Currently we support SERDES on the Topaz family in a limited way: no
IRQs and the cmode is not writable, thus the mode is determined by
strapping pins.

Marvell's examples though show how to make cmode writable on port 5 and
support SGMII autonegotiation. It is done by writing hidden registers,
for which we already have code.

This patch adds support for making the cmode for the SERDES port
writable on the Topaz family, via a new chip operation,
.port_set_cmode_writable, which is called from mv88e6xxx_port_setup_mac
just before .port_set_cmode.

SERDES IRQs are also enabled for Topaz.

Tested on Turris Mox.
Signed-off-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

This is a cosmetic update. We are removing the last underscore from
macros MV88E6XXX_PORT_STS_CMODE_100BASE_X and
MV88E6XXX_PORT_STS_CMODE_1000BASE_X. The 2500base-x version does not
have that underscore. Also PHY_INTERFACE_MODE_ macros do not have it
there.
Signed-off-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

By adding an additional serdes_get_lane implementation (for Topaz), we
can merge the implementations of other SERDES functions (powering and
IRQs). We can skip checking port numbers, since the serdes_get_lane()
methods inform if there is no lane on a port or if the lane cannot be
used for given cmode.
Signed-off-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Create a serdes_get_lane() method in the mv88e6xxx operations structure.
Use it instead of calling the different implementations.
Also change the methods so that their return value is used only for
error. The lane number is put into a place referred to by a pointer
given as argument. If the port does not have a lane, return -ENODEV.
Lanes are phy addresses, so use u8 as their type.
Signed-off-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch moves the functions operating on the hidden debug registers
into it's own file, port_hidden.c. The functions prefix is renamed from
mv88e6390_hidden_ to mv88e6xxx_port_hidden_, to be consistent with the
rest of this driver. The macros are prefixed with MV88E6XXX_ prefix, and
are changed not to use the BIT() macro nor bit shifts, since the rest of
the port.h file does not use it.

We also add the support for setting the Block Address field when
operating hidden registers. Marvell's mdio examples for SERDES settings
on Topaz use Block Address 0x7 when reading/writing hidden registers,
and although the specification says that block must be set to 0xf, those
settings are reachable only with Block Address 0x7.
Signed-off-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The mv88e6390_serdes_irq_link_sgmii IRQ handler reads the SERDES PHY
status register to determine speed, among other things. If cmode of the
port is set to 2500base-x, though, the PHY still reports 1000 Mbps (the
PHY register itself does not differentiate between 1000 Mbps and 2500
Mbps - it thinks it is running at 1000 Mbps, although clock is 2.5x
faster).
Look at the cmode and set SPEED_2500 if cmode is set to 2500base-x.
Also tell mv88e6xxx_port_setup_mac the PHY interface mode corresponding
to current cmode in terms of phy_interface_t.
Signed-off-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Xin Long says:

====================
sctp: add SCTP_ECN_SUPPORTED sockopt

This patchset is to make ecn flag per netns and endpoint and then
add SCTP_ECN_SUPPORTED sockopt, as does for other feature flags.
====================
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

SCTP_ECN_SUPPORTED sockopt will be added to allow users to change
ep ecn flag, and it's similar with other feature flags.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sysctl net.sctp.ecn_enable is added in this patch. It will allow
users to change the default sctp ecn flag, net.sctp.ecn_enable.

This feature was also required on this thread:

  http://lkml.iu.edu/hypermail/linux/kernel/0812.1/01858.htmlSigned-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch is to add ecn flag for both netns_sctp and sctp_endpoint,
net->sctp.ecn_enable is set 1 by default, and ep->ecn_enable will
be initialized with net->sctp.ecn_enable.

asoc->peer.ecn_capable will be set during negotiation only when
ep->ecn_enable is set on both sides.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fixes gcc '-Wunused-but-set-variable' warning:
drivers/net/ethernet/mediatek/mtk_eth_soc.c: In function mtk_handle_irq:
drivers/net/ethernet/mediatek/mtk_eth_soc.c:1951:6: warning: variable status set but not used [-Wunused-but-set-variable]

Fixes: 296c9120 ("net: ethernet: mediatek: Add MT7628/88 SoC support")
Signed-off-by: Mao Wenan <maowenan@huawei.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Vladimir Oltean says:

====================
Simplify DSA handling of VLAN subinterface offload

Depends on Vivien Didelot's patchset:
https://patchwork.ozlabs.org/project/netdev/list/?series=127197&state=*

This patchset removes a few strange-looking guards for -EOPNOTSUPP in
dsa_slave_vlan_rx_add_vid and dsa_slave_vlan_rx_kill_vid, making that
code path no longer possible.

It also disables the code path for the sja1105 driver, which does
support editing the VLAN table, but not hardware-accelerated VLAN
sub-interfaces, therefore the check in the DSA core would be wrong.
There was no better DSA callback to do this than .port_enable, i.e.
at ndo_open time.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The switch barely supports traffic I/O, and it does that by repurposing
VLANs when there is no bridge that is taking control of them.

Letting DSA declare this netdev feature as supported (see
dsa_slave_create) would mean that VLAN sub-interfaces created on sja1105
switch ports will be hardware offloaded. That means that
net/8021q/vlan_core.c would install the VLAN into the filter tables of
the switch, potentially interfering with the tag_8021q VLANs.

We need to prevent that from happening and not let the 8021q core
offload VLANs to the switch hardware tables. In vlan_filtering=0 modes
of operation, the switch ports can pass through VLAN-tagged frames with
no problem.
Suggested-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When adding a VLAN sub-interface on a DSA slave port, the 8021q core
checks NETIF_F_HW_VLAN_CTAG_FILTER and, if the netdev is capable of
filtering, calls .ndo_vlan_rx_add_vid or .ndo_vlan_rx_kill_vid to
configure the VLAN offloading.

DSA sets this up counter-intuitively: it always advertises this netdev
feature, but the underlying driver may not actually support VLAN table
manipulation. In that case, the DSA core is forced to ignore the error,
because not being able to offload the VLAN is still fine - and should
result in the creation of a non-accelerated VLAN sub-interface.

Change this so that the netdev feature is only advertised for switch
drivers that support VLAN manipulation, instead of checking for
-EOPNOTSUPP at runtime.
Suggested-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

René van Dorst says:

====================
net: ethernet: mediatek: convert to PHYLINK

These patches converts mediatek driver to PHYLINK API.

v3->v4:
* Phylink improvements and clean-ups after review
v2->v3:
* Phylink improvements and clean-ups after review
v1->v2:
* Rebase for mt76x8 changes
* Phylink improvements and clean-ups after review
* SGMII port doesn't support 2.5Gbit in SGMII mode only in BASE-X mode.
  Refactor the code.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch the removes the recently added mediatek,physpeed property.
Use the fixed-link property speed = <2500> to set the phy in 2.5Gbit.
See mt7622-bananapi-bpi-r64.dts for a working example.
Signed-off-by: René van Dorst <opensource@vdorst.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

* Re-add SGMII support but now with PHYLINK API support
  So the SGMII changes are more clear
* Move SGMII block setup from mtk_gmac_sgmii_path_setup() to
  mtk_mac_config()
* Merge mtk_setup_hw_path() into mtk_mac_config()
* Remove mediatek,physpeed property, fixed-link supports now any speed so
  speed = <2500>; is now valid with PHYLINK
* Demagic SGMII register values
* Use phylink state to setup fixed-link mode
Signed-off-by: René van Dorst <opensource@vdorst.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This convert the basics to PHYLINK API.
SGMII support is not in this patch.
Signed-off-by: René van Dorst <opensource@vdorst.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Vivien Didelot says:

====================
net: dsa: explicit programmation of VLAN on CPU ports

When a VLAN is programmed on a user port, every switch of the fabric also
program the CPU ports and the DSA links as part of the VLAN. To do that,
DSA makes use of bitmaps to prepare all members of a VLAN.

While this is expected for DSA links which are used as conduit between
interconnected switches, only the dedicated CPU port of the slave must be
programmed, not all CPU ports of the fabric. This may also cause problems in
other corners of DSA such as the tag_8021q.c driver, which needs to program
its ports manually, CPU port included.

We need the dsa_port_vlan_{add,del} functions and its dsa_port_vid_{add,del}
variants to simply trigger the VLAN programmation without any logic in them,
but they may currently skip the operation based on the bridge device state.

This patchset gets rid of the bitmap operations, and moves the bridge device
check as well as the explicit programmation of CPU ports where they belong,
in the slave code.

While at it, clear the VLAN flags before programming a CPU port, as it
doesn't make sense to forward the PVID flag for example for such ports.

Changes in v2: only clear the PVID flag.
====================
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When the bridge offloads a VLAN on a slave port, we also need to
program its dedicated CPU port as a member of the VLAN.

Drivers may handle the CPU port's membership as they want. For example,
Marvell as a special "Unmodified" mode to pass frames as is through
such ports.

Even though DSA expects the drivers to handle the CPU port membership,
it does not make sense to program user VLANs as PVID on the CPU port.
This patch clears this flag before programming the CPU port.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

DSA currently programs a VLAN on the CPU port implicitly after the
related notifier is received by a switch.

While we still need to do this transparent programmation of the DSA
links in the fabric, programming the CPU port this way may cause
problems in some corners such as the tag_8021q driver.

Because the dedicated CPU port is specific to a slave, make their
programmation explicit a few layers up, in the slave code.

Note that technically, DSA links have a dedicated CPU port as well,
but since they are only used as conduit between interconnected switches
of a fabric, programming them transparently this way is what we want.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The bridge VLANs are not offloaded by dsa_port_vlan_* if the port is
not bridged or if its bridge is not VLAN aware.

This is a good thing but other corners of DSA, such as the tag_8021q
driver, may need to program VLANs regardless the bridge state.

And also because bridge_dev is specific to user ports anyway, move
these checks were it belongs, one layer up in the slave code.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add dsa_slave_vlan_add and dsa_slave_vlan_del helpers to handle
SWITCHDEV_OBJ_ID_PORT_VLAN switchdev objects. Also copy the
switchdev_obj_port_vlan structure on add since we will modify it in
future patches.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently dsa_port_vid_add returns 0 if the switch returns -EOPNOTSUPP.

This function is used in the tag_8021q.c code to offload the PVID of
ports, which would simply not work if .port_vlan_add is not supported
by the underlying switch.

Do not skip -EOPNOTSUPP in dsa_port_vid_add but only when necessary,
that is to say in dsa_slave_vlan_rx_add_vid.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The bitmap operations were introduced to simplify the switch drivers
in the future, since most of them could implement the common VLAN and
MDB operations (add, del, dump) with simple functions taking all target
ports at once, and thus limiting the number of hardware accesses.

Programming an MDB or VLAN this way in a single operation would clearly
simplify the drivers a lot but would require a new get-set interface
in DSA. The usage of such bitmap from the stack also raised concerned
in the past, leading to the dynamic allocation of a new ds->_bitmap
member in the dsa_switch structure. So let's get rid of them for now.

This commit nicely wraps the ds->ops->port_{mdb,vlan}_{prepare,add}
switch operations into new dsa_switch_{mdb,vlan}_{prepare,add}
variants not using any bitmap argument anymore.

New dsa_switch_{mdb,vlan}_match helpers have been introduced to make
clear which local port of a switch must be programmed with the target
object. While the targeted user port is an obvious candidate, the
DSA links must also be programmed, as well as the CPU port for VLANs.

While at it, also remove local variables that are only used once.
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Minor conflict in r8169, bug fix had two versions in net
and net-next, take the net-next hunks.
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull NFS client bugfixes from Trond Myklebust:
 "Highlights include:

  Stable fixes:

   - Fix a page lock leak in nfs_pageio_resend()

   - Ensure O_DIRECT reports an error if the bytes read/written is 0

   - Don't handle errors if the bind/connect succeeded

   - Revert "NFSv4/flexfiles: Abort I/O early if the layout segment was
     invalidat ed"

  Bugfixes:

   - Don't refresh attributes with mounted-on-file information

   - Fix return values for nfs4_file_open() and nfs_finish_open()

   - Fix pnfs layoutstats reporting of I/O errors

   - Don't use soft RPC calls for pNFS/flexfiles I/O, and don't abort
     for soft I/O errors when the user specifies a hard mount.

   - Various fixes to the error handling in sunrpc

   - Don't report writepage()/writepages() errors twice"

* tag 'nfs-for-5.3-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  NFS: remove set but not used variable 'mapping'
  NFSv2: Fix write regression
  NFSv2: Fix eof handling
  NFS: Fix writepage(s) error handling to not report errors twice
  NFS: Fix spurious EIO read errors
  pNFS/flexfiles: Don't time out requests on hard mounts
  SUNRPC: Handle connection breakages correctly in call_status()
  Revert "NFSv4/flexfiles: Abort I/O early if the layout segment was invalidated"
  SUNRPC: Handle EADDRINUSE and ENOBUFS correctly
  pNFS/flexfiles: Turn off soft RPC calls
  SUNRPC: Don't handle errors if the bind/connect succeeded
  NFS: On fatal writeback errors, we need to call nfs_inode_remove_request()
  NFS: Fix initialisation of I/O result struct in nfs_pgio_rpcsetup
  NFS: Ensure O_DIRECT reports an error if the bytes read/written is 0
  NFSv4/pnfs: Fix a page lock leak in nfs_pageio_resend()
  NFSv4: Fix return value in nfs_finish_open()
  NFSv4: Fix return values for nfs4_file_open()
  NFS: Don't refresh attributes with mounted-on-file information

Pull ARC updates from Vineet Gupta:

 - support for Edge Triggered IRQs in ARC IDU intc

 - other fixes here and there

* tag 'arc-5.3-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc:
  arc: prefer __section from compiler_attributes.h
  dt-bindings: IDU-intc: Add support for edge-triggered interrupts
  dt-bindings: IDU-intc: Clean up documentation
  ARCv2: IDU-intc: Add support for edge-triggered interrupts
  ARC: unwind: Mark expected switch fall-throughs
  ARC: [plat-hsdk]: allow to switch between AXI DMAC port configurations
  ARC: fix typo in setup_dma_ops log message
  ARCv2: entry: early return from exception need not clear U & DE bits

Pull MFD fix from Lee Jones:
 "Identify potentially unused functions in rk808 driver when !PM"

* tag 'mfd-fixes-5.3' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd:
  mfd: rk808: Make PM function declaration static
  mfd: rk808: Mark pm functions __maybe_unused

Pull sound fixes from Takashi Iwai:
 "A collection of small fixes as usual:

   - More coverage of USB-audio descriptor sanity checks

   - A fix for mute LED regression on Conexant HD-audio codecs

   - A few device-specific fixes and quirks for USB-audio and HD-audio

   - A fix for (die-hard remaining) possible race in sequencer core

   - FireWire oxfw regression fix that was introduced in 5.3-rc1"

* tag 'sound-5.3-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: oxfw: fix to handle correct stream for PCM playback
  ALSA: seq: Fix potential concurrent access to the deleted pool
  ALSA: usb-audio: Check mixer unit bitmap yet more strictly
  ALSA: line6: Fix memory leak at line6_init_pcm() error path
  ALSA: usb-audio: Fix invalid NULL check in snd_emuusb_set_samplerate()
  ALSA: hda/ca0132 - Add new SBZ quirk
  ALSA: usb-audio: Add implicit fb quirk for Behringer UFX1604
  ALSA: hda - Fixes inverted Conexant GPIO mic mute led

Pull networking fixes from David Miller:

 1) Use 32-bit index for tails calls in s390 bpf JIT, from Ilya
    Leoshkevich.

 2) Fix missed EPOLLOUT events in TCP, from Eric Dumazet. Same fix for
    SMC from Jason Baron.

 3) ipv6_mc_may_pull() should return 0 for malformed packets, not
    -EINVAL. From Stefano Brivio.

 4) Don't forget to unpin umem xdp pages in error path of
    xdp_umem_reg(). From Ivan Khoronzhuk.

 5) Fix sta object leak in mac80211, from Johannes Berg.

 6) Fix regression by not configuring PHYLINK on CPU port of bcm_sf2
    switches. From Florian Fainelli.

 7) Revert DMA sync removal from r8169 which was causing regressions on
    some MIPS Loongson platforms. From Heiner Kallweit.

 8) Use after free in flow dissector, from Jakub Sitnicki.

 9) Fix NULL derefs of net devices during ICMP processing across
    collect_md tunnels, from Hangbin Liu.

10) proto_register() memory leaks, from Zhang Lin.

11) Set NLM_F_MULTI flag in multipart netlink messages consistently,
    from John Fastabend.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (66 commits)
  r8152: Set memory to all 0xFFs on failed reg reads
  openvswitch: Fix conntrack cache with timeout
  ipv4: mpls: fix mpls_xmit for iptunnel
  nexthop: Fix nexthop_num_path for blackhole nexthops
  net: rds: add service level support in rds-info
  net: route dump netlink NLM_F_MULTI flag missing
  s390/qeth: reject oversized SNMP requests
  sock: fix potential memory leak in proto_register()
  MAINTAINERS: Add phylink keyword to SFF/SFP/SFP+ MODULE SUPPORT
  xfrm/xfrm_policy: fix dst dev null pointer dereference in collect_md mode
  ipv4/icmp: fix rt dst dev null pointer dereference
  openvswitch: Fix log message in ovs conntrack
  bpf: allow narrow loads of some sk_reuseport_md fields with offset > 0
  bpf: fix use after free in prog symbol exposure
  bpf: fix precision tracking in presence of bpf2bpf calls
  flow_dissector: Fix potential use-after-free on BPF_PROG_DETACH
  Revert "r8169: remove not needed call to dma_sync_single_for_device"
  ipv6: propagate ipv6_add_dev's error returns out of ipv6_find_idev
  net/ncsi: Fix the payload copying for the request coming from Netlink
  qed: Add cleanup in qed_slowpath_start()
  ...

Fixes gcc '-Wunused-but-set-variable' warning:

fs/nfs/write.c: In function nfs_page_async_flush:
fs/nfs/write.c:609:24: warning: variable mapping set but not used [-Wunused-but-set-variable]

It is not use since commit aefb623c422e ("NFS: Fix
writepage(s) error handling to not report errors twice")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

Ensure we update the write result count on success, since the
RPC call itself does not do so.
Reported-by: Jan Stancek <jstancek@redhat.com>
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Tested-by: Jan Stancek <jstancek@redhat.com>

If we received a reply from the server with a zero length read and
no error, then that implies we are at eof.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

Avoids:
  ../drivers/mfd/rk808.c:771:1: warning: symbol 'rk8xx_pm_ops' \
    was not declared. Should it be static?

Fixes: 5752bc43 ("mfd: rk808: Mark pm functions __maybe_unused")
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Lee Jones <lee.jones@linaro.org>

The newly added suspend/resume functions are only used if CONFIG_PM
is enabled:

drivers/mfd/rk808.c:752:12: error: 'rk8xx_resume' defined but not used [-Werror=unused-function]
drivers/mfd/rk808.c:732:12: error: 'rk8xx_suspend' defined but not used [-Werror=unused-function]

Mark them as __maybe_unused so the compiler can silently drop them
when they are not needed.

Fixes: 586c1b41 ("mfd: rk808: Add RK817 and RK809 support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Lee Jones <lee.jones@linaro.org>

Add MODULE_FIRMWARE entries for AMDA0058 boards.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Move the call to dma_sync_single_for_cpu after calling napi_alloc_skb.
This avoids calling dma_sync_single_for_cpu w/o handing control back
to device if the memory allocation should fail.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Vlad Buslov says:

====================
Refactor cls hardware offload API to support rtnl-independent drivers

Currently, all cls API hardware offloads driver callbacks require caller
to hold rtnl lock when calling them. This patch set introduces new API
that allows drivers to register callbacks that are not dependent on rtnl
lock and unlocked classifiers to offload filters without obtaining rtnl
lock first, which is intended to allow offloading tc rules in parallel.

Recently, new rtnl registration flag RTNL_FLAG_DOIT_UNLOCKED was added.
TC rule update handlers (RTM_NEWTFILTER, RTM_DELTFILTER, etc.) are
already registered with this flag and only take rtnl lock when qdisc or
classifier requires it. Classifiers can indicate that their ops
callbacks don't require caller to hold rtnl lock by setting the
TCF_PROTO_OPS_DOIT_UNLOCKED flag. Unlocked implementation of flower
classifier is now upstreamed. However, this implementation still obtains
rtnl lock before calling hardware offloads API.

Implement following cls API changes:

- Introduce new "unlocked_driver_cb" flag to struct flow_block_offload
  to allow registering and unregistering block hardware offload
  callbacks that do not require caller to hold rtnl lock. Drivers that
  doesn't require users of its tc offload callbacks to hold rtnl lock
  sets the flag to true on block bind/unbind. Internally tcf_block is
  extended with additional lockeddevcnt counter that is used to count
  number of devices that require rtnl lock that block is bound to. When
  this counter is zero, tc_setup_cb_*() functions execute callbacks
  without obtaining rtnl lock.

- Extend cls API single hardware rule update tc_setup_cb_call() function
  with tc_setup_cb_add(), tc_setup_cb_replace(), tc_setup_cb_destroy()
  and tc_setup_cb_reoffload() functions. These new APIs are needed to
  move management of block offload counter, filter in hardware counter
  and flag from classifier implementations to cls API, which is now
  responsible for managing them in concurrency-safe manner. Access to
  cb_list from callback execution code is synchronized by obtaining new
  'cb_lock' rw_semaphore in read mode, which allows executing callbacks
  in parallel, but excludes any modifications of data from
  register/unregister code. tcf_block offloads counter type is changed
  to atomic integer to allow updating the counter concurrently.

- Extend classifier ops with new ops->hw_add() and ops->hw_del()
  callbacks which are used to notify unlocked classifiers when filter is
  successfully added or deleted to hardware without releasing cb_lock.
  This is necessary to update classifier state atomically with callback
  list traversal and updating of all relevant counters and allows
  unlocked classifiers to synchronize with concurrent reoffload without
  requiring any changes to driver callback API implementations.

New tc flow_action infrastructure is also modified to allow its user to
execute without rtnl lock protection. Function tc_setup_flow_action() is
modified to conditionally obtain rtnl lock before accessing action
state. Action data that is accessed by reference is either copied or
reference counted to prevent concurrent action overwrite from
deallocating it. New function tc_cleanup_flow_action() is introduced to
cleanup/release all such data obtained by tc_setup_flow_action().

Flower classifier (only unlocked classifier at the moment) is modified
to use new cls hardware offloads API and no longer obtains rtnl lock
before calling it.
====================
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Don't manually take rtnl lock in flower classifier before calling cls
hardware offloads API. Instead, pass rtnl lock status via 'rtnl_held'
parameter.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>