_symmetric, and
xdp_do_generic in tun_get_user need skb->data to point to the network
header. Flow dissection also needs skb->protocol set, so
eth_type_trans has to be called.

Ensure the link layer header lies in linear as eth_type_trans pulls
ETH_HLEN. Then take the same code paths for frags as for not frags.
Push the link layer header back just before calling napi_gro_frags.

By pulling up to ETH_HLEN from frag0 into linear, this disables the
frag0 optimization in the special case when IFF_NAPI_FRAGS is used
with zero length iov[0] (and thus empty skb->linear).

Fixes: 90e33d459407 ("tun: enable napi_gro_frags() for TUN/TAP driver")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Petar Penkov <ppenkov@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agocls_flower: remove mpls_opts_policy
Guillaume Nault [Sat, 30 May 2020 18:49:56 +0000 (20:49 +0200)]
cls_flower: remove mpls_opts_policy

Compiling with W=1 gives the following warning:
net/sched/cls_flower.c:731:1: warning: ‘mpls_opts_policy’ defined but not used [-Wunused-const-variable=]

The TCA_FLOWER_KEY_MPLS_OPTS contains a list of
TCA_FLOWER_KEY_MPLS_OPTS_LSE. Therefore, the attributes all have the
same type and we can't parse the list with nla_parse*() and have the
attributes validated automatically using an nla_policy.

fl_set_key_mpls_opts() properly verifies that all attributes in the
list are TCA_FLOWER_KEY_MPLS_OPTS_LSE. Then fl_set_key_mpls_lse()
uses nla_parse_nested() on all these attributes, thus verifying that
they have the NLA_F_NESTED flag. So we can safely drop the
mpls_opts_policy.

Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'bridge-mrp-Add-support-for-MRA-role'
David S. Miller [Mon, 1 Jun 2020 18:56:11 +0000 (11:56 -0700)]
Merge branch 'bridge-mrp-Add-support-for-MRA-role'

Horatiu Vultur says:

====================
bridge: mrp: Add support for MRA role

This patch series extends the MRP with the MRA role.
A node that has the MRA role can behave as a MRM or as a MRC. In case there are
multiple nodes in the topology that has the MRA role then only one node can
behave as MRM and all the others need to be have as MRC. The node that has the
higher priority(lower value) will behave as MRM.
A node that has the MRA role and behaves as MRC, it just needs to forward the
MRP_Test frames between the ring ports but also it needs to detect in case it
stops receiving MRP_Test frames. In that case it would try to behave as MRM.

v2:
 - add new patch that fixes sparse warnings
 - fix parsing of prio attribute
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobridge: mrp: Add support for role MRA
Horatiu Vultur [Sat, 30 May 2020 18:09:48 +0000 (18:09 +0000)]
bridge: mrp: Add support for role MRA

A node that has the MRA role, it can behave as MRM or MRC.

Initially it starts as MRM and sends MRP_Test frames on both ring ports.
If it detects that there are MRP_Test send by another MRM, then it
checks if these frames have a lower priority than itself. In this case
it would send MRP_Nack frames to notify the other node that it needs to
stop sending MRP_Test frames.
If it receives a MRP_Nack frame then it stops sending MRP_Test frames
and starts to behave as a MRC but it would continue to monitor the
MRP_Test frames send by MRM. If at a point the MRM stops to send
MRP_Test frames it would get the MRM role and start to send MRP_Test
frames.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobridge: mrp: Set the priority of MRP instance
Horatiu Vultur [Sat, 30 May 2020 18:09:47 +0000 (18:09 +0000)]
bridge: mrp: Set the priority of MRP instance

Each MRP instance has a priority, a lower value means a higher priority.
The priority of MRP instance is stored in MRP_Test frame in this way
all the MRP nodes in the ring can see other nodes priority.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobridge: mrp: Update MRP frame type
Horatiu Vultur [Sat, 30 May 2020 18:09:46 +0000 (18:09 +0000)]
bridge: mrp: Update MRP frame type

Replace u16/u32 with be16/be32 in the MRP frame types.
This fixes sparse warnings like:
warning: cast to restricted __be16

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: vmxnet3: fix possible buffer overflow caused by bad DMA value in vmxnet3_get_rss()
Jia-Ju Bai [Sat, 30 May 2020 02:41:50 +0000 (10:41 +0800)]
net: vmxnet3: fix possible buffer overflow caused by bad DMA value in vmxnet3_get_rss()

The value adapter->rss_conf is stored in DMA memory, and it is assigned
to rssConf, so rssConf->indTableSize can be modified at anytime by
malicious hardware. Because rssConf->indTableSize is assigned to n,
buffer overflow may occur when the code "rssConf->indTable[n]" is
executed.

To fix this possible bug, n is checked after being used.

Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoflow_dissector: work around stack frame size warning
Arnd Bergmann [Fri, 29 May 2020 20:13:58 +0000 (22:13 +0200)]
flow_dissector: work around stack frame size warning

The fl_flow_key structure is around 500 bytes, so having two of them
on the stack in one function now exceeds the warning limit after an
otherwise correct change:

net/sched/cls_flower.c:298:12: error: stack frame size of 1056 bytes in function 'fl_classify' [-Werror,-Wframe-larger-than=]

I suspect the fl_classify function could be reworked to only have one
of them on the stack and modify it in place, but I could not work out
how to do that.

As a somewhat hacky workaround, move one of them into an out-of-line
function to reduce its scope. This does not necessarily reduce the stack
usage of the outer function, but at least the second copy is removed
from the stack during most of it and does not add up to whatever is
called from there.

I now see 552 bytes of stack usage for fl_classify(), plus 528 bytes
for fl_mask_lookup().

Fixes: 58cff782cc55 ("flow_dissector: Parse multiple MPLS Label Stack Entries")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agolan743x: Added fixed link and RGMII support
Roelof Berg [Fri, 29 May 2020 19:30:02 +0000 (21:30 +0200)]
lan743x: Added fixed link and RGMII support

Microchip lan7431 is frequently connected to a phy. However, it
can also be directly connected to a MII remote peer without
any phy in between. For supporting such a phyless hardware setup
in Linux we utilized phylib, which supports a fixed-link
configuration via the device tree. And we added support for
defining the connection type R/GMII in the device tree.

New behavior:
-------------
. The automatic speed and duplex detection of the lan743x silicon
  between mac and phy is disabled. Instead phylib is used like in
  other typical Linux drivers. The usage of phylib allows to
  specify fixed-link parameters in the device tree.

. The device tree entry phy-connection-type is supported now with
  the modes RGMII or (G)MII (default).

Development state:
------------------
. Tested with fixed-phy configurations. Not yet tested in normal
  configurations with phy. Microchip kindly offered testing
  as soon as the Corona measures allow this.

. All review findings of Andrew Lunn are included

Example:
--------
&pcie {
status = "okay";

host@0 {
reg = <0 0 0 0 0>;

#address-cells = <3>;
#size-cells = <2>;

ethernet@0 {
compatible = "weyland-yutani,noscom1", "microchip,lan743x";
status = "okay";
reg = <0 0 0 0 0>;
phy-connection-type = "rgmii";

fixed-link {
speed = <100>;
full-duplex;
};
};
};
};

Signed-off-by: Roelof Berg <rberg@berg-solutions.de>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'devlink-Add-support-for-control-packet-traps'
David S. Miller [Mon, 1 Jun 2020 18:49:24 +0000 (11:49 -0700)]
Merge branch 'devlink-Add-support-for-control-packet-traps'

Ido Schimmel says:

====================
devlink: Add support for control packet traps

So far device drivers were only able to register drop and exception
packet traps with devlink. These traps are used for packets that were
either dropped by the underlying device or encountered an exception
(e.g., missing neighbour entry) during forwarding.

However, in the steady state, the majority of the packets being trapped
to the CPU are packets that are required for the correct functioning of
the control plane. For example, ARP request and IGMP query packets.

This patch set allows device drivers to register such control traps with
devlink and expose their default control plane policy to user space.
User space can then tune the packet trap policer settings according to
its needs, as with existing packet traps.

In a similar fashion to exception traps, the action associated with such
traps cannot be changed as it can easily break the control plane. Unlike
drop and exception traps, packets trapped via control traps are not
reported to the kernel's drop monitor as they are not indicative of any
problem.

Patch set overview:

Patches #1-#3 break out layer 3 exceptions to a different group to
provide better granularity. A future patch set will make this completely
configurable.

Patch #4 adds a new trap action ('mirror') that is used for packets that
are forwarded by the device and sent to the CPU. Such packets are marked
by device drivers with 'skb->offload_fwd_mark = 1' in order to prevent
the kernel from forwarding them again.

Patch #5 adds the new trap type, 'control'.

Patches #6-#8 gradually add various control traps to devlink with proper
documentation.

Patch #9 adds a few control traps to netdevsim, which are automatically
exercised by existing devlink-trap selftest.

Patches #10 performs small refactoring in mlxsw.

Patches #11-#13 change mlxsw to register its existing control traps with
devlink.

Patch #14 adds a selftest over mlxsw that exercises all the registered
control traps.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests: mlxsw: Add test for control packets
Ido Schimmel [Fri, 29 May 2020 18:36:49 +0000 (21:36 +0300)]
selftests: mlxsw: Add test for control packets

Generate packets matching the various control traps and check that the
traps' stats increase accordingly.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum_trap: Register ACL control traps
Ido Schimmel [Fri, 29 May 2020 18:36:48 +0000 (21:36 +0300)]
mlxsw: spectrum_trap: Register ACL control traps

In a similar fashion to other control traps, register ACL control traps
with devlink.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum_trap: Register layer 3 control traps
Ido Schimmel [Fri, 29 May 2020 18:36:47 +0000 (21:36 +0300)]
mlxsw: spectrum_trap: Register layer 3 control traps

In a similar fashion to layer 2 control traps, register layer 3 control
traps with devlink.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum_trap: Register layer 2 control traps
Ido Schimmel [Fri, 29 May 2020 18:36:46 +0000 (21:36 +0300)]
mlxsw: spectrum_trap: Register layer 2 control traps

In a similar fashion to other traps, register layer 2 control traps with
devlink.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum_trap: Factor out common Rx listener function
Ido Schimmel [Fri, 29 May 2020 18:36:45 +0000 (21:36 +0300)]
mlxsw: spectrum_trap: Factor out common Rx listener function

We currently have an Rx listener function for exception traps that marks
received skbs with 'offload_fwd_mark' and injects them to the kernel's
Rx path. The marking is done because all these exceptions occur during
L3 forwarding, after the packets were potentially flooded at L2.

A subsequent patch will add support for control traps. Packets received
via some of these control traps need different handling:

1. Packets might not need to be marked with 'offload_fwd_mark'. For
   example, if packet was trapped before L2 forwarding

2. Packets might not need to be injected to the kernel's Rx path. For
   example, sampled packets are reported to user space via the psample
   module

Factor out a common Rx listener function that only reports trapped
packets to devlink. Call it from mlxsw_sp_rx_no_mark_listener() and
mlxsw_sp_rx_mark_listener() that will inject the packets to the kernel's
Rx path, without and with the marking, respectively.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetdevsim: Register control traps
Ido Schimmel [Fri, 29 May 2020 18:36:44 +0000 (21:36 +0300)]
netdevsim: Register control traps

Register two control traps with devlink. The existing selftest at
tools/testing/selftests/drivers/net/netdevsim/devlink_trap.sh iterates
over all registered traps and checks that the action of non-drop traps
cannot be changed. Up until now only exception traps were tested, now
control traps will be tested as well.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: Add ACL control packet traps
Ido Schimmel [Fri, 29 May 2020 18:36:43 +0000 (21:36 +0300)]
devlink: Add ACL control packet traps

Add packet traps for packets that are sampled / trapped by ACLs, so that
capable drivers could register them with devlink. Add documentation for
every added packet trap and packet trap group.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: Add layer 3 control packet traps
Ido Schimmel [Fri, 29 May 2020 18:36:42 +0000 (21:36 +0300)]
devlink: Add layer 3 control packet traps

Add layer 3 control packet traps such as ARP and DHCP, so that capable
device drivers could register them with devlink. Add documentation for
every added packet trap and packet trap group.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: Add layer 2 control packet traps
Ido Schimmel [Fri, 29 May 2020 18:36:41 +0000 (21:36 +0300)]
devlink: Add layer 2 control packet traps

Add layer 2 control packet traps such as STP and IGMP query, so that
capable device drivers could register them with devlink. Add
documentation for every added packet trap and packet trap group.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: Add 'control' trap type
Ido Schimmel [Fri, 29 May 2020 18:36:40 +0000 (21:36 +0300)]
devlink: Add 'control' trap type

This type is used for traps that trap control packets such as ARP
request and IGMP query to the CPU.

Do not report such packets to the kernel's drop monitor as they were not
dropped by the device no encountered an exception during forwarding.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: Add 'mirror' trap action
Ido Schimmel [Fri, 29 May 2020 18:36:39 +0000 (21:36 +0300)]
devlink: Add 'mirror' trap action

The action is used by control traps such as IGMP query. The packet is
flooded by the device, but also trapped to the CPU in order for the
software bridge to mark the receiving port as a multicast router port.
Such packets are marked with 'skb->offload_fwd_mark = 1' in order to
prevent the software bridge from flooding them again.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetdevsim: Move layer 3 exceptions to exceptions trap group
Ido Schimmel [Fri, 29 May 2020 18:36:38 +0000 (21:36 +0300)]
netdevsim: Move layer 3 exceptions to exceptions trap group

The layer 3 exceptions are still subject to the same trap policer, so
nothing changes, but user space can choose to assign a different one.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: spectrum_trap: Move layer 3 exceptions to exceptions trap group
Ido Schimmel [Fri, 29 May 2020 18:36:37 +0000 (21:36 +0300)]
mlxsw: spectrum_trap: Move layer 3 exceptions to exceptions trap group

The layer 3 exceptions are still subject to the same trap policer, so
nothing changes, but user space can choose to assign a different one.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: Create dedicated trap group for layer 3 exceptions
Ido Schimmel [Fri, 29 May 2020 18:36:36 +0000 (21:36 +0300)]
devlink: Create dedicated trap group for layer 3 exceptions

Packets that hit exceptions during layer 3 forwarding must be trapped to
the CPU for the control plane to function properly. Create a dedicated
group for them, so that user space could choose to assign a different
policer for them.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Mon, 1 Jun 2020 18:46:30 +0000 (11:46 -0700)]
Merge git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for net-next
to extend ctnetlink and the flowtable infrastructure:

1) Extend ctnetlink kernel side netlink dump filtering capabilities,
   from Romain Bellan.

2) Generalise the flowtable hook parser to take a hook list.

3) Pass a hook list to the flowtable hook registration/unregistration.

4) Add a helper function to release the flowtable hook list.

5) Update the flowtable event notifier to pass a flowtable hook list.

6) Allow users to add new devices to an existing flowtables.

7) Allow users to remove devices to an existing flowtables.

8) Allow for registering a flowtable with no initial devices.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: fec: disable correct clk in the err path of fec_enet_clk_enable
Liu Xiang [Fri, 29 May 2020 15:24:56 +0000 (23:24 +0800)]
net: fec: disable correct clk in the err path of fec_enet_clk_enable

When enable clk_ref failed, clk_ptp should be disabled rather than
clk_ref itself.

Signed-off-by: Liu Xiang <liuxiang_1999@126.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: octeon: mgmt: Repair filling of RX ring
Alexander Sverdlin [Fri, 29 May 2020 12:17:10 +0000 (14:17 +0200)]
net: octeon: mgmt: Repair filling of RX ring

The removal of mips_swiotlb_ops exposed a problem in octeon_mgmt Ethernet
driver. mips_swiotlb_ops had an mb() after most of the operations and the
removal of the ops had broken the receive functionality of the driver.
My code inspection has shown no other places except
octeon_mgmt_rx_fill_ring() where an explicit barrier would be obviously
missing. The latter function however has to make sure that "ringing the
bell" doesn't happen before RX ring entry is really written.

The patch has been successfully tested on Octeon II.

Fixes: a999933db9ed ("MIPS: remove mips_swiotlb_ops")
Cc: stable@vger.kernel.org
Signed-off-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'fix-indirect-flow_block-infrastructure'
David S. Miller [Mon, 1 Jun 2020 18:42:01 +0000 (11:42 -0700)]
Merge branch 'fix-indirect-flow_block-infrastructure'

Pablo Neira Ayuso says:

====================
the indirect flow_block infrastructure, revisited

This series fixes b5140a36da78 ("netfilter: flowtable: add indr block
setup support") that adds support for the indirect block for the
flowtable. This patch crashes the kernel with the TC CT action.

[  630.908086] BUG: kernel NULL pointer dereference, address: 00000000000000f0
[  630.908233] #PF: error_code(0x0000) - not-present page
[  630.908304] PGD 800000104addd067 P4D 800000104addd067 PUD 104311d067 PMD 0
[  630.908380] Oops: 0000 [#1] SMP PTI [  630.908615] RIP: 0010:nf_flow_table_indr_block_cb+0xc0/0x190 [nf_flow_table]
[  630.908690] Code: 5b 41 5c 41 5d 41 5e 41 5f 5d c3 4c 89 75 a0 4c 89 65 a8 4d 89 ee 49 89 dd 4c 89 fe 48 c7 c7 b7 64 36 a0 31 c0 e8 ce ed d8 e0 <49> 8b b7 f0 00 00 00 48 c7 c7 c8 64      36 a0 31 c0 e8 b9 ed d8 e0 49[  630.908790] RSP: 0018:ffffc9000895f8c0 EFLAGS: 00010246
[...]
[  630.910774] Call Trace:
[  630.911192]  ? mlx5e_rep_indr_setup_block+0x270/0x270 [mlx5_core]
[  630.911621]  ? mlx5e_rep_indr_setup_block+0x270/0x270 [mlx5_core]
[  630.912040]  ? mlx5e_rep_indr_setup_block+0x270/0x270 [mlx5_core]
[  630.912443]  flow_block_cmd+0x51/0x80
[  630.912844]  __flow_indr_block_cb_register+0x26c/0x510
[  630.913265]  mlx5e_nic_rep_netdevice_event+0x9e/0x110 [mlx5_core]
[  630.913665]  notifier_call_chain+0x53/0xa0
[  630.914063]  raw_notifier_call_chain+0x16/0x20
[  630.914466]  call_netdevice_notifiers_info+0x39/0x90
[  630.914859]  register_netdevice+0x484/0x550
[  630.915256]  __ip_tunnel_create+0x12b/0x1f0 [ip_tunnel]
[  630.915661]  ip_tunnel_init_net+0x116/0x180 [ip_tunnel]
[  630.916062]  ipgre_tap_init_net+0x22/0x30 [ip_gre]
[  630.916458]  ops_init+0x44/0x110
[  630.916851]  register_pernet_operations+0x112/0x200

A workaround patch to cure this crash has been proposed. However, there
is another problem: The indirect flow_block still does not work for the
new TC CT action. The problem is that the existing flow_indr_block_entry
callback assumes you can look up for the flowtable from the netdevice to
get the flow_block. This flow_block allows you to offload the flows via
TC_SETUP_CLSFLOWER. Unfortunately, it is not possible to get the
flow_block from the TC CT flowtables because they are _not_ bound to any
specific netdevice.

= What is the indirect flow_block infrastructure?

The indirect flow_block infrastructure allows drivers to offload
tc/netfilter rules that belong to software tunnel netdevices, e.g.
vxlan.

This indirect flow_block infrastructure relates tunnel netdevices with
drivers because there is no obvious way to relate these two things
from the control plane.

= How does the indirect flow_block work before this patchset?

Front-ends register the indirect block callback through
flow_indr_add_block_cb() if they support for offloading tunnel
netdevices.

== Setting up an indirect block

1) Drivers track tunnel netdevices via NETDEV_{REGISTER,UNREGISTER} events.
   If there is a new tunnel netdevice that the driver can offload, then the
   driver invokes __flow_indr_block_cb_register() with the new tunnel
   netdevice and the driver callback. The __flow_indr_block_cb_register()
   call iterates over the list of the front-end callbacks.

2) The front-end callback sets up the flow_block_offload structure and it
   invokes the driver callback to set up the flow_block.

3) The driver callback now registers the flow_block structure and it
   returns the flow_block back to the front-end.

4) The front-end gets the flow_block object and it is now ready to
   offload rules for this tunnel netdevice.

A simplified callgraph is represented below.

        Front-end                      Driver

                                   NETDEV_REGISTER
                                         |
                     __flow_indr_block_cb_register(netdev, cb_priv, driver_cb)
                                         | [1]
            .--------------frontend_indr_block_cb(cb_priv, driver_cb)
            |
            .
   setup_flow_block_offload(bo)
            | [2]
       driver_cb(bo, cb_priv) -----------.
                                         |
                                         \/
                                  set up flow_blocks [3]
                                         |
      add rules to flow_block <----------
      TC_SETUP_CLSFLOWER [4]

== Releasing the indirect flow_block

There are two possibilities, either tunnel netdevice is removed or
a netdevice (port representor) is removed.

=== Tunnel netdevice is removed

Driver waits for the NETDEV_UNREGISTER event that announces the tunnel
netdevice removal. Then, it calls __flow_indr_block_cb_unregister() to
remove the flow_block and rules.  Callgraph is very similar to the one
described above.

=== Netdevice is removed (port representor)

Driver calls __flow_indr_block_cb_unregister() to remove the existing
netfilter/tc rule that belong to the tunnel netdevice.

= How does the indirect flow_block work after this patchset?

Drivers register the indirect flow_block setup callback through
flow_indr_dev_register() if they support for offloading tunnel
netdevices.

== Setting up an indirect flow_block

1) Frontends check if dev->netdev_ops->ndo_setup_tc is unset. If so,
   frontends call flow_indr_dev_setup_offload(). This call invokes
   the drivers' indirect flow_block setup callback.

2) The indirect flow_block setup callback sets up a flow_block structure
   which relates the tunnel netdevice and the driver.

3) The front-end uses flow_block and offload the rules.

Note that the operational to set up (non-indirect) flow_block is very
similar.

== Releasing the indirect flow_block

=== Tunnel netdevice is removed

This calls flow_indr_dev_setup_offload() to set down the flow_block and
remove the offloaded rules. This alternate path is exercised if
dev->netdev_ops->ndo_setup_tc is unset.

=== Netdevice is removed (port representor)

If a netdevice is removed, then it might need to to clean up the
offloaded tc/netfilter rules that belongs to the tunnel netdevice:

1) The driver invokes flow_indr_dev_unregister() when a netdevice is
   removed.

2) This call iterates over the existing indirect flow_blocks
   and it invokes the cleanup callback to let the front-end remove the
   tc/netfilter rules. The cleanup callback already provides the
   flow_block that the front-end needs to clean up.

        Front-end                      Driver

                                         |
                            flow_indr_dev_unregister(...)
                                         |
                         iterate over list of indirect flow_block
                               and invoke cleanup callback
                                         |
            .-----------------------------
            |
            .
   frontend_flow_block_cleanup(flow_block)
            .
            |
           \/
   remove rules to flow_block
      TC_SETUP_CLSFLOWER

= About this patchset

This patchset aims to address the existing TC CT problem while
simplifying the indirect flow_block infrastructure. Saving 300 LoC in
the flow_offload core and the drivers. The operational gets aligned with
the (non-indirect) flow_blocks logic. Patchset is composed of:

Patch #1 add nf_flow_table_gc_cleanup() which is required by the
         netfilter's flowtable new indirect flow_block approach.

Patch #2 adds the flow_block_indr object which is actually part of
         of the flow_block object. This stores the indirect flow_block
         metadata such as the tunnel netdevice owner and the cleanup
         callback (in case the tunnel netdevice goes away).

         This patch adds flow_indr_dev_{un}register() to allow drivers
         to offer netdevice tunnel hardware offload to the front-ends.
         Then, front-ends call flow_indr_dev_setup_offload() to invoke
         the drivers to set up the (indirect) flow_block.

Patch #3 add the tcf_block_offload_init() helper function, this is
         a preparation patch to adapt the tc front-end to use this
         new indirect flow_block infrastructure.

Patch #4 updates the tc and netfilter front-ends to use the new
         indirect flow_block infrastructure.

Patch #5 updates the mlx5 driver to use the new indirect flow_block
         infrastructure.

Patch #6 updates the nfp driver to use the new indirect flow_block
         infrastructure.

Patch #7 updates the bnxt driver to use the new indirect flow_block
         infrastructure.

Patch #8 removes the indirect flow_block infrastructure version 1,
         now that frontends and drivers have been translated to
         version 2 (coming in this patchset).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: remove indirect block netdev event registration
Pablo Neira Ayuso [Fri, 29 May 2020 00:25:41 +0000 (02:25 +0200)]
net: remove indirect block netdev event registration

Drivers do not register to netdev events to set up indirect blocks
anymore. Remove __flow_indr_block_cb_register() and
__flow_indr_block_cb_unregister().

The frontends set up the callbacks through flow_indr_dev_setup_block()

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_tc: update indirect block support
Pablo Neira Ayuso [Fri, 29 May 2020 00:25:40 +0000 (02:25 +0200)]
bnxt_tc: update indirect block support

Register ndo callback via flow_indr_dev_register() and
flow_indr_dev_unregister().

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonfp: update indirect block support
Pablo Neira Ayuso [Fri, 29 May 2020 00:25:39 +0000 (02:25 +0200)]
nfp: update indirect block support

Register ndo callback via flow_indr_dev_register() and
flow_indr_dev_unregister().

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlx5: update indirect block support
Pablo Neira Ayuso [Fri, 29 May 2020 00:25:38 +0000 (02:25 +0200)]
mlx5: update indirect block support

Register ndo callback via flow_indr_dev_register() and
flow_indr_dev_unregister().

No need for mlx5e_rep_indr_clean_block_privs() since flow_block_cb_free()
already releases the internal mapping via ->release callback, which in
this case is mlx5e_rep_indr_tc_block_unbind().

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: use flow_indr_dev_setup_offload()
Pablo Neira Ayuso [Fri, 29 May 2020 00:25:37 +0000 (02:25 +0200)]
net: use flow_indr_dev_setup_offload()

Update existing frontends to use flow_indr_dev_setup_offload().

This new function must be called if ->ndo_setup_tc is unset to deal
with tunnel devices.

If there is no driver that is subscribed to new tunnel device
flow_block bindings, then this function bails out with EOPNOTSUPP.

If the driver module is removed, the ->cleanup() callback removes the
entries that belong to this tunnel device. This cleanup procedures is
triggered when the device unregisters the tunnel device offload handler.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: cls_api: add tcf_block_offload_init()
Pablo Neira Ayuso [Fri, 29 May 2020 00:25:36 +0000 (02:25 +0200)]
net: cls_api: add tcf_block_offload_init()

Add a helper function to initialize the flow_block_offload structure.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: flow_offload: consolidate indirect flow_block infrastructure
Pablo Neira Ayuso [Fri, 29 May 2020 00:25:35 +0000 (02:25 +0200)]
net: flow_offload: consolidate indirect flow_block infrastructure

Tunnel devices provide no dev->netdev_ops->ndo_setup_tc(...) interface.
The tunnel device and route control plane does not provide an obvious
way to relate tunnel and physical devices.

This patch allows drivers to register a tunnel device offload handler
for the tc and netfilter frontends through flow_indr_dev_register() and
flow_indr_dev_unregister().

The frontend calls flow_indr_dev_setup_offload() that iterates over the
list of drivers that are offering tunnel device hardware offload
support and it sets up the flow block for this tunnel device.

If the driver module is removed, the indirect flow_block ends up with a
stale callback reference. The module removal path triggers the
dev_shutdown() path to remove the qdisc and the flow_blocks for the
physical devices. However, this is not useful for tunnel devices, where
relation between the physical and the tunnel device is not explicit.

This patch introduces a cleanup callback that is invoked when the driver
module is removed to clean up the tunnel device flow_block. This patch
defines struct flow_block_indr and it uses it from flow_block_cb to
store the information that front-end requires to perform the
flow_block_cb cleanup on module removal.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetfilter: nf_flowtable: expose nf_flow_table_gc_cleanup()
Pablo Neira Ayuso [Fri, 29 May 2020 00:25:34 +0000 (02:25 +0200)]
netfilter: nf_flowtable: expose nf_flow_table_gc_cleanup()

This function schedules the flow teardown state and it forces a gc run.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/sched: fix a couple of splats in the error path of tfc_gate_init()
Davide Caratti [Thu, 28 May 2020 22:05:32 +0000 (00:05 +0200)]
net/sched: fix a couple of splats in the error path of tfc_gate_init()

trying to configure TC 'act_gate' rules with invalid control actions, the
following splat can be observed:

 general protection fault, probably for non-canonical address 0xdffffc0000000002: 0000 [#1] SMP KASAN NOPTI
 KASAN: null-ptr-deref in range [0x0000000000000010-0x0000000000000017]
 CPU: 1 PID: 2143 Comm: tc Not tainted 5.7.0-rc6+ #168
 Hardware name: Red Hat KVM, BIOS 1.11.1-4.module+el8.1.0+4066+0f1aadab 04/01/2014
 RIP: 0010:hrtimer_active+0x56/0x290
 [...]
  Call Trace:
  hrtimer_try_to_cancel+0x6d/0x330
  hrtimer_cancel+0x11/0x20
  tcf_gate_cleanup+0x15/0x30 [act_gate]
  tcf_action_cleanup+0x58/0x170
  __tcf_action_put+0xb0/0xe0
  __tcf_idr_release+0x68/0x90
  tcf_gate_init+0x7c7/0x19a0 [act_gate]
  tcf_action_init_1+0x60f/0x960
  tcf_action_init+0x157/0x2a0
  tcf_action_add+0xd9/0x2f0
  tc_ctl_action+0x2a3/0x39d
  rtnetlink_rcv_msg+0x5f3/0x920
  netlink_rcv_skb+0x121/0x350
  netlink_unicast+0x439/0x630
  netlink_sendmsg+0x714/0xbf0
  sock_sendmsg+0xe2/0x110
  ____sys_sendmsg+0x5b4/0x890
  ___sys_sendmsg+0xe9/0x160
  __sys_sendmsg+0xd3/0x170
  do_syscall_64+0x9a/0x370
  entry_SYSCALL_64_after_hwframe+0x44/0xa9

this is caused by hrtimer_cancel(), running before hrtimer_init(). Fix it
ensuring to call hrtimer_cancel() only if clockid is valid, and the timer
has been initialized. After fixing this splat, the same error path causes
another problem:

 general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] SMP KASAN NOPTI
 KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
 CPU: 1 PID: 980 Comm: tc Not tainted 5.7.0-rc6+ #168
 Hardware name: Red Hat KVM, BIOS 1.11.1-4.module+el8.1.0+4066+0f1aadab 04/01/2014
 RIP: 0010:release_entry_list+0x4a/0x240 [act_gate]
 [...]
 Call Trace:
  tcf_action_cleanup+0x58/0x170
  __tcf_action_put+0xb0/0xe0
  __tcf_idr_release+0x68/0x90
  tcf_gate_init+0x7ab/0x19a0 [act_gate]
  tcf_action_init_1+0x60f/0x960
  tcf_action_init+0x157/0x2a0
  tcf_action_add+0xd9/0x2f0
  tc_ctl_action+0x2a3/0x39d
  rtnetlink_rcv_msg+0x5f3/0x920
  netlink_rcv_skb+0x121/0x350
  netlink_unicast+0x439/0x630
  netlink_sendmsg+0x714/0xbf0
  sock_sendmsg+0xe2/0x110
  ____sys_sendmsg+0x5b4/0x890
  ___sys_sendmsg+0xe9/0x160
  __sys_sendmsg+0xd3/0x170
  do_syscall_64+0x9a/0x370
  entry_SYSCALL_64_after_hwframe+0x44/0xa9

the problem is similar: tcf_action_cleanup() was trying to release a list
without initializing it first. Ensure that INIT_LIST_HEAD() is called for
every newly created 'act_gate' action, same as what was done to 'act_ife'
with commit 44c23d71599f ("net/sched: act_ife: initalize ife->metalist
earlier").

Fixes: a51c328df310 ("net: qos: introduce a gate control flow action")
CC: Ivan Vecera <ivecera@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'regmap-simple-bit-helpers'
David S. Miller [Mon, 1 Jun 2020 18:35:18 +0000 (11:35 -0700)]
Merge branch 'regmap-simple-bit-helpers'

Bartosz Golaszewski says:

====================
regmap: provide simple bitops and use them in a driver

I noticed that oftentimes I use regmap_update_bits() for simple bit
setting or clearing. In this case the fourth argument is superfluous as
it's always 0 or equal to the mask argument.

This series proposes to add simple bit operations for setting, clearing
and testing specific bits with regmap.

The second patch uses all three in a driver that got recently picked into
the net-next tree.

The patches obviously target different trees so - if you're ok with
the change itself - I propose you pick the first one into your regmap
tree for v5.8 and then I'll resend the second patch to add the first
user for these macros for v5.9.

v1 -> v2:
- convert the new macros to static inline functions

v2 -> v3:
- drop unneeded ternary operator
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ethernet: mtk-star-emac: use regmap bitops
Bartosz Golaszewski [Thu, 28 May 2020 15:45:03 +0000 (17:45 +0200)]
net: ethernet: mtk-star-emac: use regmap bitops

Shrink the code visually by replacing regmap_update_bits() with
appropriate regmap bit operations where applicable.

Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoregmap: provide helpers for simple bit operations
Bartosz Golaszewski [Thu, 28 May 2020 15:45:02 +0000 (17:45 +0200)]
regmap: provide helpers for simple bit operations

In many instances regmap_update_bits() is used for simple bit setting
and clearing. In these cases the last argument is redundant and we can
hide it with a static inline function.

This adds three new helpers for simple bit operations: set_bits,
clear_bits and test_bits (the last one defined as a regular function).

Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agocxgb4: cleanup error code in setup_sge_queues_uld()
Dan Carpenter [Thu, 28 May 2020 12:49:57 +0000 (15:49 +0300)]
cxgb4: cleanup error code in setup_sge_queues_uld()

The caller doesn't care about the error codes, they only check for zero
vs non-zero.  Still, it's better to preserve the negative error codes
from alloc_uld_rxqs() instead of changing it to 1.  We can also return
directly if there is a failure.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'Fix-infinite-loop-in-bridge-and-vxlan-modules'
David S. Miller [Mon, 1 Jun 2020 18:08:41 +0000 (11:08 -0700)]
Merge branch 'Fix-infinite-loop-in-bridge-and-vxlan-modules'

Ido Schimmel says:

====================
Fix infinite loop in bridge and vxlan modules

When suppressing invalid IPv6 Neighbour Solicitation messages, it is
possible for the bridge and vxlan modules to get stuck in an infinite
loop. See the individual changelogs for detailed explanation of the
problem and solution.

The bug was originally reported against the bridge module, but after
auditing the code base I found that the buggy code was copied from the
vxlan module. This patch set fixes both modules. Could not find more
instances of the problem.

Please consider both patches for stable releases.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovxlan: Avoid infinite loop when suppressing NS messages with invalid options
Ido Schimmel [Mon, 1 Jun 2020 12:58:55 +0000 (15:58 +0300)]
vxlan: Avoid infinite loop when suppressing NS messages with invalid options

When proxy mode is enabled the vxlan device might reply to Neighbor
Solicitation (NS) messages on behalf of remote hosts.

In case the NS message includes the "Source link-layer address" option
[1], the vxlan device will use the specified address as the link-layer
destination address in its reply.

To avoid an infinite loop, break out of the options parsing loop when
encountering an option with length zero and disregard the NS message.

This is consistent with the IPv6 ndisc code and RFC 4886 which states
that "Nodes MUST silently discard an ND packet that contains an option
with length zero" [2].

[1] https://tools.ietf.org/html/rfc4861#section-4.3
[2] https://tools.ietf.org/html/rfc4861#section-4.6

Fixes: 4b29dba9c085 ("vxlan: fix nonfunctional neigh_reduce()")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobridge: Avoid infinite loop when suppressing NS messages with invalid options
Ido Schimmel [Mon, 1 Jun 2020 12:58:54 +0000 (15:58 +0300)]
bridge: Avoid infinite loop when suppressing NS messages with invalid options

When neighbor suppression is enabled the bridge device might reply to
Neighbor Solicitation (NS) messages on behalf of remote hosts.

In case the NS message includes the "Source link-layer address" option
[1], the bridge device will use the specified address as the link-layer
destination address in its reply.

To avoid an infinite loop, break out of the options parsing loop when
encountering an option with length zero and disregard the NS message.

This is consistent with the IPv6 ndisc code and RFC 4886 which states
that "Nodes MUST silently discard an ND packet that contains an option
with length zero" [2].

[1] https://tools.ietf.org/html/rfc4861#section-4.3
[2] https://tools.ietf.org/html/rfc4861#section-4.6

Fixes: ed842faeb2bd ("bridge: suppress nd pkts on BR_NEIGH_SUPPRESS ports")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: Alla Segal <allas@mellanox.com>
Tested-by: Alla Segal <allas@mellanox.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoipv4: nexthop: Fix deadcode issue by performing a proper NULL check
Patrick Eigensatz [Mon, 1 Jun 2020 11:12:01 +0000 (13:12 +0200)]
ipv4: nexthop: Fix deadcode issue by performing a proper NULL check

After allocating the spare nexthop group it should be tested for kzalloc()
returning NULL, instead the already used nexthop group (which cannot be
NULL at this point) had been tested so far.

Additionally, if kzalloc() fails, return ERR_PTR(-ENOMEM) instead of NULL.

Coverity-id: 1463885
Reported-by: Coverity <scan-admin@coverity.com>
Signed-off-by: Patrick Eigensatz <patrickeigensatz@gmail.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetoot...
David S. Miller [Mon, 1 Jun 2020 18:01:09 +0000 (11:01 -0700)]
Merge branch 'for-upstream' of git://git./linux/kernel/git/bluetooth/bluetooth-next

Johan Hedberg says:

====================
pull request: bluetooth-next 2020-06-01

Here's one last bluetooth-next pull request for 5.8, which I hope can
still be accepted.

 - Enabled Wide-Band Speech (WBS) support for Qualcomm wcn3991
 - Multiple fixes/imprvovements to Qualcomm-based devices
 - Fix GAP/SEC/SEM/BI-10-C qualfication test case
 - Added support for Broadcom BCM4350C5 device
 - Several other smaller fixes & improvements

Please let me know if there are any issues pulling. Thanks.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoBluetooth: hci_qca: Fix QCA6390 memdump failure
Zijun Hu [Fri, 29 May 2020 14:38:31 +0000 (22:38 +0800)]
Bluetooth: hci_qca: Fix QCA6390 memdump failure

QCA6390 memdump VSE sometimes come to bluetooth driver
with wrong sequence number as illustrated as follows:
frame # in dec: frame data in hex
1396: ff fd 01 08 74 05 00 37 8f 14
1397: ff fd 01 08 75 05 00 ff bf 38
1414: ff fd 01 08 86 05 00 fb 5e 4b
1399: ff fd 01 08 77 05 00 f3 44 0a
1400: ff fd 01 08 78 05 00 ca f7 41
it is mistook for controller missing packets, so results
in page fault after overwriting memdump buffer allocated.

Fixed by ignoring QCA6390 sequence number check and
checking buffer space before writing.

Signed-off-by: Zijun Hu <zijuhu@codeaurora.org>
Tested-by: Zijun Hu <zijuhu@codeaurora.org>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
3 years agoBluetooth: btmtkuart: Use serdev_device_write_buf() instead of serdev_device_write()
Zijun Hu [Fri, 29 May 2020 15:58:56 +0000 (23:58 +0800)]
Bluetooth: btmtkuart: Use serdev_device_write_buf() instead of serdev_device_write()

serdev_device_write() is not appropriate at here because
serdev_device_write_wakeup() is not used to release completion hold
by the former at @write_wakeup member of struct serdev_device_ops.

Fix by using serdev_device_write_buf() instead of serdev_device_write().

Signed-off-by: Zijun Hu <zijuhu@codeaurora.org>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
3 years agoBluetooth: hci_qca: Improve controller ID info log level
Zijun Hu [Fri, 29 May 2020 14:46:13 +0000 (22:46 +0800)]
Bluetooth: hci_qca: Improve controller ID info log level

Controller ID info got by VSC EDL_PATCH_GETVER is very
important, so improve its log level from DEBUG to INFO.

Signed-off-by: Zijun Hu <zijuhu@codeaurora.org>
Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
3 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
David S. Miller [Mon, 1 Jun 2020 00:48:46 +0000 (17:48 -0700)]
Merge git://git./linux/kernel/git/netdev/net

xdp_umem.c had overlapping changes between the 64-bit math fix
for the calculation of npgs and the removal of the zerocopy
memory type which got rid of the chunk_size_nohdr member.

The mlx5 Kconfig conflict is a case where we just take the
net-next copy of the Kconfig entry dependency as it takes on
the ESWITCH dependency by one level of indirection which is
what the 'net' conflicting change is trying to ensure.

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge tag 'mac80211-next-for-davem-2020-05-31' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Sun, 31 May 2020 21:32:50 +0000 (14:32 -0700)]
Merge tag 'mac80211-next-for-davem-2020-05-31' of git://git./linux/kernel/git/jberg/mac80211-next

Johannes Berg says:

====================
Another set of changes, including
 * many 6 GHz changes, though it's not _quite_ complete
   (I left out scanning for now, we're still discussing)
 * allow userspace SA-query processing for operating channel
   validation
 * TX status for control port TX, for AP-side operation
 * more per-STA/TID control options
 * move to kHz for channels, for future S1G operation
 * various other small changes
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agocheckpatch/coding-style: deprecate 80-column warning
Joe Perches [Fri, 29 May 2020 23:12:21 +0000 (16:12 -0700)]
checkpatch/coding-style: deprecate 80-column warning

Yes, staying withing 80 columns is certainly still _preferred_.  But
it's not the hard limit that the checkpatch warnings imply, and other
concerns can most certainly dominate.

Increase the default limit to 100 characters.  Not because 100
characters is some hard limit either, but that's certainly a "what are
you doing" kind of value and less likely to be about the occasional
slightly longer lines.

Miscellanea:

 - to avoid unnecessary whitespace changes in files, checkpatch will no
   longer emit a warning about line length when scanning files unless
   --strict is also used

 - Add a bit to coding-style about alignment to open parenthesis

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agoMerge tag 'x86-urgent-2020-05-31' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 31 May 2020 17:45:11 +0000 (10:45 -0700)]
Merge tag 'x86-urgent-2020-05-31' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "A pile of x86 fixes:

   - Prevent a memory leak in ioperm which was caused by the stupid
     assumption that the exit cleanup is always called for current,
     which is not the case when fork fails after taking a reference on
     the ioperm bitmap.

   - Fix an arithmething overflow in the DMA code on 32bit systems

   - Fill gaps in the xstate copy with defaults instead of leaving them
     uninitialized

   - Revert: "Make __X32_SYSCALL_BIT be unsigned long" as it turned out
     that existing user space fails to build"

* tag 'x86-urgent-2020-05-31' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/ioperm: Prevent a memory leak when fork fails
  x86/dma: Fix max PFN arithmetic overflow on 32 bit systems
  copy_xstate_to_kernel(): don't leave parts of destination uninitialized
  x86/syscalls: Revert "x86/syscalls: Make __X32_SYSCALL_BIT be unsigned long"

3 years agoMerge tag 'sched-urgent-2020-05-31' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 31 May 2020 17:43:17 +0000 (10:43 -0700)]
Merge tag 'sched-urgent-2020-05-31' of git://git./linux/kernel/git/tip/tip

Pull scheduler fix from Thomas Gleixner:
 "A single scheduler fix preventing a crash in NUMA balancing.

  The current->mm check is not reliable as the mm might be temporary due
  to use_mm() in a kthread. Check for PF_KTHREAD explictly"

* tag 'sched-urgent-2020-05-31' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/fair: Don't NUMA balance for kthreads

3 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Sun, 31 May 2020 17:16:53 +0000 (10:16 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:
 "Another week, another set of bug fixes:

   1) Fix pskb_pull length in __xfrm_transport_prep(), from Xin Long.

   2) Fix double xfrm_state put in esp{4,6}_gro_receive(), also from Xin
      Long.

   3) Re-arm discovery timer properly in mac80211 mesh code, from Linus
      Lüssing.

   4) Prevent buffer overflows in nf_conntrack_pptp debug code, from
      Pablo Neira Ayuso.

   5) Fix race in ktls code between tls_sw_recvmsg() and
      tls_decrypt_done(), from Vinay Kumar Yadav.

   6) Fix crashes on TCP fallback in MPTCP code, from Paolo Abeni.

   7) More validation is necessary of untrusted GSO packets coming from
      virtualization devices, from Willem de Bruijn.

   8) Fix endianness of bnxt_en firmware message length accesses, from
      Edwin Peer.

   9) Fix infinite loop in sch_fq_pie, from Davide Caratti.

  10) Fix lockdep splat in DSA by setting lockless TX in netdev features
      for slave ports, from Vladimir Oltean.

  11) Fix suspend/resume crashes in mlx5, from Mark Bloch.

  12) Fix use after free in bpf fmod_ret, from Alexei Starovoitov.

  13) ARP retransmit timer guard uses wrong offset, from Hongbin Liu.

  14) Fix leak in inetdev_init(), from Yang Yingliang.

  15) Don't try to use inet hash and unhash in l2tp code, results in
      crashes. From Eric Dumazet"

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (77 commits)
  l2tp: add sk_family checks to l2tp_validate_socket
  l2tp: do not use inet_hash()/inet_unhash()
  net: qrtr: Allocate workqueue before kernel_bind
  mptcp: remove msk from the token container at destruction time.
  mptcp: fix race between MP_JOIN and close
  mptcp: fix unblocking connect()
  net/sched: act_ct: add nat mangle action only for NAT-conntrack
  devinet: fix memleak in inetdev_init()
  virtio_vsock: Fix race condition in virtio_transport_recv_pkt
  drivers/net/ibmvnic: Update VNIC protocol version reporting
  NFC: st21nfca: add missed kfree_skb() in an error path
  neigh: fix ARP retransmit timer guard
  bpf, selftests: Add a verifier test for assigning 32bit reg states to 64bit ones
  bpf, selftests: Verifier bounds tests need to be updated
  bpf: Fix a verifier issue when assigning 32bit reg states to 64bit ones
  bpf: Fix use-after-free in fmod_ret check
  net/mlx5e: replace EINVAL in mlx5e_flower_parse_meta()
  net/mlx5e: Fix MLX5_TC_CT dependencies
  net/mlx5e: Properly set default values when disabling adaptive moderation
  net/mlx5e: Fix arch depending casting issue in FEC
  ...

3 years agoice: Ignore EMODE when setting PHY config
Chinh T Cao [Sat, 16 May 2020 00:55:06 +0000 (17:55 -0700)]
ice: Ignore EMODE when setting PHY config

When setting the PHY cfg (CQ cmd 0x0601), if the firmware responds
with an EMODE error, software will ignore the error as it simply
means that manageability (ex: BMC) is in control of the link and that
the new setting may not be applied.

Signed-off-by: Chinh T Cao <chinh.t.cao@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: fix aRFS after flow director delete
Henry Tieman [Sat, 16 May 2020 00:55:05 +0000 (17:55 -0700)]
ice: fix aRFS after flow director delete

The logic was missing for adding back perfect flows after flow director
filter delete. The code now adds perfect flows into the HW tables after
filter delete.

Signed-off-by: Henry Tieman <henry.w.tieman@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: Use coalesce values from q_vector 0 when increasing q_vectors
Brett Creeley [Sat, 16 May 2020 00:55:04 +0000 (17:55 -0700)]
ice: Use coalesce values from q_vector 0 when increasing q_vectors

Currently when a VSI is built (i.e. reset, set channels, etc.)
the coalesce settings will be preserved in most cases. However, when the
number of q_vectors are increased the settings for the new q_vectors
will be set to the driver defaults of AIM on, Rx/Tx ITR 50, and INTRL 0.
This is causing issues with how the ethtool layer gets the current
coalesce settings since it only uses q_vector 0. So, assume that the user
set the coalesce settings globally (i.e. ethtool -C eth0) and use q_vector
0's settings for all of the new q_vectors.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: fix PCI device serial number to be lowercase values
Paul M Stillwell Jr [Sat, 16 May 2020 00:55:03 +0000 (17:55 -0700)]
ice: fix PCI device serial number to be lowercase values

Commit ceb2f00707f9 ("ice: Use pci_get_dsn()") changed the code to
use a new function to get the Device Serial Number. It also changed
the case of the filename for loading a package on a specific NIC
from lowercase to uppercase. Change the filename back to
lowercase since that is what we specified.

Fixes: ceb2f00707f9 ("ice: Use pci_get_dsn()")
Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: fix function signature style format
Bruce Allan [Sat, 16 May 2020 00:55:02 +0000 (17:55 -0700)]
ice: fix function signature style format

Where possible, cuddle multiple lines of function signatures to be
consistent throughout the code.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: Allow VF to request reset as soon as it's initialized
Brett Creeley [Sat, 16 May 2020 00:55:01 +0000 (17:55 -0700)]
ice: Allow VF to request reset as soon as it's initialized

A VF driver has the ability to request reset via VIRTCHNL_OP_RESET_VF.
This is a required step in VF driver load. Currently, the PF is only
allowing a VF to request reset using this method after the VF has
already communicated resources via VIRTCHNL_OP_GET_VF_RESOURCES.
However, this is incorrect because the VF can request reset before
requesting resources. Fix this by allowing the VF to request a reset
once it has been initialized.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: Fix inability to set channels when down
Jesse Brandeburg [Sat, 16 May 2020 00:55:00 +0000 (17:55 -0700)]
ice: Fix inability to set channels when down

Currently the driver prevents a user from doing
modprobe ice
ethtool -L eth0 combined 5
ip link set eth0 up

The ethtool command fails, because the driver is checking to see if the
interface is down before allowing the get_channels to proceed (even for
a set_channels).

Remove this check and allow the user to configure the interface
before bringing it up, which is a much better usability case.

Fixes: 87324e747fde ("ice: Implement ethtool ops for channels")
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: Always clear QRXFLXP_CNTXT before writing new value
Brett Creeley [Sat, 16 May 2020 00:54:59 +0000 (17:54 -0700)]
ice: Always clear QRXFLXP_CNTXT before writing new value

Always clear the previous value in QRXFLXP_CNTXT before writing a new
value. This will make it so re-used queues will not accidentally take the
previously configured settings.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: Reset VF for all port VLAN changes from host
Brett Creeley [Sat, 16 May 2020 00:54:58 +0000 (17:54 -0700)]
ice: Reset VF for all port VLAN changes from host

Currently the PF is modifying the VF's port VLAN on the fly when
configured via iproute. This is okay for most cases, but if the VF
already has guest VLANs configured the PF has to remove all of those
filters so only VLAN tagged traffic that matches the port VLAN will
pass. Instead of adding functionality to track which guest VLANs have
been added, just reset the VF each time port VLAN parameters are
modified.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: Update ICE_PHY_TYPE_HIGH_MAX_INDEX value
Chinh T Cao [Sat, 16 May 2020 00:51:21 +0000 (17:51 -0700)]
ice: Update ICE_PHY_TYPE_HIGH_MAX_INDEX value

As currently, we are supporting only 5 PHY_SPEEDs for phy_type_high.
Thus, we should adjust the value of ICE_PHY_TYPE_HIGH_MAX_INDEX to 5.

Signed-off-by: Chinh T Cao <chinh.t.cao@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: Increase timeout after PFR
Dan Nowlin [Sat, 16 May 2020 00:51:20 +0000 (17:51 -0700)]
ice: Increase timeout after PFR

To allow for resets during package download, increase the timeout period
after performing a PFR. The time waited is the global config lock
timeout plus the normal PFSWR timeout.

Signed-off-by: Dan Nowlin <dan.nowlin@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: Fix transmit for all software offloaded VLANs
Brett Creeley [Sat, 16 May 2020 00:51:19 +0000 (17:51 -0700)]
ice: Fix transmit for all software offloaded VLANs

Currently the driver does not recognize when there is an 802.1AD VLAN
tag right after the dmac/smac (outermost VLAN tag). If any DCB map is
applied and/or DCB is enabled this is causing the hardware to insert a
VLAN 0 tag after the 802.1AD VLAN tag that is already in the packet.
Fix this by preventing VLAN tag 0 from being added when any VLAN is
already present after dmac/smac (software offloaded) or skb (hardware
offloaded).

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: support adding 16 unicast/multicast filter on untrusted VF
Paul Greenwalt [Sat, 16 May 2020 00:51:18 +0000 (17:51 -0700)]
ice: support adding 16 unicast/multicast filter on untrusted VF

Allow untrusted VF to add 16 unicast/multicast filters. VF uses 1 filter
for the default/perm_addr/LAA MAC, 1 for broadcast, and 16 additional
unicast/multicast filters.

Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agoice: allow host to clear administratively set VF MAC
Brett Creeley [Sat, 16 May 2020 00:51:17 +0000 (17:51 -0700)]
ice: allow host to clear administratively set VF MAC

Currently a user is not allowed to clear a VF's administratively set MAC
on the PF. Fix this by allowing an all zero MAC address via "ip link set
${pf_eth} vf ${vf_id} mac 00:00:00:00:00:00".

An example use case for this would be issuing a "virsh shutdown"
command on a VM. The call to iproute mentioned above is part of this flow.
Without this change the driver incorrectly rejects clearing the VF's
administratively set MAC and prints unhelpful log messages.

Also, improve the comments surrounding this change.

Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
3 years agocfg80211: support bigger kek/kck key length
Nathan Errera [Thu, 28 May 2020 19:22:38 +0000 (21:22 +0200)]
cfg80211: support bigger kek/kck key length

With some newer AKMs, the KCK and KEK are bigger, so allow that
if the driver advertises support for it. In addition, add a new
attribute for the AKM so we can use it for offloaded rekeying.

Signed-off-by: Nathan Errera <nathan.errera@intel.com>
[reword commit message]
Link: https://lore.kernel.org/r/20200528212237.5eb58b00a5d1.I61b09d77c4f382e8d58a05dcca78096e99a6bc15@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: set short_slot for 6 GHz band
Tova Mussai [Thu, 28 May 2020 19:34:46 +0000 (21:34 +0200)]
mac80211: set short_slot for 6 GHz band

Set short slot also for 6 GHz band, just like 5 GHz.

Signed-off-by: Tova Mussai <tova.mussai@intel.com>
Link: https://lore.kernel.org/r/20200528213443.75f38e6f5efd.I272fbae402b03123f04e9ae69204eeab960c70cd@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: Consider 6 GHz band when handling power constraint
Ilan Peer [Thu, 28 May 2020 19:34:45 +0000 (21:34 +0200)]
mac80211: Consider 6 GHz band when handling power constraint

Treat it like the 5 GHz band.

Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Link: https://lore.kernel.org/r/20200528213443.889e5c9dd006.Id8ed3bb8000ba8738be5df05639415eb2e23c61a@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: accept aggregation sessions on 6 GHz
Johannes Berg [Thu, 28 May 2020 19:34:44 +0000 (21:34 +0200)]
mac80211: accept aggregation sessions on 6 GHz

On 6 GHz, stations don't have ht_supported set, but they can
still do aggregation since they must have HE, allow that.

Link: https://lore.kernel.org/r/20200528213443.776d3c891b64.Ifa099d450617b50c691832b3c4aa08959fab520a@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agocfg80211: require HE capabilities for 6 GHz band
Johannes Berg [Thu, 28 May 2020 19:34:42 +0000 (21:34 +0200)]
cfg80211: require HE capabilities for 6 GHz band

On 6 GHz band, HE capabilities must be available for all of
the interface types, otherwise we shouldn't use 6 GHz. Check
this.

Link: https://lore.kernel.org/r/20200528213443.5881cb3c8c4a.I583b54172f91f98d44af64a16c5826fe458cbb27@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agocfg80211: reject HT/VHT capabilities on 6 GHz band
Johannes Berg [Thu, 28 May 2020 19:34:41 +0000 (21:34 +0200)]
cfg80211: reject HT/VHT capabilities on 6 GHz band

On the 6 GHz band, HE should be used, but without any direct HT/VHT
capabilities, instead the HE 6 GHz band capabilities will capture
the relevant information. Reject HT/VHT capabilities here.

Link: https://lore.kernel.org/r/20200528213443.bfe89c35459a.Ibba5e066fa0087fd49d13cfee89d196ea0c68ae2@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agocfg80211: treat 6 GHz channels as valid regardless of capability
Johannes Berg [Thu, 28 May 2020 19:34:40 +0000 (21:34 +0200)]
cfg80211: treat 6 GHz channels as valid regardless of capability

If a 6 GHz channel exists, then we can probably safely assume that
the device actually supports it, and then it should support most
bandwidths.

This will probably need to be extended to check the interface type
and then dig into the HE capabilities for that though, to have the
correct bandwidth check.

Link: https://lore.kernel.org/r/20200528213443.d4864ef52e92.I82f09b2b14a56413ce20376d09967fe954a033eb@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: Add HE 6GHz capabilities element to probe request
Ilan Peer [Thu, 28 May 2020 19:34:39 +0000 (21:34 +0200)]
mac80211: Add HE 6GHz capabilities element to probe request

On 6 GHz, the 6 GHz capabilities element should be added, do that.

Signed-off-by: Ilan Peer <ilan.peer@intel.com>
[add commit message]
Link: https://lore.kernel.org/r/20200528213443.8ee764f0cde0.I2b0c66b60e11818c97c9803e04a6a197c6376243@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: use HE 6 GHz band capability and pass it to the driver
Johannes Berg [Thu, 28 May 2020 19:34:38 +0000 (21:34 +0200)]
mac80211: use HE 6 GHz band capability and pass it to the driver

In order to handle 6 GHz AP side, take the HE 6 GHz band capability
data and pass it to the driver (which needs it for A-MPDU spacing
and A-MPDU length).

Link: https://lore.kernel.org/r/1589399105-25472-6-git-send-email-rmanohar@codeaurora.org
Co-developed-by: Rajkumar Manoharan <rmanohar@codeaurora.org>
Signed-off-by: Rajkumar Manoharan <rmanohar@codeaurora.org>
Link: https://lore.kernel.org/r/20200528213443.784e4890d82f.I5f1230d5ab27e84e7bbe88e3645b24ea15a0c146@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: check the correct bit for EMA AP
Shaul Triebitz [Thu, 28 May 2020 19:34:37 +0000 (21:34 +0200)]
mac80211: check the correct bit for EMA AP

An AP supporting EMA (Enhanced Multi-BSSID advertisement) should set
bit 83 in the extended capabilities IE (9.4.2.26 in the 802.11ax D5 spec).
So the *3rd* bit of the 10th byte should be checked.
Also, in one place, the wrong byte was checked.
(cfg80211_find_ie returns a pointer to the beginning of the IE,
 so the data really starts at ie[2], so the 10th byte
 should be ie[12]. To avoid this confusion, use cfg80211_find_elem
 instead).

Signed-off-by: Shaul Triebitz <shaul.triebitz@intel.com>
Link: https://lore.kernel.org/r/20200528213443.4316121fa2a3.I9745582f8d41ad8e689dac0fefcd70b276d7c1ea@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: determine chandef from HE 6 GHz operation
Johannes Berg [Thu, 28 May 2020 19:34:36 +0000 (21:34 +0200)]
mac80211: determine chandef from HE 6 GHz operation

Support connecting to HE 6 GHz APs and mesh networks on 6 GHz,
where the HT/VHT information is missing but instead the HE 6 GHz
band capability is present, and the 6 GHz Operation information
field is used to encode the channel configuration instead of the
HT/VHT operation elements.

Also add some other bits needed to connect to 6 GHz networks.

Link: https://lore.kernel.org/r/1589399105-25472-10-git-send-email-rmanohar@codeaurora.org
Co-developed-by: Rajkumar Manoharan <rmanohar@codeaurora.org>
Signed-off-by: Rajkumar Manoharan <rmanohar@codeaurora.org>
Link: https://lore.kernel.org/r/20200528213443.25687d2695bc.I3f9747c1147480f65445f13eda5c4a5ed4e86757@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: avoid using ext NSS high BW if not supported
Johannes Berg [Thu, 28 May 2020 19:34:35 +0000 (21:34 +0200)]
mac80211: avoid using ext NSS high BW if not supported

If the AP advertises inconsistent data, namely it has CCFS1 or CCFS2,
but doesn't advertise support for 160/80+80 bandwidth or "Extended NSS
BW Support", then we cannot use any MCSes in the the higher bandwidth.
Thus, avoid connecting with higher bandwidth since it's less efficient
that way.

Link: https://lore.kernel.org/r/20200528213443.0e55d40c3ccc.I6fd0b4708ebd087e5e46466c3e91f6efbcbef668@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: do not allow HT/VHT IEs in 6 GHz mesh mode
Rajkumar Manoharan [Thu, 28 May 2020 19:34:34 +0000 (21:34 +0200)]
mac80211: do not allow HT/VHT IEs in 6 GHz mesh mode

As HT/VHT elements are not allowed in 6 GHz band, do not include
them in mesh beacon template formation.

Signed-off-by: Rajkumar Manoharan <rmanohar@codeaurora.org>
Link: https://lore.kernel.org/r/1589399105-25472-9-git-send-email-rmanohar@codeaurora.org
Link: https://lore.kernel.org/r/20200528193455.76796-2-johannes@sipsolutions.net
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: build HE operation with 6 GHz oper information
Rajkumar Manoharan [Thu, 28 May 2020 19:34:33 +0000 (21:34 +0200)]
mac80211: build HE operation with 6 GHz oper information

Add 6 GHz operation information (IEEE 802.11ax/D6.0, Figure 9-787k)
while building HE operation element for non-HE AP. This field is used to
determine channel information in the absence of HT/VHT IEs.

Signed-off-by: Rajkumar Manoharan <rmanohar@codeaurora.org>
Link: https://lore.kernel.org/r/1589399105-25472-8-git-send-email-rmanohar@codeaurora.org
[fix skb allocation size]
Link: https://lore.kernel.org/r/20200528193455.76796-1-johannes@sipsolutions.net
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 years agomac80211: add HE 6 GHz Band Capability element
Rajkumar Manoharan [Thu, 28 May 2020 19:34:32 +0000 (21:34 +0200)]
mac80211: add HE 6 GHz Band Capability element

Construct HE 6 GHz band capability element (IEEE 802.11ax/D6.0,
9.4.2.261) for association request and mesh beacon. The 6 GHz
capability information is passed by driver through iftypes caps.

Signed-off-by: Rajkumar Manoharan <rmanohar@codeaurora.org>
Link: https://lore.kernel.org/r/1589399105-25472-7-git-send-email-rmanohar@codeaurora.org
[handle SMPS, adjust for previous patches, reserve SKB space properly,
 change to handle SKB directly]
Link: https://lore.kernel.org/r/20200528213443.643aa8101111.I3f9747c1147480f65445f13eda5c4a5ed4e86757@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>