openwrt/staging/blogic.git
4 years agoMerge tag 'wireless-drivers-next-2019-12-20' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Sat, 21 Dec 2019 05:55:00 +0000 (21:55 -0800)]
Merge tag 'wireless-drivers-next-2019-12-20' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers-next patches for v5.6

First set of patches for v5.6. The biggest thing here is of course the
new driver ath11k but also new features for other drivers as well a
myriad of bug fixes.

Major changes:

ath11k

* a new driver for Qualcomm Wi-Fi 6 (IEEE 802.11ax) devices

ath10k

* significant improvements on receive throughput and firmware download
  with SDIO bus

* report signal strength for each chain also on SDIO

* set max mtu to 1500 on SDIO devices

brcmfmac

* add support for BCM4359 SDIO chipset

wil6210

* support set_multicast_to_unicast cfg80211 operation

* support set_cqm_rssi_config cfg80211 operation

wcn36xx

* disable HW_CONNECTION_MONITOR as firmware is buggy
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'hns3-next'
David S. Miller [Sat, 21 Dec 2019 05:20:39 +0000 (21:20 -0800)]
Merge branch 'hns3-next'

Huazhong Tan says:

====================
net: hns3: misc updates for -net-next

This series includes some misc updates for the HNS3 ethernet driver.

[patch 1] adds FE bit check before calling hns3_add_frag().
[patch 2] removes an unnecessary lock.
[patch 3] adds a little optimization for CMDQ uninitialization.
[patch 4] refactors the dump of FD tcams.
[patch 5] implements ndo_features_check ops.
[patch 6] adds some VF VLAN information for command "ip link show".
[patch 7] adds a debug print.
[patch 8] modifies the timing of print misc interrupt status when
handling hardware error event.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: only print misc interrupt status when handling fails
Huazhong Tan [Thu, 19 Dec 2019 06:57:47 +0000 (14:57 +0800)]
net: hns3: only print misc interrupt status when handling fails

Printing misc interrupt status of hardware error event in the
IRQ handler is unnecessary, since hclge_handle_hw_msix_error()
will print out the detail information for this hardware error
when handling success. So, this patch removes the print in
IRQ handler, and prints it when hclge_handle_hw_msix_error()
fails.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: add a log for getting chain failure in hns3_nic_uninit_vector_data()
Huazhong Tan [Thu, 19 Dec 2019 06:57:46 +0000 (14:57 +0800)]
net: hns3: add a log for getting chain failure in hns3_nic_uninit_vector_data()

Since the mapping can be overwritten, when fail to get
the chain between vector and ring, we should go on to
deal with the remaining options. For debugging, this
patch adds log info for this failure.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: add some VF VLAN information for command "ip link show"
Huazhong Tan [Thu, 19 Dec 2019 06:57:45 +0000 (14:57 +0800)]
net: hns3: add some VF VLAN information for command "ip link show"

This patch adds some VF VLAN information for command "ip link show".

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: implement ndo_features_check ops for hns3 driver
Yunsheng Lin [Thu, 19 Dec 2019 06:57:44 +0000 (14:57 +0800)]
net: hns3: implement ndo_features_check ops for hns3 driver

The function netif_skb_features() will disable the TSO feature
by using dflt_features_check() if the driver does not implement
ndo_features_check ops, which may cause performance degradation
problem when hns3 hardware can do multiple tagged TSO.

Also, the HNS3 hardware only supports checksum on the SKB with
a max header len of 480 bytes, so remove the checksum and TSO
related features when the header len is over 480 bytes.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: get FD rules location before dump in debugfs
Yufeng Mo [Thu, 19 Dec 2019 06:57:43 +0000 (14:57 +0800)]
net: hns3: get FD rules location before dump in debugfs

Currently, the dump FD tcam mode in debugfs is to query all FD tcams,
including empty rules, which is unnecessary. This patch modify to
find the position of useful rules before dump FD tcam, so that it does
not need to query empty rules.

This patch also modifies some help information in debugfs.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: optimization for CMDQ uninitialization
Huazhong Tan [Thu, 19 Dec 2019 06:57:42 +0000 (14:57 +0800)]
net: hns3: optimization for CMDQ uninitialization

When uninitializing CMDQ, HCLGE_STATE_CMD_DISABLE will
be set up firstly, then the driver does not send command
anymore. So, hclge_free_cmd_desc can be called without
holding ring->lock. hclge_destroy_cmd_queue() and
hclge_destroy_queue() are unnecessary now, so removes them,
the VF driver has implemented currently.

BTW, the VF driver should set up HCLGEVF_STATE_CMD_DISABLE
as well in the hclgevf_cmd_uninit(), just likes what the PF
driver does.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: remove useless mutex vport_cfg_mutex in the struct hclge_dev
Huazhong Tan [Thu, 19 Dec 2019 06:57:41 +0000 (14:57 +0800)]
net: hns3: remove useless mutex vport_cfg_mutex in the struct hclge_dev

Mutex vport_cfg_mutex has been used to protect uc_mac_list,
mc_mac_list and vlan_list from being modified by unloading
or reset task at the same time. But now unloading will
set up HCLGE_STATE_REMOVING flag and call cancel_work_sync to
break down this race condition, so this mutex is unnecessary.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: hns3: check FE bit before calling hns3_add_frag()
Yunsheng Lin [Thu, 19 Dec 2019 06:57:40 +0000 (14:57 +0800)]
net: hns3: check FE bit before calling hns3_add_frag()

A BD with FE bit means that it is the last BD of a packet,
currently the FE bit is checked before calling hns3_add_frag(),
which is unnecessary because the FE bit may have been checked
in some case.

This patch checks the FE bit before calling hns3_add_frag()
after processing the first BD of a SKB and adjust the location
of memcpy() to reduce duplication.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotipc: make legacy address flag readable over netlink
John Rutherford [Thu, 19 Dec 2019 05:03:57 +0000 (16:03 +1100)]
tipc: make legacy address flag readable over netlink

To enable iproute2/tipc to generate backwards compatible
printouts and validate command parameters for nodes using a
<z.c.n> node address, it needs to be able to read the legacy
address flag from the kernel.  The legacy address flag records
the way in which the node identity was originally specified.

The legacy address flag is requested by the netlink message
TIPC_NL_ADDR_LEGACY_GET.  If the flag is set the attribute
TIPC_NLA_NET_ADDR_LEGACY is set in the return message.

Signed-off-by: John Rutherford <john.rutherford@dektech.com.au>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'VSOCK-add-vsock_test-test-suite'
David S. Miller [Sat, 21 Dec 2019 05:09:21 +0000 (21:09 -0800)]
Merge branch 'VSOCK-add-vsock_test-test-suite'

Stefano Garzarella says:

====================
VSOCK: add vsock_test test suite

The vsock_diag.ko module already has a test suite but the core AF_VSOCK
functionality has no tests. This patch series adds several test cases that
exercise AF_VSOCK SOCK_STREAM socket semantics (send/recv, connect/accept,
half-closed connections, simultaneous connections).

The v1 of this series was originally sent by Stefan.

v3:
- Patch 6:
  * check the byte received in the recv_byte()
  * use send(2)/recv(2) instead of write(2)/read(2) to test also flags
    (e.g. MSG_PEEK)
- Patch 8:
  * removed unnecessary control_expectln("CLOSED") [Stefan].
- removed patches 9,10,11 added in the v2
- new Patch 9 add parameters to list and skip tests (e.g. useful for vmci
  that doesn't support half-closed socket in the host)
- new Patch 10 prints a list of options in the help
- new Patch 11 tests MSG_PEEK flags of recv(2)

v2: https://patchwork.ozlabs.org/cover/1140538/
v1: https://patchwork.ozlabs.org/cover/847998/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovsock_test: add SOCK_STREAM MSG_PEEK test
Stefano Garzarella [Wed, 18 Dec 2019 18:07:08 +0000 (19:07 +0100)]
vsock_test: add SOCK_STREAM MSG_PEEK test

Test if the MSG_PEEK flags of recv(2) works as expected.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotesting/vsock: print list of options and description
Stefano Garzarella [Wed, 18 Dec 2019 18:07:07 +0000 (19:07 +0100)]
testing/vsock: print list of options and description

Since we now have several options, in the help we print the list
of all supported options and a brief description of them.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agotesting/vsock: add parameters to list and skip tests
Stefano Garzarella [Wed, 18 Dec 2019 18:07:06 +0000 (19:07 +0100)]
testing/vsock: add parameters to list and skip tests

Some tests can fail with transports that have a slightly
different behavior, so let's add the possibility to specify
which tests to skip.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agovsock_test: wait for the remote to close the connection
Stefano Garzarella [Wed, 18 Dec 2019 18:07:05 +0000 (19:07 +0100)]
vsock_test: wait for the remote to close the connection

Before check if a send returns -EPIPE, we need to make sure the
connection is closed.
To do that, we use epoll API to wait EPOLLRDHUP or EPOLLHUP events
on the socket.

Reported-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoVSOCK: add AF_VSOCK test cases
Stefan Hajnoczi [Wed, 18 Dec 2019 18:07:04 +0000 (19:07 +0100)]
VSOCK: add AF_VSOCK test cases

The vsock_test.c program runs a test suite of AF_VSOCK test cases.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoVSOCK: add send_byte()/recv_byte() test utilities
Stefan Hajnoczi [Wed, 18 Dec 2019 18:07:03 +0000 (19:07 +0100)]
VSOCK: add send_byte()/recv_byte() test utilities

Test cases will want to transfer data.  This patch adds utility
functions to do this.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoVSOCK: add full barrier between test cases
Stefan Hajnoczi [Wed, 18 Dec 2019 18:07:02 +0000 (19:07 +0100)]
VSOCK: add full barrier between test cases

See code comment for details.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoVSOCK: extract connect/accept functions from vsock_diag_test.c
Stefan Hajnoczi [Wed, 18 Dec 2019 18:07:01 +0000 (19:07 +0100)]
VSOCK: extract connect/accept functions from vsock_diag_test.c

Many test cases will need to connect to the server or accept incoming
connections.  This patch extracts these operations into utility
functions that can be reused.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoVSOCK: extract utility functions from vsock_diag_test.c
Stefan Hajnoczi [Wed, 18 Dec 2019 18:07:00 +0000 (19:07 +0100)]
VSOCK: extract utility functions from vsock_diag_test.c

Move useful functions into a separate file in preparation for more
vsock test programs.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoVSOCK: add SPDX identifiers to vsock tests
Stefan Hajnoczi [Wed, 18 Dec 2019 18:06:59 +0000 (19:06 +0100)]
VSOCK: add SPDX identifiers to vsock tests

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoVSOCK: fix header include in vsock_diag_test
Stefan Hajnoczi [Wed, 18 Dec 2019 18:06:58 +0000 (19:06 +0100)]
VSOCK: fix header include in vsock_diag_test

The vsock_diag_test program directly included ../../../include/uapi/
headers from the source tree.  Tests are supposed to use the
usr/include/linux/ headers that have been prepared with make
headers_install instead.

Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'add-dsa-switch-support-for-ar9331'
David S. Miller [Sat, 21 Dec 2019 01:05:47 +0000 (17:05 -0800)]
Merge branch 'add-dsa-switch-support-for-ar9331'

Oleksij Rempel says:

====================
add dsa switch support for ar9331

changes v6:
- remove ag71xx changes from this patch set. It needs more work.
- ar9331: fix register definition and add ASCII art switch documentation.

changes v6:
- rebase against net-next

changes v5:
- remote support for port5. The effort of using this port is
  questionable. Currently, it is better to not use it at all, then
  adding buggy support.
- remove port enable call back. There is nothing what we actually need
  to enable.
- rebase it against v5.5-rc1

changes v4:
- ag71xx: ag71xx_mac_validate fix always false comparison (&& -> ||)
- tag_ar9331: use skb_pull_rcsum() instead of skb_pull().
- tag_ar9331: drop skb_set_mac_header()

changes v3:
- ag71xx: ag71xx_mac_config: ignore MLO_AN_INBAND mode. It is not
  supported by HW and SW.
- ag71xx: ag71xx_mac_validate: return all supported bits on
  PHY_INTERFACE_MODE_NA

changes v2:
- move Atheros AR9331 TAG format to separate patch
- use netdev_warn_once in the tag driver to reduce potential message spam
- typo fixes
- reorder tag driver alphabetically
- configure switch to maximal frame size
- use mdiobus_read/write
- fail if mdio sub node is not found
- add comment for post reset state
- remove deprecated comment about device id
- remove phy-handle option for node with fixed-link
- ag71xx: set 1G support only for GMII mode

This patch series provides dsa switch support for Atheros ar9331 WiSoC.
As side effect ag71xx needed to be ported to phylink to make the switch
driver (as well phylink based) work properly.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: add support for Atheros AR9331 built-in switch
Oleksij Rempel [Wed, 18 Dec 2019 08:02:15 +0000 (09:02 +0100)]
net: dsa: add support for Atheros AR9331 built-in switch

Provide basic support for Atheros AR9331 built-in switch. So far it
works as port multiplexer without any hardware offloading support.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: dsa: add support for Atheros AR9331 TAG format
Oleksij Rempel [Wed, 18 Dec 2019 08:02:14 +0000 (09:02 +0100)]
net: dsa: add support for Atheros AR9331 TAG format

Add support for tag format used in Atheros AR9331 built-in switch.

Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMIPS: ath79: ar9331: add ar9331-switch node
Oleksij Rempel [Wed, 18 Dec 2019 08:02:13 +0000 (09:02 +0100)]
MIPS: ath79: ar9331: add ar9331-switch node

Add switch node supported by dsa ar9331 driver.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agodt-bindings: net: dsa: qca, ar9331 switch documentation
Oleksij Rempel [Wed, 18 Dec 2019 08:02:12 +0000 (09:02 +0100)]
dt-bindings: net: dsa: qca, ar9331 switch documentation

Atheros AR9331 has built-in 5 port switch. The switch can be configured
to use all 5 or 4 ports. One of built-in PHYs can be used by first built-in
ethernet controller or to be used directly by the switch over second ethernet
controller.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoNFC: pn544: Adjust indentation in pn544_hci_check_presence
Nathan Chancellor [Wed, 18 Dec 2019 01:21:52 +0000 (18:21 -0700)]
NFC: pn544: Adjust indentation in pn544_hci_check_presence

Clang warns

../drivers/nfc/pn544/pn544.c:696:4: warning: misleading indentation;
statement is not part of the previous 'if' [-Wmisleading-indentation]
                 return nfc_hci_send_cmd(hdev, NFC_HCI_RF_READER_A_GATE,
                 ^
../drivers/nfc/pn544/pn544.c:692:3: note: previous statement is here
                if (target->nfcid1_len != 4 && target->nfcid1_len != 7 &&
                ^
1 warning generated.

This warning occurs because there is a space after the tab on this line.
Remove it so that the indentation is consistent with the Linux kernel
coding style and clang no longer warns.

Fixes: da052850b911 ("NFC: Add pn544 presence check for different targets")
Link: https://github.com/ClangBuiltLinux/linux/issues/814
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'bcmgenet-Turn-on-offloads-by-default'
David S. Miller [Fri, 20 Dec 2019 02:11:10 +0000 (18:11 -0800)]
Merge branch 'bcmgenet-Turn-on-offloads-by-default'

Doug Berger says:

====================
net: bcmgenet: Turn on offloads by default

This commit stack is based on Florian's commit 4e8aedfe78c7 ("net:
systemport: Turn on offloads by default") and enables the offloads for
the bcmgenet driver by default.

The first commit adds support for the HIGHDMA feature to the driver.

The second converts the Tx checksum implementation to use the generic
hardware logic rather than the deprecated IP centric methods.

The third modifies the Rx checksum implementation to use the hardware
offload to compute the complete checksum rather than filtering out bad
packets detected by the hardware's IP centric implementation. This may
increase processing load by passing bad packets to the network stack,
but it provides for more flexible handling of packets by the network
stack without requiring software computation of the checksum.

The remaining commits mirror the extensions Florian made to the sysport
driver to retain symmetry with that driver and to make the benefits of
the hardware offloads more ubiquitous.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: bcmgenet: Add software counters to track reallocations
Doug Berger [Wed, 18 Dec 2019 00:51:15 +0000 (16:51 -0800)]
net: bcmgenet: Add software counters to track reallocations

When inserting the TSB, keep track of how many times we had to do
it and if there was a failure in doing so, this helps profile the
driver for possibly incorrect headroom settings.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: bcmgenet: Be drop monitor friendly while re-allocating headroom
Doug Berger [Wed, 18 Dec 2019 00:51:14 +0000 (16:51 -0800)]
net: bcmgenet: Be drop monitor friendly while re-allocating headroom

During bcmgenet_put_tx_csum() make sure we differentiate a SKB
headroom re-allocation failure from the normal swap and replace
path.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: bcmgenet: Turn on offloads by default
Doug Berger [Wed, 18 Dec 2019 00:51:13 +0000 (16:51 -0800)]
net: bcmgenet: Turn on offloads by default

We can turn on the RX/TX checksum offloads and the scatter/gather
features by default and make sure that those are properly reflected
back to e.g: stacked devices such as VLAN.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: bcmgenet: Utilize bcmgenet_set_features() during resume/open
Doug Berger [Wed, 18 Dec 2019 00:51:12 +0000 (16:51 -0800)]
net: bcmgenet: Utilize bcmgenet_set_features() during resume/open

During driver resume and open, the HW may have lost its context/state,
utilize bcmgenet_set_features() to make sure we do restore the correct
set of features that were previously configured.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: bcmgenet: Refactor bcmgenet_set_features()
Doug Berger [Wed, 18 Dec 2019 00:51:11 +0000 (16:51 -0800)]
net: bcmgenet: Refactor bcmgenet_set_features()

In preparation for unconditionally enabling TX and RX checksum
offloads, refactor bcmgenet_set_features() a bit such that
__netdev_update_features() during register_netdev() can make sure
that features are correctly programmed during network device
registration.

Since we can now be called during register_netdev() with clocks
gated, we need to temporarily turn them on/off in order to have a
successful register programming.

We also move the CRC forward setting read into
bcmgenet_set_features() since priv->crc_fwd_en matters while
turning on RX checksum offload, that way we are guaranteed they
are in sync in case we ever add support for NETIF_F_RXFCS at some
point in the future.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: bcmgenet: use CHECKSUM_COMPLETE for NETIF_F_RXCSUM
Doug Berger [Wed, 18 Dec 2019 00:51:10 +0000 (16:51 -0800)]
net: bcmgenet: use CHECKSUM_COMPLETE for NETIF_F_RXCSUM

This commit updates the Rx checksum offload behavior of the driver
to use the more generic CHECKSUM_COMPLETE method that supports all
protocols over the CHECKSUM_UNNECESSARY method that only applies
to some protocols known by the hardware.

This behavior is perceived to be superior.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: bcmgenet: enable NETIF_F_HW_CSUM feature
Doug Berger [Wed, 18 Dec 2019 00:51:09 +0000 (16:51 -0800)]
net: bcmgenet: enable NETIF_F_HW_CSUM feature

The GENET hardware should be capable of generating IP checksums
using the NETIF_F_HW_CSUM feature, so switch to using that feature
instead of the depricated NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: bcmgenet: enable NETIF_F_HIGHDMA flag
Doug Berger [Wed, 18 Dec 2019 00:51:08 +0000 (16:51 -0800)]
net: bcmgenet: enable NETIF_F_HIGHDMA flag

This commit configures the DMA masks for the GENET driver and
sets the NETIF_F_HIGHDMA flag to report support of the feature.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: systemport: Set correct DMA mask
Florian Fainelli [Wed, 18 Dec 2019 00:29:50 +0000 (16:29 -0800)]
net: systemport: Set correct DMA mask

SYSTEMPORT is capabable of doing up to 40-bit of physical addresses, set
an appropriate DMA mask to permit that.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'nfp-tls-implement-the-stream-sync-RX-resync'
David S. Miller [Fri, 20 Dec 2019 01:46:51 +0000 (17:46 -0800)]
Merge branch 'nfp-tls-implement-the-stream-sync-RX-resync'

Jakub Kicinski says:

====================
nfp: tls: implement the stream sync RX resync

This small series adds support for using the device
in stream scan RX resync mode which improves the RX
resync success rate. Without stream scan it's pretty
much impossible to successfully resync a continuous
stream.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: tls: implement the stream sync RX resync
Jakub Kicinski [Tue, 17 Dec 2019 22:12:02 +0000 (14:12 -0800)]
nfp: tls: implement the stream sync RX resync

The simple RX resync strategy controlled by the kernel does not
guarantee as good results as if the device helps by detecting
the potential record boundaries and keeping track of them.

We've called this strategy stream scan in the tls-offload doc.

Implement this strategy for the NFP. The device sends a request
for record boundary confirmation, which is then recorded in
per-TLS socket state and responded to once record is reached.
Because the device keeps track of records passing after the
request was sent the response is not as latency sensitive as
when kernel just tries to tell the device the information
about the next record.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet/tls: add helper for testing if socket is RX offloaded
Jakub Kicinski [Tue, 17 Dec 2019 22:12:01 +0000 (14:12 -0800)]
net/tls: add helper for testing if socket is RX offloaded

There is currently no way for driver to reliably check that
the socket it has looked up is in fact RX offloaded. Add
a helper. This allows drivers to catch misbehaving firmware.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: pass packet pointer to nfp_net_parse_meta()
Jakub Kicinski [Tue, 17 Dec 2019 22:12:00 +0000 (14:12 -0800)]
nfp: pass packet pointer to nfp_net_parse_meta()

Make nfp_net_parse_meta() take a packet pointer and return
a drop/no drop decision. Right now it returns the end of
metadata and caller compares it to the packet pointer.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'nfp-ipv6-tunnel'
David S. Miller [Fri, 20 Dec 2019 01:37:13 +0000 (17:37 -0800)]
Merge branch 'nfp-ipv6-tunnel'

John Hurley says:

====================
Add ipv6 tunnel support to NFP

The following patches add support for IPv6 tunnel offload to the NFP
driver.

Patches 1-2 do some code tidy up and prepare existing code for reuse in
IPv6 tunnels.
Patches 3-4 handle IPv6 tunnel decap (match) rules.
Patches 5-8 handle encap (action) rules.
Patch 9 adds IPv6 support to the merge and pre-tunnel rule functions.

v1->v2:
- fix compiler warning when building without CONFIG_IPV6 set -
  Jakub Kicinski (patch 7)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: update flow merge code to support IPv6 tunnels
John Hurley [Tue, 17 Dec 2019 21:57:24 +0000 (21:57 +0000)]
nfp: flower: update flow merge code to support IPv6 tunnels

Both pre-tunnel match rules and flow merge functions parse compiled
match/action fields for validation.

Update these validation functions to include IPv6 match and action fields.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: support ipv6 tunnel keep-alive messages from fw
John Hurley [Tue, 17 Dec 2019 21:57:23 +0000 (21:57 +0000)]
nfp: flower: support ipv6 tunnel keep-alive messages from fw

FW sends an update of IPv6 tunnels that are active in a given period. Use
this information to update the kernel table so that neighbour entries do
not time out when active on the NIC.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: handle notifiers for ipv6 route changes
John Hurley [Tue, 17 Dec 2019 21:57:22 +0000 (21:57 +0000)]
nfp: flower: handle notifiers for ipv6 route changes

A notifier is used to track route changes in the kernel. If a change is
made to a route that is offloaded to fw then an update is sent to the NIC.
The driver tracks all routes that are offloaded to determine if a kernel
change is of interest.

Extend the notifier to track IPv6 route changes and create a new list that
stores offloaded IPv6 routes. Modify the IPv4 route helper functions to
accept varying address lengths. This way, the same core functions can be
used to handle IPv4 and IPv6.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: handle ipv6 tunnel no neigh request
John Hurley [Tue, 17 Dec 2019 21:57:21 +0000 (21:57 +0000)]
nfp: flower: handle ipv6 tunnel no neigh request

When fw does not know the next hop for an IPv6 tunnel, it sends a request
to the driver.

Handle this request by doing a route lookup on the IPv6 address and
offloading the next hop to the fw neighbour table.

Similar functions already exist to handle IPv4 no neighbour requests. To
avoid confusion, append these functions with the _ipv4 tag. There is no
change in functionality with this.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: modify pre-tunnel and set tunnel action for ipv6
John Hurley [Tue, 17 Dec 2019 21:57:20 +0000 (21:57 +0000)]
nfp: flower: modify pre-tunnel and set tunnel action for ipv6

The IPv4 set tunnel action allows the setting of tunnel metadata such as
the TTL and ToS values. The pre-tunnel action includes the destination IP
address and is used to calculate the next hop from from the neighbour
table.

Much of the IPv4 tunnel actions can be reused for IPv6 tunnels. Change the
names of associated functions and structs to remove the IPv4 identifier
and make minor modifcations to support IPv6 tunnel actions.

Ensure the pre-tunnel action contains the IPv6 address along with an
identifying flag when an IPv6 tunnel action is required.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: offload list of IPv6 tunnel endpoint addresses
John Hurley [Tue, 17 Dec 2019 21:57:19 +0000 (21:57 +0000)]
nfp: flower: offload list of IPv6 tunnel endpoint addresses

Fw requires a list of IPv6 addresses that are used as tunnel endpoints to
enable correct decap of tunneled packets.

Store a list of IPv6 endpoints used in rules with a ref counter to track
how many times it is in use. Offload the entire list any time a new IPv6
address is added or when an address is removed (ref count is 0).

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: compile match for IPv6 tunnels
John Hurley [Tue, 17 Dec 2019 21:57:18 +0000 (21:57 +0000)]
nfp: flower: compile match for IPv6 tunnels

IPv6 tunnel matches are now supported by firmware. Modify the NFP driver
to compile these match rules. IPv6 matches are handled similar to IPv4
tunnels with the difference the address length. The type of tunnel is
indicated by the same bitmap that is used in IPv4 with an extra bit
signifying that the IPv6 variation should be used.

Only compile IPv6 tunnel matches when the fw features symbol indicated
that they are compatible with the currently loaded fw.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: move udp tunnel key match compilation to helper function
John Hurley [Tue, 17 Dec 2019 21:57:17 +0000 (21:57 +0000)]
nfp: flower: move udp tunnel key match compilation to helper function

IPv4 UDP and GRE tunnel match rule compile helpers share functions for
compiling fields such as IP addresses. However, they handle fields such
tunnel IDs differently.

Create new helper functions for compiling GRE and UDP tunnel key data.
This is in preparation for supporting IPv6 tunnels where these new
functions can be reused.

This patch does not change functionality.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonfp: flower: pass flow rule pointer directly to match functions
John Hurley [Tue, 17 Dec 2019 21:57:16 +0000 (21:57 +0000)]
nfp: flower: pass flow rule pointer directly to match functions

In kernel 5.1, the flow offload API was introduced along with a helper
function to extract the flow_rule from the TC offload struct. Each of the
match helper functions are passed the offload struct and extract the flow
rule to a local variable.

Simplify the code while also removing the extra compat and local variable
calls by extracting the rule once in the main match handler, and passing
a reference to the rule direct to each helper.

This patch does not change driver functionality.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agohdlcdrv: replace unnecessary assertion in hdlcdrv_register
Aditya Pakki [Tue, 17 Dec 2019 21:06:19 +0000 (15:06 -0600)]
hdlcdrv: replace unnecessary assertion in hdlcdrv_register

In hdlcdrv_register, failure to register the driver causes a crash.
The three callers of hdlcdrv_register all pass valid pointers and
do not fail. The patch eliminates the unnecessary BUG_ON assertion.

Signed-off-by: Aditya Pakki <pakki001@umn.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: mvpp2: cycle comphy to power it down
Russell King [Tue, 17 Dec 2019 15:47:36 +0000 (15:47 +0000)]
net: mvpp2: cycle comphy to power it down

Presently, at boot time, the comphys are enabled. For firmware
compatibility reasons, the comphy driver does not power down the
comphys at boot. Consequently, the ethernet comphys are left active
until the network interfaces are brought through an up/down cycle.

If the port is never used, the port wastes power needlessly. Arrange
for the ethernet comphys to be cycled by the mvpp2 driver as if the
interface went through an up/down cycle during driver probe, thereby
powering them down.

This saves:
  270mW per 10G SFP+ port on the Macchiatobin Single Shot (eth0/eth1)
  370mW per 10G PHY port on the Macchiatobin Double Shot (eth0/eth1)
  160mW on the SFP port on either Macchiatobin flavour (eth3)

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: sfp: report error on failure to read sfp soft status
Russell King [Tue, 17 Dec 2019 13:50:29 +0000 (13:50 +0000)]
net: sfp: report error on failure to read sfp soft status

Report a rate-limited error if we fail to read the SFP soft status,
and preserve the current status in that case. This avoids I2C bus
errors from triggering a link flap.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'phylib-consolidation'
David S. Miller [Thu, 19 Dec 2019 20:52:34 +0000 (12:52 -0800)]
Merge branch 'phylib-consolidation'

Russell King says:

====================
phylib consolidation

Over the last few releases, there has been a push to clean up and
consolidate the phylib code. Some cases have been missed, and this
series catches those cases.

1. Remove redundant .aneg_done initialisers; calling genphy_aneg_done()
   for clause 22 PHYs is the default when .aneg_done is not set.

2. Some PHY drivers manually set phydev->pause and phydev->asym_pause,
   but we have a helper for this - phy_resolve_aneg_pause(), introduced
   in 2d880b8709c0 ("net: phy: extract pause mode").  Use this in the
   lxt, marvell and uPD60620 drivers.

   Incidentally, this brings up the question whether marvell fiber mode
   is correctly interpreting and advertising the pause parameters.

3. Add a genphy_check_and_restart_aneg() helper, which complements the
   clause 45 version of this. This will be useful for PHY drivers that
   open code this logic (e.g. marvell.c)

4. Add a genphy_read_status_fixed() helper to read the fixed-mode
   status from a clause 22 PHY.  lxt and marvell both contain copies
   of this code, so convert them over.

5. Arrange marvell driver to use genphy_read_lpa() for copper mode.
   This needs some rearrangement of the code in
   marvell_read_status_page_an(), but preserves using the PHY specific
   status register to derive the current negotiation results.

6. Simplify the marvell driver so we can use the
   genphy_read_status_fixed() helper directly rather than
   marvell_read_status_page_fixed().

7. Use positive logic in the marvell driver to determine the link
   state, and get rid of the REGISTER_LINK_STATUS definition; we
   already have a definition for this.

8. The marvell driver reads the PHY specific status register multiple
   times when determining the status: once in marvell_update_link()
   and again in marvell_read_status_page_an(). This is a waste;
   rearrange to read the status register once, and pass its value into
   marvell_read_status_page_an().  We preserve using
   genphy_update_link() for the copper side.

9. The marvell driver was using private clause 37 definitions, but we
   have clause 37 definitions in uapi/linux/mii.h. Use the generic
   definitions.

10. Switch the marvell driver to use phy_modify_changed() to modify
    the fiber advertisement.

11. Switch the marvell driver to use genphy_check_and_restart_aneg()
    introduced above rather than open-coding this functionality.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: use genphy_check_and_restart_aneg()
Russell King [Tue, 17 Dec 2019 13:39:52 +0000 (13:39 +0000)]
net: phy: marvell: use genphy_check_and_restart_aneg()

Use the helper to check and restart autonegotiation for the marvell
fiber page negotiation setting.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: use phy_modify_changed()
Russell King [Tue, 17 Dec 2019 13:39:47 +0000 (13:39 +0000)]
net: phy: marvell: use phy_modify_changed()

Use phy_modify_changed() to change the fiber advertisement register
rather than open coding this functionality.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: use existing clause 37 definitions
Russell King [Tue, 17 Dec 2019 13:39:42 +0000 (13:39 +0000)]
net: phy: marvell: use existing clause 37 definitions

Use existing clause 37 advertising/link partner definitions rather than
private ones for the advertisement registers.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: consolidate phy status reading
Russell King [Tue, 17 Dec 2019 13:39:36 +0000 (13:39 +0000)]
net: phy: marvell: consolidate phy status reading

marvell_read_status_page_an() always reads the PHY status register, but
marvell_update_link() has already done this.  Rather than wastefully
reading the register twice in quick succession, read it once in
marvell_read_status_page() and use the result for both.

This makes marvell_update_link() rather pointless, so move it into
marvell_read_status_page().

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: use positive logic for link state
Russell King [Tue, 17 Dec 2019 13:39:31 +0000 (13:39 +0000)]
net: phy: marvell: use positive logic for link state

Rather than using negative logic:

if (there is no link)
set link = 0
else
set link = 1

use the more natural positive logic:

if (there is link)
set link = 1
else
set link = 0

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: initialise link partner state earlier
Russell King [Tue, 17 Dec 2019 13:39:26 +0000 (13:39 +0000)]
net: phy: marvell: initialise link partner state earlier

Move the initialisation of the link partner state earlier, inside
marvell_read_status_page(), so we don't have the same initialisation
scattered amongst the other files.  This is in a similar place to
the genphy implementation, so would result in the same behaviour if
a PHY read error occurs.

This allows us to get rid of marvell_read_status_page_fixed(), which
became a pointless wrapper around genphy_read_status_fixed().

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: marvell: rearrange to use genphy_read_lpa()
Russell King [Tue, 17 Dec 2019 13:39:21 +0000 (13:39 +0000)]
net: phy: marvell: rearrange to use genphy_read_lpa()

Rearrange the Marvell PHY driver to use genphy_read_lpa() rather than
open-coding this functionality.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: provide and use genphy_read_status_fixed()
Russell King [Tue, 17 Dec 2019 13:39:16 +0000 (13:39 +0000)]
net: phy: provide and use genphy_read_status_fixed()

There are two drivers and generic code which contain exactly the same
code to read the status of a PHY operating without autonegotiation
enabled. Rather than duplicate this code, provide a helper to read
this information.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: add genphy_check_and_restart_aneg()
Russell King [Tue, 17 Dec 2019 13:39:11 +0000 (13:39 +0000)]
net: phy: add genphy_check_and_restart_aneg()

Add a helper for restarting autonegotiation(), similar to the clause 45
variant.  Use it in __genphy_config_aneg()

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: use phy_resolve_aneg_pause()
Russell King [Tue, 17 Dec 2019 13:39:06 +0000 (13:39 +0000)]
net: phy: use phy_resolve_aneg_pause()

Several drivers code their own version of this, working from the LPA
register, after setting the ethtool link partner advertisement bitmask.
Use the generic function instead.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: phy: remove redundant .aneg_done initialisers
Russell King [Tue, 17 Dec 2019 13:39:00 +0000 (13:39 +0000)]
net: phy: remove redundant .aneg_done initialisers

Remove initialisers that set .aneg_done to genphy_aneg_done - this is
the default for clause 22 PHYs, so the initialiser is redundant.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge ath-next from git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/ath.git
Kalle Valo [Thu, 19 Dec 2019 16:27:36 +0000 (18:27 +0200)]
Merge ath-next from git://git./linux/kernel/git/kvalo/ath.git

ath.git patches for v5.6. Major changes:

wil6210

* support set_multicast_to_unicast cfg80211 operation

* support set_cqm_rssi_config cfg80211 operation

wcn36xx

* disable HW_CONNECTION_MONITOR as firmware is buggy

4 years agoath11k: Use sizeof_field() instead of FIELD_SIZEOF()
Kees Cook [Thu, 19 Dec 2019 00:58:02 +0000 (16:58 -0800)]
ath11k: Use sizeof_field() instead of FIELD_SIZEOF()

The FIELD_SIZEOF() macro was redundant, and is being removed from the
kernel. Since commit c593642c8be0 ("treewide: Use sizeof_field() macro")
this is one of the last users of the old macro, so replace it.

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agoath11k: explicitly cast wmi commands to their correct struct type
John Crispin [Thu, 5 Dec 2019 06:26:45 +0000 (07:26 +0100)]
ath11k: explicitly cast wmi commands to their correct struct type

Three of the WMI command handlers were not casting to the right data type.
Lets make the code consistent with the other handlers.

Signed-off-by: John Crispin <john@phrozen.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agowil6210: add support for set_cqm_rssi_config
Dedy Lansky [Wed, 18 Dec 2019 18:10:24 +0000 (20:10 +0200)]
wil6210: add support for set_cqm_rssi_config

set_cqm_rssi_config() is used by the kernel to configure connection
quality monitor RSSI threshold.
wil6210 uses WMI_SET_LINK_MONITOR_CMDID to set the RSSI threshold to
FW which in turn reports RSSI threshold changes with
WMI_LINK_MONITOR_EVENTID.

Signed-off-by: Dedy Lansky <dlansky@codeaurora.org>
Signed-off-by: Maya Erez <merez@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agowil6210: support set_multicast_to_unicast cfg80211 operation
Ahmad Masri [Wed, 18 Dec 2019 18:10:21 +0000 (20:10 +0200)]
wil6210: support set_multicast_to_unicast cfg80211 operation

Wil6210 AP has a separate ring for transmitting multicast packets,
multicast packets are transmitted without an ack from the receiver side.
Therefore, 802.11 spec defines some low MCS rates for multicat packets.
However, there is no guarantee that these packets were really received
and handled on the client side.

Some applications that rely on multicast packets, may prefer to
transmit these packets as a unicast to ensure reliability, and also
to ensure better performance with high MCS rates.
multicast to unicast is done by duplicating multicast packets to all
clients and changing the DA (multicast) to the MAC address of the
client.
see NL80211_CMD_SET_MULTICAST_TO_UNICAST for more info.

Signed-off-by: Ahmad Masri <amasri@codeaurora.org>
Signed-off-by: Maya Erez <merez@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agowil6210: fix MID valid bits in Rx status message
Dedy Lansky [Wed, 18 Dec 2019 18:10:18 +0000 (20:10 +0200)]
wil6210: fix MID valid bits in Rx status message

Fix incorrect definitions of MAC ID bits inside Rx status message.

Signed-off-by: Dedy Lansky <dlansky@codeaurora.org>
Signed-off-by: Maya Erez <merez@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agowil6210: reduce ucode_debug memory region
Dedy Lansky [Wed, 18 Dec 2019 18:10:16 +0000 (20:10 +0200)]
wil6210: reduce ucode_debug memory region

ucode_debug memory region defined as 4K bytes. Fix this according to
Talyn device memory map.

Signed-off-by: Dedy Lansky <dlansky@codeaurora.org>
Signed-off-by: Maya Erez <merez@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agowil6210: add verification for cid upper bound
Alexei Avshalom Lazar [Wed, 18 Dec 2019 18:10:13 +0000 (20:10 +0200)]
wil6210: add verification for cid upper bound

max_assoc_sta can receive values (from the user or from the FW)
that are higher than WIL6210_MAX_CID.
Verify that cid doesn't exceed the upper bound of WIL6210_MAX_CID.

Signed-off-by: Alexei Avshalom Lazar <ailizaro@codeaurora.org>
Signed-off-by: Maya Erez <merez@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agowil6210: take mem_lock for writing in crash dump collection
Alexei Avshalom Lazar [Wed, 18 Dec 2019 18:10:10 +0000 (20:10 +0200)]
wil6210: take mem_lock for writing in crash dump collection

On some crash dump cases mem_lock is already taken, error
returns and crash dump copy fails.
In this case wait until mem_lock available instead of failing
the operation.
Also take the mem_lock for writing to prevent other threads from
altering the state of the device while collecting crash dump.

Signed-off-by: Alexei Avshalom Lazar <ailizaro@codeaurora.org>
Signed-off-by: Maya Erez <merez@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agowil6210: minimize the time that mem_lock is held
Alexei Avshalom Lazar [Wed, 18 Dec 2019 18:10:07 +0000 (20:10 +0200)]
wil6210: minimize the time that mem_lock is held

mem_lock is taken for the entire wil_reset().
Optimize this by taking mem_lock just before device is
being reset and release the lock after FW download.

Signed-off-by: Alexei Avshalom Lazar <ailizaro@codeaurora.org>
Signed-off-by: Maya Erez <merez@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agowil6210: dump Rx status message on errors
Ahmad Masri [Wed, 18 Dec 2019 18:10:04 +0000 (20:10 +0200)]
wil6210: dump Rx status message on errors

Dump all the Rx status message on different errors to allow more
visibility of the case.

Signed-off-by: Ahmad Masri <amasri@codeaurora.org>
Signed-off-by: Maya Erez <merez@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
4 years agonet: stmmac: tc: Fix TAPRIO division operation
Jose Abreu [Wed, 18 Dec 2019 22:55:01 +0000 (23:55 +0100)]
net: stmmac: tc: Fix TAPRIO division operation

For ARCHs that don't support 64 bits division we need to use the
helpers.

Fixes: b60189e0392f ("net: stmmac: Integrate EST with TAPRIO scheduler API")
Signed-off-by: Jose Abreu <Jose.Abreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 'ETS-qdisc'
David S. Miller [Wed, 18 Dec 2019 21:32:30 +0000 (13:32 -0800)]
Merge branch 'ETS-qdisc'

Petr Machata says:

====================
Add a new Qdisc, ETS

The IEEE standard 802.1Qaz (and 802.1Q-2014) specifies four principal
transmission selection algorithms: strict priority, credit-based shaper,
ETS (bandwidth sharing), and vendor-specific. All these have their
corresponding knobs in DCB. But DCB does not have interfaces to configure
RED and ECN, unlike Qdiscs.

In the Qdisc land, strict priority is implemented by PRIO. Credit-based
transmission selection algorithm can then be modeled by having e.g. TBF or
CBS Qdisc below some of the PRIO bands. ETS would then be modeled by
placing a DRR Qdisc under the last PRIO band.

The problem with this approach is that DRR on its own, as well as the
combination of PRIO and DRR, are tricky to configure and tricky to offload
to 802.1Qaz-compliant hardware. This is due to several reasons:

- As any classful Qdisc, DRR supports adding classifiers to decide in which
  class to enqueue packets. Unlike PRIO, there's however no fallback in the
  form of priomap. A way to achieve classification based on packet priority
  is e.g. like this:

    # tc filter add dev swp1 root handle 1: \
basic match 'meta(priority eq 0)' flowid 1:10

  Expressing the priomap in this manner however forces drivers to deep dive
  into the classifier block to parse the individual rules.

  A possible solution would be to extend the classes with a "defmap" a la
  split / defmap mechanism of CBQ, and introduce this as a last resort
  classification. However, unlike priomap, this doesn't have the guarantee
  of covering all priorities. Traffic whose priority is not covered is
  dropped by DRR as unclassified. But ASICs tend to implement dropping in
  the ACL block, not in scheduling pipelines. The need to treat these
  configurations correctly (if only to decide to not offload at all)
  complicates a driver.

  It's not clear how to retrofit priomap with all its benefits to DRR
  without changing it beyond recognition.

- The interplay between PRIO and DRR is also causing problems. 802.1Qaz has
  all ETS TCs as a last resort. Switch ASICs that support ETS at all are
  likely to handle ETS traffic this way as well. However, the Linux model
  is more generic, allowing the DRR block in any band. Drivers would need
  to be careful to handle this case correctly, otherwise the offloaded
  model might not match the slow-path one.

  In a similar vein, PRIO and DRR need to agree on the list of priorities
  assigned to DRR. This is doubly problematic--the user needs to take care
  to keep the two in sync, and the driver needs to watch for any holes in
  DRR coverage and treat the traffic correctly, as discussed above.

  Note that at the time that DRR Qdisc is added, it has no classes, and
  thus any priorities assigned to that PRIO band are not covered. Thus this
  case is surprisingly rather common, and needs to be handled gracefully by
  the driver.

- Similarly due to DRR flexibility, when a Qdisc (such as RED) is attached
  below it, it is not immediately clear which TC the class represents. This
  is unlike PRIO with its straightforward classid scheme. When DRR is
  combined with PRIO, the relationship between classes and TCs gets even
  more murky.

  This is a problem for users as well: the TC mapping is rather important
  for (devlink) shared buffer configuration and (ethtool) counters.

So instead, this patch set introduces a new Qdisc, which is based on
802.1Qaz wording. It is PRIO-like in how it is configured, meaning one
needs to specify how many bands there are, how many are strict and how many
are ETS, quanta for the latter, and priomap.

The new Qdisc operates like the PRIO / DRR combo would when configured as
per the standard. The strict classes, if any, are tried for traffic first.
When there's no traffic in any of the strict queues, the ETS ones (if any)
are treated in the same way as in DRR.

The chosen interface makes the overall system both reasonably easy to
configure, and reasonably easy to offload. The extra code to support ETS in
mlxsw (which already supports PRIO) is about 150 lines, of which perhaps 20
lines is bona fide new business logic.

Credit-based shaping transmission selection algorithm can be configured by
adding a CBS Qdisc under one of the strict bands (e.g. TBF can be used to a
similar effect as well). As a non-work-conserving Qdisc, CBS can't be
hooked under the ETS bands. This is detected and handled identically to DRR
Qdisc at runtime. Note that offloading CBS is not subject of this patchset.

The patchset proceeds in four stages:

- Patches #1-#3 are cleanups.
- Patches #4 and #5 contain the new Qdisc.
- Patches #6 and #7 update mlxsw to offload the new Qdisc.
- Patches #8-#10 add selftests for ETS.

Examples:

- Add a Qdisc with 6 bands, 3 strict and 3 ETS with 45%-30%-25% weights:

    # tc qdisc add dev swp1 root handle 1: \
ets strict 3 quanta 4500 3000 2500 priomap 0 1 1 1 2 3 4 5
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 6 strict 3 quanta 4500 3000 2500 priomap 0 1 1 1 2 3 4 5 5 5 5 5 5 5 5 5

- Tweak quantum of one of the classes of the previous Qdisc:

    # tc class ch dev swp1 classid 1:4 ets quantum 1000
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 6 strict 3 quanta 1000 3000 2500 priomap 0 1 1 1 2 3 4 5 5 5 5 5 5 5 5 5
    # tc class ch dev swp1 classid 1:3 ets quantum 1000
    Error: Strict bands do not have a configurable quantum.

- Purely strict Qdisc with 1:1 mapping between priorities and TCs:

    # tc qdisc add dev swp1 root handle 1: \
ets strict 8 priomap 7 6 5 4 3 2 1 0
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 8 strict 8 priomap 7 6 5 4 3 2 1 0 7 7 7 7 7 7 7 7

- Use "bands" to specify number of bands explicitly. Underspecified bands
  are implicitly ETS and their quantum is taken from MTU. The following
  thus gives each band the same weight:

    # tc qdisc add dev swp1 root handle 1: \
ets bands 8 priomap 7 6 5 4 3 2 1 0
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 8 quanta 1514 1514 1514 1514 1514 1514 1514 1514 priomap 7 6 5 4 3 2 1 0 7 7 7 7 7 7 7 7

v2:
- This addresses points raised by David Miller.
- Patch #4:
    - sch_ets.c: Add a comment with description of the Qdisc and the
      dequeuing algorithm.
    - Kconfig: Add a high-level description to the help blurb.

v1:
- No changes, first upstream submission after RFC.

v3 (internal):
- This addresses review from Jiri Pirko.
- Patch #3:
    - Rename to _HR_ instead of to _HIERARCHY_.
- Patch #4:
    - pkt_sched.h: Keep all the TCA_ETS_ constants in one enum.
    - pkt_sched.h: Rename TCA_ETS_BANDS to _NBANDS, _STRICT to _NSTRICT,
      _BAND_QUANTUM to _QUANTA_BAND and _PMAP_BAND to _PRIOMAP_BAND.
    - sch_ets.c: Update to reflect the above changes. Add a new policy,
      ets_class_policy, which is used when parsing class changes.
      Currently that policy is the same as the quanta policy, but that
      might change.
    - sch_ets.c: Move MTU handling from ets_quantum_parse() to the one
      caller that makes use of it.
    - sch_ets.c: ets_qdisc_priomap_parse(): WARN_ON_ONCE on invalid
      attribute instead of returning an extack.
- Patch #6:
    - __mlxsw_sp_qdisc_ets_replace(): Pass the weights argument to this
      function in this patch already. Drop the weight computation.
    - mlxsw_sp_qdisc_prio_replace(): Rename "quanta" to "zeroes" and
      pass for the abovementioned "weights".
    - mlxsw_sp_qdisc_prio_graft(): Convert to a wrapper around
      __mlxsw_sp_qdisc_ets_graft(), instead of invoking the latter
      directly from mlxsw_sp_setup_tc_prio().
    - Update to follow the _HIERARCHY_ -> _HR_ renaming.
- Patch #7:
    - __mlxsw_sp_qdisc_ets_replace(): The "weights" argument passing and
      weight computation removal are now done in a previous patch.
    - mlxsw_sp_setup_tc_ets(): Drop case TC_ETS_REPLACE, which is handled
      earlier in the function.
- Patch #3 (iproute2):
    - Add an example output to the commit message.
    - tc-ets.8: Fix output of two examples.
    - tc-ets.8: Describe default values of "bands", "quanta".
    - q_ets.c: A number of fixes in error messages.
    - q_ets.c: Comment formatting: /*padding*/ -> /* padding */
    - q_ets.c: parse_nbands: Move duplicate checking to callers.
    - q_ets.c: Don't accept both "quantum" and "quanta" as equivalent.

v2 (internal):
- This addresses review from Ido Schimmel and comments from Alexander
  Kushnarov.
- Patch #2:
    - s/coment/comment in the commit message.
- Patch #4:
    - sch_ets: ets_class_is_strict(), ets_class_id(): Constify an argument
    - ets_class_find(): RXTify
- Patch #3 (iproute2):
    - tc-ets.8: some spelling fixes
    - tc-ets.8: add another example
    - tc.8: add an ETS to "CLASSFUL QDISCS" section

v1 (internal):
- This addresses RFC reviews from Ido Schimmel and Roman Mashak, bugs found
  by Alexander Petrovskiy and myself, and other improvements.
- Patch #2:
    - Expand the explanation with an explicit example.
- Patch #4:
    - Kconfig: s/sch_drr/sch_ets/
    - sch_ets: Reorder includes to be in alphabetical order
    - sch_ets: ets_quantum_parse(): Rename the return-pointer argument
      from pquantum to quantum, and use it directly, not going through a
      local temporary.
    - sch_ets: ets_qdisc_quanta_parse(): Convert syntax of function
      argument "quanta" from an array to a pointer.
    - sch_ets: ets_qdisc_priomap_parse(): Likewise with "priomap".
    - sch_ets: ets_qdisc_quanta_parse(), ets_qdisc_priomap_parse(): Invoke
      __nla_validate_nested directly instead of nl80211_validate_nested().
    - sch_ets: ets_qdisc_quanta_parse(): WARN_ON_ONCE on invalid attribute
      instead of returning an extack.
    - sch_ets: ets_qdisc_change(): Make the last band the default one for
      unmentioned priomap priorities.
    - sch_ets: Fix a panic when an offloaded child in a bandwidth-sharing
      band notified its ETS parent.
    - sch_ets: When ungrafting, add the newly-created invisible FIFO to
      the Qdisc hash
- Patch #5:
    - pkt_cls.h: Note that quantum=0 signifies a strict band.
    - Fix error path handling when ets_offload_dump() fails.
- Patch #6:
    - __mlxsw_sp_qdisc_ets_replace(): Convert syntax of function arguments
      "quanta" and "priomap" from arrays to pointers.
- Patch #7:
    - __mlxsw_sp_qdisc_ets_replace(): Convert syntax of function argument
      "weights" from an array to a pointer.
- Patch #9:
    - mlxsw/sch_ets.sh: Add a comment explaining packet prioritization.
    - Adjust the whole suite to allow testing of traffic classifiers
      in addition to testing priomap.
- Patch #10:
    - Add a number of new tests to test default priomap band, overlarge
      number of bands, zeroes in quanta, and altogether missing quanta.
- Patch #1 (iproute2):
    - State motivation for inclusion of this patch in the patcheset in the
      commit message.
- Patch #3 (iproute2):
    - tc-ets.8: it is now December
    - tc-ets.8: explain inactivity WRT using non-WC Qdiscs under ETS band
    - tc-ets.8: s/flow/band in explanation of quantum
    - tc-ets.8: explain what happens with priorities not covered by priomap
    - tc-ets.8: default priomap band is now the last one
    - q_ets.c: ets_parse_opt(): Remove unnecessary initialization of
      priomap and quanta.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: qdiscs: Add test coverage for ETS Qdisc
Petr Machata [Wed, 18 Dec 2019 14:55:24 +0000 (14:55 +0000)]
selftests: qdiscs: Add test coverage for ETS Qdisc

Add TDC coverage for the new ETS Qdisc.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: forwarding: sch_ets: Add test coverage for ETS Qdisc
Petr Machata [Wed, 18 Dec 2019 14:55:22 +0000 (14:55 +0000)]
selftests: forwarding: sch_ets: Add test coverage for ETS Qdisc

This tests the newly-added ETS Qdisc. It runs two to three streams of
traffic, each with a different priority. ETS Qdisc is supposed to allocate
bandwidth according to the DRR algorithm and given weights. After running
the traffic for a while, counters are compared for each stream to check
that the expected ratio is in fact observed.

In order for the DRR process to kick in, a traffic bottleneck must exist in
the first place. In slow path, such bottleneck can be implemented by
wrapping the ETS Qdisc inside a TBF or other shaper. This might however
make the configuration unoffloadable. Instead, on HW datapath, the
bottleneck would be set up by lowering port speed and configuring shared
buffer suitably.

Therefore the test is structured as a core component that implements the
testing, with two wrapper scripts that implement the details of slow path
resp. fast path configuration.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoselftests: forwarding: Move start_/stop_traffic from mlxsw to lib.sh
Petr Machata [Wed, 18 Dec 2019 14:55:21 +0000 (14:55 +0000)]
selftests: forwarding: Move start_/stop_traffic from mlxsw to lib.sh

These two functions are used for starting several streams of traffic, and
then stopping them later. They will be handy for the test coverage of ETS
Qdisc. Move them from mlxsw-specific qos_lib.sh to the generic lib.sh.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_qdisc: Support offloading of ETS Qdisc
Petr Machata [Wed, 18 Dec 2019 14:55:19 +0000 (14:55 +0000)]
mlxsw: spectrum_qdisc: Support offloading of ETS Qdisc

Handle TC_SETUP_QDISC_ETS, add a new ops structure for the ETS Qdisc.
Invoke the extended prio handlers implemented in the previous patch. For
stats ops, invoke directly the prio callbacks, which are not sensitive to
differences between PRIO and ETS.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_qdisc: Generalize PRIO offload to support ETS
Petr Machata [Wed, 18 Dec 2019 14:55:17 +0000 (14:55 +0000)]
mlxsw: spectrum_qdisc: Generalize PRIO offload to support ETS

Thanks to the similarity between PRIO and ETS it is possible to simply
reuse most of the code for offloading PRIO Qdisc. Extract the common
functionality into separate functions, making the current PRIO handlers
thin API adapters.

Extend the new functions to pass quanta for individual bands, which allows
configuring a subset of bands as WRR. Invoke mlxsw_sp_port_ets_set() as
appropriate to de/configure WRR-ness and weight of individual bands.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: sch_ets: Make the ETS qdisc offloadable
Petr Machata [Wed, 18 Dec 2019 14:55:15 +0000 (14:55 +0000)]
net: sch_ets: Make the ETS qdisc offloadable

Add hooks at appropriate points to make it possible to offload the ETS
Qdisc.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: sch_ets: Add a new Qdisc
Petr Machata [Wed, 18 Dec 2019 14:55:13 +0000 (14:55 +0000)]
net: sch_ets: Add a new Qdisc

Introduces a new Qdisc, which is based on 802.1Q-2014 wording. It is
PRIO-like in how it is configured, meaning one needs to specify how many
bands there are, how many are strict and how many are dwrr, quanta for the
latter, and priomap.

The new Qdisc operates like the PRIO / DRR combo would when configured as
per the standard. The strict classes, if any, are tried for traffic first.
When there's no traffic in any of the strict queues, the ETS ones (if any)
are treated in the same way as in DRR.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum: Rename MLXSW_REG_QEEC_HIERARCY_* enumerators
Petr Machata [Wed, 18 Dec 2019 14:55:11 +0000 (14:55 +0000)]
mlxsw: spectrum: Rename MLXSW_REG_QEEC_HIERARCY_* enumerators

These enums want to be named MLXSW_REG_QEEC_HIERARCHY_, but due to a typo
lack the second H. That is confusing and complicates searching.

But actually the enumerators should be named _HR_, because that is how
their enum type is called. So rename them as appropriate.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agomlxsw: spectrum_qdisc: Clarify a comment
Petr Machata [Wed, 18 Dec 2019 14:55:10 +0000 (14:55 +0000)]
mlxsw: spectrum_qdisc: Clarify a comment

Expand the comment at mlxsw_sp_qdisc_prio_graft() to make the problem that
this function is trying to handle clearer.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agonet: pkt_cls: Clarify a comment
Petr Machata [Wed, 18 Dec 2019 14:55:08 +0000 (14:55 +0000)]
net: pkt_cls: Clarify a comment

The bit about negating HW backlog left me scratching my head. Clarify the
comment.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agosch_cake: drop unused variable tin_quantum_prio
Kevin 'ldir' Darbyshire-Bryant [Wed, 18 Dec 2019 14:05:13 +0000 (14:05 +0000)]
sch_cake: drop unused variable tin_quantum_prio

Turns out tin_quantum_prio isn't used anymore and is a leftover from a
previous implementation of diffserv tins.  Since the variable isn't used
in any calculations it can be eliminated.

Drop variable and places where it was set.  Rename remaining variable
and consolidate naming of intermediate variables that set it.

Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agoMerge branch 's390-next'
David S. Miller [Wed, 18 Dec 2019 20:34:57 +0000 (12:34 -0800)]
Merge branch 's390-next'

Julian Wiedmann says:

====================
s390/qeth: features 2019-12-18

please apply the following patch series to your net-next tree.
Nothing major, just the usual mix of small improvements and cleanups.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: make use of napi_schedule_irqoff()
Julian Wiedmann [Wed, 18 Dec 2019 16:34:50 +0000 (17:34 +0100)]
s390/qeth: make use of napi_schedule_irqoff()

qeth_qdio_start_poll() is called from the qdio layer's IRQ handler,
while IRQs are masked.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: consolidate helpers for capability checking
Julian Wiedmann [Wed, 18 Dec 2019 16:34:49 +0000 (17:34 +0100)]
s390/qeth: consolidate helpers for capability checking

Convert the old code to use struct qeth_ipa_caps, and while at it remove
all unused helper macros.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: stop yielding the ip_lock during IPv4 registration
Julian Wiedmann [Wed, 18 Dec 2019 16:34:48 +0000 (17:34 +0100)]
s390/qeth: stop yielding the ip_lock during IPv4 registration

As commit df2a2a5225cc ("s390/qeth: convert IP table spinlock to mutex")
converted the ip_lock to a mutex, we no longer have to yield it while
the subsequent IO sleep-waits for completion.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: don't raise NETDEV_REBOOT event from L3 offline path
Julian Wiedmann [Wed, 18 Dec 2019 16:34:47 +0000 (17:34 +0100)]
s390/qeth: don't raise NETDEV_REBOOT event from L3 offline path

This is a leftover from back when a recovery action didn't go through
dev_close(), and was meant to shoot down all remaining af_iucv sockets
on the interface.

Now that the offline path always calls dev_close(), the
NETDEV_GOING_DOWN event from __dev_close_many() is sufficient and this
hack can be removed.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: remove open-coded inet_make_mask()
Julian Wiedmann [Wed, 18 Dec 2019 16:34:46 +0000 (17:34 +0100)]
s390/qeth: remove open-coded inet_make_mask()

Use inet_make_mask() to replace some complicated bit-fiddling.

Also use the right data types to replace some raw memcpy calls with
proper assignments.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: clean up L3 sysfs code
Julian Wiedmann [Wed, 18 Dec 2019 16:34:45 +0000 (17:34 +0100)]
s390/qeth: clean up L3 sysfs code

Consolidate some duplicated code for adding RXIP/VIPA addresses, and
move the locking to where it's actually needed.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
4 years agos390/qeth: overhaul L3 IP address dump code
Julian Wiedmann [Wed, 18 Dec 2019 16:34:44 +0000 (17:34 +0100)]
s390/qeth: overhaul L3 IP address dump code

The current code that dumps the RXIP/VIPA/IPATO addresses via sysfs
first checks whether the buffer still provides sufficient space to hold
another formatted address.
But the maximum length of an formatted IPv4 address is 15 characters,
not 12. So we underestimate the max required length and if the buffer
was previously filled to _just_ the right level, a formatted address can
end up being truncated.

Revamp these code paths to use the _actually_ required length of the
formatted IP address, and while at it suppress a gratuitous newline.

Also use scnprintf() to format the output. In case of a truncation, this
would allow us to return the number of characters that were actually
written.

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