platform/kernel/linux-starfive.git
18 months agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf-next
David S. Miller [Wed, 18 Jan 2023 13:19:48 +0000 (13:19 +0000)]
Merge git://git./linux/kernel/git/netfilter/nf-next

Florian Westphal says:

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

following patch set includes netfilter updates for your *net-next* tree.

1. Replace pr_debug use with nf_log infra for debugging in sctp
   conntrack.
2. Remove pr_debug calls, they are either useless or we have better
   options in place.
3. Avoid repeated load of ct->status in some spots.
   Some bit-flags cannot change during the lifeetime of
   a connection, so no need to re-fetch those.
4. Avoid uneeded nesting of rcu_read_lock during tuple lookup.
5. Remove the CLUSTERIP target.  Marked as obsolete for years,
   and we still have WARN splats wrt. races of the out-of-band
   /proc interface installed by this target.
6. Add static key to nf_tables to avoid the retpoline mitigation
   if/else if cascade provided the cpu doesn't need the retpoline thunk.
7. add nf_tables objref calls to the retpoline mitigation workaround.
8. Split parts of nft_ct.c that do not need symbols exported by
   the conntrack modules and place them in nf_tables directly.
   This allows to avoid indirect call for 'ct status' checks.
9. Add 'destroy' commands to nf_tables.  They are identical
   to the existing 'delete' commands, but do not indicate
   an error if the referenced object (set, chain, rule...)
   did not exist, from Fernando.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'tsnep-xdp-support'
David S. Miller [Wed, 18 Jan 2023 13:17:06 +0000 (13:17 +0000)]
Merge branch 'tsnep-xdp-support'

Gerhard Engleder says:

====================
tsnep: XDP support

Implement XDP support for tsnep driver. I tried to follow existing
drivers like igb/igc as far as possible. Some prework was already done
in previous patch series, so in this series only actual XDP stuff is
included.

Thanks for the NetDev 0x14 slides "Add XDP support on a NIC driver".

Some commits contain changes not related to XDP but found during review
of XDP support patches.

v5:
- fix spelling of 'subtract' in commit message (Alexander Duyck)
- call txq_trans_cond_update() only if TX is complete (Alexander Duyck)
- remove const from static functions (Alexander Duyck)
- replace TX spin_lock with __netif_tx_lock (Alexander Duyck)
- use xdp_return_frame_rx_napi() instead of xdp_return_frame_bulk() (Alexander Duyck)
- eliminate __TSNEP_DOWN (Alexander Duyck)
- introduce single function for xdp_rxq and napi init (Alexander Duyck)
- use TX queue of pair instead of expensive processor id modulo for XDP_TX (Alexander Duyck)
- eliminate processor id modulo in tsnep_netdev_xdp_xmit (Alexander Duyck)
- use bitmap for TX type and add fragment type (Alexander Duyck)
- always use XDP_PACKET_HEADROOM and DMA_BIDIRECTIONAL

v4:
- remove process context from spin_lock_bh commit message (Alexander Lobakin)
- move tsnep_adapter::state to prevent 4 byte hole (Alexander Lobakin)
- braces for bitops in combination logical ops (Alexander Lobakin)
- make various pointers const (Alexander Lobakin)
- '!i' instead of 'i == 0' (Alexander Lobakin)
- removed redundant braces (Alexander Lobakin)
- squash variables into same line if same type (Alexander Lobakin)
- use fact that ::skb and ::xdpf use same slot for simplification (Alexander Lobakin)
- use u32 for smp_processor_id() (Alexander Lobakin)
- don't add $(tsnep-y) to $(tsnep-objs) (Alexander Lobakin)
- use rev xmas tree in tsnep_netdev_open() (Alexander Lobakin)
- do not move tsnep_queue::napi (Alexander Lobakin)
- call xdp_init_buff() only once (Alexander Lobakin)
- get nq and tx only once for XDP TX (Alexander Lobakin)
- move XDP BPF program setup to end of patch series (Alexander Lobakin)
- check for XDP state change and prevent redundant down-ups (Alexander Lobakin)
- access tsnep_adapter::xdp_prog only with READ_ONCE in RX path (Alexander Lobakin)
- forward NAPI budget to napi_consume_skb() (Alexander Lobakin)
- fix errno leftover in tsnep_xdp_xmit_back() (Dan Carpenter)
- eliminate tsnep_xdp_is_enabled() by setting RX offset during init

v3:
- use spin_lock_bh for TX (Paolo Abeni)
- add comment for XDP TX descriptor available check (Maciej Fijalkowski)
- return value bool for tsnep_xdp_xmit_frame_ring() (Saeed Mahameed)
- do not print DMA mapping error (Saeed Mahameed)
- use reverse xmas tree variable declaration (Saeed Mahameed)
- move struct xdp_rxq_info to end of struct tsnep_rx (Maciej Fijalkowski)
- check __TSNEP_DOWN flag on close to prevent double free (Saeed Mahameed)
- describe TSNEP_RX_INLINE_METADATA_SIZE in comment (Maciej Fijalkowski)
- substract TSNEP_RX_INLINE_METADATA_SIZE after DMA sync (Maciej Fijalkowski)
- use enum tsnep_tx_type for tsnep_xdp_tx_map (Saeed Mahameed)
- use nxmit as loop iterator in tsnep_netdev_xdp_xmit (Saeed Mahameed)
- stop netdev in tsnep_netdev_close() which is called during BPF prog setup

v2:
- move tsnep_xdp_xmit_back() to commit where it is used (Paolo Abeni)
- remove inline from tsnep_rx_offset() (Paolo Abeni)
- remove inline from tsnep_rx_offset_xdp() (Paolo Abeni)
- simplify tsnep_xdp_run_prog() call by moving xdp_status update to it (Paolo Abeni)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Support XDP BPF program setup
Gerhard Engleder [Mon, 16 Jan 2023 20:24:58 +0000 (21:24 +0100)]
tsnep: Support XDP BPF program setup

Implement setup of BPF programs for XDP RX path with command
XDP_SETUP_PROG of ndo_bpf(). This is the final step for XDP RX path
support.

There is no need to reinit the RX queues as they are always prepared for
XDP.

Additionally remove $(tsnep-y) from $(tsnep-objs) because it is added
automatically.

Test results with A53 1.2GHz:

XDP_DROP (samples/bpf/xdp1)
proto 17:     883878 pkt/s

XDP_TX (samples/bpf/xdp2)
proto 17:     255693 pkt/s

XDP_REDIRECT (samples/bpf/xdpsock)
 sock0@eth2:0 rxdrop xdp-drv
                   pps            pkts           1.00
rx                 855,582        5,404,523
tx                 0              0

XDP_REDIRECT (samples/bpf/xdp_redirect)
eth2->eth1         613,267 rx/s   0 err,drop/s   613,272 xmit/s

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Add XDP RX support
Gerhard Engleder [Mon, 16 Jan 2023 20:24:57 +0000 (21:24 +0100)]
tsnep: Add XDP RX support

If BPF program is set up, then run BPF program for every received frame
and execute the selected action.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Add RX queue info for XDP support
Gerhard Engleder [Mon, 16 Jan 2023 20:24:56 +0000 (21:24 +0100)]
tsnep: Add RX queue info for XDP support

Register xdp_rxq_info with page_pool memory model. This is needed for
XDP buffer handling.

Additionally fix error path by removing call of tsnep_phy_close() after
failed tsnep_phy_open().

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Prepare RX buffer for XDP support
Gerhard Engleder [Mon, 16 Jan 2023 20:24:55 +0000 (21:24 +0100)]
tsnep: Prepare RX buffer for XDP support

Always reserve XDP_PACKET_HEADROOM in front of RX buffer. Similar DMA
direction is always set to DMA_BIDIRECTIONAL. This eliminates the need
for RX queue reconfiguration during BPF program setup. The RX queue is
always prepared for XDP.

No negative impact of DMA_BIDIRECTIONAL was measured.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Subtract TSNEP_RX_INLINE_METADATA_SIZE once
Gerhard Engleder [Mon, 16 Jan 2023 20:24:54 +0000 (21:24 +0100)]
tsnep: Subtract TSNEP_RX_INLINE_METADATA_SIZE once

Subtract size of metadata in front of received data only once. This
simplifies the RX code.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Add XDP TX support
Gerhard Engleder [Mon, 16 Jan 2023 20:24:53 +0000 (21:24 +0100)]
tsnep: Add XDP TX support

Implement ndo_xdp_xmit() for XDP TX support. Support for fragmented XDP
frames is included.

Also some braces and logic cleanups are done in normal TX path to keep
both TX paths in sync.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Do not print DMA mapping error
Gerhard Engleder [Mon, 16 Jan 2023 20:24:52 +0000 (21:24 +0100)]
tsnep: Do not print DMA mapping error

Printing in data path shall be avoided. DMA mapping error is already
counted in stats so printing is not necessary.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Forward NAPI budget to napi_consume_skb()
Gerhard Engleder [Mon, 16 Jan 2023 20:24:51 +0000 (21:24 +0100)]
tsnep: Forward NAPI budget to napi_consume_skb()

NAPI budget must be forwarded to napi_consume_skb(). It is used to
detect non-NAPI context.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agotsnep: Replace TX spin_lock with __netif_tx_lock
Gerhard Engleder [Mon, 16 Jan 2023 20:24:50 +0000 (21:24 +0100)]
tsnep: Replace TX spin_lock with __netif_tx_lock

TX spin_lock can be eliminated, because the normal TX path is already
protected with __netif_tx_lock and this lock can be used for access to
queue outside of normal TX path too.

Signed-off-by: Gerhard Engleder <gerhard@engleder-embedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: ftmac100: handle netdev flags IFF_PROMISC and IFF_ALLMULTI
Sergei Antonov [Mon, 16 Jan 2023 18:27:16 +0000 (21:27 +0300)]
net: ftmac100: handle netdev flags IFF_PROMISC and IFF_ALLMULTI

When netdev->flags has IFF_PROMISC or IFF_ALLMULTI, set the
corresponding bits in the MAC Control Register (MACCR).

This change is based on code from the ftgmac100 driver, see
ftgmac100_start_hw() in ftgmac100.c

Signed-off-by: Sergei Antonov <saproj@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoipv6: Remove extra counter pull before gc
Tanmay Bhushan [Mon, 16 Jan 2023 14:55:00 +0000 (15:55 +0100)]
ipv6: Remove extra counter pull before gc

Per cpu entries are no longer used in consideration
for doing gc or not. Remove the extra per cpu entries
pull to directly check for time and perform gc.

Signed-off-by: Tanmay Bhushan <007047221b@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'am65-cpts-PPS'
David S. Miller [Wed, 18 Jan 2023 12:56:38 +0000 (12:56 +0000)]
Merge branch 'am65-cpts-PPS'

Siddharth Vadapalli says:

====================
Add PPS support to am65-cpts driver

The CPTS hardware doesn't support PPS signal generation. Using the GenFx
(periodic signal generator) function, it is possible to model a PPS signal
followed by routing it via the time sync router to the CPTS_HWy_TS_PUSH
(hardware time stamp) input, in order to generate timestamps at 1 second
intervals.

This series adds driver support for enabling PPS signal generation.
Additionally, the documentation for the am65-cpts driver is updated with
the bindings for the "ti,pps" property, which is used to inform the
pair [CPTS_HWy_TS_PUSH, GenFx] to the cpts driver.

Changes from v1:
1. Drop device-tree patches.
2. Address Roger's comments on the:
   "net: ethernet: ti: am65-cpts: add pps support" patch.
3. Collect Reviewed-by tag from Rob Herring.

v1:
https://lore.kernel.org/r/20230111114429.1297557-1-s-vadapalli@ti.com/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: ethernet: ti: am65-cpts: adjust pps following ptp changes
Grygorii Strashko [Mon, 16 Jan 2023 08:55:34 +0000 (14:25 +0530)]
net: ethernet: ti: am65-cpts: adjust pps following ptp changes

When CPTS clock is sync/adjusted by running linuxptp (ptp4l) it will cause
PPS jitter as Genf running PPS is not adjusted.

The same PPM adjustment has to be applied to GenF as to PHC clock to
correct PPS length and keep them in sync.

Testing:
 Master:
  ptp4l -P -2 -H -i eth0 -l 6 -m -q -p /dev/ptp1 -f ptp.cfg &
  testptp -d /dev/ptp1 -P 1
  ppstest /dev/pps0

 Slave:
  linuxptp/ptp4l -P -2 -H -i eth0 -l 6 -m -q -p /dev/ptp1 -f ptp1.cfg -s &
    <port 1: UNCALIBRATED to SLAVE on MASTER_CLOCK_SELECTED;>
  testptp -d /dev/ptp1 -P 1
  ppstest /dev/pps0

Master log:
source 0 - assert 620.000000689, sequence: 530
source 0 - assert 621.000000689, sequence: 531
source 0 - assert 622.000000689, sequence: 532
source 0 - assert 623.000000689, sequence: 533
source 0 - assert 624.000000689, sequence: 534
source 0 - assert 625.000000689, sequence: 535
source 0 - assert 626.000000689, sequence: 536
source 0 - assert 627.000000689, sequence: 537
source 0 - assert 628.000000689, sequence: 538
source 0 - assert 629.000000689, sequence: 539
source 0 - assert 630.000000689, sequence: 540
source 0 - assert 631.000000689, sequence: 541
source 0 - assert 632.000000689, sequence: 542
source 0 - assert 633.000000689, sequence: 543
source 0 - assert 634.000000689, sequence: 544
source 0 - assert 635.000000689, sequence: 545

Slave log:
source 0 - assert 620.000000706, sequence: 252
source 0 - assert 621.000000709, sequence: 253
source 0 - assert 622.000000707, sequence: 254
source 0 - assert 623.000000707, sequence: 255
source 0 - assert 624.000000706, sequence: 256
source 0 - assert 625.000000705, sequence: 257
source 0 - assert 626.000000709, sequence: 258
source 0 - assert 627.000000709, sequence: 259
source 0 - assert 628.000000707, sequence: 260
source 0 - assert 629.000000706, sequence: 261
source 0 - assert 630.000000710, sequence: 262
source 0 - assert 631.000000708, sequence: 263
source 0 - assert 632.000000705, sequence: 264
source 0 - assert 633.000000710, sequence: 265
source 0 - assert 634.000000708, sequence: 266
source 0 - assert 635.000000707, sequence: 267

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Reviewed-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: ethernet: ti: am65-cpts: add pps support
Grygorii Strashko [Mon, 16 Jan 2023 08:55:33 +0000 (14:25 +0530)]
net: ethernet: ti: am65-cpts: add pps support

CPTS doesn't have HW support for PPS ("pulse per second”) signal
generation, but it can be modeled by using Time Sync Router and routing
GenFx (periodic signal generator) output to CPTS_HWy_TS_PUSH (hardware time
stamp) input, and configuring GenFx to generate 1sec pulses.

     +------------------------+
     |          CPTS          |
     |                        |
 +--->CPTS_HW4_PUSH      GENFx+---+
 |   |                        |   |
 |   +------------------------+   |
 |                                |
 +--------------------------------+

Add corresponding support to am65-cpts driver. The DT property "ti,pps"
has to be used to enable PPS support and configure pair
[CPTS_HWy_TS_PUSH, GenFx].

Once enabled, PPS can be tested using ppstest tool:
 # ./ppstest /dev/pps0

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Reviewed-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-binding: net: ti: am65x-cpts: add 'ti,pps' property
Grygorii Strashko [Mon, 16 Jan 2023 08:55:32 +0000 (14:25 +0530)]
dt-binding: net: ti: am65x-cpts: add 'ti,pps' property

Add the ti,pps property used to indicate the pair of HWx_TS_PUSH input and
the TS_GENFy output.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Siddharth Vadapalli <s-vadapalli@ti.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'stmmac-imx93'
David S. Miller [Wed, 18 Jan 2023 12:47:40 +0000 (12:47 +0000)]
Merge branch 'stmmac-imx93'

Clark Wang says:

====================
stmmac: Add eqos and fec support for imx93

This patchset add imx93 support for dwmac-imx glue driver.
There are some changes of GPR implement.
And add fec and eqos nodes for imx93 dts.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoarm64: dts: imx93-11x11-evk: enable fec function
Clark Wang [Fri, 13 Jan 2023 03:33:47 +0000 (11:33 +0800)]
arm64: dts: imx93-11x11-evk: enable fec function

Enable FEC function for imx93-11x11-evk board.

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoarm64: dts: imx93-11x11-evk: enable eqos
Clark Wang [Fri, 13 Jan 2023 03:33:46 +0000 (11:33 +0800)]
arm64: dts: imx93-11x11-evk: enable eqos

Enable EQoS function for imx93-11x11-evk board.

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoarm64: dts: imx93: add FEC support
Clark Wang [Fri, 13 Jan 2023 03:33:45 +0000 (11:33 +0800)]
arm64: dts: imx93: add FEC support

Add FEC node for imx93 platform.

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoarm64: dts: imx93: add eqos support
Clark Wang [Fri, 13 Jan 2023 03:33:44 +0000 (11:33 +0800)]
arm64: dts: imx93: add eqos support

Add EQoS node for imx93 platform.

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: fec: add mx93 description
Clark Wang [Fri, 13 Jan 2023 03:33:43 +0000 (11:33 +0800)]
dt-bindings: net: fec: add mx93 description

Add mx93 compatible string for fec driver.

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: add mx93 description
Clark Wang [Fri, 13 Jan 2023 03:33:42 +0000 (11:33 +0800)]
dt-bindings: add mx93 description

Add mx93 compatible string for eqos driver.

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: stmmac: add imx93 platform support
Clark Wang [Fri, 13 Jan 2023 03:33:41 +0000 (11:33 +0800)]
net: stmmac: add imx93 platform support

Add imx93 platform support for dwmac-imx driver.

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonetfilter: nf_tables: add support to destroy operation
Fernando Fernandez Mancera [Mon, 2 Jan 2023 14:42:34 +0000 (15:42 +0100)]
netfilter: nf_tables: add support to destroy operation

Introduce NFT_MSG_DESTROY* message type. The destroy operation performs a
delete operation but ignoring the ENOENT errors.

This is useful for the transaction semantics, where failing to delete an
object which does not exist results in aborting the transaction.

This new command allows the transaction to proceed in case the object
does not exist.

Signed-off-by: Fernando Fernandez Mancera <ffmancera@riseup.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agonetfilter: nf_tables: avoid retpoline overhead for some ct expression calls
Florian Westphal [Tue, 3 Jan 2023 12:47:17 +0000 (13:47 +0100)]
netfilter: nf_tables: avoid retpoline overhead for some ct expression calls

nft_ct expression cannot be made builtin to nf_tables without also
forcing the conntrack itself to be builtin.

However, this can be avoided by splitting retrieval of a few
selector keys that only need to access the nf_conn structure,
i.e. no function calls to nf_conntrack code.

Many rulesets start with something like
"ct status established,related accept"

With this change, this no longer requires an indirect call, which
gives about 1.8% more throughput with a simple conntrack-enabled
forwarding test (retpoline thunk used).

Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agonetfilter: nf_tables: avoid retpoline overhead for objref calls
Florian Westphal [Tue, 3 Jan 2023 12:47:16 +0000 (13:47 +0100)]
netfilter: nf_tables: avoid retpoline overhead for objref calls

objref expression is builtin, so avoid calls to it for
RETOLINE=y builds.

Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agonetfilter: nf_tables: add static key to skip retpoline workarounds
Florian Westphal [Tue, 3 Jan 2023 12:47:15 +0000 (13:47 +0100)]
netfilter: nf_tables: add static key to skip retpoline workarounds

If CONFIG_RETPOLINE is enabled nf_tables avoids indirect calls for
builtin expressions.

On newer cpus indirect calls do not go through the retpoline thunk
anymore, even for RETPOLINE=y builds.

Just like with the new tc retpoline wrappers:
Add a static key to skip the if / else if cascade if the cpu
does not require retpolines.

Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agonetfilter: ip_tables: remove clusterip target
Florian Westphal [Thu, 5 Jan 2023 19:22:02 +0000 (20:22 +0100)]
netfilter: ip_tables: remove clusterip target

Marked as 'to be removed soon' since kernel 4.1 (2015).
Functionality was superseded by the 'cluster' match, added in kernel
2.6.30 (2009).

clusterip_tg_check still has races that can give

 proc_dir_entry 'ipt_CLUSTERIP/10.1.1.2' already registered

followed by a WARN splat.

Remove it instead of trying to fix this up again.
clusterip uapi header is left as-is for now.

Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agonetfilter: conntrack: move rcu read lock to nf_conntrack_find_get
Florian Westphal [Fri, 16 Dec 2022 01:46:28 +0000 (02:46 +0100)]
netfilter: conntrack: move rcu read lock to nf_conntrack_find_get

Move rcu_read_lock/unlock to nf_conntrack_find_get(), this avoids
nested rcu_read_lock call from resolve_normal_ct().

Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agonetfilter: conntrack: avoid reload of ct->status
Florian Westphal [Mon, 2 Jan 2023 11:46:12 +0000 (12:46 +0100)]
netfilter: conntrack: avoid reload of ct->status

Compiler can't merge the two test_bit() calls, so load ct->status
once and use non-atomic accesses.

This is fine because IPS_EXPECTED or NAT_CLASH are either set at ct
creation time or not at all, but compiler can't know that.

Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agonetfilter: conntrack: remove pr_debug calls
Florian Westphal [Mon, 2 Jan 2023 11:46:11 +0000 (12:46 +0100)]
netfilter: conntrack: remove pr_debug calls

Those are all useless or dubious.
getorigdst() is called via setsockopt, so return value/errno will
already indicate an appropriate error.

For other pr_debug calls there are better replacements, such as
slab/slub debugging or 'conntrack -E' (ctnetlink events).

Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agonetfilter: conntrack: sctp: use nf log infrastructure for invalid packets
Florian Westphal [Mon, 2 Jan 2023 11:46:10 +0000 (12:46 +0100)]
netfilter: conntrack: sctp: use nf log infrastructure for invalid packets

The conntrack logging facilities include useful info such as in/out
interface names and packet headers.

Use those in more places instead of pr_debug calls.
Furthermore, several pr_debug calls can be removed, they are useless
on production machines due to the sheer volume of log messages.

Signed-off-by: Florian Westphal <fw@strlen.de>
18 months agoMerge branch 'net-mdio-continue-separating-c22-and-c45'
Jakub Kicinski [Wed, 18 Jan 2023 03:34:10 +0000 (19:34 -0800)]
Merge branch 'net-mdio-continue-separating-c22-and-c45'

Michael Walle says:

====================
net: mdio: Continue separating C22 and C45

I've picked this older series from Andrew up and rebased it onto
the latest net-next.

This is the third (and hopefully last) patch set in the series which
separates the C22 and C45 MDIO bus transactions at the API level to the
MDIO bus drivers.

The first patch is a newer version of the former "net: dsa: Separate C22
and C45 MDIO bus transaction methods", which only contains the mt7530
changes. Although posted as v1, because this is a new series, there
is a changelog included in the patch comment section.

The last patch is a new one, which isn't from Andrew's tree.
====================

Link: https://lore.kernel.org/r/20230116-net-next-c45-seperation-part-3-v1-0-0c53afa56aad@walle.cc
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: ethernet: renesas: rswitch: C45 only transactions
Michael Walle [Mon, 16 Jan 2023 23:52:27 +0000 (00:52 +0100)]
net: ethernet: renesas: rswitch: C45 only transactions

The rswitch MDIO bus driver only supports C45 transfers. Update the
function names to make this clear, pass the mmd as a parameter, and
register the accessors to the _c45 ops of the bus driver structure.

Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: dsa: sja1105: Separate C22 and C45 transactions for T1 MDIO bus
Andrew Lunn [Mon, 16 Jan 2023 23:52:26 +0000 (00:52 +0100)]
net: dsa: sja1105: Separate C22 and C45 transactions for T1 MDIO bus

The T1 MDIO bus driver can perform both C22 and C45 transfers. Create
separate functions for each and register the C45 versions using the
new API calls where appropriate.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: dsa: sja1105: C45 only transactions for PCS
Andrew Lunn [Mon, 16 Jan 2023 23:52:25 +0000 (00:52 +0100)]
net: dsa: sja1105: C45 only transactions for PCS

The sja1105 MDIO bus driver only supports C45 transfers. Update the
function names to make this clear, pass the mmd as a parameter, and
register the accessors to the _c45 ops of the bus driver structure.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoamd-xgbe: Replace MII_ADDR_C45 with XGBE_ADDR_C45
Andrew Lunn [Mon, 16 Jan 2023 23:52:24 +0000 (00:52 +0100)]
amd-xgbe: Replace MII_ADDR_C45 with XGBE_ADDR_C45

The xgbe driver reuses MII_ADDR_C45 for its own purpose. The values
derived with it are never passed to phylib or a linux MDIO bus driver.
In order that MII_ADDR_C45 can be removed, add an XGBE specific

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoamd-xgbe: Separate C22 and C45 transactions
Andrew Lunn [Mon, 16 Jan 2023 23:52:23 +0000 (00:52 +0100)]
amd-xgbe: Separate C22 and C45 transactions

The xgbe MDIO bus driver can perform both C22 and C45 transfers, when
using its MDIO bus hardware. The SFP I2C mdio bus driver only supports
C22. Create separate functions for each and register the C45 versions
using the new API calls where appropriate.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: hns: Separate C22 and C45 transactions
Andrew Lunn [Mon, 16 Jan 2023 23:52:22 +0000 (00:52 +0100)]
net: hns: Separate C22 and C45 transactions

The hns MDIO bus driver can perform both C22 and C45 transfers.
Create separate functions for each and register the C45 versions using
the new API calls where appropriate.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoixgbe: Use C45 mdiobus accessors
Andrew Lunn [Mon, 16 Jan 2023 23:52:21 +0000 (00:52 +0100)]
ixgbe: Use C45 mdiobus accessors

When performing a C45 bus transaction, make use of the c45 variants of
the bus read/write helpers. The ability to pass a special register
value is being removed to clean up the mdio bus driver API.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoixgbe: Separate C22 and C45 transactions
Andrew Lunn [Mon, 16 Jan 2023 23:52:20 +0000 (00:52 +0100)]
ixgbe: Separate C22 and C45 transactions

The ixgbe MDIO bus driver can perform both C22 and C45 transfers.
Create separate functions for each and register the C45 versions using
the new API calls where appropriate.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: macb: Separate C22 and C45 transactions
Andrew Lunn [Mon, 16 Jan 2023 23:52:19 +0000 (00:52 +0100)]
net: macb: Separate C22 and C45 transactions

The macb MDIO bus driver can perform both C22 and C45 transfers.
Create separate functions for each and register the C45 versions using
the new API calls where appropriate.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: nixge: Separate C22 and C45 transactions
Andrew Lunn [Mon, 16 Jan 2023 23:52:18 +0000 (00:52 +0100)]
net: nixge: Separate C22 and C45 transactions

The nixge MDIO bus driver can perform both C22 and C45 transfers.
Create separate functions for each and register the C45 versions using
the new API calls where appropriate.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: sxgbe: Separate C22 and C45 transactions
Andrew Lunn [Mon, 16 Jan 2023 23:52:17 +0000 (00:52 +0100)]
net: sxgbe: Separate C22 and C45 transactions

The sxgdb MDIO bus driver can perform both C22 and C45 transfers.
Create separate functions for each and register the C45 versions using
the new API calls where appropriate.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: dsa: mt7530: Separate C22 and C45 MDIO bus transactions
Andrew Lunn [Mon, 16 Jan 2023 23:52:16 +0000 (00:52 +0100)]
net: dsa: mt7530: Separate C22 and C45 MDIO bus transactions

mt7530 does support C45, but its uses a mix of registering its MDIO
bus and providing its private MDIO bus to the DSA core, too. This makes
the change a bit more complex.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: wangxun: clean up the code
Mengyuan Lou [Mon, 16 Jan 2023 10:38:39 +0000 (18:38 +0800)]
net: wangxun: clean up the code

Convert various mult-bit fields to be defined using GENMASK/FIELD_PREP.
Simplify the code with the ternary operator.

Signed-off-by: Mengyuan Lou <mengyuanlou@net-swift.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230116103839.84087-1-mengyuanlou@net-swift.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoMerge tag 'i2c-fwnode-api-2023017' of https://git.kernel.org/pub/scm/linux/kernel...
Jakub Kicinski [Wed, 18 Jan 2023 03:16:28 +0000 (19:16 -0800)]
Merge tag 'i2c-fwnode-api-2023017' of https://git./linux/kernel/git/wsa/linux

Wolfram Sang says:

====================
Immutable branch adding fwnode API to the I2C core

I2C changes requested by Russell King.
This allows him to rework SFP code further.

* tag 'i2c-fwnode-api-2023017' of https://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: add fwnode APIs
====================

Link: https://lore.kernel.org/r/Y8ZhI4g0wsvpjokd@ninjato/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoMerge tag 'regmap-mdio-c45-rework' of https://git.kernel.org/pub/scm/linux/kernel...
Jakub Kicinski [Wed, 18 Jan 2023 03:09:42 +0000 (19:09 -0800)]
Merge tag 'regmap-mdio-c45-rework' of https://git./linux/kernel/git/broonie/regmap

Mark Brown says:

====================
regmap: Rework regmap_mdio_c45_{read|write} for new C45 API.

This reworks the regmap MDIO handling of C45 addresses in
preparation for some forthcoming updates to the networking code.

* tag 'regmap-mdio-c45-rework' of https://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: Rework regmap_mdio_c45_{read|write} for new C45 API.
====================

Link: https://lore.kernel.org/r/Y8VjkgcWHjR9TzNw@sirena.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: mdio: cavium: Remove unneeded simicolons
Andrew Lunn [Sun, 15 Jan 2023 16:42:03 +0000 (17:42 +0100)]
net: mdio: cavium: Remove unneeded simicolons

The recent refactoring to split C22 and C45 introduced two unneeded
semiconons which the kernel test bot reported. Remove them.

Reported-by: kernel test robot <lkp@intel.com>
Fixes: 93641ecbaa1f ("net: mdio: cavium: Separate C22 and C45 transactions")
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Michael Walle <michael@walle.cc>
Link: https://lore.kernel.org/r/20230115164203.510615-1-andrew@lunn.ch
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agoinet: fix fast path in __inet_hash_connect()
Pietro Borrello [Sat, 14 Jan 2023 13:11:41 +0000 (13:11 +0000)]
inet: fix fast path in __inet_hash_connect()

__inet_hash_connect() has a fast path taken if sk_head(&tb->owners) is
equal to the sk parameter.
sk_head() returns the hlist_entry() with respect to the sk_node field.
However entries in the tb->owners list are inserted with respect to the
sk_bind_node field with sk_add_bind_node().
Thus the check would never pass and the fast path never execute.

This fast path has never been executed or tested as this bug seems
to be present since commit 1da177e4c3f4 ("Linux-2.6.12-rc2"), thus
remove it to reduce code complexity.

Signed-off-by: Pietro Borrello <borrello@diag.uniroma1.it>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230112-inet_hash_connect_bind_head-v3-1-b591fd212b93@diag.uniroma1.it
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agoMerge branch 'net-ethernet-mtk_wed-introduce-reset-support'
Paolo Abeni [Tue, 17 Jan 2023 10:36:47 +0000 (11:36 +0100)]
Merge branch 'net-ethernet-mtk_wed-introduce-reset-support'

Lorenzo Bianconi says:

====================
net: ethernet: mtk_wed: introduce reset support

Introduce proper reset integration between ethernet and wlan drivers in order
to schedule wlan driver reset when ethernet/wed driver is resetting.
Introduce mtk_hw_reset_monitor work in order to detect possible DMA hangs.
====================

Link: https://lore.kernel.org/r/cover.1673715298.git.lorenzo@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agonet: ethernet: mtk_wed: add reset/reset_complete callbacks
Lorenzo Bianconi [Sat, 14 Jan 2023 17:01:32 +0000 (18:01 +0100)]
net: ethernet: mtk_wed: add reset/reset_complete callbacks

Introduce reset and reset_complete wlan callback to schedule WLAN driver
reset when ethernet/wed driver is resetting.

Tested-by: Daniel Golle <daniel@makrotopia.org>
Co-developed-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agonet: ethernet: mtk_eth_soc: add dma checks to mtk_hw_reset_check
Lorenzo Bianconi [Sat, 14 Jan 2023 17:01:31 +0000 (18:01 +0100)]
net: ethernet: mtk_eth_soc: add dma checks to mtk_hw_reset_check

Introduce mtk_hw_check_dma_hang routine to monitor possible dma hangs.

Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Tested-by: Daniel Golle <daniel@makrotopia.org>
Co-developed-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agonet: ethernet: mtk_eth_soc: align reset procedure to vendor sdk
Lorenzo Bianconi [Sat, 14 Jan 2023 17:01:30 +0000 (18:01 +0100)]
net: ethernet: mtk_eth_soc: align reset procedure to vendor sdk

Avoid to power-down the ethernet chip during hw reset and align reset
procedure to vendor sdk.

Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Tested-by: Daniel Golle <daniel@makrotopia.org>
Co-developed-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agonet: ethernet: mtk_eth_soc: introduce mtk_hw_warm_reset support
Lorenzo Bianconi [Sat, 14 Jan 2023 17:01:29 +0000 (18:01 +0100)]
net: ethernet: mtk_eth_soc: introduce mtk_hw_warm_reset support

Introduce mtk_hw_warm_reset utility routine. This is a preliminary patch
to align reset procedure to vendor sdk and avoid to power down the chip
during hw reset.

Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Tested-by: Daniel Golle <daniel@makrotopia.org>
Co-developed-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agonet: ethernet: mtk_eth_soc: introduce mtk_hw_reset utility routine
Lorenzo Bianconi [Sat, 14 Jan 2023 17:01:28 +0000 (18:01 +0100)]
net: ethernet: mtk_eth_soc: introduce mtk_hw_reset utility routine

This is a preliminary patch to add Wireless Ethernet Dispatcher reset
support.

Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Tested-by: Daniel Golle <daniel@makrotopia.org>
Co-developed-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Sujuan Chen <sujuan.chen@mediatek.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agoMerge branch 'net-use-kmem_cache_free_bulk-in-kfree_skb_list'
Paolo Abeni [Tue, 17 Jan 2023 09:27:31 +0000 (10:27 +0100)]
Merge branch 'net-use-kmem_cache_free_bulk-in-kfree_skb_list'

Jesper Dangaard Brouer says:

====================
net: use kmem_cache_free_bulk in kfree_skb_list

The kfree_skb_list function walks SKB (via skb->next) and frees them
individually to the SLUB/SLAB allocator (kmem_cache). It is more
efficient to bulk free them via the kmem_cache_free_bulk API.

Netstack NAPI fastpath already uses kmem_cache bulk alloc and free
APIs for SKBs.

The kfree_skb_list call got an interesting optimization in commit
520ac30f4551 ("net_sched: drop packets after root qdisc lock is
released") that can create a list of SKBs "to_free" e.g. when qdisc
enqueue fails or deliberately chooses to drop . It isn't a normal data
fastpath, but the situation will likely occur when system/qdisc are
under heavy workloads, thus it makes sense to use a faster API for
freeing the SKBs.

E.g. the (often distro default) qdisc fq_codel will drop batches of
packets from fattest elephant flow, default capped at 64 packets (but
adjustable via tc argument drop_batch).

Performance measurements done in [1]:
 [1] https://github.com/xdp-project/xdp-project/blob/master/areas/mem/kfree_skb_list01.org
====================

Link: https://lore.kernel.org/r/167361788585.531803.686364041841425360.stgit@firesoul
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agonet: kfree_skb_list use kmem_cache_free_bulk
Jesper Dangaard Brouer [Fri, 13 Jan 2023 13:52:04 +0000 (14:52 +0100)]
net: kfree_skb_list use kmem_cache_free_bulk

The kfree_skb_list function walks SKB (via skb->next) and frees them
individually to the SLUB/SLAB allocator (kmem_cache). It is more
efficient to bulk free them via the kmem_cache_free_bulk API.

This patches create a stack local array with SKBs to bulk free while
walking the list. Bulk array size is limited to 16 SKBs to trade off
stack usage and efficiency. The SLUB kmem_cache "skbuff_head_cache"
uses objsize 256 bytes usually in an order-1 page 8192 bytes that is
32 objects per slab (can vary on archs and due to SLUB sharing). Thus,
for SLUB the optimal bulk free case is 32 objects belonging to same
slab, but runtime this isn't likely to occur.

The expected gain from using kmem_cache bulk alloc and free API
have been assessed via a microbencmark kernel module[1].

The module 'slab_bulk_test01' results at bulk 16 element:
 kmem-in-loop Per elem: 109 cycles(tsc) 30.532 ns (step:16)
 kmem-bulk    Per elem: 64 cycles(tsc) 17.905 ns (step:16)

More detailed description of benchmarks avail in [2].

[1] https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm
[2] https://github.com/xdp-project/xdp-project/blob/master/areas/mem/kfree_skb_list01.org

V2: rename function to kfree_skb_add_bulk.

Reviewed-by: Saeed Mahameed <saeed@kernel.org>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agonet: fix call location in kfree_skb_list_reason
Jesper Dangaard Brouer [Fri, 13 Jan 2023 13:51:59 +0000 (14:51 +0100)]
net: fix call location in kfree_skb_list_reason

The SKB drop reason uses __builtin_return_address(0) to give the call
"location" to trace_kfree_skb() tracepoint skb:kfree_skb.

To keep this stable for compilers kfree_skb_reason() is annotated with
__fix_address (noinline __noclone) as fixed in commit c205cc7534a9
("net: skb: prevent the split of kfree_skb_reason() by gcc").

The function kfree_skb_list_reason() invoke kfree_skb_reason(), which
cause the __builtin_return_address(0) "location" to report the
unexpected address of kfree_skb_list_reason.

Example output from 'perf script':
 kpktgend_0  1337 [000]    81.002597: skb:kfree_skb: skbaddr=0xffff888144824700 protocol=2048 location=kfree_skb_list_reason+0x1e reason: QDISC_DROP

Patch creates an __always_inline __kfree_skb_reason() helper call that
is called from both kfree_skb_list() and kfree_skb_list_reason().
Suggestions for solutions that shares code better are welcome.

As preparation for next patch move __kfree_skb() invocation out of
this helper function.

Reviewed-by: Saeed Mahameed <saeed@kernel.org>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agodevlink: remove some unnecessary code
Dan Carpenter [Fri, 13 Jan 2023 07:35:43 +0000 (10:35 +0300)]
devlink: remove some unnecessary code

This code checks if (attrs[DEVLINK_ATTR_TRAP_POLICER_ID]) twice.  Once
at the start of the function and then a couple lines later.  Delete the
second check since that one must be true.

Because the second condition is always true, it means the:

policer_item = group_item->policer_item;

assignment is immediately over-written.  Delete that as well.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://lore.kernel.org/r/Y8EJz8oxpMhfiPUb@kili
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agosched: add new attr TCA_EXT_WARN_MSG to report tc extact message
Hangbin Liu [Fri, 13 Jan 2023 03:43:53 +0000 (11:43 +0800)]
sched: add new attr TCA_EXT_WARN_MSG to report tc extact message

We will report extack message if there is an error via netlink_ack(). But
if the rule is not to be exclusively executed by the hardware, extack is not
passed along and offloading failures don't get logged.

In commit 81c7288b170a ("sched: cls: enable verbose logging") Marcelo
made cls could log verbose info for offloading failures, which helps
improving Open vSwitch debuggability when using flower offloading.

It would also be helpful if userspace monitor tools, like "tc monitor",
could log this kind of message, as it doesn't require vswitchd log level
adjusment. Let's add a new tc attributes to report the extack message so
the monitor program could receive the failures. e.g.

  # tc monitor
  added chain dev enp3s0f1np1 parent ffff: chain 0
  added filter dev enp3s0f1np1 ingress protocol all pref 49152 flower chain 0 handle 0x1
    ct_state +trk+new
    not_in_hw
          action order 1: gact action drop
           random type none pass val 0
           index 1 ref 1 bind 1

  Warning: mlx5_core: matching on ct_state +new isn't supported.

In this patch I only report the extack message on add/del operations.
It doesn't look like we need to report the extack message on get/dump
operations.

Note this message not only reporte to multicast groups, it could also
be reported unicast, which may affect the current usersapce tool's behaivor.

Suggested-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://lore.kernel.org/r/20230113034353.2766735-1-liuhangbin@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agoi2c: add fwnode APIs
Russell King (Oracle) [Wed, 11 Jan 2023 10:54:21 +0000 (10:54 +0000)]
i2c: add fwnode APIs

Add fwnode APIs for finding and getting I2C adapters, which will be
used by the SFP code. These are passed the fwnode corresponding to
the adapter, and return the I2C adapter. It is the responsibility of
the caller to find the appropriate fwnode.

We keep the DT and ACPI interfaces, but where appropriate, recode them
to use the fwnode interfaces internally.

Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: Wolfram Sang <wsa@kernel.org>
18 months agoMerge branch 'dt-bindings-ocelot-switches'
David S. Miller [Mon, 16 Jan 2023 18:42:55 +0000 (18:42 +0000)]
Merge branch 'dt-bindings-ocelot-switches'

Colin Foster says

====================
dt-binding preparation for ocelot switches

Ocelot switches have the abilitiy to be used internally via
memory-mapped IO or externally via SPI or PCIe. This brings up issues
for documentation, where the same chip might be accessed internally in a
switchdev manner, or externally in a DSA configuration. This patch set
is perparation to bring DSA functionality to the VSC7512, utilizing as
much as possible with an almost identical VSC7514 chip.

This patch set changed quite a bit from v2, so I'll omit the background
of how those sets came to be. Rob offered a lot of very useful guidance.
My thanks.

At the end of the day, with this patch set, there should be a framework
to document Ocelot switches (and any switch) in scenarios where they can
be controlled internally (ethernet-switch) or externally (dsa-switch).

---

v6 -> v7
  * Add Reviewed / Acked on patch 1
  * Clean up descriptions on Ethernet / DSA switch port bindings

v5 -> v6
  * Rebase so it applies to net-next cleanly.
  * No other changes - during the last submission round I said I'd
    submit v6 with a change to move $dsa-port.yaml to outside the allOf
    list. In retrospect that wasn't the right thing to do, because later
    in the patch series the $dsa-port.yaml is removed outright. So I
    believe the submission in v5 to keep "type: object" was correct.

v4 -> v5
  * Sync DSA maintainers with MAINTAINERS file (new patch 1)
  * Undo move of port description of mediatek,mt7530.yaml (patch 4)
  * Move removal of "^(ethernet-)?switch(@.*)?$" in dsa.yaml from patch 4
    to patch 8
  * Add more consistent capitalization in title lines and better Ethernet
    switch port description. (patch 8)

v3 -> v4
  * Renamed "base" to "ethernet-ports" to avoid confusion with the concept
    of a base class.
  * Squash ("dt-bindings: net: dsa: mediatek,mt7530: fix port description location")
    patch into ("dt-bindings: net: dsa: utilize base definitions for standard dsa
    switches")
  * Corrections to fix confusion about additonalProperties vs unevaluatedProperties.
    See specific patches for details.

v2 -> v3
  * Restructured everything to use a "base" iref for devices that don't
    have additional properties, and simply a "ref" for devices that do.
  * New patches to fix up brcm,sf2, qca8k, and mt7530
  * Fix unevaluatedProperties errors from previous sets (see specific
    patches for more detail)
  * Removed redundant "Device Tree Binding" from titles, where applicable.

v1 -> v2
  * Two MFD patches were brought into the MFD tree, so are dropped
  * Add first patch 1/6 to allow DSA devices to add ports and port
    properties
  * Test qca8k against new dt-bindings and fix warnings. (patch 2/6)
  * Add tags (patch 3/6)
  * Fix vsc7514 refs and properties
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: mscc,vsc7514-switch: utilize generic ethernet-switch.yaml
Colin Foster [Thu, 12 Jan 2023 17:56:13 +0000 (07:56 -1000)]
dt-bindings: net: mscc,vsc7514-switch: utilize generic ethernet-switch.yaml

Several bindings for ethernet switches are available for non-dsa switches
by way of ethernet-switch.yaml. Remove these duplicate entries and utilize
the common bindings for the VSC7514.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: add generic ethernet-switch-port binding
Colin Foster [Thu, 12 Jan 2023 17:56:12 +0000 (07:56 -1000)]
dt-bindings: net: add generic ethernet-switch-port binding

The dsa-port.yaml binding had several references that can be common to all
ethernet ports, not just dsa-specific ones. Break out the generic bindings
to ethernet-switch-port.yaml they can be used by non-dsa drivers.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: add generic ethernet-switch
Colin Foster [Thu, 12 Jan 2023 17:56:11 +0000 (07:56 -1000)]
dt-bindings: net: add generic ethernet-switch

The dsa.yaml bindings had references that can apply to non-dsa switches. To
prevent duplication of this information, keep the dsa-specific information
inside dsa.yaml and move the remaining generic information to the newly
created ethernet-switch.yaml.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: dsa: mediatek,mt7530: remove unnecessary dsa-port reference
Colin Foster [Thu, 12 Jan 2023 17:56:10 +0000 (07:56 -1000)]
dt-bindings: net: dsa: mediatek,mt7530: remove unnecessary dsa-port reference

dsa.yaml contains a reference to dsa-port.yaml, so a duplicate reference to
the binding isn't necessary. Remove this unnecessary reference.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: dsa: qca8k: utilize shared dsa.yaml
Colin Foster [Thu, 12 Jan 2023 17:56:09 +0000 (07:56 -1000)]
dt-bindings: net: dsa: qca8k: utilize shared dsa.yaml

The dsa.yaml binding contains duplicated bindings for address and size
cells, as well as the reference to dsa-port.yaml. Instead of duplicating
this information, remove the reference to dsa-port.yaml and include the
full reference to dsa.yaml.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: dsa: allow additional ethernet-port properties
Colin Foster [Thu, 12 Jan 2023 17:56:08 +0000 (07:56 -1000)]
dt-bindings: net: dsa: allow additional ethernet-port properties

Explicitly allow additional properties for both the ethernet-port and
ethernet-ports properties. This specifically will allow the qca8k.yaml
binding to use shared properties.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: dsa: utilize base definitions for standard dsa switches
Colin Foster [Thu, 12 Jan 2023 17:56:07 +0000 (07:56 -1000)]
dt-bindings: net: dsa: utilize base definitions for standard dsa switches

DSA switches can fall into one of two categories: switches where all ports
follow standard '(ethernet-)?port' properties, and switches that have
additional properties for the ports.

The scenario where DSA ports are all standardized can be handled by
switches with a reference to the new 'dsa.yaml#/$defs/ethernet-ports'.

The scenario where DSA ports require additional properties can reference
'$dsa.yaml#' directly. This will allow switches to reference these standard
definitions of the DSA switch, but add additional properties under the port
nodes.

Suggested-by: Rob Herring <robh@kernel.org>
Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Acked-by: Alvin Ĺ ipraga <alsi@bang-olufsen.dk> # realtek
Acked-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: dsa: qca8k: remove address-cells and size-cells from switch node
Colin Foster [Thu, 12 Jan 2023 17:56:06 +0000 (07:56 -1000)]
dt-bindings: net: dsa: qca8k: remove address-cells and size-cells from switch node

The children of the switch node don't have a unit address, and therefore
should not need the #address-cells or #size-cells entries. Fix the example
schemas accordingly.

Suggested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: net: dsa: sf2: fix brcm,use-bcm-hdr documentation
Colin Foster [Thu, 12 Jan 2023 17:56:05 +0000 (07:56 -1000)]
dt-bindings: net: dsa: sf2: fix brcm,use-bcm-hdr documentation

The property use-bcm-hdr was documented as an entry under the ports node
for the bcm_sf2 DSA switch. This property is actually evaluated for each
port. Correct the documentation to match the actual behavior and properly
reference dsa-port.yaml for additional properties of the node.

Suggested-by: Rob Herring <robh@kernel.org>
Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agodt-bindings: dsa: sync with maintainers
Colin Foster [Thu, 12 Jan 2023 17:56:04 +0000 (07:56 -1000)]
dt-bindings: dsa: sync with maintainers

The MAINTAINERS file has Andrew Lunn, Florian Fainelli, and Vladimir Oltean
listed as the maintainers for generic dsa bindings. Update dsa.yaml and
dsa-port.yaml accordingly.

Signed-off-by: Colin Foster <colin.foster@in-advantage.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Acked-by: Rob Herring <robh@kernel.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'net-microchip-vcap-rules'
David S. Miller [Mon, 16 Jan 2023 13:45:17 +0000 (13:45 +0000)]
Merge branch 'net-microchip-vcap-rules'

Steen Hegelund says:

====================
net: microchip: Add support for two classes of VCAP rules

This adds support for two classes of VCAP rules:

- Permanent rules (added e.g. for PTP support)
- TC user rules (added by the TC userspace tool)

For this to work the VCAP Loopups must be enabled from boot, so that the
"internal" clients like PTP can add rules that are always active.

When the TC tool add a flower filter the VCAP rule corresponding to this
filter will be disabled (kept in memory) until a TC matchall filter creates
a link from chain 0 to the chain (lookup) where the flower filter was
added.

When the flower filter is enabled it will be written to the appropriate
VCAP lookup and become active in HW.

Likewise the flower filter will be disabled if there is no link from chain
0 to the chain of the filter (lookup), and when that happens the
corresponding VCAP rule will be read from the VCAP instance and stored in
memory until it is deleted or enabled again.

Version History:
================
v4      Removed a leftover 'Fixes' tag from v2.  No functional changes.

v3      Removed the change that allowed rules to always be added in the
        LAN996x even though the lookups are not enabled (Horatiu Vultur).
        This was sent separately to net instead.

        Removed the 'Fixes' tags due to the patch sent to net by Horatiu
        Vultur.

        Added a check for validity of the chain source when enabling a
        lookup.

v2      Adding a missing goto exit in vcap_add_rule (Dan Carpenter).
        Added missing checks for error returns in vcap_enable_rule.

v1      Initial version
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: microchip: vcap api: Enable/Disable rules via chains in VCAP HW
Steen Hegelund [Sat, 14 Jan 2023 13:42:42 +0000 (14:42 +0100)]
net: microchip: vcap api: Enable/Disable rules via chains in VCAP HW

This supports that individual rules are enabled and disabled via chain
information.
This is done by keeping disabled rules in the VCAP list (cached) until they
are enabled, and only at this time are the rules written to the VCAP HW.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: microchip: vcap api: Add a storage state to a VCAP rule
Steen Hegelund [Sat, 14 Jan 2023 13:42:41 +0000 (14:42 +0100)]
net: microchip: vcap api: Add a storage state to a VCAP rule

This allows a VCAP rule to be in one of 3 states:

- permanently stored in the VCAP HW (for rules that must always be present)
- enabled (stored in HW) when the corresponding lookup has been enabled
- disabled (stored in SW) when the lookup is disabled

This way important VCAP rules can be added even before the user enables the
VCAP lookups using a TC matchall filter.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: microchip: vcap api: Check chains when adding a tc flower filter
Steen Hegelund [Sat, 14 Jan 2023 13:42:40 +0000 (14:42 +0100)]
net: microchip: vcap api: Check chains when adding a tc flower filter

This changes the way the chain information verified when adding a new tc
flower filter.

When adding a flower filter it is now checked that the filter contains a
goto action to one of the IS2 VCAP lookups, except for the last lookup
which may omit this goto action.

It is also checked if you attempt to add multiple matchall filters to
enable the same VCAP lookup.  This will be rejected.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: microchip: vcap api: Use src and dst chain id to chain VCAP lookups
Steen Hegelund [Sat, 14 Jan 2023 13:42:39 +0000 (14:42 +0100)]
net: microchip: vcap api: Use src and dst chain id to chain VCAP lookups

This adds both the source and destination chain id to the information kept
for enabled port lookups.
This allows enabling and disabling a chain of lookups by walking the chain
information for a port.

This changes the way that VCAP lookups are enabled from userspace: instead
of one matchall rule that enables all the 4 Sparx5 IS2 lookups, you need a
matchall rule per lookup.

In practice that is done by adding one matchall rule in chain 0 to goto IS2
Lookup 0, and then for each lookup you add a rule per lookup (low priority)
that does a goto to the next lookup chain.

Examples:

If you want IS2 Lookup 0 to be enabled you add the same matchall filter as
before:

tc filter add dev eth12 ingress chain 0 prio 1000 handle 1000 matchall \
       skip_sw action goto chain 8000000

If you also want to enable lookup 1 to 3 in IS2 and chain them you need
to add the following matchall filters:

tc filter add dev eth12 ingress chain 8000000 prio 1000 handle 1000 \
    matchall skip_sw action goto chain 8100000

tc filter add dev eth12 ingress chain 8100000 prio 1000 handle 1000 \
    matchall skip_sw action goto chain 8200000

tc filter add dev eth12 ingress chain 8200000 prio 1000 handle 1000 \
    matchall skip_sw action goto chain 8300000

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: microchip: vcap api: Convert multi-word keys/actions when encoding
Steen Hegelund [Sat, 14 Jan 2023 13:42:38 +0000 (14:42 +0100)]
net: microchip: vcap api: Convert multi-word keys/actions when encoding

The conversion to the platform specific multi-word format is moved from the
key/action add functions to the encoding key/action.
This allows rules that are disabled (not in VCAP HW) to use the same format
for keys/actions as rules that have just been read from VCAP HW.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: microchip: vcap api: Always enable VCAP lookups
Steen Hegelund [Sat, 14 Jan 2023 13:42:37 +0000 (14:42 +0100)]
net: microchip: vcap api: Always enable VCAP lookups

This changes the VCAP lookups state to always be enabled so that it is
possible to add "internal" VCAP rules that must be available even though
the user has not yet enabled the VCAP chains via a TC matchall filter.

The API callback to enable and disable VCAP lookups is therefore removed.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: microchip: sparx5: Reset VCAP counter for new rules
Steen Hegelund [Sat, 14 Jan 2023 13:42:36 +0000 (14:42 +0100)]
net: microchip: sparx5: Reset VCAP counter for new rules

When a rule counter is external to the VCAP such as the Sparx5 IS2 counters
are, then this counter must be reset when a new rule is created.

Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: microchip: vcap api: Erase VCAP cache before encoding rule
Steen Hegelund [Sat, 14 Jan 2023 13:42:35 +0000 (14:42 +0100)]
net: microchip: vcap api: Erase VCAP cache before encoding rule

For consistency the VCAP cache area is erased just before the new rule is
being encoded.

Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agor8169: reset bus if NIC isn't accessible after tx timeout
Heiner Kallweit [Fri, 13 Jan 2023 22:46:19 +0000 (23:46 +0100)]
r8169: reset bus if NIC isn't accessible after tx timeout

ASPM issues may result in the NIC not being accessible any longer.
In this case disabling ASPM may not work. Therefore detect this case
by checking whether register reads return ~0, and try to make the
NIC accessible again by resetting the secondary bus.

v2:
- add exception handling for the case that pci_reset_bus() fails

Suggested-by: Alexander Duyck <alexander.duyck@gmail.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: dsa: mv88e6xxx: Enable PTP receive for mv88e6390
Kurt Kanzenbach [Fri, 13 Jan 2023 15:12:58 +0000 (16:12 +0100)]
net: dsa: mv88e6xxx: Enable PTP receive for mv88e6390

The switch receives management traffic such as STP and LLDP. However, PTP
messages are not received, only transmitted.

Ideally, the switch would trap all PTP messages to the management CPU. This
particular switch has a PTP block which identifies PTP messages and traps them
to a dedicated port. There is a register to program this destination. This is
not used at the moment.

Therefore, program it to the same port as the MGMT traffic is trapped to. This
allows to receive PTP messages as soon as timestamping is enabled.

In addition, the datasheet mentions that this register is not valid e.g., for
6190 variants. So, add a new PTP operation which is added for the 6390 and 6290
devices.

Tested simply like this on Marvell 88E6390, revision 1:

|/ # ptp4l -2 -i lan4 --tx_timestamp_timeout=40 -m
|[...]
|ptp4l[147.450]: master offset         56 s2 freq   +1262 path delay       413
|ptp4l[148.450]: master offset         22 s2 freq   +1244 path delay       434
|ptp4l[149.450]: master offset          5 s2 freq   +1234 path delay       446
|ptp4l[150.451]: master offset          3 s2 freq   +1233 path delay       451
|ptp4l[151.451]: master offset          1 s2 freq   +1232 path delay       451
|ptp4l[152.451]: master offset         -3 s2 freq   +1229 path delay       451
|ptp4l[153.451]: master offset          9 s2 freq   +1240 path delay       451

Link: https://lore.kernel.org/r/CAFSKS=PJBpvtRJxrR4sG1hyxpnUnQpiHg4SrUNzAhkWnyt9ivg@mail.gmail.com
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio/vsock: replace virtio_vsock_pkt with sk_buff
Bobby Eshleman [Fri, 13 Jan 2023 22:21:37 +0000 (22:21 +0000)]
virtio/vsock: replace virtio_vsock_pkt with sk_buff

This commit changes virtio/vsock to use sk_buff instead of
virtio_vsock_pkt. Beyond better conforming to other net code, using
sk_buff allows vsock to use sk_buff-dependent features in the future
(such as sockmap) and improves throughput.

This patch introduces the following performance changes:

Tool: Uperf
Env: Phys Host + L1 Guest
Payload: 64k
Threads: 16
Test Runs: 10
Type: SOCK_STREAM
Before: commit b7bfaa761d760 ("Linux 6.2-rc3")

Before
------
g2h: 16.77Gb/s
h2g: 10.56Gb/s

After
-----
g2h: 21.04Gb/s
h2g: 10.76Gb/s

Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next...
David S. Miller [Mon, 16 Jan 2023 13:21:11 +0000 (13:21 +0000)]
Merge branch '10GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2023-01-13 (ixgbe)

This series contains updates to ixgbe driver only.

Jesse resolves warning for RCU pointer by no longer restoring old
pointer.

Sebastian adds waiting for updating of link info on devices utilizing
crosstalk fix to avoid false link state.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoregmap: Rework regmap_mdio_c45_{read|write} for new C45 API.
Andrew Lunn [Mon, 16 Jan 2023 11:15:09 +0000 (12:15 +0100)]
regmap: Rework regmap_mdio_c45_{read|write} for new C45 API.

The MDIO subsystem is getting rid of MII_ADDR_C45 and thus also
encoding associated encoding of the C45 device address and register
address into one value. regmap-mdio also uses this encoding for the
C45 bus.

Move to the new C45 helpers for MDIO access and provide regmap-mdio
helper macros.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Link: https://lore.kernel.org/r/20230116111509.4086236-1-michael@walle.cc
Signed-off-by: Mark Brown <broonie@kernel.org>
18 months agounix: Improve locking scheme in unix_show_fdinfo()
Kirill Tkhai [Sat, 14 Jan 2023 09:35:02 +0000 (12:35 +0300)]
unix: Improve locking scheme in unix_show_fdinfo()

After switching to TCP_ESTABLISHED or TCP_LISTEN sk_state, alive SOCK_STREAM
and SOCK_SEQPACKET sockets can't change it anymore (since commit 3ff8bff704f4
"unix: Fix race in SOCK_SEQPACKET's unix_dgram_sendmsg()").

Thus, we do not need to take lock here.

Signed-off-by: Kirill Tkhai <tkhai@ya.ru>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'virtio-net-xdp-multi-buffer'
David S. Miller [Mon, 16 Jan 2023 11:15:49 +0000 (11:15 +0000)]
Merge branch 'virtio-net-xdp-multi-buffer'

Heng Qi says:

====================
virtio-net: support multi buffer xdp

Changes since PATCH v4:
- Make netdev_warn() in [PATCH 2/10] independent from [PATCH 3/10].

Changes since PATCH v3:
- Separate fix patch [2/10] for MTU calculation of single buffer xdp.
  Note that this patch needs to be backported to the stable branch.

Changes since PATCH v2:
- Even if single buffer xdp has a hole mechanism, there will be no
  problem (limiting mtu and turning off GUEST GSO), so there is no
  need to backport "[PATCH 1/9]";
- Modify calculation of MTU for single buffer xdp in virtnet_xdp_set();
- Make truesize in mergeable mode return to literal meaning;
- Add some comments for legibility;

Changes since RFC:
- Using headroom instead of vi->xdp_enabled to avoid re-reading
  in add_recvbuf_mergeable();
- Disable GRO_HW and keep linearization for single buffer xdp;
- Renamed to virtnet_build_xdp_buff_mrg();
- pr_debug() to netdev_dbg();
- Adjusted the order of the patch series.

Currently, virtio net only supports xdp for single-buffer packets
or linearized multi-buffer packets. This patchset supports xdp for
multi-buffer packets, then larger MTU can be used if xdp sets the
xdp.frags. This does not affect single buffer handling.

In order to build multi-buffer xdp neatly, we integrated the code
into virtnet_build_xdp_buff_mrg() for xdp. The first buffer is used
for prepared xdp buff, and the rest of the buffers are added to
its skb_shared_info structure. This structure can also be
conveniently converted during XDP_PASS to get the corresponding skb.

Since virtio net uses comp pages, and bpf_xdp_frags_increase_tail()
is based on the assumption of the page pool,
(rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag))
is negative in most cases. So we didn't set xdp_rxq->frag_size in
virtnet_open() to disable the tail increase.

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

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: support multi-buffer xdp
Heng Qi [Sat, 14 Jan 2023 08:22:29 +0000 (16:22 +0800)]
virtio-net: support multi-buffer xdp

Driver can pass the skb to stack by build_skb_from_xdp_buff().

Driver forwards multi-buffer packets using the send queue
when XDP_TX and XDP_REDIRECT, and clears the reference of multi
pages when XDP_DROP.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: remove xdp related info from page_to_skb()
Heng Qi [Sat, 14 Jan 2023 08:22:28 +0000 (16:22 +0800)]
virtio-net: remove xdp related info from page_to_skb()

For the clear construction of xdp_buff, we remove the xdp processing
interleaved with page_to_skb(). Now, the logic of xdp and building
skb from xdp are separate and independent.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: build skb from multi-buffer xdp
Heng Qi [Sat, 14 Jan 2023 08:22:27 +0000 (16:22 +0800)]
virtio-net: build skb from multi-buffer xdp

This converts the xdp_buff directly to a skb, including
multi-buffer and single buffer xdp. We'll isolate the
construction of skb based on xdp from page_to_skb().

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: transmit the multi-buffer xdp
Heng Qi [Sat, 14 Jan 2023 08:22:26 +0000 (16:22 +0800)]
virtio-net: transmit the multi-buffer xdp

This serves as the basis for XDP_TX and XDP_REDIRECT
to send a multi-buffer xdp_frame.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: construct multi-buffer xdp in mergeable
Heng Qi [Sat, 14 Jan 2023 08:22:25 +0000 (16:22 +0800)]
virtio-net: construct multi-buffer xdp in mergeable

Build multi-buffer xdp using virtnet_build_xdp_buff_mrg().

For the prefilled buffer before xdp is set, we will probably use
vq reset in the future. At the same time, virtio net currently
uses comp pages, and bpf_xdp_frags_increase_tail() needs to calculate
the tailroom of the last frag, which will involve the offset of the
corresponding page and cause a negative value, so we disable tail
increase by not setting xdp_rxq->frag_size.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: build xdp_buff with multi buffers
Heng Qi [Sat, 14 Jan 2023 08:22:24 +0000 (16:22 +0800)]
virtio-net: build xdp_buff with multi buffers

Support xdp for multi buffer packets in mergeable mode.

Putting the first buffer as the linear part for xdp_buff,
and the rest of the buffers as non-linear fragments to struct
skb_shared_info in the tailroom belonging to xdp_buff.

Let 'truesize' return to its literal meaning, that is, when
xdp is set, it includes the length of headroom and tailroom.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: update bytes calculation for xdp_frame
Heng Qi [Sat, 14 Jan 2023 08:22:23 +0000 (16:22 +0800)]
virtio-net: update bytes calculation for xdp_frame

Update relative record value for xdp_frame as basis
for multi-buffer xdp transmission.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: set up xdp for multi buffer packets
Heng Qi [Sat, 14 Jan 2023 08:22:22 +0000 (16:22 +0800)]
virtio-net: set up xdp for multi buffer packets

When the xdp program sets xdp.frags, which means it can process
multi-buffer packets over larger MTU, so we continue to support xdp.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovirtio-net: fix calculation of MTU for single-buffer xdp
Heng Qi [Sat, 14 Jan 2023 08:22:21 +0000 (16:22 +0800)]
virtio-net: fix calculation of MTU for single-buffer xdp

When single-buffer xdp is loaded, the size of the buffer filled each time
is 'sz = (PAGE_SIZE - headroom - tailroom)', which is the maximum packet
length that the driver allows the device to pass in. Otherwise, the packet
with a length greater than sz will come in, so num_buf will be greater than
or equal to 2, and xdp_linearize_page() will be performed and the packet
will be dropped because the total length is greater than PAGE_SIZE. So the
maximum value of MTU for single-buffer xdp is 'max_sz = sz - ETH_HLEN'.

Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>