af_unix: Remove io_uring code for GC.

Since commit 705318a99a13 ("io_uring/af_unix: disable sending
io_uring over sockets"), io_uring's unix socket cannot be passed
via SCM_RIGHTS, so it does not contribute to cyclic reference and
no longer be candidate for garbage collection.

Also, commit 6e5e6d274956 ("io_uring: drop any code related to
SCM_RIGHTS") cleaned up SCM_RIGHTS code in io_uring.

Let's do it in AF_UNIX as well by reverting commit 0091bfc81741
("io_uring/af_unix: defer registered files gc to io_uring release")
and commit 10369080454d ("net: reclaim skb->scm_io_uring bit").

Signed-off-by: Kuniyuki Iwashima <[email protected]>
Acked-by: Jens Axboe <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

af_unix: Replace BUG_ON() with WARN_ON_ONCE().

This is a prep patch for the last patch in this series so that
checkpatch will not warn about BUG_ON().

Signed-off-by: Kuniyuki Iwashima <[email protected]>
Acked-by: Jens Axboe <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: bridge: Use KMEM_CACHE instead of kmem_cache_create

commit 0a31bd5f2bbb ("KMEM_CACHE(): simplify slab cache creation")
introduces a new macro.
Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
to simplify the creation of SLAB caches.

Signed-off-by: Kunwu Chan <[email protected]>
Acked-by: Nikolay Aleksandrov <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: ipv4: Simplify the allocation of slab caches in inet_initpeers

commit 0a31bd5f2bbb ("KMEM_CACHE(): simplify slab cache creation")
introduces a new macro.
Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
to simplify the creation of SLAB caches.

Signed-off-by: Kunwu Chan <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Merge branch 'net-phy-split-at803x'

Christian Marangi says:

====================
net: phy: split at803x

This is the last patchset of a long series of cleanup and
preparation to make at803x better maintainable and permit
the addition of other QCOM PHY Families.

A shared library modules is created since many QCOM PHY share
similar/exact implementation and are reused.

This series doesn't introduce any new code but just move the
function around and introduce a new module for all the functions
that are shared between the 3 different PHY family.

Since the drivers are actually detached, new probe function are
introduced that allocate the specific priv struct for the PHYs.

After this patch, qca808x will be further generalized as LED
and cable test function are also used by the QCA807x PHYs.
This is just for reference and the additional function move will
be done on the relates specific series.

This is also needed in preparation for the introduction of
qca807x PHYs family and PHY package concept.
====================

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: phy: qcom: detach qca808x PHY driver from at803x

Almost all the QCA8081 PHY driver OPs are specific and only some of them
use the generic at803x.

To make the at803x code slimmer, move all the specific qca808x regs and
functions to a dedicated PHY driver.

Probe function and priv struct is reworked to allocate and use only the
qca808x specific data. Unused data from at803x PHY driver are dropped
from at803x priv struct.

Also a new Kconfig is introduced QCA808X_PHY, to compile the newly
introduced PHY driver for QCA8081 PHY.

As the Kconfig name starts with Qualcomm the same order is kept.

Signed-off-by: Christian Marangi <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: phy: qcom: move additional functions to shared library

Move additional functions to shared library in preparation for qca808x
PHY Family to be detached from at803x driver.

Only the shared defines are moved to the shared qcom.h header.

Signed-off-by: Christian Marangi <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: phy: qcom: deatch qca83xx PHY driver from at803x

Deatch qca83xx PHY driver from at803x.

The QCA83xx PHYs implement specific function and doesn't use generic
at803x so it can be detached from the driver and moved to a dedicated
one.

Probe function and priv struct is reimplemented to allocate and use
only the qca83xx specific data. Unused data from at803x PHY driver
are dropped from at803x priv struct.

This is to make slimmer PHY drivers instead of including lots of bloat
that would never be used in specific SoC.

A new Kconfig flag QCA83XX_PHY is introduced to compile the new
introduced PHY driver.

As the Kconfig name starts with Qualcomm the same order is kept.

Signed-off-by: Christian Marangi <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: phy: qcom: create and move functions to shared library

Create and move functions to shared library in preparation for qca83xx
PHY Family to be detached from at803x driver.

Only the shared defines are moved to the shared qcom.h header.

Signed-off-by: Christian Marangi <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: phy: move at803x PHY driver to dedicated directory

In preparation for addition of other Qcom PHY and to tidy things up,
move the at803x PHY driver to dedicated directory.

The same order in the Kconfig selection is saved.

Signed-off-by: Christian Marangi <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Merge branch 'prevent-nullptr-exceptions-in-isr'

Andre Werner says:

====================
Prevent nullptr exceptions in ISR

In case phydev->irq is modified unconditionally to a valid IRQ, handling
the IRQ may lead to a nullptr exception if no interrupt handler is
registered to the phy driver. phy_interrupt calls a
phy_device->handle_interrupt unconditionally. And interrupts are enabled
in phy_connect_direct if phydev->irq is not equal to PHY_POLL or
PHY_MAC_INTERRUPT, so it does not check for a phy driver providing an ISR.

Adding an additonal check for a valid interrupt handler in phy_attach_direct
function, and falling back to polling mode if not, should prevent for
such nullptr exceptions.

Moreover, the ADIN1100 phy driver is extended with an interrupt handler
for changes in the link status.
====================

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: phy: adin1100: Add interrupt support for link change

An interrupt handler was added to the driver as well as functions
to enable interrupts at the phy.

There are several interrupts maskable at the phy, but only link change
interrupts are handled by the driver yet.

Signed-off-by: Andre Werner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: phy: phy_device: Prevent nullptr exceptions on ISR

If phydev->irq is set unconditionally, check
for valid interrupt handler or fall back to polling mode to prevent
nullptr exceptions in interrupt service routine.

Signed-off-by: Andre Werner <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

ptp: lan743x: Use spin_lock instead of spin_lock_bh

lan743x_ptp_request_tx_timestamp uses spin_lock_bh, but it is
only called from lan743x_tx_xmit_frame where all IRQs are
already disabled.

This fixes the "IRQs not enabled as expected" warning.

Signed-off-by: Lucas Tanure <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

dpll: move xa_erase() call in to match dpll_pin_alloc() error path order

This is cosmetics. Move the call of xa_erase() in dpll_pin_put()
so the order of cleanup calls matches the error path of
dpll_pin_alloc().

Signed-off-by: Jiri Pirko <[email protected]>
Reviewed-by: Vadim Fedorenko <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

selftests/net: calibrate txtimestamp

The test sends packets and compares enqueue, transmit and Ack
timestamps with expected values. It installs netem delays to increase
latency between these points.

The test proves flaky in virtual environment (vng). Increase the
delays to reduce variance. Scale measurement tolerance accordingly.

Time sensitive tests are difficult to calibrate. Increasing delays 10x
also increases runtime 10x, for one. And it may still prove flaky at
some rate.

Signed-off-by: Willem de Bruijn <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Merge tag 'nf-next-24-01-29' of https://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf-next
Florian Westphal says:

====================
nf-next pr 2024-01-29

This batch contains updates for your *next* tree.

First three changes, from Phil Sutter, allow userspace to define
a table that is exclusively owned by a daemon (via netlink socket
aliveness) without auto-removing this table when the userspace program
exits.  Such table gets marked as orphaned and a restarting management
daemon may re-attach/reassume ownership.

Next patch, from Pablo, passes already-validated flags variable around
rather than having called code re-fetch it from netlnik message.

Patches 5 and 6 update ipvs and nf_conncount to use the recently
introduced KMEM_CACHE() macro.

Last three patches, from myself, tweak kconfig logic a little to
permit a kernel configuration that can run iptables-over-nftables
but not classic (setsockopt) iptables.

Such builds lack the builtin-filter/mangle/raw/nat/security tables,
the set/getsockopt interface and the "old blob format"
interpreter/traverser.  For now, this is 'oldconfig friendly', users
need to manually deselect existing config options for this.
====================

Signed-off-by: David S. Miller <[email protected]>

Merge branch 'qca_spi-improvements'

Stefan Wahren says:

====================
qca_spi: collection of improvements

This series contains a wild collection of improvements for the
qca_spi driver. This is a follow-up series to the recent bugfixes [1].

Patch 1, 2 & 3 in this series is the initially intended rework of
netdev_open/close. Patch 4 & 10 are minor functional improvements and
the rest is clean-up.

Changes in V4:
- keep thread handling in netdev_open, but improve this by two
  separate patches ( 1 & 2 )
- add missing link to bugfix series

Changes in V3:
- rebase on net-next-20240124
- improve commit message in Patch 8
- add Reviewed-by's from Jacob Keller
- add Patch 13 for mailmap as suggested by Jacob
- add Patch 14 for MAINTAINERS entry

Changes in V2:
- rebase on net-next-20231218
- improve cover letter

[1] - https://lore.kernel.org/netdev/20231206141222[email protected]/
====================

Signed-off-by: David S. Miller <[email protected]>

MAINTAINERS: add entry for qca7k driver(s)

Since upstreaming i contributed a lot to this driver(s),
so add myself as a maintainer.

Signed-off-by: Stefan Wahren <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

mailmap: add entry for Stefan Wahren

Add a .mailmap entry because my old i2se.com address will be
deactivated in the near future.

Signed-off-by: Stefan Wahren <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_7k: Replace old mail address

The company I2SE has been acquired a long time ago. Switch to
my private mail address before the I2SE account is deactivated.

Signed-off-by: Stefan Wahren <[email protected]>
Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_7k: Replace BSD boilerplate with SPDX

According to MODULE_LICENSE the driver is under a dual license.
So replace the BSD license text with the proper SPDX tag.

Signed-off-by: Stefan Wahren <[email protected]>
Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Adjust log of SPI_REG_RDBUF_BYTE_AVA

All known SPI registers of the QCA700x are 16 bit long. So adjust
the formater width accordingly.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Log expected signature in error case

Most of the users doesn't know the expected signature of the QCA700x.
So provide it within the error message. Btw use lowercase for hex as
in the rest of the driver.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Improve calculation of RX buffer size

There are two points with the calculation of RX buffer size which are
not optimal:
1. dev->mtu is a mutual parameter and it's currently initialized with
   QCAFRM_MAX_MTU. But for RX buffer size calculation we always need the
   maximum possible MTU. So better use the define directly.
2. This magic number 4 represent the hardware generated frame length
   which is specific to SPI. We better replace this with the suitable
   define.

There is no functional change.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Introduce QCASPI_RX_MAX_FRAMES

Currently qca_spi reserves enough space for 4 complete Ethernet over SPI
frames in the receive buffer. Unfortunately this is hidden under a magic
number. So replace it with a more self explaining define.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Add QCASPI prefix to ring defines

All defines in qca_spi.h except of the two ring limit defines have
a QCASPI prefix. Since the name is quite generic add the QCASPI prefix
to avoid possible name conflicts.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_7k_common: Drop unused len from qcafrm_handle

This member is never used. So drop it.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_7k_common: Drop unnecessary function description

qcafrm_fsm_decode has the almost the same function description in
qca_7k_common.c. So drop the comment here.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Avoid skb_copy_expand in TX path

The skb spare room needs to be expanded for SPI header, footer
and possible padding within the TX path. So announce the necessary
space in order to avoid expensive skb_copy_expand calls.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Improve SPI IRQ handling

The functions qcaspi_netdev_open/close are responsible of request &
free of the SPI interrupt, which wasn't the best choice because
allocation problems are discovered not during probe. So let us split
IRQ allocation & enabling, so we can take advantage of a device
managed IRQ.

Signed-off-by: Stefan Wahren <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Improve SPI thread creation

Directly storing the result of kthread_run within the private
driver data makes it harder to identify if the pointer has a
running thread or not. So better use a local variable for
the result check and we don't have to care about error pointer
in the rest of the code.

Signed-off-by: Stefan Wahren <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

qca_spi: Add check for kthread_stop

We better not rely on that spi_thread points to a running
thread. So add an check for this.

Signed-off-by: Stefan Wahren <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: lan966x: debugfs: Fix showing the port keyset

On lan966x, it is possible to use debugfs to print different information
about the VCAPs. Information like, if it is enabled, how the ports are
configured, print the actual rules. The issue is that when printing how
the ports are configured for IS1 lookups, it was parsing the wrong
register to get this information. The fix consists in reading the
correct register that contains this information.

Signed-off-by: Horatiu Vultur <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'ethtool-EEE'

Heiner Kallweit says:

====================
ethtool: switch EEE netlink interface to use EEE linkmode bitmaps

So far only 32bit legacy bitmaps are passed to userspace. This makes
it impossible to manage EEE linkmodes beyond bit 32, e.g. manage EEE
for 2500BaseT and 5000BaseT. This series adds support for passing
full linkmode bitmaps between kernel and userspace.

Fortunately the netlink-based part of ethtool is quite smart and no
changes are needed in ethtool. However this applies to the netlink
interface only, the ioctl interface for now remains restricted to
legacy bitmaps.

Next step will be adding support for the c45 EEE2 standard registers
(3.21, 7.62, 7.63) to the genphy_c45 functions dealing with EEE.
I have a follow-up series for this ready to be submitted.

v2:
- now as RFC
- adopt suggestion from Andrew to start with struct ethtool_keee
  being an identical copy of ethtool_eee, and switch all users
v3:
- switch from RFC to net-next
- add patch 4, and reuse old names in patch 5
- rebase patch 1
v4:
- fix missing replacement in patch 4
====================

Signed-off-by: David S. Miller <[email protected]>

net: phy: c45: change genphy_c45_ethtool_[get|set]_eee to use EEE linkmode bitmaps

Change genphy_c45_ethtool_[get|set]_eee to use EEE linkmode bitmaps.
This is a prerequisite for adding support for EEE modes beyond bit 31.

Signed-off-by: Heiner Kallweit <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

ethtool: add linkmode bitmap support to struct ethtool_keee

Add linkmode bitmap members to struct ethtool_keee, but keep the legacy
u32 bitmaps for compatibility with existing drivers.
Use linkmode "supported" not being empty as indicator that a user wants
to use the linkmode bitmap members instead of the legacy bitmaps.

Signed-off-by: Heiner Kallweit <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

ethtool: add suffix _u32 to legacy bitmap members of struct ethtool_keee

This is in preparation of using the existing names for linkmode
bitmaps.

Suggested-by: Andrew Lunn <[email protected]>
Signed-off-by: Heiner Kallweit <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

ethtool: adjust struct ethtool_keee to kernel needs

This patch changes the following in struct ethtool_keee
- remove member cmd, it's not needed on kernel side
- remove reserved fields
- switch the semantically boolean members to type bool

We don't have to change any user of the boolean members due to the
implicit casting from/to bool. A small change is needed where a
pointer to bool members is used, in addition remove few now unneeded
double negations.

Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: Heiner Kallweit <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

ethtool: switch back from ethtool_keee to ethtool_eee for ioctl

In order to later extend struct ethtool_keee, we have to decouple it
from the userspace format represented by struct ethtool_eee.
Therefore switch back to struct ethtool_eee, representing the userspace
format, and add conversion between ethtool_eee and ethtool_keee.
Struct ethtool_keee will be changed in follow-up patches, therefore
don't do a *keee = *eee here.
Member cmd isn't copied, because it's not used, and we'll remove
it in the next patch of this series. In addition omit setting cmd
to ETHTOOL_GEEE in the ioctl response, userspace ethtool isn't
interested in it.

Signed-off-by: Heiner Kallweit <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

ethtool: replace struct ethtool_eee with a new struct ethtool_keee on kernel side

In order to pass EEE link modes beyond bit 32 to userspace we have to
complement the 32 bit bitmaps in struct ethtool_eee with linkmode
bitmaps. Therefore, similar to ethtool_link_settings and
ethtool_link_ksettings, add a struct ethtool_keee. In a first step
it's an identical copy of ethtool_eee. This patch simply does a
s/ethtool_eee/ethtool_keee/g for all users.
No functional change intended.

Suggested-by: Andrew Lunn <[email protected]>
Signed-off-by: Heiner Kallweit <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'stmmac-EST'

Rohan G says:

====================
net: stmmac: EST conformance support

This patchset enables support for queueMaxSDU and transmission overrun
counters which are required for Qbv conformance.
====================

Signed-off-by: David S. Miller <[email protected]>

net: stmmac: Report taprio offload status

Report taprio offload status. This includes per txq and global
counters of window_drops and tx_overruns.

Window_drops count include count of frames dropped because of
queueMaxSDU setting and HLBF error. Transmission overrun counter
inform the user application whether any packets are currently being
transmitted on a particular queue during a gate-close event.DWMAC IPs
takes care Transmission overrun won't happen hence this is always 0.

Signed-off-by: Rohan G Thomas <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: stmmac: est: Per Tx-queue error count for HLBF

Keep per Tx-queue error count on Head-Of-Line Blocking due to frame
size(HLBF) error. The MAC raises HLBF error on one or more queues
when none of the time Intervals of open-gates in the GCL is greater
than or equal to the duration needed for frame transmission and by
default drops those packets that causes HLBF error. EST_FRM_SZ_ERR
register provides the One Hot encoded Queue numbers that have the
Frame Size related error.

Signed-off-by: Rohan G Thomas <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: stmmac: Offload queueMaxSDU from tc-taprio

Add support for configuring queueMaxSDU. As DWMAC IPs doesn't support
queueMaxSDU table handle this in the SW. The maximum 802.3 frame size
that is allowed to be transmitted by any queue is queueMaxSDU +
16 bytes (i.e. 6 bytes SA + 6 bytes DA + 4 bytes FCS).

Inspired from intel i225 driver.

Signed-off-by: Rohan G Thomas <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

selftests/net: calibrate fq_band_pktlimit

This test validates per-band packet limits in FQ. Packets are dropped
rather than enqueued if the limit for their band is reached.

This test is timing sensitive. It queues packets in FQ with a future
delivery time to fill the qdisc.

The test failed in a virtual environment (vng). Increase the delays
to make it more tolerant to environments with timing variance.

Signed-off-by: Willem de Bruijn <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'selftests-TEST_INCLUDES'

Benjamin Poirier says:

====================
selftests: Add TEST_INCLUDES directive and adjust tests to use it

After commit 25ae948b4478 ("selftests/net: add lib.sh") but before commit
2114e83381d3 ("selftests: forwarding: Avoid failures to source
net/lib.sh"), some net selftests encountered errors when they were being
exported and run. This was because the new net/lib.sh was not exported
along with the tests. The errors were crudely avoided by duplicating some
content between net/lib.sh and net/forwarding/lib.sh in 2114e83381d3.

In order to restore the sourcing of net/lib.sh from net/forwarding/lib.sh
and remove the duplicated content, this series introduces a new selftests
Makefile variable to list extra files to export from other directories and
makes use of it to avoid reintroducing the errors mentioned above.

v2:
* "selftests: Introduce Makefile variable to list shared bash scripts"
Fix rst syntax in Documentation/dev-tools/kselftest.rst (Jakub Kicinski)

v1:
* "selftests: Introduce Makefile variable to list shared bash scripts"
Changed TEST_INCLUDES to take relative paths, like other TEST_* variables.
Paths are adjusted accordingly in the subsequent patches. (Vladimir Oltean)

* selftests: bonding: Change script interpreter
  selftests: forwarding: Remove executable bits from lib.sh
Removed from this series, submitted separately.

Since commit 2114e83381d3 ("selftests: forwarding: Avoid failures to source
net/lib.sh") resolved the test errors, this version of the series is
focused on removing the duplication that was added in that commit. Directly
rebasing the series would reintroduce the problems that 2114e83381d3
avoided before fixing them again. In order to prevent such breakage partway
through the series, patches are reordered and content changed slightly but
there is no diff at the end compared with the simple rebasing approach. I
have dropped most review tags on account of this reordering.

RFC:
https://lore.kernel.org/netdev/20231222135836[email protected]/

Link: https://lore.kernel.org/netdev/ZXu7dGj7F9Ng8iIX@Laptop-X1/
====================

Signed-off-by: David S. Miller <[email protected]>

selftests: forwarding: Remove duplicated lib.sh content

commit 25ae948b4478 ("selftests/net: add lib.sh") added net/lib.sh to
contain code shared by tests under net/ and net/forwarding/. However, this
caused issues with selftests from directories other than net/forwarding/,
in particular those under drivers/net/. Those issues were avoided in a
simple way by duplicating some content in commit 2114e83381d3 ("selftests:
forwarding: Avoid failures to source net/lib.sh").

In order to remove the duplicated content, restore the inclusion of
net/lib.sh from net/forwarding/lib.sh but with the following changes:

* net/lib.sh is imported through the net_forwarding_dir path
The original expression "source ../lib.sh" would look for lib.sh in the
directory above the script file's one, which did not work for tests under
drivers/net/.

* net/lib.sh is added to TEST_INCLUDES
Since net/forwarding/lib.sh now sources net/lib.sh, both of those files
must be exported along with tests which source net/forwarding/lib.sh.

Suggested-by: Hangbin Liu <[email protected]>
Reviewed-by: Hangbin Liu <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Benjamin Poirier <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

selftests: forwarding: Redefine relative_path variable

The following code which is part of lib.sh:
relative_path="${BASH_SOURCE%/*}"
if [[ "$relative_path" == "${BASH_SOURCE}" ]]; then
       relative_path="."
fi

reimplements functionality that is part of `dirname`:
$ dirname ""
.

To avoid this duplication, replace "relative_path" by "net_forwarding_dir",
a new variable defined using dirname.

Furthermore, to avoid the potential confusion about what "relative_path" is
about (cwd, test script directory or test library directory), define
"net_forwarding_dir" as the absolute path to net/forwarding/.

Tested-by: Petr Machata <[email protected]>
Reviewed-by: Hangbin Liu <[email protected]>
Signed-off-by: Benjamin Poirier <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

selftests: dsa: Replace test symlinks by wrapper script

The dsa tests which are symlinks of tests from net/forwarding/ (like
tc_actions.sh) become regular files after export (because `rsync
--copy-unsafe-links` is used) and expect to source lib.sh
(net/forwarding/lib.sh) from the same directory.

In the last patch of this series, net/forwarding/lib.sh will source lib.sh
from its parent directory (ie. net/lib.sh). This would not work for dsa
tests because net/lib.sh is not present under drivers/net/.

Since the tests in net/forwarding/ are not meant to be copied and run from
another directory, as a preparation for that last patch, replace the test
symlinks by a wrapper script which runs the original tests under
net/forwarding/. Following from that, the links to shared library scripts
in dsa/ are no longer used so remove them and add all the original files
needed from parent directories to TEST_INCLUDES.

Suggested-by: Hangbin Liu <[email protected]>
Reviewed-by: Petr Machata <[email protected]>
Tested-by: Petr Machata <[email protected]>
Reviewed-by: Hangbin Liu <[email protected]>
Signed-off-by: Benjamin Poirier <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

selftests: team: Add shared library scripts to TEST_INCLUDES

In order to avoid duplicated files when both the team and bonding tests are
exported together, add lag_lib.sh to TEST_INCLUDES.

Do likewise for net/forwarding/lib.sh regarding team and forwarding tests.

Reviewed-by: Petr Machata <[email protected]>
Tested-by: Petr Machata <[email protected]>
Reviewed-by: Hangbin Liu <[email protected]>
Signed-off-by: Benjamin Poirier <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

selftests: bonding: Add net/forwarding/lib.sh to TEST_INCLUDES

In order to avoid duplicated files when both the bonding and forwarding
tests are exported together, add net/forwarding/lib.sh to TEST_INCLUDES and
include it via its relative path.

Reviewed-by: Petr Machata <[email protected]>
Tested-by: Petr Machata <[email protected]>
Reviewed-by: Hangbin Liu <[email protected]>
Signed-off-by: Benjamin Poirier <[email protected]>
Reviewed-by: Jay Vosburgh <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

selftests: Introduce Makefile variable to list shared bash scripts

Some tests written in bash source other files in a parent directory. For
example, drivers/net/bonding/dev_addr_lists.sh sources
net/forwarding/lib.sh. If a subset of tests is exported and run outside the
source tree (for example by using `make -C tools/testing/selftests gen_tar
TARGETS="drivers/net/bonding"`), these other files must be made available
as well.

Commit ae108c48b5d2 ("selftests: net: Fix cross-tree inclusion of scripts")
addressed this problem by symlinking and copying the sourced files but this
only works for direct dependencies. Commit 25ae948b4478 ("selftests/net:
add lib.sh") changed net/forwarding/lib.sh to source net/lib.sh. As a
result, that latter file must be included as well when the former is
exported. This was not handled and was reverted in commit 2114e83381d3
("selftests: forwarding: Avoid failures to source net/lib.sh"). In order to
allow reinstating the inclusion of net/lib.sh from net/forwarding/lib.sh,
add a mechanism to list dependent files in a new Makefile variable and
export them. This allows sourcing those files using the same expression
whether tests are run in-tree or exported.

Dependencies are not resolved recursively so transitive dependencies must
be listed in TEST_INCLUDES. For example, if net/forwarding/lib.sh sources
net/lib.sh; the Makefile related to a test that sources
net/forwarding/lib.sh from a parent directory must list:
TEST_INCLUDES := \
../../../net/forwarding/lib.sh \
../../../net/lib.sh

v2:
Fix rst syntax in Documentation/dev-tools/kselftest.rst (Jakub Kicinski)

v1 (from RFC):
* changed TEST_INCLUDES to take relative paths, like other TEST_* variables
  (Vladimir Oltean)
* preserved common "$(MAKE) OUTPUT=... -C ... target" ordering in Makefile
  (Petr Machata)

Signed-off-by: Benjamin Poirier <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'stmmac-jh7100'

Cristian Ciocaltea says:

====================
StarFive DWMAC support for JH7100

This is just a subset of the initial patch series [1] adding networking
support for StarFive JH7100 SoC.

[1]: https://lore.kernel.org/lkml/20231218214451.2345691[email protected]/

Changes in v4:
 - Rebased series onto next-20240125
 - Added R-b tag from Rob in PATCH 1
 - v3:
   https://lore.kernel.org/lkml/20231222101001.2541758[email protected]/

Changes in v3:
 - Optimized jh7110 resets & reset-names properties (Rob)
 - Added R-b tag from Jacob in PATCH 1
 - v2:
   https://lore.kernel.org/lkml/20231220002824.2462655[email protected]/

Changes in v2:
 - Add the missing binding patch (Conor)
 - v1:
   https://lore.kernel.org/lkml/20231219231040.2459358[email protected]/
====================

Signed-off-by: David S. Miller <[email protected]>

net: stmmac: dwmac-starfive: Add support for JH7100 SoC

Add a missing quirk to enable support for the StarFive JH7100 SoC.

Additionally, for greater flexibility in operation, allow using the
rgmii-rxid and rgmii-txid phy modes.

Co-developed-by: Emil Renner Berthing <[email protected]>
Signed-off-by: Emil Renner Berthing <[email protected]>
Signed-off-by: Cristian Ciocaltea <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

dt-bindings: net: starfive,jh7110-dwmac: Add JH7100 SoC compatible

The Synopsys DesignWare MAC found on StarFive JH7100 SoC is mostly
similar to the newer JH7110, but it requires only two interrupts and a
single reset line, which is 'ahb' instead of the commonly used
'stmmaceth'.

Since the common binding 'snps,dwmac' allows selecting 'ahb' only in
conjunction with 'stmmaceth', extend the logic to also permit exclusive
usage of the 'ahb' reset name.  This ensures the following use cases are
supported:

  JH7110: reset-names = "stmmaceth", "ahb";
  JH7100: reset-names = "ahb";
  other:  reset-names = "stmmaceth";

Also note the need to use a different dwmac fallback, as v5.20 applies
to JH7110 only, while JH7100 relies on v3.7x.

Additionally, drop the reset description items from top-level binding as
they are already provided by the included snps,dwmac schema.

Signed-off-by: Cristian Ciocaltea <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
Reviewed-by: Krzysztof Kozlowski <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

mlxsw: remove I2C_CLASS_HWMON from drivers w/o detect and address_list

Class-based I2C probing requires detect() and address_list to be
set in the I2C client driver, see checks in i2c_detect().
It's misleading to declare I2C_CLASS_HWMON support if this
precondition isn't met.

Signed-off-by: Heiner Kallweit <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

xdp: Remove usage of the deprecated ida_simple_xx() API

ida_alloc() and ida_free() should be preferred to the deprecated
ida_simple_get() and ida_simple_remove().

Note that the upper limit of ida_simple_get() is exclusive, but the one of
ida_alloc_range() is inclusive. So a -1 has been added when needed.

Signed-off-by: Christophe JAILLET <[email protected]>
Link: https://lore.kernel.org/r/8e889d18a6c881b09db4650d4b30a62d76f4fe77.1705734073.git.christophe.jaillet@wanadoo.fr
Signed-off-by: Jakub Kicinski <[email protected]>

net: rds: Simplify the allocation of slab caches in rds_conn_init

Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
to simplify the creation of SLAB caches.

Signed-off-by: Kunwu Chan <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Merge branch 'mlxsw-refactor-reference-counting-code'

Petr Machata says:

====================
mlxsw: Refactor reference counting code

Amit Cohen writes:

This set converts all reference counters defined as 'unsigned int' to
refcount_t type. The reference counting of LAGs can be simplified, so first
refactor the related code and then change the type of the reference
counter.

Patch set overview:
Patches #1-#4 are preparations for LAG refactor
Patch #5 refactors LAG code and change the type of reference counter
Patch #6 converts the remaining reference counters in mlxsw driver
====================

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Paolo Abeni <[email protected]>

mlxsw: Use refcount_t for reference counting

mlxsw driver uses 'unsigned int' for reference counters in several
structures. Instead, use refcount_t type which allows us to catch overflow
and underflow issues. Change the type of the counters and use the
appropriate API.

Signed-off-by: Amit Cohen <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

mlxsw: spectrum: Refactor LAG create and destroy code

mlxsw_sp stores an array of LAGs. When a port joins a LAG, in case that
this LAG is already in use, we only have to increase the reference counter.
Otherwise, we have to search for an unused LAG ID and configure it in
hardware. When a port leaves a LAG, we have to destroy it only for the last
user. This code can be simplified, for such requirements we usually add
get() and put() functions which create and destroy the object.

Add mlxsw_sp_lag_{get,put}() and use them. These functions take care of
the reference counter and hardware configuration if needed. Change the
reference counter to refcount_t type which catches overflow and underflow
issues.

Signed-off-by: Amit Cohen <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

mlxsw: spectrum: Search for free LAD ID once

Currently, the function mlxsw_sp_lag_index_get() is called twice - first
as part of NETDEV_PRECHANGEUPPER event and later as part of
NETDEV_CHANGEUPPER. This function will be changed in the next patch. To
simplify the code, call it only once as part of NETDEV_CHANGEUPPER
event and set an error message using 'extack' in case of failure.

Signed-off-by: Amit Cohen <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

mlxsw: spectrum: Query max_lag once

The maximum number of LAGs is queried from core several times. It is
used to allocate LAG array, and then to iterate over it. In addition, it
is used for PGT initialization. To simplify the code, instead of
querying it several times, store the value as part of 'mlxsw_sp' and use
it.

Signed-off-by: Amit Cohen <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

mlxsw: spectrum: Remove mlxsw_sp_lag_get()

A next patch will add mlxsw_sp_lag_{get,put}() functions to handle LAG
reference counting and create/destroy it only for first user/last user.
Remove mlxsw_sp_lag_get() function and access LAG array directly.

Signed-off-by: Amit Cohen <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

mlxsw: spectrum: Change mlxsw_sp_upper to LAG structure

The structure mlxsw_sp_upper is used only as LAG. Rename it to
mlxsw_sp_lag and move it to spectrum.c file, as it is used only there.
Move the function mlxsw_sp_lag_get() with the structure.

Signed-off-by: Amit Cohen <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

selftests: forwarding: Add missing config entries

The config file contains a partial kernel configuration to be used by
`virtme-configkernel --custom'. The presumption is that the config file
contains all Kconfig options needed by the selftests from the directory.

In net/forwarding/config, many are missing, which manifests as spurious
failures when running the selftests, with messages about unknown device
types, qdisc kinds or classifier actions. Add the missing configurations.

Tested the resulting configuration using virtme-ng as follows:

 # vng -b -f tools/testing/selftests/net/forwarding/config
 # vng --user root
 (within the VM:)
 # make -C tools/testing/selftests TARGETS=net/forwarding run_tests

Signed-off-by: Petr Machata <[email protected]>
Link: https://lore.kernel.org/r/025abded7ff9cea5874a7fe35dcd3fd41bf5e6ac.1706286755.git.petrm@nvidia.com
Signed-off-by: Paolo Abeni <[email protected]>

net: micrel: Fix set/get PHC time for lan8814

When setting or getting PHC time, the higher bits of the second time (>32
bits) they were ignored. Meaning that setting some time in the future like
year 2150, it was failing to set this.

The issue can be reproduced like this:

 # phc_ctl /dev/ptp1 set 10000000000
 phc_ctl[12.290]: set clock time to 10000000000.000000000 or Sat Nov 20 17:46:40 2286

 # phc_ctl /dev/ptp1 get
 phc_ctl[15.309]: clock time is 1410065411.018055420 or Sun Sep  7 04:50:11 2014

Signed-off-by: Horatiu Vultur <[email protected]>
Reviewed-by: Maxime Chevallier <[email protected]>
Reviewed-by: Divya Koppera <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Paolo Abeni <[email protected]>

net/tun: use reciprocal_scale

Use the inline function reciprocal_scale rather than open coding
the scale optimization.  Also, remove unnecessary initializations.
Resulting compiled code is unchanged (according to godbolt).

Signed-off-by: Stephen Hemminger <[email protected]>
Reviewed-by: Willem de Bruijn <[email protected]>
Acked-by: Jason Wang <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Paolo Abeni <[email protected]>

Merge branch 'ice-fix-timestamping-in-reset-process'

Tony Nguyen says:

====================
ice: fix timestamping in reset process

Karol Kolacinski says:

PTP reset process has multiple places where timestamping can end up in
an incorrect state.

This series introduces a proper state machine for PTP and refactors
a large part of the code to ensure that timestamping does not break.

The following are changes since commit 91374ba537bd60caa9ae052c9f1c0fe055b39149:
  net: dsa: mt7530: support OF-based registration of switch MDIO bus
and are available in the git repository at:
  git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue 100GbE
====================

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Paolo Abeni <[email protected]>

ice: stop destroying and reinitalizing Tx tracker during reset

The ice driver currently attempts to destroy and re-initialize the Tx
timestamp tracker during the reset flow. The release of the Tx tracker
only happened during CORE reset or GLOBAL reset. The ice_ptp_rebuild()
function always calls the ice_ptp_init_tx function which will allocate
a new tracker data structure, resulting in memory leaks during PF reset.

Certainly the driver should not be allocating a new tracker without
removing the old tracker data, as this results in a memory leak.
Additionally, there's no reason to remove the tracker memory during a
reset. Remove this logic from the reset and rebuild flow. Instead of
releasing the Tx tracker, flush outstanding timestamps just before we
reset the PHY timestamp block in ice_ptp_cfg_phy_interrupt().

Signed-off-by: Jacob Keller <[email protected]>
Signed-off-by: Karol Kolacinski <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Tested-by: Pucha Himasekhar Reddy <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

ice: factor out ice_ptp_rebuild_owner()

The ice_ptp_reset() function uses a goto to skip past clock owner
operations if performing a PF reset or if the device is not the clock
owner. This is a bit confusing. Factor this out into
ice_ptp_rebuild_owner() instead.

The ice_ptp_reset() function is called by ice_rebuild() to restore PTP
functionality after a device reset. Follow the convention set by the
ice_main.c file and rename this function to ice_ptp_rebuild(), in the
same way that we have ice_prepare_for_reset() and
ice_ptp_prepare_for_reset().

Signed-off-by: Jacob Keller <[email protected]>
Signed-off-by: Karol Kolacinski <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Tested-by: Pucha Himasekhar Reddy <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

ice: rename ice_ptp_tx_cfg_intr

The ice_ptp_tx_cfg_intr() function sends a control queue message to
configure the PHY timestamp interrupt block. This is a very similar name
to a function which is used to configure the MAC Other Interrupt Cause
Enable register.

Rename this function to ice_ptp_cfg_phy_interrupt in order to make it
more obvious to the reader what action it performs, and distinguish it
from other similarly named functions.

Signed-off-by: Jacob Keller <[email protected]>
Signed-off-by: Karol Kolacinski <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Tested-by: Pucha Himasekhar Reddy <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

ice: don't check has_ready_bitmap in E810 functions

E810 hardware does not have a Tx timestamp ready bitmap. Don't check
has_ready_bitmap in E810-specific functions.
Add has_ready_bitmap check in ice_ptp_process_tx_tstamp() to stop
relying on the fact that ice_get_phy_tx_tstamp_ready() returns all 1s.

Signed-off-by: Jacob Keller <[email protected]>
Signed-off-by: Karol Kolacinski <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Tested-by: Pucha Himasekhar Reddy <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

ice: rename verify_cached to has_ready_bitmap

The tx->verify_cached flag is used to inform the Tx timestamp tracking
code whether it needs to verify the cached Tx timestamp value against
a previous captured value. This is necessary on E810 hardware which does
not have a Tx timestamp ready bitmap.

In addition, we currently rely on the fact that the
ice_get_phy_tx_tstamp_ready() function returns all 1s for E810 hardware.
Instead of introducing a brand new flag, rename and verify_cached to
has_ready_bitmap, inverting the relevant checks.

Signed-off-by: Jacob Keller <[email protected]>
Signed-off-by: Karol Kolacinski <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Tested-by: Pucha Himasekhar Reddy <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

ice: pass reset type to PTP reset functions

The ice_ptp_prepare_for_reset() and ice_ptp_reset() functions currently
check the pf->flags ICE_FLAG_PFR_REQ bit to determine if the current
reset is a PF reset or not.

This is problematic, because it is possible that a PF reset and a higher
level reset (CORE reset, GLOBAL reset, EMP reset) are requested
simultaneously. In that case, the driver performs the highest level
reset requested. However, the ICE_FLAG_PFR_REQ flag will still be set.

The main driver reset functions take an enum ice_reset_req indicating
which reset is actually being performed. Pass this data into the PTP
functions and rely on this instead of relying on the driver flags.

This ensures that the PTP code performs the proper level of reset that
the driver is actually undergoing.

Signed-off-by: Jacob Keller <[email protected]>
Signed-off-by: Karol Kolacinski <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Tested-by: Pucha Himasekhar Reddy <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

ice: introduce PTP state machine

Add PTP state machine so that the driver can correctly identify PTP
state around resets.
When the driver got information about ungraceful reset, PTP was not
prepared for reset and it returned error. When this situation occurs,
prepare PTP before rebuilding its structures.

Signed-off-by: Jacob Keller <[email protected]>
Co-developed-by: Karol Kolacinski <[email protected]>
Signed-off-by: Karol Kolacinski <[email protected]>
Reviewed-by: Jacob Keller <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Tested-by: Pucha Himasekhar Reddy <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

net: tcp: accept old ack during closing

For now, the packet with an old ack is not accepted if we are in
FIN_WAIT1 state, which can cause retransmission. Taking the following
case as an example:

    Client                               Server
      |                                    |
  FIN_WAIT1(Send FIN, seq=10)          FIN_WAIT1(Send FIN, seq=20, ack=10)
      |                                    |
      |                                Send ACK(seq=21, ack=11)
   Recv ACK(seq=21, ack=11)
      |
   Recv FIN(seq=20, ack=10)

In the case above, simultaneous close is happening, and the FIN and ACK
packet that send from the server is out of order. Then, the FIN will be
dropped by the client, as it has an old ack. Then, the server has to
retransmit the FIN, which can cause delay if the server has set the
SO_LINGER on the socket.

Old ack is accepted in the ESTABLISHED and TIME_WAIT state, and I think
it should be better to keep the same logic.

In this commit, we accept old ack in FIN_WAIT1/FIN_WAIT2/CLOSING/LAST_ACK
states. Maybe we should limit it to FIN_WAIT1 for now?

Signed-off-by: Menglong Dong <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Reviewed-by: Eric Dumazet <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Merge branch 'mt7530-dsa-subdriver-improvements-act-i'

Arınç ÜNAL says:

====================
MT7530 DSA Subdriver Improvements Act I

This patch series simplifies the MT7530 DSA subdriver and improves the
logic of the support for MT7530, MT7531, and the switch on the MT7988 SoC.

I have done a simple ping test to confirm basic communication on all switch
ports on MCM and standalone MT7530, and MT7531 switch with this patch
series applied.

MT7621 Unielec, MCM MT7530:

rgmii-only-gmac0-mt7621-unielec-u7621-06-16m.dtb
gmac0-and-gmac1-mt7621-unielec-u7621-06-16m.dtb

tftpboot 0x80008000 mips-uzImage.bin; tftpboot 0x83000000 mips-rootfs.cpio.uboot; tftpboot 0x83f00000 $dtb; bootm 0x80008000 0x83000000 0x83f00000

MT7622 Bananapi, MT7531:

gmac0-and-gmac1-mt7622-bananapi-bpi-r64.dtb

tftpboot 0x40000000 arm64-Image; tftpboot 0x45000000 arm64-rootfs.cpio.uboot; tftpboot 0x4a000000 $dtb; booti 0x40000000 0x45000000 0x4a000000

MT7623 Bananapi, standalone MT7530:

rgmii-only-gmac0-mt7623n-bananapi-bpi-r2.dtb
gmac0-and-gmac1-mt7623n-bananapi-bpi-r2.dtb

tftpboot 0x80008000 arm-zImage; tftpboot 0x83000000 arm-rootfs.cpio.uboot; tftpboot 0x83f00000 $dtb; bootz 0x80008000 0x83000000 0x83f00000

This patch series is the continuation of the patch series linked below.

https://lore.kernel.org/r/20230522121532[email protected]

v2: https://lore.kernel.org/r/20231227044347[email protected]
v1: https://lore.kernel.org/r/20231118123205[email protected]

Signed-off-by: Arınç ÜNAL <[email protected]>
====================

Link: https://lore.kernel.org/r/20240122-for-netnext-mt7530-improvements-1-v3-0-042401f2b279@arinc9.com
Signed-off-by: Jakub Kicinski <[email protected]>

net: dsa: mt7530: do not run mt7530_setup_port5() if port 5 is disabled

There's no need to run all the code on mt7530_setup_port5() if port 5 is
disabled. The only case for calling mt7530_setup_port5() from
mt7530_setup() is when PHY muxing is enabled. That is because port 5 is not
defined as a port on the devicetree, therefore, it cannot be controlled by
phylink.

Because of this, run mt7530_setup_port5() if priv->p5_intf_sel is
P5_INTF_SEL_PHY_P0 or P5_INTF_SEL_PHY_P4. Remove the P5_DISABLED case from
mt7530_setup_port5().

Stop initialising the interface variable as the remaining cases will always
call mt7530_setup_port5() with it initialised.

Signed-off-by: Arınç ÜNAL <[email protected]>
Reviewed-by: Vladimir Oltean <[email protected]>
Link: https://lore.kernel.org/r/20240122-for-netnext-mt7530-improvements-1-v3-7-042401f2b279@arinc9.com
Signed-off-by: Jakub Kicinski <[email protected]>

net: dsa: mt7530: do not set priv->p5_interface on mt7530_setup_port5()

Running mt7530_setup_port5() from mt7530_setup() used to handle all cases
of configuring port 5, including phylink.

Setting priv->p5_interface under mt7530_setup_port5() makes sure that
mt7530_setup_port5() from mt753x_phylink_mac_config() won't run.

The commit ("net: dsa: mt7530: improve code path for setting up port 5")
makes so that mt7530_setup_port5() from mt7530_setup() runs only on
non-phylink cases.

Get rid of unnecessarily setting priv->p5_interface under
mt7530_setup_port5() as port 5 phylink configuration will be done by
running mt7530_setup_port5() from mt753x_phylink_mac_config() now.

Signed-off-by: Arınç ÜNAL <[email protected]>
Reviewed-by: Vladimir Oltean <[email protected]>
Link: https://lore.kernel.org/r/20240122-for-netnext-mt7530-improvements-1-v3-6-042401f2b279@arinc9.com
Signed-off-by: Jakub Kicinski <[email protected]>

net: dsa: mt7530: improve code path for setting up port 5

There're two code paths for setting up port 5:

mt7530_setup()
-> mt7530_setup_port5()

mt753x_phylink_mac_config()
-> mt753x_mac_config()
   -> mt7530_mac_config()
      -> mt7530_setup_port5()

Currently mt7530_setup_port5() from mt7530_setup() always runs. If port 5
is used as a CPU, DSA, or user port, mt7530_setup_port5() from
mt753x_phylink_mac_config() won't run. That is because priv->p5_interface
set on mt7530_setup_port5() will match state->interface on
mt753x_phylink_mac_config() which will stop running mt7530_setup_port5()
again.

Therefore, mt7530_setup_port5() will never run from
mt753x_phylink_mac_config().

Address this by not running mt7530_setup_port5() from mt7530_setup() if
port 5 is used as a CPU, DSA, or user port. This driver isn't in the
dsa_switches_apply_workarounds[] array so phylink will always be present.

To keep the cases where port 5 isn't controlled by phylink working as
before, preserve the mt7530_setup_port5() call from mt7530_setup().

Do not set priv->p5_intf_sel to P5_DISABLED. It is already set to that when
"priv" is allocated.

Move setting the interface to a more specific location. It's supposed to be
overwritten if PHY muxing is detected.

Improve the comment which explains the process.

Signed-off-by: Arınç ÜNAL <[email protected]>
Reviewed-by: Vladimir Oltean <[email protected]>
Link: https://lore.kernel.org/r/20240122-for-netnext-mt7530-improvements-1-v3-5-042401f2b279@arinc9.com
Signed-off-by: Jakub Kicinski <[email protected]>

net: dsa: mt7530: improve comments regarding switch ports

There's no logic to numerically order the CPU ports. Just state the port
number instead.

Remove the irrelevant PHY muxing information from
mt7530_mac_port_get_caps(). Explain the supported MII modes instead.

Remove the out of place PHY muxing information from
mt753x_phylink_mac_config(). The function is for MT7530, MT7531, and the
switch on the MT7988 SoC but there's no PHY muxing on MT7531 or the switch
on the MT7988 SoC.

These comments were gradually introduced with the commits below.
commit ca366d6c889b ("net: dsa: mt7530: Convert to PHYLINK API")
commit 38f790a80560 ("net: dsa: mt7530: Add support for port 5")
commit 88bdef8be9f6 ("net: dsa: mt7530: Extend device data ready for adding
a new hardware")
commit c288575f7810 ("net: dsa: mt7530: Add the support of MT7531 switch")

Signed-off-by: Arınç ÜNAL <[email protected]>
Acked-by: Daniel Golle <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Reviewed-by: Vladimir Oltean <[email protected]>
Link: https://lore.kernel.org/r/20240122-for-netnext-mt7530-improvements-1-v3-4-042401f2b279@arinc9.com
Signed-off-by: Jakub Kicinski <[email protected]>

net: dsa: mt7530: store port 5 SGMII capability of MT7531

Introduce the p5_sgmii field to store the information for whether port 5
has got SGMII or not. Instead of reading the MT7531_TOP_SIG_SR register
multiple times, the register will be read once and the value will be
stored on the p5_sgmii field. This saves unnecessary reads of the
register.

Move the comment about MT7531AE and MT7531BE to mt7531_setup(), where the
switch is identified.

Get rid of mt7531_dual_sgmii_supported() now that priv->p5_sgmii stores the
information. Address the code where mt7531_dual_sgmii_supported() is used.

Get rid of mt7531_is_rgmii_port() which just prints the opposite of
priv->p5_sgmii.

Instead of calling mt7531_pll_setup() then returning, do not call it if
port 5 is SGMII.

Remove P5_INTF_SEL_GMAC5_SGMII. The p5_interface_select enum is supposed to
represent the mode that port 5 is being used in, not the hardware
information of port 5. Set p5_intf_sel to P5_INTF_SEL_GMAC5 instead, if
port 5 is not dsa_is_unused_port().

Signed-off-by: Arınç ÜNAL <[email protected]>
Acked-by: Daniel Golle <[email protected]>
Reviewed-by: Vladimir Oltean <[email protected]>
Link: https://lore.kernel.org/r/20240122-for-netnext-mt7530-improvements-1-v3-3-042401f2b279@arinc9.com
Signed-off-by: Jakub Kicinski <[email protected]>

net: dsa: mt7530: use p5_interface_select as data type for p5_intf_sel

Use the p5_interface_select enumeration as the data type for the
p5_intf_sel field. This ensures p5_intf_sel can only take the values
defined in the p5_interface_select enumeration.

Remove the explicit assignment of 0 to P5_DISABLED as the first enum item
is automatically assigned 0.

Signed-off-by: Arınç ÜNAL <[email protected]>
Acked-by: Daniel Golle <[email protected]>
Reviewed-by: Vladimir Oltean <[email protected]>
Link: https://lore.kernel.org/r/20240122-for-netnext-mt7530-improvements-1-v3-2-042401f2b279@arinc9.com
Signed-off-by: Jakub Kicinski <[email protected]>

net: dsa: mt7530: always trap frames to active CPU port on MT7530

On the MT7530 switch, the CPU_PORT field indicates which CPU port to trap
frames to, regardless of the affinity of the inbound user port.

When multiple CPU ports are in use, if the DSA conduit interface is down,
trapped frames won't be passed to the conduit interface.

To make trapping frames work including this case, implement
ds->ops->conduit_state_change() on this subdriver and set the CPU_PORT
field to the numerically smallest CPU port whose conduit interface is up.
Introduce the active_cpu_ports field to store the information of the active
CPU ports. Correct the macros, CPU_PORT is bits 4 through 6 of the
register.

Add a comment to explain frame trapping for this switch.

Currently, the driver doesn't support the use of multiple CPU ports so this
is not necessarily a bug fix.

Suggested-by: Vladimir Oltean <[email protected]>
Suggested-by: Russell King (Oracle) <[email protected]>
Signed-off-by: Arınç ÜNAL <[email protected]>
Reviewed-by: Vladimir Oltean <[email protected]>
Link: https://lore.kernel.org/r/20240122-for-netnext-mt7530-improvements-1-v3-1-042401f2b279@arinc9.com
Signed-off-by: Jakub Kicinski <[email protected]>

netfilter: ebtables: allow xtables-nft only builds

Same patch as previous one, but for ebtables.

To build a kernel that only supports ebtables-nft, the builtin tables
need to be disabled, i.e.:

CONFIG_BRIDGE_EBT_BROUTE=n
CONFIG_BRIDGE_EBT_T_FILTER=n
CONFIG_BRIDGE_EBT_T_NAT=n

The ebtables specific extensions can then be used nftables'
NFT_COMPAT interface.

Signed-off-by: Florian Westphal <[email protected]>

netfilter: xtables: allow xtables-nft only builds

Add hidden IP(6)_NF_IPTABLES_LEGACY symbol.

When any of the "old" builtin tables are enabled the "old" iptables
interface will be supported.

To disable the old set/getsockopt interface the existing options
for the builtin tables need to be turned off:

CONFIG_IP_NF_IPTABLES=m
CONFIG_IP_NF_FILTER is not set
CONFIG_IP_NF_NAT is not set
CONFIG_IP_NF_MANGLE is not set
CONFIG_IP_NF_RAW is not set
CONFIG_IP_NF_SECURITY is not set

Same for CONFIG_IP6_NF_ variants.

This allows to build a kernel that only supports ip(6)tables-nft
(iptables-over-nftables api).

In the future the _LEGACY symbol will become visible and the select
statements will be turned into 'depends on', but for now be on safe side
so "make oldconfig" won't break things.

Signed-off-by: Florian Westphal <[email protected]>

netfilter: arptables: allow xtables-nft only builds

Allows to build kernel that supports the arptables mangle target
via nftables' compat infra but without the arptables get/setsockopt
interface or the old arptables filter interpreter.

IOW, setting IP_NF_ARPFILTER=n will break arptables-legacy, but
arptables-nft will continue to work as long as nftables compat
support is enabled.

Signed-off-by: Florian Westphal <[email protected]>
Reviewed-by: Phil Sutter <[email protected]>

ipvs: Simplify the allocation of ip_vs_conn slab caches

Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
to simplify the creation of SLAB caches.

Signed-off-by: Kunwu Chan <[email protected]>
Acked-by: Simon Horman <[email protected]>
Signed-off-by: Florian Westphal <[email protected]>

netfilter: nf_conncount: Use KMEM_CACHE instead of kmem_cache_create()

Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
to simplify the creation of SLAB caches.

Signed-off-by: Kunwu Chan <[email protected]>
Signed-off-by: Florian Westphal <[email protected]>

netfilter: nf_tables: pass flags to set backend selection routine

No need to refetch the flag from the netlink attribute, pass the
existing flags variable which already provide validated flags.

Signed-off-by: Pablo Neira Ayuso <[email protected]>
Signed-off-by: Florian Westphal <[email protected]>

netfilter: nf_tables: Implement table adoption support

Allow a new process to take ownership of a previously owned table,
useful mostly for firewall management services restarting or suspending
when idle.

By extending __NFT_TABLE_F_UPDATE, the on/off/on check in
nf_tables_updtable() also covers table adoption, although it is actually
not needed: Table adoption is irreversible because nf_tables_updtable()
rejects attempts to drop NFT_TABLE_F_OWNER so table->nlpid setting can
happen just once within the transaction.

If the transaction commences, table's nlpid and flags fields are already
set and no further action is required. If it aborts, the table returns
to orphaned state.

Signed-off-by: Phil Sutter <[email protected]>
Signed-off-by: Florian Westphal <[email protected]>

netfilter: nf_tables: Introduce NFT_TABLE_F_PERSIST

This companion flag to NFT_TABLE_F_OWNER requests the kernel to keep the
table around after the process has exited. It marks such table as
orphaned (by dropping OWNER flag but keeping PERSIST flag in place),
which opens it for other processes to manipulate. For the sake of
simplicity, PERSIST flag may not be altered though.

Signed-off-by: Phil Sutter <[email protected]>
Signed-off-by: Florian Westphal <[email protected]>

netfilter: uapi: Document NFT_TABLE_F_OWNER flag

Add at least this one-liner describing the obvious.

Fixes: 6001a930ce03 ("netfilter: nftables: introduce table ownership")
Signed-off-by: Phil Sutter <[email protected]>
Signed-off-by: Florian Westphal <[email protected]>

net: free altname using an RCU callback

We had to add another synchronize_rcu() in recent fix.
Bite the bullet and add an rcu_head to netdev_name_node,
free from RCU.

Note that name_node does not hold any reference on dev
to which it points, but there must be a synchronize_rcu()
on device removal path, so we should be fine.

Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Reviewed-by: Eric Dumazet <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

dt-bindings: nfc: ti,trf7970a: fix usage example

The TRF7970A is a SPI device, not I2C.

Signed-off-by: Tobias Schramm <[email protected]>
Reviewed-by: Krzysztof Kozlowski <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

ptp: add FemtoClock3 Wireless as ptp hardware clock

The RENESAS FemtoClock3 Wireless is a high-performance jitter attenuator,
frequency translator, and clock synthesizer. The device is comprised of 3
digital PLLs (DPLL) to track CLKIN inputs and three independent low phase
noise fractional output dividers (FOD) that output low phase noise clocks.

FemtoClock3 supports one Time Synchronization (Time Sync) channel to enable
an external processor to control the phase and frequency of the Time Sync
channel and to take phase measurements using the TDC. Intended applications
are synchronization using the precision time protocol (PTP) and
synchronization with 0.5 Hz and 1 Hz signals from GNSS.

Signed-off-by: Min Li <[email protected]>
Acked-by: Lee Jones <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

ptp: introduce PTP_CLOCK_EXTOFF event for the measured external offset

This change is for the PHC devices that can measure the phase offset
between PHC signal and the external signal, such as the 1PPS signal of
GNSS. Reporting PTP_CLOCK_EXTOFF to user space will be piggy-backed to
the existing ptp_extts_event so that application such as ts2phc can
poll the external offset the same way as extts. Hence, ts2phc can use
the offset to achieve the alignment between PHC and the external signal
by the help of either SW or HW filters.

Signed-off-by: Min Li <[email protected]>
Acked-by: Richard Cochran <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'net-module-description'

Breno Leitao says:

====================
Fix MODULE_DESCRIPTION() for net (p3)

There are hundreds of network modules that misses MODULE_DESCRIPTION(),
causing a warning when compiling with W=1. Example:

        WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/net/arcnet/com90io.o
        WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/net/arcnet/arc-rimi.o
        WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/net/arcnet/com20020.o

This part3 of the patchset focus on the missing ethernet drivers, which
is now warning free. This also fixes net/pcs and ieee802154.
====================

Signed-off-by: David S. Miller <[email protected]>