platform/kernel/linux-starfive.git
3 years agonet: phy: Add 25G BASE-R interface mode
Steen Hegelund [Fri, 11 Jun 2021 12:54:51 +0000 (14:54 +0200)]
net: phy: Add 25G BASE-R interface mode

Add 25gbase-r phy interface mode

Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: Bjarni Jonasson <bjarni.jonasson@microchip.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodt-bindings: net: Add 25G BASE-R phy interface
Steen Hegelund [Fri, 11 Jun 2021 12:54:50 +0000 (14:54 +0200)]
dt-bindings: net: Add 25G BASE-R phy interface

Add 25gbase-r PHY interface mode.

Signed-off-by: Steen Hegelund <steen.hegelund@microchip.com>
Signed-off-by: Bjarni Jonasson <bjarni.jonasson@microchip.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 's390-net-updates'
David S. Miller [Sat, 12 Jun 2021 20:06:34 +0000 (13:06 -0700)]
Merge branch 's390-net-updates'

Julian Wiedmann says:

====================
s390/iucv: updates 2021-06-11

please apply the following iucv patches to netdev's net-next tree.

This cleans up a pattern of forward declarations in two iucv drivers,
so that they stop causing compile warnings with gcc11.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/netiuvc: get rid of forward declarations
Heiko Carstens [Fri, 11 Jun 2021 07:45:02 +0000 (09:45 +0200)]
s390/netiuvc: get rid of forward declarations

Move netiucv_handler to get rid of forward declarations and gcc11
compile warnings:

drivers/s390/net/netiucv.c:518:65: warning: argument 2 of type ‘u8[16]’ {aka ‘unsigned char[16]’} with mismatched bound [-Warray-parameter=]
  518 | static void netiucv_callback_connack(struct iucv_path *path, u8 ipuser[16])
      |                                                              ~~~^~~~~~~~~~
drivers/s390/net/netiucv.c:122:58: note: previously declared as ‘u8 *’ {aka ‘unsigned char *’}
  122 | static void netiucv_callback_connack(struct iucv_path *, u8 *);

Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/af_iucv: clean up some forward declarations
Julian Wiedmann [Fri, 11 Jun 2021 07:45:01 +0000 (09:45 +0200)]
net/af_iucv: clean up some forward declarations

The forward declarations for the iucv_handler callbacks are causing
various compile warnings with gcc-11. Reshuffle the code to get rid
of these prototypes.

Reported-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'ipa-sysfs'
David S. Miller [Fri, 11 Jun 2021 21:13:18 +0000 (14:13 -0700)]
Merge branch 'ipa-sysfs'

Alex Elder says:

====================
net: ipa: introduce ipa_syfs.c

This series (its last patch, actually) creates a new source file,
"ipa_syfs.c", to contain functions and data that expose to user
space information known by the IPA driver via device attributes.

The directory containing these files on supported systems is:
    /sys/devices/platform/soc@0/1e40000.ipa

And within that direcftory, the following files and directories
are added:
    .
    |-- feature
    |   |-- rx_offload          Type of checksum offload supported
    |   `-- tx_offload
    |   . . .
    |-- modem
    |   |-- rx_endpoint_id      IPA endpoint IDs for the embedded modem
    |   `-- tx_endpoint_id
    |   . . .
    |-- version                 IPA hardware version (informational)
        . . .

The first patch just makes endpoint validation unconditional, as
suggested by Leon Romanovsky.  The second just ensures the version
defined in configuration data is valid, so the version attribute
doesn't have to handle unrecognized version numbers.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipa: introduce sysfs code
Alex Elder [Fri, 11 Jun 2021 20:39:40 +0000 (15:39 -0500)]
net: ipa: introduce sysfs code

Add IPA device attributes to expose information known by the IPA
driver about the hardware and its configuration.

All pointers used to display these attribute values (i.e., IPA
pointer and endpoint pointers) will have been initialized by the
time IPA probe has completed, so they may be safely dereferenced.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipa: introduce ipa_version_valid()
Alex Elder [Fri, 11 Jun 2021 20:39:39 +0000 (15:39 -0500)]
net: ipa: introduce ipa_version_valid()

Define and use a new function that just validates the version
defined in configuration data.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipa: make endpoint data validation unconditional
Alex Elder [Fri, 11 Jun 2021 20:39:38 +0000 (15:39 -0500)]
net: ipa: make endpoint data validation unconditional

The cost of validating the endpoint configuration data is not all
that high, so just do it unconditionally, rather than doing so only
when IPA_VALIDATAION is defined.

Suggested-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoibmvnic: fix kernel build warning in strncpy
Lijun Pan [Fri, 11 Jun 2021 18:33:53 +0000 (13:33 -0500)]
ibmvnic: fix kernel build warning in strncpy

drivers/net/ethernet/ibm/ibmvnic.c: In function ‘handle_vpd_rsp’:
drivers/net/ethernet/ibm/ibmvnic.c:4393:3: warning: ‘strncpy’ output truncated before terminating nul copying 3 bytes from a string of the same length [-Wstringop-truncation]
 4393 |   strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
      |   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Signed-off-by: Lijun Pan <lijunp213@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'sja1105-xpcs'
David S. Miller [Fri, 11 Jun 2021 20:43:56 +0000 (13:43 -0700)]
Merge branch 'sja1105-xpcs'

Vladimir Oltean says:

====================
Port the SJA1105 DSA driver to XPCS

As requested when adding support for the NXP SJA1110, the SJA1105 driver
could make use of the common XPCS driver, to eliminate some hardware
specific code duplication.

This series modifies the XPCS driver so that it can accommodate the XPCS
instantiation from NXP switches, and the SJA1105 driver so it can expose
what the XPCS driver expects.

Tested on NXP SJA1105S and SJA1110A.

Changes in v3:
None. This is a resend of v2 which had "changes requested" even though
there was no direct feedback.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: plug in support for 2500base-x
Vladimir Oltean [Fri, 11 Jun 2021 20:05:31 +0000 (23:05 +0300)]
net: dsa: sja1105: plug in support for 2500base-x

The MAC treats 2500base-x same as SGMII (yay for that) except that it
must be set to a different speed.

Extend all places that check for SGMII to also check for 2500base-x.

Also add the missing 2500base-x compatibility matrix entry for SJA1110D.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: SGMII and 2500base-x on the SJA1110 are 'special'
Vladimir Oltean [Fri, 11 Jun 2021 20:05:30 +0000 (23:05 +0300)]
net: dsa: sja1105: SGMII and 2500base-x on the SJA1110 are 'special'

For the xMII Mode Parameters Table to be properly configured for SGMII
mode on SJA1110, we need to set the "special" bit, since SGMII is
officially bitwise coded as 0b0011 in SJA1105 (decimal 3, equal to
XMII_MODE_SGMII), and as 0b1011 in SJA1110 (decimal 11).

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: register the PCS MDIO bus for SJA1110
Vladimir Oltean [Fri, 11 Jun 2021 20:05:29 +0000 (23:05 +0300)]
net: dsa: sja1105: register the PCS MDIO bus for SJA1110

On the SJA1110, the PCS of each SERDES-capable port is accessed through
a different memory window which is 0x100 bytes in size, denoted by
"pcs_base".

In each PCS register access window, the XPCS MMDs are accessed in an
indirect way: in pages/banks of up to 0x100 addresses each. Changing the
page/bank is done by writing to a special register at the end of the
access window.

The MDIO register map accessed indirectly through the indirect banked
method described above is similar to what SJA1105 has: upper 5 bits are
the MMD, lower 16 bits are the MDIO address within that MMD.

Since the PHY ID reported by the XPCS inside SJA1110 is also all zeroes
(like SJA1105), we need to trap those reads and return a fake PHY ID so
that the xpcs driver can apply some specific fixups for our integration.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: sja1105: migrate to xpcs for SGMII
Vladimir Oltean [Fri, 11 Jun 2021 20:05:28 +0000 (23:05 +0300)]
net: dsa: sja1105: migrate to xpcs for SGMII

There is a desire to use the generic driver for the Synopsys XPCS
located in drivers/net/pcs, and to achieve that, the sja1105 driver must
expose an MDIO bus for the SGMII PCS, because the XPCS probes as an
mdio_device.

In preparation of the SJA1110 which in fact has a different access
procedure for the SJA1105, we register this PCS MDIO bus once in the
common code, but we implement function pointers for the read and write
methods. In this patch there is a single implementation for them.

There is exactly one MDIO bus for the PCS, this will contain all PCSes
at MDIO addresses equal to the port number.

We delete a bunch of hardware support code because the xpcs driver
already does what we need.

We need to hack up the MDIO reads for the PHY ID, since our XPCS
instantiation returns zeroes and there are some specific fixups which
need to be applied by the xpcs driver.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pcs: xpcs: export xpcs_do_config and xpcs_link_up
Vladimir Oltean [Fri, 11 Jun 2021 20:05:27 +0000 (23:05 +0300)]
net: pcs: xpcs: export xpcs_do_config and xpcs_link_up

The sja1105 hardware has a quirk in that some changes require a switch
reset, which loses all configuration. When the reset is initiated,
everything needs to be reprogrammed, including the MACs and the PCS.
This is currently done in sja1105_static_config_reload() - we manually
call sja1105_adjust_port_config(), sja1105_sgmii_pcs_config() and
sja1105_sgmii_pcs_force_speed() which are all internal functions.

There is a desire for sja1105 to use the common xpcs driver, and that
means that the equivalents of those functions, xpcs_do_config() and
xpcs_link_up() respectively, will no longer be local functions.

Forcing phylink to retrigger a resolve somehow, say by doing dev_close()
followed by dev_open() is not really an option, because the CPU port
might have a PCS as well, and there is no net device which we can close
and reopen for that. Additionally, the dev_close/dev_open sequence might
force a renegotiation of the copper-side link for SGMII ports connected
to a PHY, and this is undesirable as well, because the switch reset is
much quicker than a PHY autoneg, so we would have a lot more downtime.

The only solution I see is for the sja1105 driver to keep doing what
it's doing, and that means we need to export the equivalents from xpcs
for sja1105_sgmii_pcs_config and sja1105_sgmii_pcs_force_speed, and call
them directly in sja1105_static_config_reload(). This will be done
during the conversion patch.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pcs: xpcs: add support for NXP SJA1110
Vladimir Oltean [Fri, 11 Jun 2021 20:05:26 +0000 (23:05 +0300)]
net: pcs: xpcs: add support for NXP SJA1110

The NXP SJA1110 switch integrates its own, non-Synopsys PMA, but it
manages it through the register space of the XPCS itself, in a small
register window inside MDIO_MMD_VEND2 from address 0x8030 to 0x806e.

This coincides with where the registers for the default Synopsys PMA
are, but the register definitions are of course not the same.

This situation is an odd hardware quirk, but the simplest way to manage
it is to drive the SJA1110's PMA from within the XPCS driver.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pcs: xpcs: add support for NXP SJA1105
Vladimir Oltean [Fri, 11 Jun 2021 20:05:25 +0000 (23:05 +0300)]
net: pcs: xpcs: add support for NXP SJA1105

The NXP SJA1105 DSA switch integrates a Synopsys SGMII XPCS on port 4.
The generic code works fine, except there is an integration issue which
needs to be dealt with: in this switch, the XPCS is integrated with a
PMA that has the TX lane polarity inverted by default (PLUS is MINUS,
MINUS is PLUS).

To obtain normal non-inverted behavior, the TX lane polarity must be
inverted in the PCS, via the DIGITAL_CONTROL_2 register.

We introduce a pma_config() method in xpcs_compat which is called by the
phylink_pcs_config() implementation.

Also, the NXP SJA1105 returns all zeroes in the PHY ID registers 2 and 3.
We need to hack up an ad-hoc PHY ID (OUI is zero, device ID is 1) in
order for the XPCS driver to recognize it. This PHY ID is added to the
public include/linux/pcs/pcs-xpcs.h for that reason (for the sja1105
driver to be able to use it in a later patch).

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pcs: xpcs: also ignore phy id if it's all ones
Vladimir Oltean [Fri, 11 Jun 2021 20:05:24 +0000 (23:05 +0300)]
net: pcs: xpcs: also ignore phy id if it's all ones

xpcs_get_id() searches multiple MMDs for a known PHY ID, starting with
MDIO_MMD_PCS (3). However not all integrators might have implemented
that MMD on their MDIO bus. For example, the NXP SJA1105 and SJA1110
switches only implement vendor-specific MMD 1 and 2.

When there is nothing on an MDIO bus at a certain address, traditionally
the bus returns 0xffff, which means that the bus remained in its default
pull-up state for the duration of the MDIO transaction. The 0xffff value
is widely used in drivers/net/phy/phy_device.c (see get_phy_c22_id for
example) to denote a missing device.

So it makes sense for the xpcs to ignore this value as well, and
continue its search, eventually finding the proper PHY ID in the
vendor-specific MMDs.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pcs: xpcs: add support for sgmii with no inband AN
Vladimir Oltean [Fri, 11 Jun 2021 20:05:23 +0000 (23:05 +0300)]
net: pcs: xpcs: add support for sgmii with no inband AN

In fixed-link use cases, the XPCS can disable the clause 37 in-band
autoneg process, disable the "Automatic Speed Mode Change after CL37 AN"
setting, and force operation in a speed dictated by management.

Add support for this operating mode.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pcs: xpcs: move register bit descriptions to a header file
Vladimir Oltean [Fri, 11 Jun 2021 20:05:22 +0000 (23:05 +0300)]
net: pcs: xpcs: move register bit descriptions to a header file

Vendors which integrate the Designware XPCS might modify a few things
here and there, and to support those, it's best to create separate C
files in order to not clutter up the main pcs-xpcs.c.

Because the vendor files might want to access the common xpcs registers
too, let's move them in a header file which is local to this driver and
can be included by vendor files as appropriate.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: stmmac: reduce indentation when calling stmmac_xpcs_setup
Vladimir Oltean [Fri, 11 Jun 2021 20:05:21 +0000 (23:05 +0300)]
net: stmmac: reduce indentation when calling stmmac_xpcs_setup

There is no reason to embed an if within an if, we can just logically
AND the two conditions.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Wong Vee Khee <vee.khee.wong@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: stmmac: reverse Christmas tree notation in stmmac_xpcs_setup
Vladimir Oltean [Fri, 11 Jun 2021 20:05:20 +0000 (23:05 +0300)]
net: stmmac: reverse Christmas tree notation in stmmac_xpcs_setup

Reorder the variable declarations in descending line length order,
according to the networking coding style.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Wong Vee Khee <vee.khee.wong@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pcs: xpcs: rename mdio_xpcs_args to dw_xpcs
Vladimir Oltean [Fri, 11 Jun 2021 20:05:19 +0000 (23:05 +0300)]
net: pcs: xpcs: rename mdio_xpcs_args to dw_xpcs

The struct mdio_xpcs_args is reminiscent of when a similarly named
struct mdio_xpcs_ops existed. Now that that is removed, we can shorten
the name to dw_xpcs (dw for DesignWare).

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'rmnet-checksums-part-1'
David S. Miller [Fri, 11 Jun 2021 20:37:50 +0000 (13:37 -0700)]
Merge branch 'rmnet-checksums-part-1'

Alex Elder says:

====================
net: qualcomm: rmnet: MAPv4 download checksum cleanup, part 1

I'm posting a large series an two smaller parts; this is part 1.

The RMNet driver handles MAP (or QMAP) protocol traffic.  There are
several versions of this protocol.  Version 1 supports multiplexing,
as well as aggregation of packets in a single buffer.  Version 4
adds the ability to perform checksum offload.  And version 5
implements checksum offload in a different way from version 4.

This series involves only MAPv4 protocol checksum offload, and only
in the download (RX) direction.  It affects handling of checksums
computed by hardware for UDP datagrams and TCP segments, carried
over both IPv4 and IPv6.

MAP packets arriving on an RMNet port implementing MAPv4 checksum
offload are passed to rmnet_map_checksum_downlink_packet() for
handling.

The packet is then passed to rmnet_map_ipv4_dl_csum_trailer() or
rmnet_map_ipv6_dl_csum_trailer(), depending contents of the MAP
payload.  These two functions interpret checksum metadata to
determine whether the checksum in the received packet matches that
calculated by the hardware.

It is these two functions that are the subject of this series (parts
1 and 2).  The bulk of these functions are transformed--in a lot of
small steps--from an extremely difficult-to-follow block of checksum
processing code into a fairly simple, heavily commented equivalent.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qualcomm: rmnet: avoid unnecessary IPv6 byte-swapping
Alex Elder [Fri, 11 Jun 2021 19:05:29 +0000 (14:05 -0500)]
net: qualcomm: rmnet: avoid unnecessary IPv6 byte-swapping

In the previous patch IPv4 download checksum offload code was
updated to avoid unnecessary byte swapping, based on properties of
the Internet checksum algorithm.  This patch makes comparable
changes to the IPv6 download checksum offload handling.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qualcomm: rmnet: avoid unnecessary byte-swapping
Alex Elder [Fri, 11 Jun 2021 19:05:28 +0000 (14:05 -0500)]
net: qualcomm: rmnet: avoid unnecessary byte-swapping

Internet checksums are used for IPv4 header checksum, as well as TCP
segment and UDP datagram checksums.  Such a checksum represents the
negated sum of adjacent pairs of bytes, using ones' complement
arithmetic.

One property of the Internet checkum is byte order independence [1].
Specifically, the sum of byte-swapped pairs is equal to the result
of byte swapping the sum of those same pairs when not byte-swapped.

So for example if a, b, c, d, y, and z are hexadecimal digits, and
PLUS represents ones' complement addition:
    If: ab PLUS cd = yz
    Then: ba PLUS dc = zy

For this reason, there is no need to swap the order of bytes in the
checksum value held in a message header, nor the one in the QMAPv4
trailer, in order to operate on them.

In other words, we can determine whether the hardware-computed
checksum matches the one in the message header without any byte
swaps.

(This patch leaves in place all existing type casts.)

[1] https://tools.ietf.org/html/rfc1071

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qualcomm: rmnet: clarify a bit of code
Alex Elder [Fri, 11 Jun 2021 19:05:27 +0000 (14:05 -0500)]
net: qualcomm: rmnet: clarify a bit of code

In rmnet_map_ipv6_dl_csum_trailer() there is an especially involved
line of code that determines the ones' complement sum of the IPv6
packet header (in host byte order).  Simplify that by storing the
result of computing just the header checksum in a local variable,
then using that in the original assignment.

Use the size of the IPv6 header structure as the number of bytes to
checksum, rather than computing the offset to the transport header.
And use ip_fast_csum() rather than ipa_compute_csum(), knowing that
the size of an IPv6 header (40 bytes) is a multiple of 4 bytes
greater than 16.

Add some comments to match rmnet_map_ipv4_dl_csum_trailer().

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qualcomm: rmnet: IPv4 header has zero checksum
Alex Elder [Fri, 11 Jun 2021 19:05:26 +0000 (14:05 -0500)]
net: qualcomm: rmnet: IPv4 header has zero checksum

In rmnet_map_ipv4_dl_csum_trailer(), an illegal checksum subtraction
is done, subtracting hdr_csum (in host byte order) from csum_value (in
network byte order).  Despite being illegal, it generally works,
because it turns out the value subtracted is (or should be) always 0,
which has the same representation in either byte order.

Doing illegal operations is not good form though, so fix this by
verifying the IP header checksum early in that function.  If its
checksum is non-zero, the packet will be bad, so just return an
error.  This will cause the packet to passed to the IP layer where
it can be dropped.

Thereafter, there is no need subtract the IP header checksum from
the checksum value in the trailer because we know it is zero.
Add a comment explaining this.

This type of packet error is different from other types, so add a
new statistics counter to track this condition.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qualcomm: rmnet: simplify rmnet_map_get_csum_field()
Alex Elder [Fri, 11 Jun 2021 19:05:25 +0000 (14:05 -0500)]
net: qualcomm: rmnet: simplify rmnet_map_get_csum_field()

The checksum fields of the TCP and UDP header structures already
have type __sum16.  We don't support any other protocol headers, so
we can simplify rmnet_map_get_csum_field(), getting rid of the local
variable entirely and just returning the appropriate address.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qualcomm: rmnet: get rid of some local variables
Alex Elder [Fri, 11 Jun 2021 19:05:24 +0000 (14:05 -0500)]
net: qualcomm: rmnet: get rid of some local variables

The value passed as an argument to rmnet_map_ipv4_ul_csum_header()
is always an IPv4 header.  Rather than using a local variable, just
have the type of the argument reflect the proper type.

In rmnet_map_ipv6_ul_csum_header() things are defined a little
differently, but make the same basic change there.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qualcomm: rmnet: eliminate some ifdefs
Alex Elder [Fri, 11 Jun 2021 19:05:23 +0000 (14:05 -0500)]
net: qualcomm: rmnet: eliminate some ifdefs

If IPV6 is not enabled in the kernel configuration, the RMNet
checksum code indicates a buffer containing an IPv6 packet is not
supported.  The same thing happens if a buffer contains something
other than an IPv4 or IPv6 packet.

We can rearrange things a bit in two functions so that some #ifdef
calls can simply be eliminated.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qualcomm: rmnet: use ip_is_fragment()
Alex Elder [Fri, 11 Jun 2021 19:05:22 +0000 (14:05 -0500)]
net: qualcomm: rmnet: use ip_is_fragment()

In rmnet_map_ipv4_dl_csum_trailer() use ip_is_fragment() to
determine whether a socket buffer contains a packet fragment.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next...
David S. Miller [Fri, 11 Jun 2021 20:35:17 +0000 (13:35 -0700)]
Merge branch '100GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
Jake Keller says:

====================
100GbE Intel Wired LAN Driver Updates 2021-06-11

Extend the ice driver to support basic PTP clock functionality for E810
devices.

This includes some tangential work required to setup the sideband queue and
driver shared parameters as well.

This series only supports E810-based devices. This is because other devices
based on the E822 MAC use a different and more complex PHY.

The low level device functionality is kept within ice_ptp_hw.c and is
designed to be extensible for supporting E822 devices in a future series.

This series also only supports very basic functionality including the
ptp_clock device and timestamping. Support for configuring periodic outputs
and external input timestamps will be implemented in a future series.

There are a couple of potential "what? why?" bits in this series I want to
point out:

1) the PTP hardware functionality is shared between multiple functions. This
means that the same clock registers are shared across multiple PFs. In order
to avoid contention or clashing between PFs, firmware assigns "ownership" to
one PF, while other PFs are merely "associated" with the timer. Because we
share the hardware resource, only the clock owner will allocate and register
a PTP clock device. Other PFs determine the appropriate PTP clock index to
report by using a firmware interface to read a shared parameter that is set
by the owning PF.

2) the ice driver uses its own kthread instead of using do_aux_work. This is
because the periodic and asynchronous tasks are necessary for all PFs, but
only one PF will allocate the clock.

The series is broken up into functional pieces to allow easy review.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'virtio-vsock-seqpacket'
David S. Miller [Fri, 11 Jun 2021 20:32:47 +0000 (13:32 -0700)]
Merge branch 'virtio-vsock-seqpacket'

Arseny Krasnov says:

====================
virtio/vsock: introduce SOCK_SEQPACKET support

This patchset implements support of SOCK_SEQPACKET for virtio
transport.
As SOCK_SEQPACKET guarantees to save record boundaries, so to
do it, new bit for field 'flags' was added: SEQ_EOR. This bit is
set to 1 in last RW packet of message.
Now as  packets of one socket are not reordered neither on vsock
nor on vhost transport layers, such bit allows to restore original
message on receiver's side. If user's buffer is smaller than message
length, when all out of size data is dropped.
Maximum length of datagram is limited by 'peer_buf_alloc' value.
Implementation also supports 'MSG_TRUNC' flags.
Tests also implemented.

Thanks to stsp2@yandex.ru for encouragements and initial design
recommendations.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovirtio/vsock: update trace event for SEQPACKET
Arseny Krasnov [Fri, 11 Jun 2021 11:14:20 +0000 (14:14 +0300)]
virtio/vsock: update trace event for SEQPACKET

Add SEQPACKET socket type to vsock trace event.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovsock_test: add SOCK_SEQPACKET tests
Arseny Krasnov [Fri, 11 Jun 2021 11:14:04 +0000 (14:14 +0300)]
vsock_test: add SOCK_SEQPACKET tests

Implement two tests of SOCK_SEQPACKET socket: first sends data by
several 'write()'s and checks that number of 'read()' were same.
Second test checks MSG_TRUNC flag. Cases for connect(), bind(),
etc. are not tested, because it is same as for stream socket.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovsock/loopback: enable SEQPACKET for transport
Arseny Krasnov [Fri, 11 Jun 2021 11:13:50 +0000 (14:13 +0300)]
vsock/loopback: enable SEQPACKET for transport

Add SEQPACKET ops for loopback transport and 'seqpacket_allow()'
callback.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovhost/vsock: support SEQPACKET for transport
Arseny Krasnov [Fri, 11 Jun 2021 11:13:37 +0000 (14:13 +0300)]
vhost/vsock: support SEQPACKET for transport

When received packet is copied to guests's rx queue, data buffers
of rx queue could be smaller that data buffer of input packet, so
data of input packet is copied to each rx buffer, thus each rx
buffer will be a packet with dynamically created header. Fields
of such header are initialized from header of input packet(except
length field which value is depends on number of bytes copied to
rx buffer). But in SEQPACKET case, we also need to take care of
record delimeter bit: if input packet has this bit set, we don't
copy it to header of packet in rx buffer, except case when such
rx buffer is last part of input packet. Otherwise, we will get
sequence of packets with delimeter bit set, thus braking record
bounds.
Also remove ignore of non-stream type of packets, handle SEQPACKET
feature bit.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovirtio/vsock: enable SEQPACKET for transport
Arseny Krasnov [Fri, 11 Jun 2021 11:13:22 +0000 (14:13 +0300)]
virtio/vsock: enable SEQPACKET for transport

To make transport work with SOCK_SEQPACKET add two things:
1) SOCK_SEQPACKET ops for virtio transport and 'seqpacket_allow()'
   callback.
2) Handling of SEQPACKET bit: guest tries to negotiate it with vhost,
   so feature will be enabled only if bit is negotiated with device.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovirtio/vsock: rest of SOCK_SEQPACKET support
Arseny Krasnov [Fri, 11 Jun 2021 11:13:06 +0000 (14:13 +0300)]
virtio/vsock: rest of SOCK_SEQPACKET support

Small updates to make SOCK_SEQPACKET work:
1) Send SHUTDOWN on socket close for SEQPACKET type.
2) Set SEQPACKET packet type during send.
3) Set 'VIRTIO_VSOCK_SEQ_EOR' bit in flags for last
   packet of message.
4) Implement data check function for SEQPACKET.
5) Check for max datagram size.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovirtio/vsock: add SEQPACKET receive logic
Arseny Krasnov [Fri, 11 Jun 2021 11:12:53 +0000 (14:12 +0300)]
virtio/vsock: add SEQPACKET receive logic

Update current receive logic for SEQPACKET support: performs
check for packet and socket types on receive(if mismatch, then
reset connection). Increment EOR counter on receive. Also if
buffer of new packet was appended to buffer of last packet in
rx queue, update flags of last packet with flags of new packet.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovirtio/vsock: dequeue callback for SOCK_SEQPACKET
Arseny Krasnov [Fri, 11 Jun 2021 11:12:38 +0000 (14:12 +0300)]
virtio/vsock: dequeue callback for SOCK_SEQPACKET

Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovirtio/vsock: defines and constants for SEQPACKET
Arseny Krasnov [Fri, 11 Jun 2021 11:12:22 +0000 (14:12 +0300)]
virtio/vsock: defines and constants for SEQPACKET

Add set of defines and constants for SOCK_SEQPACKET support
in vsock.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovirtio/vsock: simplify credit update function API
Arseny Krasnov [Fri, 11 Jun 2021 11:12:08 +0000 (14:12 +0300)]
virtio/vsock: simplify credit update function API

This function is static and 'hdr' arg was always NULL.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agovirtio/vsock: set packet's type in virtio_transport_send_pkt_info()
Arseny Krasnov [Fri, 11 Jun 2021 11:11:31 +0000 (14:11 +0300)]
virtio/vsock: set packet's type in virtio_transport_send_pkt_info()

There is no need to set type of packet which differs from type
of socket, so move passing type of packet from 'info' structure
to  'virtio_transport_send_pkt_info()' function. Since at current
time only stream type is supported, set it directly in 'virtio_
transport_send_pkt_info()', so callers don't need to set it.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoaf_vsock: update comments for stream sockets
Arseny Krasnov [Fri, 11 Jun 2021 11:11:18 +0000 (14:11 +0300)]
af_vsock: update comments for stream sockets

Replace 'stream' to 'connection oriented' in comments as
SEQPACKET is also connection oriented.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoaf_vsock: rest of SEQPACKET support
Arseny Krasnov [Fri, 11 Jun 2021 11:11:04 +0000 (14:11 +0300)]
af_vsock: rest of SEQPACKET support

Add socket ops for SEQPACKET type and .seqpacket_allow() callback
to query transports if they support SEQPACKET. Also split path
for data check for STREAM and SEQPACKET branches.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoaf_vsock: implement send logic for SEQPACKET
Arseny Krasnov [Fri, 11 Jun 2021 11:10:49 +0000 (14:10 +0300)]
af_vsock: implement send logic for SEQPACKET

Update current stream enqueue function for SEQPACKET
support:
1) Call transport's seqpacket enqueue callback.
2) Return value from enqueue function is whole record length or error
   for SOCK_SEQPACKET.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoaf_vsock: implement SEQPACKET receive loop
Arseny Krasnov [Fri, 11 Jun 2021 11:10:34 +0000 (14:10 +0300)]
af_vsock: implement SEQPACKET receive loop

Add receive loop for SEQPACKET. It looks like receive loop for
STREAM, but there are differences:
1) It doesn't call notify callbacks.
2) It doesn't care about 'SO_SNDLOWAT' and 'SO_RCVLOWAT' values, because
   there is no sense for these values in SEQPACKET case.
3) It waits until whole record is received.
4) It processes and sets 'MSG_TRUNC' flag.

So to avoid extra conditions for two types of socket inside one loop, two
independent functions were created.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoaf_vsock: separate receive data loop
Arseny Krasnov [Fri, 11 Jun 2021 11:10:21 +0000 (14:10 +0300)]
af_vsock: separate receive data loop

Some code in receive data loop could be shared between SEQPACKET
and STREAM sockets, while another part is type specific, so move STREAM
specific data receive logic to '__vsock_stream_recvmsg()' dedicated
function, while checks, that will be same for both STREAM and SEQPACKET
sockets, stays in 'vsock_connectible_recvmsg()'.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoaf_vsock: separate wait data loop
Arseny Krasnov [Fri, 11 Jun 2021 11:10:07 +0000 (14:10 +0300)]
af_vsock: separate wait data loop

Wait loop for data could be shared between SEQPACKET and STREAM
sockets, so move it to dedicated function. While moving the code
around, let's update an old comment.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoaf_vsock: update functions for connectible socket
Arseny Krasnov [Fri, 11 Jun 2021 11:09:47 +0000 (14:09 +0300)]
af_vsock: update functions for connectible socket

Prepare af_vsock.c for SEQPACKET support: rename some functions such
as setsockopt(), getsockopt(), connect(), recvmsg(), sendmsg() in general
manner, because they are shared with stream sockets.

Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'octeontx2-trusted-vf'
David S. Miller [Fri, 11 Jun 2021 20:21:11 +0000 (13:21 -0700)]
Merge branch 'octeontx2-trusted-vf'

Naveen Mamindlapalli says:

====================
octeontx2: Add trusted VF support

This series adds support for trusted VF. The trusted VF support
allows VFs to perform priviliged operations such as setting VF
interface in promiscuous mode, all-multicast mode and also
changing the VF MAC address even if it was asssigned by PF.

Patches #1 and #2 provides the necessary functionality for supporting
promiscuous and multicast packets on both the PF and VF.

Patches #3 and #4 enable trusted VF configuration support.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-pf: add support for ndo_set_vf_trust
Hariprasad Kelam [Fri, 11 Jun 2021 09:42:05 +0000 (15:12 +0530)]
octeontx2-pf: add support for ndo_set_vf_trust

Add support for setting a VF as a trusted VF by PF admin. Trusted VF
feature allows VFs to perform priviliged operations such as enabling
VF promiscuous mode, all-multicast mode and changing the VF MAC address
even if it was assigned by PF.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <Sunil.Goutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-af: add new mailbox to configure VF trust mode
Hariprasad Kelam [Fri, 11 Jun 2021 09:42:04 +0000 (15:12 +0530)]
octeontx2-af: add new mailbox to configure VF trust mode

Add new mailbox to enable PF to configure VF as trusted VF.
Trusted VF feature allows VFs to perform priviliged operations
such as enabling VF promiscuous mode, all-multicast mode and
changing the VF MAC address configured by PF. Refactored the
VF interface flags maintained by the AF driver such that the
flags do not overlap for various configurations.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <Sunil.Goutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-nicvf: add ndo_set_rx_mode support for multicast & promisc
Naveen Mamindlapalli [Fri, 11 Jun 2021 09:42:03 +0000 (15:12 +0530)]
octeontx2-nicvf: add ndo_set_rx_mode support for multicast & promisc

Add ndo_set_rx_mode callback handler to configure promisc, multicast and
allmulti options for VF driver. Also, modified PF driver ndo_set_rx_mode
handler to support multicast and promisc mode independently.

Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <Sunil.Goutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-af: add support for multicast/promisc packet replication feature
Naveen Mamindlapalli [Fri, 11 Jun 2021 09:42:02 +0000 (15:12 +0530)]
octeontx2-af: add support for multicast/promisc packet replication feature

Currently, multicast packet filtering is accomplished by installing
MCAM rule that matches all-multicast MAC address and has its
NPC_RX_ACTION set to unicast to PF. Similarly promisc feature is
achieved by installing MCAM rule that matches all the traffic received
by the channel and unicast the packets to PF. This approach only applies
to PF and is not scalable across VFs.

This patch adds support for PF/VF multicast and promisc feature by
reserving NIX_RX_MCE_S entries from the global MCE list allocated
during NIX block initialization. The NIX_RX_MCE_S entries create a
linked list with a flag indicating the end of the list, and each entry
points to a PF_FUNC (either PF or VF). When a packet NPC_RX_ACTION is
set to MCAST, the corresponding NIX_RX_MCE_S list is traversed and the
packet is queued to each PF_FUNC available on the list.

The PF or VF driver adds the multicast/promisc packet match entry and
updates the MCE list with correspondng PF_FUNC. When a PF or VF interface
is disabled, the corresponding NIX_RX_MCE_S entry is removed from the
MCE list and the MCAM entry will be disabled if the list is empty.

Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <Sunil.Goutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: stmmac: Fix potential integer overflow
Wong Vee Khee [Fri, 11 Jun 2021 09:02:38 +0000 (17:02 +0800)]
net: stmmac: Fix potential integer overflow

The commit d96febedfde2 ("net: stmmac: arrange Tx tail pointer update
to stmmac_flush_tx_descriptors") introduced the following coverity
warning:-

  1. Unintentional integer overflow (OVERFLOW_BEFORE_WIDEN)
     overflow_before_widen: Potentially overflowing expression
     'tx_q->cur_tx * desc_size' with type 'unsigned int' (32 bits,
     unsigned) is evaluated using 32-bit arithmetic, and then used in a
     context that expects an expression of type dma_addr_t (64 bits,
     unsigned).

Fixed this by assigning tx_tail_addr to dma_addr_t type, as dma_addr_t
datatype is decided by CONFIG_ARCH_DMA_ADDR_T_64_BIT.

Fixes: d96febedfde2 ("net: stmmac: arrange Tx tail pointer update to stmmac_flush_tx_descriptors")
Signed-off-by: Wong Vee Khee <vee.khee.wong@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mdio: mscc-miim: Use devm_platform_get_and_ioremap_resource()
Yang Yingliang [Fri, 11 Jun 2021 08:04:09 +0000 (16:04 +0800)]
net: mdio: mscc-miim: Use devm_platform_get_and_ioremap_resource()

Use devm_platform_get_and_ioremap_resource() to simplify
code.

Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'dpaa2-ACPI'
David S. Miller [Fri, 11 Jun 2021 20:08:53 +0000 (13:08 -0700)]
Merge branch 'dpaa2-ACPI'

Ioana Ciornei says:

====================
ACPI support for dpaa2 driver

This patch set provides ACPI support to DPAA2 network drivers.

It also introduces new fwnode based APIs to support phylink and phy
layers
    Following functions are defined:
      phylink_fwnode_phy_connect()
      fwnode_mdiobus_register_phy()
      fwnode_get_phy_id()
      fwnode_phy_find_device()
      device_phy_find_device()
      fwnode_get_phy_node()
      fwnode_mdio_find_device()
      acpi_get_local_address()

    First one helps in connecting phy to phylink instance.
    Next three helps in getting phy_id and registering phy to mdiobus
    Next two help in finding a phy on a mdiobus.
    Next one helps in getting phy_node from a fwnode.
    Last one is used to get local address from _ADR object.

    Corresponding OF functions are refactored.

Tested-on: LX2160ARDB

Changes in v9:
 - merged some minimal changes requested in the wording of the commit
   messages
 - fixed some build problems in patch 8/15 by moving the removal of
   of_find_mii_timestamper from patch 8/15 to 9/15.

Changes in v8:
 - fixed some checkpatch warnings/checks
 - included linux/fwnode_mdio.h in fwnode_mdio.c (fixed the build warnings)
 - added fwnode_find_mii_timestamper() and
   fwnode_mdiobus_phy_device_register() in order to get rid of the cycle
   dependency.
 - change to 'depends on (ACPI || OF) || COMPILE_TEST (for FWNODE_MDIO)
 - remove the fwnode_mdiobus_register from fwnode_mdio.c since it
   introduces a cycle of dependencies.

Changes in v7:
- correct fwnode_mdio_find_device() description
- check NULL in unregister_mii_timestamper()
- Call unregister_mii_timestamper() without NULL check
- Create fwnode_mdio.c and move fwnode_mdiobus_register_phy()
- include fwnode_mdio.h
- Include headers directly used in acpi_mdio.c
- Move fwnode_mdiobus_register() to fwnode_mdio.c
- Include fwnode_mdio.h
- Alphabetically sort header inclusions
- remove unnecassary checks

Changes in v6:
- Minor cleanup
- fix warning for function parameter of fwnode_mdio_find_device()
- Initialize mii_ts to NULL
- use GENMASK() and ACPI_COMPANION_SET()
- some cleanup
- remove unwanted header inclusion
- remove OF check for fixed-link
- use dev_fwnode()
- remove useless else
- replace of_device_is_available() to fwnode_device_is_available()

Changes in v5:
- More cleanup
- Replace fwnode_get_id() with acpi_get_local_address()
- add missing MODULE_LICENSE()
- replace fwnode_get_id() with OF and ACPI function calls
- replace fwnode_get_id() with OF and ACPI function calls

Changes in v4:
- More cleanup
- Improve code structure to handle all cases
- Remove redundant else from fwnode_mdiobus_register()
- Cleanup xgmac_mdio_probe()
- call phy_device_free() before returning

Changes in v3:
- Add more info on legacy DT properties "phy" and "phy-device"
- Redefine fwnode_phy_find_device() to follow of_phy_find_device()
- Use traditional comparison pattern
- Use GENMASK
- Modified to retrieve reg property value for ACPI as well
- Resolved compilation issue with CONFIG_ACPI = n
- Added more info into documentation
- Use acpi_mdiobus_register()
- Avoid unnecessary line removal
- Remove unused inclusion of acpi.h

Changes in v2:
- Updated with more description in document
- use reverse christmas tree ordering for local variables
- Refactor OF functions to use fwnode functions
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dpaa2-mac: Add ACPI support for DPAA2 MAC driver
Calvin Johnson [Fri, 11 Jun 2021 10:54:01 +0000 (13:54 +0300)]
net: dpaa2-mac: Add ACPI support for DPAA2 MAC driver

Modify dpaa2_mac_get_node() to get the dpmac fwnode from either
DT or ACPI.

Modify dpaa2_mac_get_if_mode() to get interface mode from dpmac_node
which is a fwnode.

Modify dpaa2_pcs_create() to create pcs from dpmac_node fwnode.

Modify dpaa2_mac_connect() to support ACPI along with DT.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Rafael J. Wysocki <rafael@kernel.org> # from the ACPI side
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phylink: Refactor phylink_of_phy_connect()
Calvin Johnson [Fri, 11 Jun 2021 10:54:00 +0000 (13:54 +0300)]
net: phylink: Refactor phylink_of_phy_connect()

Refactor phylink_of_phy_connect() to use phylink_fwnode_phy_connect().

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phylink: introduce phylink_fwnode_phy_connect()
Calvin Johnson [Fri, 11 Jun 2021 10:53:59 +0000 (13:53 +0300)]
net: phylink: introduce phylink_fwnode_phy_connect()

Define phylink_fwnode_phy_connect() to connect phy specified by
a fwnode to a phylink instance.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/fsl: Use [acpi|of]_mdiobus_register
Calvin Johnson [Fri, 11 Jun 2021 10:53:58 +0000 (13:53 +0300)]
net/fsl: Use [acpi|of]_mdiobus_register

Depending on the device node type, call the specific OF or ACPI
mdiobus_register function.

Note: For both ACPI and DT cases, endianness of MDIO controllers
need to be specified using the "little-endian" property.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mdio: Add ACPI support code for mdio
Calvin Johnson [Fri, 11 Jun 2021 10:53:57 +0000 (13:53 +0300)]
net: mdio: Add ACPI support code for mdio

Define acpi_mdiobus_register() to Register mii_bus and create PHYs for
each ACPI child node.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Rafael J. Wysocki <rafael@kernel.org>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoACPI: utils: Introduce acpi_get_local_address()
Calvin Johnson [Fri, 11 Jun 2021 10:53:56 +0000 (13:53 +0300)]
ACPI: utils: Introduce acpi_get_local_address()

Introduce a wrapper around the _ADR evaluation.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Rafael J. Wysocki <rafael@kernel.org>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoof: mdio: Refactor of_mdiobus_register_phy()
Calvin Johnson [Fri, 11 Jun 2021 10:53:55 +0000 (13:53 +0300)]
of: mdio: Refactor of_mdiobus_register_phy()

Refactor of_mdiobus_register_phy() to use fwnode_mdiobus_register_phy().
Also, remove the of_find_mii_timestamper() since the fwnode variant is
used instead.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mdiobus: Introduce fwnode_mdiobus_register_phy()
Calvin Johnson [Fri, 11 Jun 2021 10:53:54 +0000 (13:53 +0300)]
net: mdiobus: Introduce fwnode_mdiobus_register_phy()

Introduce fwnode_mdiobus_register_phy() to register PHYs on the
mdiobus. From the compatible string, identify whether the PHY is
c45 and based on this create a PHY device instance which is
registered on the mdiobus.

Along with fwnode_mdiobus_register_phy() also introduce
fwnode_find_mii_timestamper() and fwnode_mdiobus_phy_device_register()
since they are needed.
While at it, also use the newly introduced fwnode operation in
of_mdiobus_phy_device_register().

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mii_timestamper: check NULL in unregister_mii_timestamper()
Calvin Johnson [Fri, 11 Jun 2021 10:53:53 +0000 (13:53 +0300)]
net: mii_timestamper: check NULL in unregister_mii_timestamper()

Callers of unregister_mii_timestamper() currently check for NULL
value of mii_ts before calling it.

Place the NULL check inside unregister_mii_timestamper() and update
the callers accordingly.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Suggested-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoof: mdio: Refactor of_get_phy_id()
Calvin Johnson [Fri, 11 Jun 2021 10:53:52 +0000 (13:53 +0300)]
of: mdio: Refactor of_get_phy_id()

With the introduction of fwnode_get_phy_id(), refactor of_get_phy_id()
to use fwnode equivalent.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: Introduce fwnode_get_phy_id()
Calvin Johnson [Fri, 11 Jun 2021 10:53:51 +0000 (13:53 +0300)]
net: phy: Introduce fwnode_get_phy_id()

Extract phy_id from compatible string. This will be used by
fwnode_mdiobus_register_phy() to create phy device using the
phy_id.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoof: mdio: Refactor of_phy_find_device()
Calvin Johnson [Fri, 11 Jun 2021 10:53:50 +0000 (13:53 +0300)]
of: mdio: Refactor of_phy_find_device()

Refactor of_phy_find_device() to use fwnode_phy_find_device().

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: Introduce phy related fwnode functions
Calvin Johnson [Fri, 11 Jun 2021 10:53:49 +0000 (13:53 +0300)]
net: phy: Introduce phy related fwnode functions

Define fwnode_phy_find_device() to iterate an mdiobus and find the
phy device of the provided phy fwnode. Additionally define
device_phy_find_device() to find phy device of provided device.

Define fwnode_get_phy_node() to get phy_node using named reference.

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: Introduce fwnode_mdio_find_device()
Calvin Johnson [Fri, 11 Jun 2021 10:53:48 +0000 (13:53 +0300)]
net: phy: Introduce fwnode_mdio_find_device()

Define fwnode_mdio_find_device() to get a pointer to the
mdio_device from fwnode passed to the function.

Refactor of_mdio_find_device() to use fwnode_mdio_find_device().

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoDocumentation: ACPI: DSD: Document MDIO PHY
Calvin Johnson [Fri, 11 Jun 2021 10:53:47 +0000 (13:53 +0300)]
Documentation: ACPI: DSD: Document MDIO PHY

Introduce a mechanism based on generic ACPI _DSD device properties
definition [1] to get PHYs registered on a MDIO bus and provide them to
be connected to MAC.

[1] http://www.uefi.org/sites/default/files/resources/_DSD-device-properties-UUID.pdf

Describe properties "phy-handle" and "phy-mode".

Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Rafael J. Wysocki <rafael@kernel.org>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: stmmac: Fix unused values warnings
Wong Vee Khee [Fri, 11 Jun 2021 07:11:43 +0000 (15:11 +0800)]
net: stmmac: Fix unused values warnings

The commit 8532f613bc78 ("net: stmmac: introduce MSI Interrupt routines
for mac, safety, RX & TX") introduced the converity warnings:-

  1. Unused value (UNUSED_VALUE)
     assigned_value: Assigning value REQ_IRQ_ERR_MAC to irq_err here,
     but that stored value is not used.

  2. Unused value (UNUSED_VALUE)
     assigned_value: Assigning value REQ_IRQ_ERR_NO to irq_err here,
     but that stored value is overwritten before it can used.

  3. Unused value (UNUSED_VALUE)
     assigned_value: Assigning value REQ_IRQ_ERR_WOL to irq_err here,
     but that stored value is not used.

Fixed these by removing the unnecessary value assignments.

Fixes: 8532f613bc78 ("net: stmmac: introduce MSI Interrupt routines for mac, safety, RX & TX")
Signed-off-by: Wong Vee Khee <vee.khee.wong@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: usb: asix: ax88772: manage PHY PM from MAC
Oleksij Rempel [Fri, 11 Jun 2021 03:55:59 +0000 (05:55 +0200)]
net: usb: asix: ax88772: manage PHY PM from MAC

Take over PHY power management, otherwise PHY framework will try to
access ASIX MDIO bus before MAC resume was completed.

Fixes: e532a096be0e ("net: usb: asix: ax88772: add phylib support")
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
Reported-by: Jon Hunter <jonathanh@nvidia.com>
Suggested-by: Heiner Kallweit <hkallweit1@gmail.com>
Tested-by: Jon Hunter <jonathanh@nvidia.com>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'pc300too'
David S. Miller [Fri, 11 Jun 2021 19:58:12 +0000 (12:58 -0700)]
Merge branch 'pc300too'

Peng Li says:

====================
net: pc300too: clean up some code style issues

This patchset clean up some code style issues.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pc300too: fix the comments style issue
Peng Li [Fri, 11 Jun 2021 03:36:22 +0000 (11:36 +0800)]
net: pc300too: fix the comments style issue

Networking block comments don't use an empty /* line,
use /* Comment...
This patch fixes the comments style issues.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pc300too: add some required spaces
Peng Li [Fri, 11 Jun 2021 03:36:21 +0000 (11:36 +0800)]
net: pc300too: add some required spaces

Add spaces required before the open parenthesis '('.
Add spaces required after that close brace '}'.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pc300too: replace comparison to NULL with "!card->plxbase"
Peng Li [Fri, 11 Jun 2021 03:36:20 +0000 (11:36 +0800)]
net: pc300too: replace comparison to NULL with "!card->plxbase"

According to the chackpatch.pl, comparison to NULL could
be written "!card->plxbase".

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pc300too: remove redundant initialization for statics
Peng Li [Fri, 11 Jun 2021 03:36:19 +0000 (11:36 +0800)]
net: pc300too: remove redundant initialization for statics

Should not initialise statics to 0.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pc300too: move out assignment in if condition
Peng Li [Fri, 11 Jun 2021 03:36:18 +0000 (11:36 +0800)]
net: pc300too: move out assignment in if condition

Should not use assignment in if condition.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pc300too: fix the code style issue about "foo * bar"
Peng Li [Fri, 11 Jun 2021 03:36:17 +0000 (11:36 +0800)]
net: pc300too: fix the code style issue about "foo * bar"

Fix the checkpatch error as "foo * bar" and should be "foo *bar".

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pc300too: add blank line after declarations
Peng Li [Fri, 11 Jun 2021 03:36:16 +0000 (11:36 +0800)]
net: pc300too: add blank line after declarations

This patch fixes the checkpatch error about missing a blank line
after declarations.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: pc300too: remove redundant blank lines
Peng Li [Fri, 11 Jun 2021 03:36:15 +0000 (11:36 +0800)]
net: pc300too: remove redundant blank lines

This patch removes some redundant blank lines.

Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: devres: Correct a grammatical error
gushengxian [Fri, 11 Jun 2021 01:33:33 +0000 (09:33 +0800)]
net: devres: Correct a grammatical error

Correct a grammatical error.

Signed-off-by: gushengxian <gushengxian@yulong.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agor8169: avoid link-up interrupt issue on RTL8106e if user enables ASPM
Heiner Kallweit [Thu, 10 Jun 2021 20:56:59 +0000 (22:56 +0200)]
r8169: avoid link-up interrupt issue on RTL8106e if user enables ASPM

It has been reported that on RTL8106e the link-up interrupt may be
significantly delayed if the user enables ASPM L1. Per default ASPM
is disabled. The change leaves L1 enabled on the PCIe link (thus still
allowing to reach higher package power saving states), but the
NIC won't actively trigger it.

Reported-by: Koba Ko <koba.ko@canonical.com>
Tested-by: Koba Ko <koba.ko@canonical.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonfc: fdp: remove unnecessary labels
wengjianfeng [Thu, 10 Jun 2021 02:46:16 +0000 (10:46 +0800)]
nfc: fdp: remove unnecessary labels

Some labels are meaningless, so we delete them and use the
return statement instead of the goto statement.

Signed-off-by: wengjianfeng <wengjianfeng@yulong.com>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 's390-qeyj-next'
David S. Miller [Fri, 11 Jun 2021 19:49:15 +0000 (12:49 -0700)]
Merge branch 's390-qeyj-next'

Julian Wiedmann says:

====================
s390/qeth: updates 2021-06-11

please apply the following patch series for qeth to netdev's net-next tree.

This enables TX NAPI for those devices that didn't use it previously, so
that we can eventually rip out the qdio layer's internal interrupt
machinery.

Other than that it's just the normal mix of minor improvements and
cleanups.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: Consider dependency on SWITCHDEV module
Alexandra Winter [Fri, 11 Jun 2021 07:33:41 +0000 (09:33 +0200)]
s390/qeth: Consider dependency on SWITCHDEV module

Without the SWITCHDEV module, the bridgeport attribute LEARNING_SYNC
of the physical device (self) does not provide any functionality.
Instead of calling the no-op stub version of the switchdev functions,
fail the setting of the attribute with an appropriate message.

While at it, also add an error message for the 'not supported by HW'
case.

Signed-off-by: Alexandra Winter <wintera@linux.ibm.com>
Reviewed-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: shrink TX buffer struct
Julian Wiedmann [Fri, 11 Jun 2021 07:33:40 +0000 (09:33 +0200)]
s390/qeth: shrink TX buffer struct

Convert the large boolean array into a bitmap, this substantially
reduces the struct's size. While at it also clarify the naming.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: remove TX buffer's pointer to its queue
Julian Wiedmann [Fri, 11 Jun 2021 07:33:39 +0000 (09:33 +0200)]
s390/qeth: remove TX buffer's pointer to its queue

qeth_tx_complete_buf() is the only remaining user of buf->q, and the
callers can easily provide this as a parameter instead.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: remove QAOB's pointer to its TX buffer
Julian Wiedmann [Fri, 11 Jun 2021 07:33:38 +0000 (09:33 +0200)]
s390/qeth: remove QAOB's pointer to its TX buffer

Maintaining a pointer inside the aob's user-definable area is fragile
and unnecessary. At this stage we only need it to overload the buffer's
state field, and to access the buffer's TX queue.

The first part is easily solved by tracking the aob's state within the
aob itself. This also feels much cleaner and self-contained.
For enabling the access to the associated TX queue, we can store the
queue's index in the aob.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: consolidate completion of pending TX buffers
Julian Wiedmann [Fri, 11 Jun 2021 07:33:37 +0000 (09:33 +0200)]
s390/qeth: consolidate completion of pending TX buffers

With commit 396c100472dd ("s390/qdio: let driver manage the QAOB")
a pending TX buffer now has access to its associated QAOB during
TX completion processing. We can thus reduce the amount of work & state
propagation that needs to be done by qeth_qdio_handle_aob().

Move all this logic into the respective TX completion paths. Doing so
even allows us to determine more precise TX_NOTIFY_* values via
qeth_compute_cq_notification(aob->aorc, ...).

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: use ethtool_sprintf()
Julian Wiedmann [Fri, 11 Jun 2021 07:33:36 +0000 (09:33 +0200)]
s390/qeth: use ethtool_sprintf()

Use a recently introduced helper to fill our ethtool stats strings.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: unify the tracking of active cmds on ccw device
Julian Wiedmann [Fri, 11 Jun 2021 07:33:35 +0000 (09:33 +0200)]
s390/qeth: unify the tracking of active cmds on ccw device

We have one field to track _whether_ a cmd is active on a ccw device
('irq_pending'), and one to track _which_ cmd it is ('active_cmd').

Get rid of the irq_pending field, by testing active_cmd for NULL.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: also use TX NAPI for non-IQD devices
Julian Wiedmann [Fri, 11 Jun 2021 07:33:34 +0000 (09:33 +0200)]
s390/qeth: also use TX NAPI for non-IQD devices

Set scan_threshold = 0 to opt out from the qdio layer's internal tasklet
& timer mechanism for TX completions, and replace it with the TX NAPI
infrastructure that qeth already uses for IQD devices. This avoids the
fragile logic in qdio_check_output_queue(), enables tighter integration
and gives us more tuning options via ethtool in the future.

For now we continue to apply the same policy as the qdio layer:
scan for completions if 32 TX buffers are in use, or after 1 sec.
A re-scan is done after 10 sec, but only if no TX interrupt is pending.

With scan_threshold = 0 we no longer get TX completion scans from
within qdio_get_next_buffers(). So trigger these manually in qeth_poll()
and in the RX path switch to the equivalent qdio_inspect_queue().

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agos390/qeth: count TX completion interrupts
Julian Wiedmann [Fri, 11 Jun 2021 07:33:33 +0000 (09:33 +0200)]
s390/qeth: count TX completion interrupts

While the qdio layer already tracks the number of HW interrupts for a
device, there's value in understanding how many of them have been
raised due to our TX completion logic.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>