linux-2.6-microblaze.git
3 years agomptcp: use batch snmp operations in mptcp_seq_show()
Eric Dumazet [Thu, 30 Sep 2021 01:03:33 +0000 (18:03 -0700)]
mptcp: use batch snmp operations in mptcp_seq_show()

Using snmp_get_cpu_field_batch() allows for better cpu cache
utilization, especially on hosts with large number of cpus.

Also remove special handling when mptcp mibs where not yet
allocated.

I chose to use temporary storage on the stack to keep this patch simple.
We might in the future use the storage allocated in netstat_seq_show().

Combined with prior patch (inlining snmp_get_cpu_field)
time to fetch and output mptcp counters on a 256 cpu host [1]
goes from 75 usec to 16 usec.

[1] L1 cache size is 32KB, it is not big enough to hold all dataset.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: snmp: inline snmp_get_cpu_field()
Eric Dumazet [Thu, 30 Sep 2021 01:03:32 +0000 (18:03 -0700)]
net: snmp: inline snmp_get_cpu_field()

This trivial function is called ~90,000 times on 256 cpus hosts,
when reading /proc/net/netstat. And this number keeps inflating.

Inlining it saves many cycles.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/mlx4_en: Add XDP_REDIRECT statistics
Joshua Roys [Thu, 30 Sep 2021 02:30:23 +0000 (22:30 -0400)]
net/mlx4_en: Add XDP_REDIRECT statistics

Add counters for XDP REDIRECT success and failure. This brings the
redirect path in line with metrics gathered via the other XDP paths.

Signed-off-by: Joshua Roys <roysjosh@gmail.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoixgbe: let the xdpdrv work with more than 64 cpus
Jason Xing [Wed, 29 Sep 2021 17:56:05 +0000 (10:56 -0700)]
ixgbe: let the xdpdrv work with more than 64 cpus

Originally, ixgbe driver doesn't allow the mounting of xdpdrv if the
server is equipped with more than 64 cpus online. So it turns out that
the loading of xdpdrv causes the "NOMEM" failure.

Actually, we can adjust the algorithm and then make it work through
mapping the current cpu to some xdp ring with the protect of @tx_lock.

Here are some numbers before/after applying this patch with xdp-example
loaded on the eth0X:

As client (tx path):
                     Before    After
TCP_STREAM send-64   734.14    714.20
TCP_STREAM send-128  1401.91   1395.05
TCP_STREAM send-512  5311.67   5292.84
TCP_STREAM send-1k   9277.40   9356.22 (not stable)
TCP_RR     send-1    22559.75  21844.22
TCP_RR     send-128  23169.54  22725.13
TCP_RR     send-512  21670.91  21412.56

As server (rx path):
                     Before    After
TCP_STREAM send-64   1416.49   1383.12
TCP_STREAM send-128  3141.49   3055.50
TCP_STREAM send-512  9488.73   9487.44
TCP_STREAM send-1k   9491.17   9356.22 (not stable)
TCP_RR     send-1    23617.74  23601.60
...

Notice: the TCP_RR mode is unstable as the official document explains.

I tested many times with different parameters combined through netperf.
Though the result is not that accurate, I cannot see much influence on
this patch. The static key is places on the hot path, but it actually
shouldn't cause a huge regression theoretically.

Co-developed-by: Shujin Li <lishujin@kuaishou.com>
Signed-off-by: Shujin Li <lishujin@kuaishou.com>
Signed-off-by: Jason Xing <xingwanli@kuaishou.com>
Tested-by: Sandeep Penigalapati <sandeep.penigalapati@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'SO_RESEVED_MEM'
David S. Miller [Thu, 30 Sep 2021 12:36:46 +0000 (13:36 +0100)]
Merge branch 'SO_RESEVED_MEM'

Wei Wang says:

====================
net: add new socket option SO_RESERVE_MEM

This patch series introduces a new socket option SO_RESERVE_MEM.
This socket option provides a mechanism for users to reserve a certain
amount of memory for the socket to use. When this option is set, kernel
charges the user specified amount of memory to memcg, as well as
sk_forward_alloc. This amount of memory is not reclaimable and is
available in sk_forward_alloc for this socket.
With this socket option set, the networking stack spends less cycles
doing forward alloc and reclaim, which should lead to better system
performance, with the cost of an amount of pre-allocated and
unreclaimable memory, even under memory pressure.
With a tcp_stream test with 10 flows running on a simulated 100ms RTT
link, I can see the cycles spent in __sk_mem_raise_allocated() dropping
by ~0.02%. Not a whole lot, since we already have logic in
sk_mem_uncharge() to only reclaim 1MB when sk_forward_alloc has more
than 2MB free space. But on a system suffering memory pressure
constently, the savings should be more.

The first patch is the implementation of this socket option. The
following 2 patches change the tcp stack to make use of this reserved
memory when under memory pressure. This makes the tcp stack behavior
more flexible when under memory pressure, and provides a way for user to
control the distribution of the memory among its sockets.
With a TCP connection on a simulated 100ms RTT link, the default
throughput under memory pressure is ~500Kbps. With SO_RESERVE_MEM set to
100KB, the throughput under memory pressure goes up to ~3.5Mbps.

Change since v2:
- Added description for new field added in struct sock in patch 1
Change since v1:
- Added performance stats in cover letter and rebased
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agotcp: adjust rcv_ssthresh according to sk_reserved_mem
Wei Wang [Wed, 29 Sep 2021 17:25:13 +0000 (10:25 -0700)]
tcp: adjust rcv_ssthresh according to sk_reserved_mem

When user sets SO_RESERVE_MEM socket option, in order to utilize the
reserved memory when in memory pressure state, we adjust rcv_ssthresh
according to the available reserved memory for the socket, instead of
using 4 * advmss always.

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agotcp: adjust sndbuf according to sk_reserved_mem
Wei Wang [Wed, 29 Sep 2021 17:25:12 +0000 (10:25 -0700)]
tcp: adjust sndbuf according to sk_reserved_mem

If user sets SO_RESERVE_MEM socket option, in order to fully utilize the
reserved memory in memory pressure state on the tx path, we modify the
logic in sk_stream_moderate_sndbuf() to set sk_sndbuf according to
available reserved memory, instead of MIN_SOCK_SNDBUF, and adjust it
when new data is acked.

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: add new socket option SO_RESERVE_MEM
Wei Wang [Wed, 29 Sep 2021 17:25:11 +0000 (10:25 -0700)]
net: add new socket option SO_RESERVE_MEM

This socket option provides a mechanism for users to reserve a certain
amount of memory for the socket to use. When this option is set, kernel
charges the user specified amount of memory to memcg, as well as
sk_forward_alloc. This amount of memory is not reclaimable and is
available in sk_forward_alloc for this socket.
With this socket option set, the networking stack spends less cycles
doing forward alloc and reclaim, which should lead to better system
performance, with the cost of an amount of pre-allocated and
unreclaimable memory, even under memory pressure.

Note:
This socket option is only available when memory cgroup is enabled and we
require this reserved memory to be charged to the user's memcg. We hope
this could avoid mis-behaving users to abused this feature to reserve a
large amount on certain sockets and cause unfairness for others.

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: marvell10g: add downshift tunable support
Russell King [Wed, 29 Sep 2021 15:28:53 +0000 (16:28 +0100)]
net: phy: marvell10g: add downshift tunable support

Add support for the downshift tunable for the Marvell 88x3310 PHY.
Downshift is only usable with firmware 0.3.5.0 and later.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-af: Remove redundant initialization of variable pin
Colin Ian King [Wed, 29 Sep 2021 13:27:53 +0000 (14:27 +0100)]
octeontx2-af: Remove redundant initialization of variable pin

The variable pin is being initialized with a value that is never
read, it is being updated later on in only one case of a switch
statement.  The assignment is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: macb: ptp: Switch to gettimex64() interface
Lars-Peter Clausen [Wed, 29 Sep 2021 12:07:39 +0000 (14:07 +0200)]
net: macb: ptp: Switch to gettimex64() interface

The macb PTP support currently implements the `gettime64` callback to allow
to retrieve the hardware clock time. Update the implementation to provide
the `gettimex64` callback instead.

The difference between the two is that with `gettime64` a snapshot of the
system clock is taken before and after invoking the callback. Whereas
`gettimex64` expects the callback itself to take the snapshots.

To get the time from the macb Ethernet core multiple register accesses have
to be done. Only one of which will happen at the time reported by the
function. This leads to a non-symmetric delay and adds a slight offset
between the hardware and system clock time when using the `gettime64`
method. This offset can be a few 100 nanoseconds. Switching to the
`gettimex64` method allows for a more precise correlation of the hardware
and system clocks and results in a lower offset between the two.

On a Xilinx ZynqMP system `phc2sys` reports a delay of 1120 ns before and
300 ns after the patch. With the latter being mostly symmetric.

Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodissector: do not set invalid PPP protocol
Boris Sukholitko [Wed, 29 Sep 2021 11:32:23 +0000 (14:32 +0300)]
dissector: do not set invalid PPP protocol

The following flower filter fails to match non-PPP_IP{V6} packets
wrapped in PPP_SES protocol:

tc filter add dev eth0 ingress protocol ppp_ses flower \
        action simple sdata hi64

The reason is that proto local variable is being set even when
FLOW_DISSECT_RET_OUT_BAD status is returned.

The fix is to avoid setting proto variable if the PPP protocol is unknown.

Signed-off-by: Boris Sukholitko <boris.sukholitko@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: rtl8366rb: Use core filtering tracking
Linus Walleij [Wed, 29 Sep 2021 11:23:22 +0000 (13:23 +0200)]
net: dsa: rtl8366rb: Use core filtering tracking

We added a state variable to track whether a certain port
was VLAN filtering or not, but we can just inquire the DSA
core about this.

Cc: Vladimir Oltean <olteanv@gmail.com>
Cc: Mauri Sandberg <sandberg@mailfence.com>
Cc: DENG Qingfang <dqfext@gmail.com>
Cc: Alvin Šipraga <alsi@bang-olufsen.dk>
Cc: Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-pf: Add XDP support to netdev PF
Geetha sowjanya [Wed, 29 Sep 2021 09:54:56 +0000 (15:24 +0530)]
octeontx2-pf: Add XDP support to netdev PF

Adds XDP_PASS, XDP_TX, XDP_DROP and XDP_REDIRECT support
for netdev PF.

Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-af: Adjust LA pointer for cpt parse header
Kiran Kumar K [Wed, 29 Sep 2021 05:58:31 +0000 (11:28 +0530)]
octeontx2-af: Adjust LA pointer for cpt parse header

In case of ltype NPC_LT_LA_CPT_HDR, LA pointer is pointing to the
start of cpt parse header. Since cpt parse header has veriable
length padding, this will be a problem for DMAC extraction. Adding
KPU profile changes to adjust the LA pointer to start at ether header
in case of cpt parse header by
   - Adding ptr advance in pkind 58 to a fixed value 40
   - Adding variable length offset 7 and mask 7 (pad len in
     CPT_PARSE_HDR).
Also added the missing static declaration for npc_set_var_len_offset_pkind
function.

Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet_sched: Use struct_size() and flex_array_size() helpers
Gustavo A. R. Silva [Tue, 28 Sep 2021 19:31:07 +0000 (14:31 -0500)]
net_sched: Use struct_size() and flex_array_size() helpers

Make use of the struct_size() and flex_array_size() helpers instead of
an open-coded version, in order to avoid any potential type mistakes
or integer overflows that, in the worse scenario, could lead to heap
overflows.

Link: https://github.com/KSPP/linux/issues/160
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Link: https://lore.kernel.org/r/20210928193107.GA262595@embeddedor
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agodevlink: Add missed notifications iterators
Leon Romanovsky [Wed, 29 Sep 2021 14:18:20 +0000 (17:18 +0300)]
devlink: Add missed notifications iterators

The commit mentioned in Fixes line missed a couple of notifications that
were registered before devlink_register() and should be delayed too.

As such, the too early placed WARN_ON() check spotted it.

WARNING: CPU: 1 PID: 6540 at net/core/devlink.c:5158 devlink_nl_region_notify+0x184/0x1e0 net/core/devlink.c:5158
Modules linked in:
CPU: 1 PID: 6540 Comm: syz-executor.0 Not tainted 5.15.0-rc2-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:devlink_nl_region_notify+0x184/0x1e0 net/core/devlink.c:5158
Code: 38 41 b8 c0 0c 00 00 31 d2 48 89 ee 4c 89 e7 e8 72 1a 26 00 48 83 c4 08 5b 5d 41 5c 41 5d 41 5e e9 01 bd 41 fa
e8 fc bc 41 fa <0f> 0b e9 f7 fe ff ff e8 f0 bc 41 fa 0f 0b eb da 4c 89 e7 e8 c4 18
RSP: 0018:ffffc90002d6f658 EFLAGS: 00010293
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
RDX: ffff88801f08d580 RSI: ffffffff87344e94 RDI: 0000000000000003
RBP: ffff88801ee42100 R08: 0000000000000000 R09: 0000000000000000
R10: ffffffff87344d8a R11: 0000000000000000 R12: ffff88801c1dc000
R13: 0000000000000000 R14: 000000000000002c R15: ffff88801c1dc070
FS:  0000555555e8e400(0000) GS:ffff8880b9d00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000055dd7c590310 CR3: 0000000069a09000 CR4: 00000000003506e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 devlink_region_create+0x39f/0x4c0 net/core/devlink.c:10327
 nsim_dev_dummy_region_init drivers/net/netdevsim/dev.c:481 [inline]
 nsim_dev_probe+0x5f6/0x1150 drivers/net/netdevsim/dev.c:1479
 call_driver_probe drivers/base/dd.c:517 [inline]
 really_probe+0x245/0xcc0 drivers/base/dd.c:596
 __driver_probe_device+0x338/0x4d0 drivers/base/dd.c:751
 driver_probe_device+0x4c/0x1a0 drivers/base/dd.c:781
 __device_attach_driver+0x20b/0x2f0 drivers/base/dd.c:898
 bus_for_each_drv+0x15f/0x1e0 drivers/base/bus.c:427
 __device_attach+0x228/0x4a0 drivers/base/dd.c:969
 bus_probe_device+0x1e4/0x290 drivers/base/bus.c:487
 device_add+0xc35/0x21b0 drivers/base/core.c:3359
 nsim_bus_dev_new drivers/net/netdevsim/bus.c:435 [inline]
 new_device_store+0x48b/0x770 drivers/net/netdevsim/bus.c:302
 bus_attr_store+0x72/0xa0 drivers/base/bus.c:122
 sysfs_kf_write+0x110/0x160 fs/sysfs/file.c:139
 kernfs_fop_write_iter+0x342/0x500 fs/kernfs/file.c:296
 call_write_iter include/linux/fs.h:2163 [inline]
 new_sync_write+0x429/0x660 fs/read_write.c:507
 vfs_write+0x7cf/0xae0 fs/read_write.c:594
 ksys_write+0x12d/0x250 fs/read_write.c:647
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x7f328409d3ef
Code: 89 54 24 18 48 89 74 24 10 89 7c 24 08 e8 99 fd ff ff 48 8b 54 24 18 48 8b 74 24 10 41 89 c0 8b 7c 24 08 b8 01
00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 31 44 89 c7 48 89 44 24 08 e8 cc fd ff ff 48
RSP: 002b:00007ffdc6851140 EFLAGS: 00000293 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007f328409d3ef
RDX: 0000000000000003 RSI: 00007ffdc6851190 RDI: 0000000000000004
RBP: 0000000000000004 R08: 0000000000000000 R09: 00007ffdc68510e0
R10: 0000000000000000 R11: 0000000000000293 R12: 00007f3284144971
R13: 00007ffdc6851190 R14: 0000000000000000 R15: 00007ffdc6851860

Fixes: cf530217408e ("devlink: Notify users when objects are accessible")
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/2ed1159291f2a589b013914f2b60d8172fc525c1.1632925030.git.leonro@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agogve: Use kvcalloc() instead of kvzalloc()
Gustavo A. R. Silva [Tue, 28 Sep 2021 21:38:05 +0000 (16:38 -0500)]
gve: Use kvcalloc() instead of kvzalloc()

Use 2-factor argument form kvcalloc() instead of kvzalloc().

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/ipv4/datagram.c: remove superfluous header files from datagram.c
Mianhan Liu [Wed, 29 Sep 2021 05:31:09 +0000 (13:31 +0800)]
net/ipv4/datagram.c: remove superfluous header files from datagram.c

datagram.c hasn't use any macro or function declared in linux/ip.h.
Thus, these files can be removed from datagram.c safely without
affecting the compilation of the net/ipv4 module

Signed-off-by: Mianhan Liu <liumh1@shanghaitech.edu.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/dsa/tag_ksz.c: remove superfluous headers
Mianhan Liu [Wed, 29 Sep 2021 06:41:06 +0000 (14:41 +0800)]
net/dsa/tag_ksz.c: remove superfluous headers

tag_ksz.c hasn't use any macro or function declared in linux/slab.h.
Thus, these files can be removed from tag_ksz.c safely without
affecting the compilation of the ./net/dsa module

Signed-off-by: Mianhan Liu <liumh1@shanghaitech.edu.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/dsa/tag_8021q.c: remove superfluous headers
Mianhan Liu [Wed, 29 Sep 2021 06:36:11 +0000 (14:36 +0800)]
net/dsa/tag_8021q.c: remove superfluous headers

tag_8021q.c hasn't use any macro or function declared in linux/if_bridge.h.
Thus, these files can be removed from tag_8021q.c safely without
affecting the compilation of the ./net/dsa module

Signed-off-by: Mianhan Liu <liumh1@shanghaitech.edu.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/mlx4: Use array_size() helper in copy_to_user()
Gustavo A. R. Silva [Tue, 28 Sep 2021 20:17:33 +0000 (15:17 -0500)]
net/mlx4: Use array_size() helper in copy_to_user()

Use array_size() helper instead of the open-coded version in
copy_to_user(). These sorts of multiplication factors need
to be wrapped in array_size().

Link: https://github.com/KSPP/linux/issues/160
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bridge: Use array_size() helper in copy_to_user()
Gustavo A. R. Silva [Tue, 28 Sep 2021 20:12:39 +0000 (15:12 -0500)]
net: bridge: Use array_size() helper in copy_to_user()

Use array_size() helper instead of the open-coded version in
copy_to_user(). These sorts of multiplication factors need
to be wrapped in array_size().

Link: https://github.com/KSPP/linux/issues/160
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoethtool: ioctl: Use array_size() helper in copy_{from,to}_user()
Gustavo A. R. Silva [Tue, 28 Sep 2021 19:57:35 +0000 (14:57 -0500)]
ethtool: ioctl: Use array_size() helper in copy_{from,to}_user()

Use array_size() helper instead of the open-coded version in
copy_{from,to}_user().  These sorts of multiplication factors
need to be wrapped in array_size().

Link: https://github.com/KSPP/linux/issues/160
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'mctp-core-updates'
David S. Miller [Wed, 29 Sep 2021 10:00:12 +0000 (11:00 +0100)]
Merge branch 'mctp-core-updates'

Matt Johnston says:

====================
Updates to MCTP core

This series adds timeouts for MCTP tags (a limited resource), and a few
other improvements to the MCTP core.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: Warn if pointer is set for a wrong dev type
Matt Johnston [Wed, 29 Sep 2021 07:26:14 +0000 (15:26 +0800)]
mctp: Warn if pointer is set for a wrong dev type

Should not occur but is a sanity check.

May help tracking down Trinity reported issue
https://lore.kernel.org/lkml/20210913030701.GA5926@xsang-OptiPlex-9020/

Signed-off-by: Matt Johnston <matt@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: Set route MTU via netlink
Matt Johnston [Wed, 29 Sep 2021 07:26:13 +0000 (15:26 +0800)]
mctp: Set route MTU via netlink

A route's RTAX_MTU can be set in nested RTAX_METRICS

Signed-off-by: Matt Johnston <matt@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodoc/mctp: Add a little detail about kernel internals
Jeremy Kerr [Wed, 29 Sep 2021 07:26:12 +0000 (15:26 +0800)]
doc/mctp: Add a little detail about kernel internals

Describe common flows and refcounting behaviour.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: Do inits as a subsys_initcall
Jeremy Kerr [Wed, 29 Sep 2021 07:26:11 +0000 (15:26 +0800)]
mctp: Do inits as a subsys_initcall

In a future change, we'll want to provide a registration call for
mctp-specific devices. This requires us to have the networks established
before device driver inits, so run the core init as a subsys_initcall.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: Add tracepoints for tag/key handling
Jeremy Kerr [Wed, 29 Sep 2021 07:26:10 +0000 (15:26 +0800)]
mctp: Add tracepoints for tag/key handling

The tag allocation, release and bind events are somewhat opaque outside
the kernel; this change adds a few tracepoints to assist in
instrumentation and debugging.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: Implement a timeout for tags
Jeremy Kerr [Wed, 29 Sep 2021 07:26:09 +0000 (15:26 +0800)]
mctp: Implement a timeout for tags

Currently, a MCTP (local-eid,remote-eid,tag) tuple is allocated to a
socket on send, and only expires when the socket is closed.

This change introduces a tag timeout, freeing the tuple after a fixed
expiry - currently six seconds. This is greater than (but close to) the
max response timeout in upper-layer bindings.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: Add refcounts to mctp_dev
Jeremy Kerr [Wed, 29 Sep 2021 07:26:08 +0000 (15:26 +0800)]
mctp: Add refcounts to mctp_dev

Currently, we tie the struct mctp_dev lifetime to the underlying struct
net_device, and hold/put that device as a proxy for a separate mctp_dev
refcount. This works because we're not holding any references to the
mctp_dev that are different from the netdev lifetime.

In a future change we'll break that assumption though, as we'll need to
hold mctp_dev references in a workqueue, which might live past the
netdev unregister notification.

In order to support that, this change introduces a refcount on the
mctp_dev, currently taken by the net_device->mctp_ptr reference, and
released on netdev unregister events. We can then use this for future
references that might outlast the net device.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: locking, lifetime and validity changes for sk_keys
Jeremy Kerr [Wed, 29 Sep 2021 07:26:07 +0000 (15:26 +0800)]
mctp: locking, lifetime and validity changes for sk_keys

We will want to invalidate sk_keys in a future change, which will
require a boolean flag to mark invalidated items in the socket & net
namespace lists. We'll also need to take a reference to keys, held over
non-atomic contexts, so we need a refcount on keys also.

This change adds a validity flag (currently always true) and refcount to
struct mctp_sk_key.  With a refcount on the keys, using RCU no longer
makes much sense; we have exact indications on the lifetime of keys. So,
we also change the RCU list traversal to a locked implementation.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: Allow local delivery to the null EID
Jeremy Kerr [Wed, 29 Sep 2021 07:26:06 +0000 (15:26 +0800)]
mctp: Allow local delivery to the null EID

We may need to receive packets addressed to the null EID (==0), but
addressed to us at the physical layer.

This change adds a lookup for local routes when we see a packet
addressed to EID 0, and a local phys address.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomctp: Allow MCTP on tun devices
Matt Johnston [Wed, 29 Sep 2021 07:26:05 +0000 (15:26 +0800)]
mctp: Allow MCTP on tun devices

Allowing TUN is useful for testing, to route packets to userspace or to
tunnel between machines.

Signed-off-by: Matt Johnston <matt@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: phy: micrel: Add support for LAN8804 PHY
Horatiu Vultur [Tue, 28 Sep 2021 18:45:19 +0000 (20:45 +0200)]
net: phy: micrel: Add support for LAN8804 PHY

The LAN8804 PHY has same features as that of LAN8814 PHY except that it
doesn't support 1588, SyncE or Q-USGMII.

This PHY is found inside the LAN966X switches.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
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/nex
David S. Miller [Wed, 29 Sep 2021 09:30:56 +0000 (10:30 +0100)]
Merge branch '100GbE' of git://git./linux/kernel/git/tnguy/nex
t-queue

Tony Nguyen says:

====================
100GbE Intel Wired LAN Driver Updates 2021-09-28

This series contains updates to ice driver only.

Dave adds support for QoS DSCP allowing for DSCP to TC mapping via APP
TLVs.

Ani adds enforcement of DSCP to only supported devices with the
introduction of a feature bitmap and corrects messaging of unsupported
modules based on link mode.

Jake refactors devlink info functions to be void as the functions no
longer return errors.

Jeff fixes a macro name to properly reflect the value.

Len Baker converts a kzalloc allocation to, the preferred, kcalloc.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'octeontx2-ptp-vf'
David S. Miller [Wed, 29 Sep 2021 09:27:33 +0000 (10:27 +0100)]
Merge branch 'octeontx2-ptp-vf'

Subbaraya Sundeep <sbhatta@marvell.com>

====================
octeontx2: Add PTP support for VFs

PTP is a shared hardware block which can prepend
RX timestamps to packets before directing packets to
PFs or VFs and can notify the TX timestamps to PFs or VFs
via TX completion queue descriptors. Hence adding PTP
support for VFs is exactly similar to PFs with minimal changes.
This patchset adds that PTP support for VFs.

Patch 1 - When an interface is set in promisc/multicast
the same setting is not retained when changing mtu or channels.
This is due to toggling of the interface by driver but not
calling set_rx_mode in the down-up sequence. Since setting
an interface to multicast properly is required for ptp this is
addressed in this patch.

Patch 2 - Changes in VF driver for registering timestamping
ethtool ops and ndo_ioctl.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-nicvf: Add PTP hardware clock support to NIX VF
Naveen Mamindlapalli [Tue, 28 Sep 2021 17:43:46 +0000 (23:13 +0530)]
octeontx2-nicvf: Add PTP hardware clock support to NIX VF

This patch adds PTP PHC support to NIX VF interfaces. This enables
a VF to run PTP master/slave instance. PTP block being a shared
hardware resource it is recommended to avoid running multiple
PTP instances in the system which will impact the PTP clock
accuracy.

Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-pf: Enable promisc/allmulti match MCAM entries.
Rakesh Babu [Tue, 28 Sep 2021 17:43:45 +0000 (23:13 +0530)]
octeontx2-pf: Enable promisc/allmulti match MCAM entries.

Whenever the interface is brought up/down then set_rx_mode
function is called by the stack which enables promisc/allmulti
MCAM entries. But there are cases when driver brings
interface down and then up such as while changing number
of channels. In these cases promisc/allmulti MCAM entries
are left disabled as set_rx_mode callback is not called.
This patch enables these MCAM entries in all such cases.

Signed-off-by: Rakesh Babu <rsaladi2@marvell.com>
Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: qrtr: combine nameservice into main module
Luca Weiss [Tue, 28 Sep 2021 17:11:57 +0000 (19:11 +0200)]
net: qrtr: combine nameservice into main module

Previously with CONFIG_QRTR=m a separate ns.ko would be built which
wasn't done on purpose and should be included in qrtr.ko.

Rename qrtr.c to af_qrtr.c so we can build a qrtr.ko with both af_qrtr.c
and ns.c.

Signed-off-by: Luca Weiss <luca@z3ntu.xyz>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Tested-By: Steev Klimaszewski <steev@kali.org>
Reviewed-by: Manivannan Sadhasivam <mani@kernel.org>
Link: https://lore.kernel.org/r/20210928171156.6353-1-luca@z3ntu.xyz
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ipv4: remove superfluous header files from fib_notifier.c
Mianhan Liu [Tue, 28 Sep 2021 16:40:11 +0000 (00:40 +0800)]
net: ipv4: remove superfluous header files from fib_notifier.c

fib_notifier.c hasn't use any macro or function declared
in net/netns/ipv4.h.

Thus, these files can be removed from fib_notifier.c safely
without affecting the compilation of the net/ipv4 module.

Signed-off-by: Mianhan Liu <liumh1@shanghaitech.edu.cn>
Link: https://lore.kernel.org/r/20210928164011.1454-1-liumh1@shanghaitech.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoice: Prefer kcalloc over open coded arithmetic
Len Baker [Sun, 5 Sep 2021 06:50:20 +0000 (08:50 +0200)]
ice: Prefer kcalloc over open coded arithmetic

As noted in the "Deprecated Interfaces, Language Features, Attributes,
and Conventions" documentation [1], size calculations (especially
multiplication) should not be performed in memory allocator (or similar)
function arguments due to the risk of them overflowing. This could lead
to values wrapping around and a smaller allocation being made than the
caller was expecting. Using those allocations could lead to linear
overflows of heap memory and other misbehaviors.

In this case this is not actually dynamic sizes: both sides of the
multiplication are constant values. However it is best to refactor this
anyway, just to keep the open-coded math idiom out of code.

So, use the purpose specific kcalloc() function instead of the argument
size * count in the kzalloc() function.

[1] https://www.kernel.org/doc/html/v5.14/process/deprecated.html#open-coded-arithmetic-in-allocator-arguments

Signed-off-by: Len Baker <len.baker@gmx.com>
Reviewed-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
3 years agoice: Fix macro name for IPv4 fragment flag
Jeff Guo [Fri, 16 Jul 2021 22:16:44 +0000 (15:16 -0700)]
ice: Fix macro name for IPv4 fragment flag

In IPv4 header, fragment flags indicate whether the packet needs
to be fragmented or not. The value 0x20 represents MF (More Fragment); fix
the macro name to match this.

Signed-off-by: Ting Xu <ting.xu@intel.com>
Signed-off-by: Jeff Guo <jia.guo@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
3 years agoice: refactor devlink getter/fallback functions to void
Jacob Keller [Fri, 20 Aug 2021 15:09:50 +0000 (08:09 -0700)]
ice: refactor devlink getter/fallback functions to void

After commit a8f89fa27773 ("ice: do not abort devlink info if board
identifier can't be found"), the getter/fallback() functions no longer
report an error. Convert the interface to a void so that it is no
longer possible to add a version field that is fatal. This makes
sense, because we should not fail to report other versions just
because one of the version pieces could not be found.

Finally, clean up the getter functions line wrapping so that none of
them take more than 80 columns, as is the usual style for networking
files.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
3 years agoice: Fix link mode handling
Anirudh Venkataramanan [Fri, 16 Jul 2021 22:16:39 +0000 (15:16 -0700)]
ice: Fix link mode handling

The messaging for unsupported module detection is different for
lenient mode and strict mode. Update the code to print the right
messaging for a given link mode.

Media topology conflict is not an error in lenient mode, so return
an error code only if not in lenient mode.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
3 years agoice: Add feature bitmap, helpers and a check for DSCP
Anirudh Venkataramanan [Fri, 16 Jul 2021 22:16:41 +0000 (15:16 -0700)]
ice: Add feature bitmap, helpers and a check for DSCP

DSCP a.k.a L3 QoS is only supported on certain devices. To enforce this,
this patch introduces a bitmap of features and helper functions.

The feature bitmap is set based on device IDs on driver init. Currently,
DSCP is the only feature in this bitmap, but there will be more in the
future. In the DCB netlink flow, check if the feature bit is set before
exercising DSCP.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
3 years agoice: Add DSCP support
Dave Ertman [Fri, 6 Aug 2021 20:53:56 +0000 (13:53 -0700)]
ice: Add DSCP support

Implement code to handle submission of APP TLV's
containing DSCP to TC mapping.

The first such mapping received on an interface
will cause that PF to switch to L3 DSCP QoS mode,
apply the default config for that mode, and apply
the received mapping.

Only one such mapping will be allowed per DSCP value,
and when the last DSCP mapping is deleted, the PF
will switch back into L2 VLAN QoS mode, applying the
appropriate default QoS settings.

L3 DSCP QoS mode will only be allowed in SW DCBx
mode, in other words, when the FW LLDP engine is
disabled.  Commands that break this mutual exclusivity
will be blocked.

Co-developed-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
3 years agogve: DQO: avoid unused variable warnings
Arnd Bergmann [Tue, 28 Sep 2021 14:15:13 +0000 (16:15 +0200)]
gve: DQO: avoid unused variable warnings

The use of dma_unmap_addr()/dma_unmap_len() in the driver causes
multiple warnings when these macros are defined as empty, e.g.
in an ARCH=i386 allmodconfig build:

drivers/net/ethernet/google/gve/gve_tx_dqo.c: In function 'gve_tx_add_skb_no_copy_dqo':
drivers/net/ethernet/google/gve/gve_tx_dqo.c:494:40: error: unused variable 'buf' [-Werror=unused-variable]
  494 |                 struct gve_tx_dma_buf *buf =

This is not how the NEED_DMA_MAP_STATE macros are meant to work,
as they rely on never using local variables or a temporary structure
like gve_tx_dma_buf.

Remote the gve_tx_dma_buf definition and open-code the contents
in all places to avoid the warning. This causes some rather long
lines but otherwise ends up making the driver slightly smaller.

Fixes: a57e5de476be ("gve: DQO: Add TX path")
Link: https://lore.kernel.org/netdev/20210723231957.1113800-1-bcf@google.com/
Link: https://lore.kernel.org/netdev/20210721151100.2042139-1-arnd@kernel.org/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-pf: Use hardware register for CQE count
Geetha sowjanya [Tue, 28 Sep 2021 05:55:26 +0000 (11:25 +0530)]
octeontx2-pf: Use hardware register for CQE count

Current driver uses software CQ head pointer to poll on CQE
header in memory to determine if CQE is valid. Software needs
to make sure, that the reads of the CQE do not get re-ordered
so much that it ends up with an inconsistent view of the CQE.
To ensure that DMB barrier after read to first CQE cacheline
and before reading of the rest of the CQE is needed.
But having barrier for every CQE read will impact the performance,
instead use hardware CQ head and tail pointers to find the
valid number of CQEs.

Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'octeontx2-af-external-ptp-clock'
David S. Miller [Tue, 28 Sep 2021 12:50:38 +0000 (13:50 +0100)]
Merge branch 'octeontx2-af-external-ptp-clock'

Hariprasad Kelam says:

====================
Externel ptp clock support

Externel ptp support is required in a scenario like connecting
a external timing device to the chip for time synchronization.
This series of patches adds support to ptp driver to use external
clock and enables PTP config in CN10K MAC block (RPM). Currently
PTP configuration is left unchanged in FLR handler these patches
addresses the same.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
,

3 years agoocteontx2-af: Add external ptp input clock
Yi Guo [Tue, 28 Sep 2021 11:31:01 +0000 (17:01 +0530)]
octeontx2-af: Add external ptp input clock

PTP hardware block can be configured to utilize
the external clock. Also the current ptp timestamp
can be captured when external trigger is applied on
a gpio pin. These features are required in scenarios
like connecting a external timing device to the chip
for time synchronization. The timing device provides
the clock and trigger(PPS signal) to the PTP block.
This patch does the following:
1. configures PTP block to use external clock
frequency and timestamp capture on external event.
2. sends PTP_REQ_EXTTS events to kernel ptp phc susbsytem
with captured timestamps
3. aligns PPS edge to adjusted ptp clock in the ptp device
by setting the PPS_THRESH to the reminder of the last
timestamp value captured by external PPS

Signed-off-by: Yi Guo <yig@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-af: Use ptp input clock info from firmware data
Subbaraya Sundeep [Tue, 28 Sep 2021 11:31:00 +0000 (17:01 +0530)]
octeontx2-af: Use ptp input clock info from firmware data

The input clock frequency of PTP block is figured
out from hardware reset block currently. The firmware
data already has this info in sclk. Hence simplify
ptp driver to use sclk from firmware data.

Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-af: cn10k: RPM hardware timestamp configuration
Hariprasad Kelam [Tue, 28 Sep 2021 11:30:59 +0000 (17:00 +0530)]
octeontx2-af: cn10k: RPM hardware timestamp configuration

MAC on CN10K support hardware timestamping such that 8 bytes addition
header is prepended to incoming packets. This patch does necessary
configuration to enable Hardware time stamping upon receiving request
from PF netdev interfaces.

Timestamp configuration is different on MAC (CGX) Octeontx2 silicon
and MAC (RPM) OcteonTX3 CN10k. Based on silicon variant appropriate
fn() pointer is called. Refactor MAC specific mbox messages to remove
unnecessary gaps in mboxids.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2-af: Reset PTP config in FLR handler
Harman Kalra [Tue, 28 Sep 2021 11:30:58 +0000 (17:00 +0530)]
octeontx2-af: Reset PTP config in FLR handler

Upon receiving ptp config request from netdev interface , Octeontx2 MAC
block CGX is configured to append timestamp to every incoming packet
and NPC config is updated with DMAC offset change.

Currently this configuration is not reset in FLR handler. This patch
resets the same.

Signed-off-by: Harman Kalra <hkalra@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/tls: support SM4 CCM algorithm
Tianjia Zhang [Tue, 28 Sep 2021 06:28:43 +0000 (14:28 +0800)]
net/tls: support SM4 CCM algorithm

The IV of CCM mode has special requirements, this patch supports CCM
mode of SM4 algorithm.

Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipv6: use ipv6-y directly instead of ipv6-objs
Masahiro Yamada [Mon, 27 Sep 2021 14:28:40 +0000 (23:28 +0900)]
net: ipv6: use ipv6-y directly instead of ipv6-objs

Kbuild supports <modname>-y as well as <modname>-objs.
This simplifies the Makefile.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipv6: squash $(ipv6-offload) in Makefile
Masahiro Yamada [Mon, 27 Sep 2021 14:28:39 +0000 (23:28 +0900)]
net: ipv6: squash $(ipv6-offload) in Makefile

Assign the objects directly to obj-$(CONFIG_INET).

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: stmmac: fix off-by-one error in sanity check
Arnd Bergmann [Mon, 27 Sep 2021 13:58:29 +0000 (15:58 +0200)]
net: stmmac: fix off-by-one error in sanity check

My previous patch had an off-by-one error in the added sanity
check, the arrays are MTL_MAX_{RX,TX}_QUEUES long, so if that
index is that number, it has overflown.

The patch silenced the warning anyway because the strings could
no longer overlap with the input, but they could still overlap
with other fields.

Fixes: 3e0d5699a975 ("net: stmmac: fix gcc-10 -Wrestrict warning")
Reported-by: Russell King (Oracle) <linux@armlinux.org.uk>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoam65-cpsw: avoid null pointer arithmetic
Arnd Bergmann [Mon, 27 Sep 2021 09:37:57 +0000 (11:37 +0200)]
am65-cpsw: avoid null pointer arithmetic

clang warns about arithmetic on NULL pointers:

drivers/net/ethernet/ti/am65-cpsw-ethtool.c:71:2: error: performing pointer subtraction with a null pointer has undefined behavior [-Werror,-Wnull-pointer-subtraction]
        AM65_CPSW_REGDUMP_REC(AM65_CPSW_REGDUMP_MOD_NUSS, 0x0, 0x1c),
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/ethernet/ti/am65-cpsw-ethtool.c:64:29: note: expanded from macro 'AM65_CPSW_REGDUMP_REC'
        .hdr.len = (((u32 *)(end)) - ((u32 *)(start)) + 1) * sizeof(u32) * 2 + \
                                   ^ ~~~~~~~~~~~~~~~~

The expression here is easily changed to a calculation based on integers
that is no less readable.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mac80211: check return value of rhashtable_init
MichelleJin [Mon, 27 Sep 2021 03:34:57 +0000 (03:34 +0000)]
net: mac80211: check return value of rhashtable_init

When rhashtable_init() fails, it returns -EINVAL.
However, since error return value of rhashtable_init is not checked,
it can cause use of uninitialized pointers.
So, fix unhandled errors of rhashtable_init.

Signed-off-by: MichelleJin <shjy180909@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipv6: check return value of rhashtable_init
MichelleJin [Mon, 27 Sep 2021 03:34:56 +0000 (03:34 +0000)]
net: ipv6: check return value of rhashtable_init

When rhashtable_init() fails, it returns -EINVAL.
However, since error return value of rhashtable_init is not checked,
it can cause use of uninitialized pointers.
So, fix unhandled errors of rhashtable_init.

Signed-off-by: MichelleJin <shjy180909@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/mlx5e: check return value of rhashtable_init
MichelleJin [Mon, 27 Sep 2021 03:34:55 +0000 (03:34 +0000)]
net/mlx5e: check return value of rhashtable_init

When rhashtable_init() fails, it returns -EINVAL.
However, since error return value of rhashtable_init is not checked,
it can cause use of uninitialized pointers.
So, fix unhandled errors of rhashtable_init.

Signed-off-by: MichelleJin <shjy180909@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'bcmgenet-flow-control'
David S. Miller [Mon, 27 Sep 2021 15:35:06 +0000 (16:35 +0100)]
Merge branch 'bcmgenet-flow-control'

Florian Fainelli says:

====================
net: bcmgenet: support for flow control

This patch series adds support for flow control to the GENET driver, the
first 2 patches remove superfluous code, the 3rd one does re-organize
code a little bit and the 4th one ads the support for flow control
proper.
====================

3 years agonet: bcmgenet: add support for ethtool flow control
Doug Berger [Sun, 26 Sep 2021 03:21:14 +0000 (20:21 -0700)]
net: bcmgenet: add support for ethtool flow control

This commit extends the supported ethtool operations to allow MAC
level flow control to be configured for the bcmgenet driver.

The ethtool utility can be used to change the configuration of
auto-negotiated symmetric and asymmetric modes as well as manually
configuring support for RX and TX Pause frames individually.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bcmgenet: pull mac_config from adjust_link
Doug Berger [Sun, 26 Sep 2021 03:21:13 +0000 (20:21 -0700)]
net: bcmgenet: pull mac_config from adjust_link

This commit separates out the MAC configuration that occurs on a
PHY state change into a function named bcmgenet_mac_config().

This allows the function to be called directly elsewhere.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bcmgenet: remove old link state values
Doug Berger [Sun, 26 Sep 2021 03:21:12 +0000 (20:21 -0700)]
net: bcmgenet: remove old link state values

The PHY state machine has been fixed to only call the adjust_link
callback when the link state has changed. Therefore the old link
state variables are no longer needed to detect a change in link
state.

This commit effectively reverts
commit 5ad6e6c50899 ("net: bcmgenet: improve bcmgenet_mii_setup()")

Signed-off-by: Doug Berger <opendmb@gmail.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: bcmgenet: remove netif_carrier_off from adjust_link
Doug Berger [Sun, 26 Sep 2021 03:21:11 +0000 (20:21 -0700)]
net: bcmgenet: remove netif_carrier_off from adjust_link

The bcmgenet_mii_setup() function is registered as the adjust_link
callback from the phylib for the GENET driver.

The phylib always sets the netif_carrier according to phydev->link
prior to invoking the adjust_link callback, so there is no need to
repeat that in the link down case within the network driver.

Signed-off-by: Doug Berger <opendmb@gmail.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'rtl8366-cleanups'
David S. Miller [Mon, 27 Sep 2021 15:33:57 +0000 (16:33 +0100)]
Merge branch 'rtl8366-cleanups'

Linus Walleij says:

====================
RTL8366(RB) cleanups part 1

This is a first set of patches making the RTL8366RB work out of
the box with a default OpenWrt userspace.

We achieve bridge port isolation with the first patch, and the
next 5 patches removes the very weird VLAN set-up with one
VLAN with PVID per port that has been in this driver in all
vendor trees and in OpenWrt for years.

The switch is now managed the way a modern bridge/DSA switch
shall be managed.

After these patches are merged, I will send the next set which
adds new features, some which have circulated before.

ChangeLog v4->v5:
- Drop the patch disabling 4K VLAN.
- Drop the patch forcing VLAN0 untagged.
- Fix a semantic bug in the filer enablement code.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: rtl8366: Drop and depromote pointless prints
Linus Walleij [Sat, 25 Sep 2021 22:59:29 +0000 (00:59 +0200)]
net: dsa: rtl8366: Drop and depromote pointless prints

We don't need a message for every VLAN association, dbg
is fine. The message about adding the DSA or CPU
port to a VLAN is directly misleading, this is perfectly
fine.

Cc: Vladimir Oltean <olteanv@gmail.com>
Cc: Mauri Sandberg <sandberg@mailfence.com>
Cc: DENG Qingfang <dqfext@gmail.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: rtl8366: Fix a bug in deleting VLANs
Linus Walleij [Sat, 25 Sep 2021 22:59:28 +0000 (00:59 +0200)]
net: dsa: rtl8366: Fix a bug in deleting VLANs

We were checking that the MC (member config) was != 0
for some reason, all we need to check is that the config
has no ports, i.e. no members. Then it can be recycled.
This must be some misunderstanding.

Fixes: 4ddcaf1ebb5e ("net: dsa: rtl8366: Properly clear member config")
Cc: Mauri Sandberg <sandberg@mailfence.com>
Cc: DENG Qingfang <dqfext@gmail.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: rtl8366rb: Fix off-by-one bug
Linus Walleij [Sat, 25 Sep 2021 22:59:27 +0000 (00:59 +0200)]
net: dsa: rtl8366rb: Fix off-by-one bug

The max VLAN number with non-4K VLAN activated is 15, and the
range is 0..15. Not 16.

The impact should be low since we by default have 4K VLAN and
thus have 4095 VLANs to play with in this switch. There will
not be a problem unless the code is rewritten to only use
16 VLANs.

Fixes: d8652956cf37 ("net: dsa: realtek-smi: Add Realtek SMI driver")
Cc: Mauri Sandberg <sandberg@mailfence.com>
Cc: DENG Qingfang <dqfext@gmail.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: rtl8366rb: Rewrite weird VLAN filering enablement
Linus Walleij [Sat, 25 Sep 2021 22:59:26 +0000 (00:59 +0200)]
net: dsa: rtl8366rb: Rewrite weird VLAN filering enablement

While we were defining one VLAN per port for isolating the ports
the port_vlan_filtering() callback was implemented to enable a
VLAN on the port + 1. This function makes no sense, not only is
it incomplete as it only enables the VLAN, it doesn't do what
the callback is supposed to do, which is to selectively enable
and disable filtering on a certain port.

Implement the correct callback: we have two registers dealing
with filtering on the RTL9366RB, so we implement an ASIC-specific
callback and implement filering using the register bit that makes
the switch drop frames if the port is not in the VLAN member set.

The DSA documentation Documentation/networking/switchdev.rst states:

  When the bridge has VLAN filtering enabled and a PVID is not
  configured on the ingress port, untagged and 802.1p tagged
  packets must be dropped. When the bridge has VLAN filtering
  enabled and a PVID exists on the ingress port, untagged and
  priority-tagged packets must be accepted and forwarded according
  to the bridge's port membership of the PVID VLAN. When the
  bridge has VLAN filtering disabled, the presence/lack of a
  PVID should not influence the packet forwarding decision.

To comply with this, we add two arrays of bool in the RTL8366RB
state that keeps track of if filtering and PVID is enabled or
not for each port. We then add code such that whenever filtering
or PVID changes, we update the filter according to the
specification.

Cc: Vladimir Oltean <olteanv@gmail.com>
Cc: Mauri Sandberg <sandberg@mailfence.com>
Cc: Alvin Šipraga <alsi@bang-olufsen.dk>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: DENG Qingfang <dqfext@gmail.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: rtl8366: Drop custom VLAN set-up
Linus Walleij [Sat, 25 Sep 2021 22:59:25 +0000 (00:59 +0200)]
net: dsa: rtl8366: Drop custom VLAN set-up

This hacky default VLAN setup was done in order to direct
packets to the right ports and provide port isolation, both
which we now support properly using custom tags and proper
bridge port isolation.

We can drop the custom VLAN code and leave all VLAN handling
alone, as users expect things to be. We can also drop
ds->configure_vlan_while_not_filtering = false; and let
the core deal with any VLANs it wants.

Cc: Mauri Sandberg <sandberg@mailfence.com>
Cc: DENG Qingfang <dqfext@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: rtl8366rb: Support bridge offloading
DENG Qingfang [Sat, 25 Sep 2021 22:59:24 +0000 (00:59 +0200)]
net: dsa: rtl8366rb: Support bridge offloading

Use port isolation registers to configure bridge offloading.

Tested on the D-Link DIR-685, switching between ports and
sniffing ports to make sure no packets leak.

Cc: Vladimir Oltean <olteanv@gmail.com>
Cc: Mauri Sandberg <sandberg@mailfence.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: DENG Qingfang <dqfext@gmail.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'devlink_register-last'
David S. Miller [Mon, 27 Sep 2021 15:32:21 +0000 (16:32 +0100)]
Merge branch 'devlink_register-last'

Leon Romanovsky says:

====================
Move devlink_register to be last devlink command

This is second version of patch series
https://lore.kernel.org/netdev/cover.1628599239.git.leonro@nvidia.com/

The main change is addition of delayed notification logic that will
allowed us to delete devlink_params_publish API (future series will
remove it completely) and conversion of all drivers to have devlink_register
being last commend.

The series itself is pretty straightforward, except liquidio driver
which performs initializations in various workqueues without proper
locks. That driver doesn't hole device_lock and it is clearly broken
for any parallel driver core flows (modprobe + devlink + PCI reset will
100% crash it).

In order to annotate devlink_register() will lockdep of holding
device_lock, I added workaround in this driver.

Thanks

----------------------
From previous cover letter:
Hi Dave and Jakub,

This series prepares code to remove devlink_reload_enable/_disable API
and in order to do, we move all devlink_register() calls to be right
before devlink_reload_enable().

The best place for such a call should be right before exiting from
the probe().

This is done because devlink_register() opens devlink netlink to the
users and gives them a venue to issue commands before initialization
is finished.

1. Some drivers were aware of such "functionality" and tried to protect
themselves with extra locks, state machines and devlink_reload_enable().
Let's assume that it worked for them, but I'm personally skeptical about
it.

2. Some drivers copied that pattern, but without locks and state
machines. That protected them from reload flows, but not from any _set_
routines.

3. And all other drivers simply didn't understand the implications of early
devlink_register() and can be seen as "broken".
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: dsa: Move devlink registration to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:23:01 +0000 (14:23 +0300)]
net: dsa: Move devlink registration to be last devlink command

This change prevents from users to access device before devlink
is fully configured.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agostaging: qlge: Move devlink registration to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:23:00 +0000 (14:23 +0300)]
staging: qlge: Move devlink registration to be last devlink command

This change prevents from users to access device before devlink is
fully configured.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoptp: ocp: Move devlink registration to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:59 +0000 (14:22 +0300)]
ptp: ocp: Move devlink registration to be last devlink command

This change prevents from users to access device before devlink is
fully configured.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: wwan: iosm: Move devlink_register to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:58 +0000 (14:22 +0300)]
net: wwan: iosm: Move devlink_register to be last devlink command

This change prevents from users to access device before devlink is
fully configured. Indirectly this change fixes the commit mentioned
below where devlink_unregister() was prematurely removed.

Fixes: db4278c55fa5 ("devlink: Make devlink_register to be void")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetdevsim: Move devlink registration to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:57 +0000 (14:22 +0300)]
netdevsim: Move devlink registration to be last devlink command

This change prevents from users to access device before devlink is
fully configured.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ethernet: ti: Move devlink registration to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:56 +0000 (14:22 +0300)]
net: ethernet: ti: Move devlink registration to be last devlink command

This change prevents from users to access device before devlink is
fully configured.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoqed: Move devlink registration to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:55 +0000 (14:22 +0300)]
qed: Move devlink registration to be last devlink command

This change prevents from users to access device before devlink is
fully configured.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoionic: Move devlink registration to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:54 +0000 (14:22 +0300)]
ionic: Move devlink registration to be last devlink command

This change prevents from users to access device before devlink is
fully configured.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonfp: Move delink_register to be last command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:53 +0000 (14:22 +0300)]
nfp: Move delink_register to be last command

Open user space access to the devlink after driver is probed.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: mscc: ocelot: delay devlink registration to the end
Leon Romanovsky [Sat, 25 Sep 2021 11:22:52 +0000 (14:22 +0300)]
net: mscc: ocelot: delay devlink registration to the end

Open access to the devlink interface when the driver fully initialized.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agomlxsw: core: Register devlink instance last
Leon Romanovsky [Sat, 25 Sep 2021 11:22:51 +0000 (14:22 +0300)]
mlxsw: core: Register devlink instance last

Make sure that devlink is open to receive user input when all
parameters are initialized.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/mlx5: Accept devlink user input after driver initialization complete
Leon Romanovsky [Sat, 25 Sep 2021 11:22:50 +0000 (14:22 +0300)]
net/mlx5: Accept devlink user input after driver initialization complete

The change of devlink_alloc() to accept device makes sure that device
is fully initialized and device_register() does nothing except allowing
users to use that devlink instance.

Such change ensures that no user input will be usable till that point and
it eliminates the need to worry about internal locking as long as devlink_register
is called last since all accesses to the devlink are during initialization.

This change fixes the following lockdep warning.

 ======================================================
 WARNING: possible circular locking dependency detected
 5.14.0-rc2+ #27 Not tainted
 ------------------------------------------------------
 devlink/265 is trying to acquire lock:
 ffff8880133c2bc0 (&dev->intf_state_mutex){+.+.}-{3:3}, at: mlx5_unload_one+0x1e/0xa0 [mlx5_core]
 but task is already holding lock:
 ffffffff8362b468 (devlink_mutex){+.+.}-{3:3}, at: devlink_nl_pre_doit+0x2b/0x8d0
 which lock already depends on the new lock.
 the existing dependency chain (in reverse order) is:

 -> #1 (devlink_mutex){+.+.}-{3:3}:
        __mutex_lock+0x149/0x1310
        devlink_register+0xe7/0x280
        mlx5_devlink_register+0x118/0x480 [mlx5_core]
        mlx5_init_one+0x34b/0x440 [mlx5_core]
        probe_one+0x480/0x6e0 [mlx5_core]
        pci_device_probe+0x2a0/0x4a0
        really_probe+0x1cb/0xba0
        __driver_probe_device+0x18f/0x470
        driver_probe_device+0x49/0x120
        __driver_attach+0x1ce/0x400
        bus_for_each_dev+0x11e/0x1a0
        bus_add_driver+0x309/0x570
        driver_register+0x20f/0x390
        0xffffffffa04a0062
        do_one_initcall+0xd5/0x400
        do_init_module+0x1c8/0x760
        load_module+0x7d9d/0xa4b0
        __do_sys_finit_module+0x118/0x1a0
        do_syscall_64+0x3d/0x90
        entry_SYSCALL_64_after_hwframe+0x44/0xae

 -> #0 (&dev->intf_state_mutex){+.+.}-{3:3}:
        __lock_acquire+0x2999/0x5a40
        lock_acquire+0x1a9/0x4a0
        __mutex_lock+0x149/0x1310
        mlx5_unload_one+0x1e/0xa0 [mlx5_core]
        mlx5_devlink_reload_down+0x185/0x2b0 [mlx5_core]
        devlink_reload+0x1f2/0x640
        devlink_nl_cmd_reload+0x6c3/0x10d0
        genl_family_rcv_msg_doit+0x1e9/0x2f0
        genl_rcv_msg+0x27f/0x4a0
        netlink_rcv_skb+0x11e/0x340
        genl_rcv+0x24/0x40
        netlink_unicast+0x433/0x700
        netlink_sendmsg+0x6fb/0xbe0
        sock_sendmsg+0xb0/0xe0
        __sys_sendto+0x192/0x240
        __x64_sys_sendto+0xdc/0x1b0
        do_syscall_64+0x3d/0x90
        entry_SYSCALL_64_after_hwframe+0x44/0xae

 other info that might help us debug this:

  Possible unsafe locking scenario:

        CPU0                    CPU1
        ----                    ----
   lock(devlink_mutex);
                                lock(&dev->intf_state_mutex);
                                lock(devlink_mutex);
   lock(&dev->intf_state_mutex);

  *** DEADLOCK ***

 3 locks held by devlink/265:
  #0: ffffffff836371d0 (cb_lock){++++}-{3:3}, at: genl_rcv+0x15/0x40
  #1: ffffffff83637288 (genl_mutex){+.+.}-{3:3}, at: genl_rcv_msg+0x31a/0x4a0
  #2: ffffffff8362b468 (devlink_mutex){+.+.}-{3:3}, at: devlink_nl_pre_doit+0x2b/0x8d0

 stack backtrace:
 CPU: 0 PID: 265 Comm: devlink Not tainted 5.14.0-rc2+ #27
 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
 Call Trace:
  dump_stack_lvl+0x45/0x59
  check_noncircular+0x268/0x310
  ? print_circular_bug+0x460/0x460
  ? __kernel_text_address+0xe/0x30
  ? alloc_chain_hlocks+0x1e6/0x5a0
  __lock_acquire+0x2999/0x5a40
  ? lockdep_hardirqs_on_prepare+0x3e0/0x3e0
  ? add_lock_to_list.constprop.0+0x6c/0x530
  lock_acquire+0x1a9/0x4a0
  ? mlx5_unload_one+0x1e/0xa0 [mlx5_core]
  ? lock_release+0x6c0/0x6c0
  ? lockdep_hardirqs_on_prepare+0x3e0/0x3e0
  ? lock_is_held_type+0x98/0x110
  __mutex_lock+0x149/0x1310
  ? mlx5_unload_one+0x1e/0xa0 [mlx5_core]
  ? lock_is_held_type+0x98/0x110
  ? mlx5_unload_one+0x1e/0xa0 [mlx5_core]
  ? find_held_lock+0x2d/0x110
  ? mutex_lock_io_nested+0x1160/0x1160
  ? mlx5_lag_is_active+0x72/0x90 [mlx5_core]
  ? lock_downgrade+0x6d0/0x6d0
  ? do_raw_spin_lock+0x12e/0x270
  ? rwlock_bug.part.0+0x90/0x90
  ? mlx5_unload_one+0x1e/0xa0 [mlx5_core]
  mlx5_unload_one+0x1e/0xa0 [mlx5_core]
  mlx5_devlink_reload_down+0x185/0x2b0 [mlx5_core]
  ? netlink_broadcast_filtered+0x308/0xac0
  ? mlx5_devlink_info_get+0x1f0/0x1f0 [mlx5_core]
  ? __build_skb_around+0x110/0x2b0
  ? __alloc_skb+0x113/0x2b0
  devlink_reload+0x1f2/0x640
  ? devlink_unregister+0x1e0/0x1e0
  ? security_capable+0x51/0x90
  devlink_nl_cmd_reload+0x6c3/0x10d0
  ? devlink_nl_cmd_get_doit+0x1e0/0x1e0
  ? devlink_nl_pre_doit+0x72/0x8d0
  genl_family_rcv_msg_doit+0x1e9/0x2f0
  ? __lock_acquire+0x15e2/0x5a40
  ? genl_family_rcv_msg_attrs_parse.constprop.0+0x240/0x240
  ? mutex_lock_io_nested+0x1160/0x1160
  ? security_capable+0x51/0x90
  genl_rcv_msg+0x27f/0x4a0
  ? genl_get_cmd+0x3c0/0x3c0
  ? lock_acquire+0x1a9/0x4a0
  ? devlink_nl_cmd_get_doit+0x1e0/0x1e0
  ? lock_release+0x6c0/0x6c0
  netlink_rcv_skb+0x11e/0x340
  ? genl_get_cmd+0x3c0/0x3c0
  ? netlink_ack+0x930/0x930
  genl_rcv+0x24/0x40
  netlink_unicast+0x433/0x700
  ? netlink_attachskb+0x750/0x750
  ? __alloc_skb+0x113/0x2b0
  netlink_sendmsg+0x6fb/0xbe0
  ? netlink_unicast+0x700/0x700
  ? netlink_unicast+0x700/0x700
  sock_sendmsg+0xb0/0xe0
  __sys_sendto+0x192/0x240
  ? __x64_sys_getpeername+0xb0/0xb0
  ? do_sys_openat2+0x10a/0x370
  ? down_write_nested+0x150/0x150
  ? do_user_addr_fault+0x215/0xd50
  ? __x64_sys_openat+0x11f/0x1d0
  ? __x64_sys_open+0x1a0/0x1a0
  __x64_sys_sendto+0xdc/0x1b0
  ? syscall_enter_from_user_mode+0x1d/0x50
  do_syscall_64+0x3d/0x90
  entry_SYSCALL_64_after_hwframe+0x44/0xae
 RIP: 0033:0x7f50b50b6b3a
 Code: d8 64 89 02 48 c7 c0 ff ff ff ff eb b8 0f 1f 00 f3 0f 1e fa 41 89 ca 64 8b 04 25 18 00 00 00 85 c0 75 15 b8 2c 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 76 c3 0f 1f 44 00 00 55 48 83 ec 30 44 89 4c
 RSP: 002b:00007fff6c0d3f38 EFLAGS: 00000246 ORIG_RAX: 000000000000002c
 RAX: ffffffffffffffda RBX: 0000000000000005 RCX: 00007f50b50b6b3a
 RDX: 0000000000000038 RSI: 000055763ac08440 RDI: 0000000000000003
 RBP: 000055763ac08410 R08: 00007f50b5192200 R09: 000000000000000c
 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
 R13: 0000000000000000 R14: 000055763ac08410 R15: 000055763ac08440
 mlx5_core 0000:00:09.0: firmware version: 4.8.9999
 mlx5_core 0000:00:09.0: 0.000 Gb/s available PCIe bandwidth (8.0 GT/s PCIe x255 link)
 mlx5_core 0000:00:09.0 eth1: Link up

Fixes: a6f3b62386a0 ("net/mlx5: Move devlink registration before interfaces load")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/mlx4: Move devlink_register to be the last initialization command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:49 +0000 (14:22 +0300)]
net/mlx4: Move devlink_register to be the last initialization command

Refactor the code to make sure that devlink_register() is the last
command during initialization stage.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/prestera: Split devlink and traps registrations to separate routines
Leon Romanovsky [Sat, 25 Sep 2021 11:22:48 +0000 (14:22 +0300)]
net/prestera: Split devlink and traps registrations to separate routines

Separate devlink registrations and traps registrations so devlink will
be registered when driver is fully initialized.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoocteontx2: Move devlink registration to be last devlink command
Leon Romanovsky [Sat, 25 Sep 2021 11:22:47 +0000 (14:22 +0300)]
octeontx2: Move devlink registration to be last devlink command

This change prevents from users to access device before devlink is fully
configured. This change allows us to delete call to devlink_params_publish()
and impossible check during unregister flow.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoice: Open devlink when device is ready
Leon Romanovsky [Sat, 25 Sep 2021 11:22:46 +0000 (14:22 +0300)]
ice: Open devlink when device is ready

Move devlink_registration routine to be the last command, when the
device is fully initialized.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: hinic: Open device for the user access when it is ready
Leon Romanovsky [Sat, 25 Sep 2021 11:22:45 +0000 (14:22 +0300)]
net: hinic: Open device for the user access when it is ready

Move devlink registration to be the last command in device activation,
so it opens the driver to accept such devlink commands from the user
when it is fully initialized.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodpaa2-eth: Register devlink instance at the end of probe
Leon Romanovsky [Sat, 25 Sep 2021 11:22:44 +0000 (14:22 +0300)]
dpaa2-eth: Register devlink instance at the end of probe

Move devlink_register to be the last command in the initialization
sequence.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoliquidio: Overcome missing device lock protection in init/remove flows
Leon Romanovsky [Sat, 25 Sep 2021 11:22:43 +0000 (14:22 +0300)]
liquidio: Overcome missing device lock protection in init/remove flows

The liquidio driver is broken by design. It initialize PCI devices
in separate delayed works. It causes to the situation where device lock
is dropped during initialize and remove sequences.

That lock is part of driver/core and needed to protect from races during
init, destroy and bus invocations.

In addition to lack of locking protection, it has incorrect order of
destroy flows and very questionable synchronization scheme based on
atomic_t.

This change doesn't fix that driver but makes sure that rest of the
netdev subsystem doesn't suffer from such basic protection by adding
device_lock over devlink_*() APIs and by moving devlink_register()
to be last command in setup_nic_devices().

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Register devlink instance at the end devlink configuration
Leon Romanovsky [Sat, 25 Sep 2021 11:22:42 +0000 (14:22 +0300)]
bnxt_en: Register devlink instance at the end devlink configuration

Move devlink_register() to be last command in devlink configuration
sequence, so no user space access will be possible till devlink instance
is fully operable. As part of this change, the devlink_params_publish
call is removed as not needed.

This change fixes forgotten devlink_params_unpublish() too.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodevlink: Notify users when objects are accessible
Leon Romanovsky [Sat, 25 Sep 2021 11:22:41 +0000 (14:22 +0300)]
devlink: Notify users when objects are accessible

The devlink core code notified users about add/remove objects without
relation if this object can be accessible or not. In this patch we unify
such user visible notifications in one place.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agocxgb: avoid open-coded offsetof()
Arnd Bergmann [Mon, 27 Sep 2021 12:16:04 +0000 (14:16 +0200)]
cxgb: avoid open-coded offsetof()

clang-14 does not like the custom offsetof() macro in vsc7326:

drivers/net/ethernet/chelsio/cxgb/vsc7326.c:597:3: error: performing pointer subtraction with a null pointer has undefined behavior [-Werror,-Wnull-pointer-subtraction]
                HW_STAT(RxUnicast, RxUnicastFramesOK),
                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/ethernet/chelsio/cxgb/vsc7326.c:594:56: note: expanded from macro 'HW_STAT'
        { reg, (&((struct cmac_statistics *)NULL)->stat_name) - (u64 *)NULL }

Rewrite this to use the version provided by the kernel.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: stmmac: fix gcc-10 -Wrestrict warning
Arnd Bergmann [Mon, 27 Sep 2021 10:02:44 +0000 (12:02 +0200)]
net: stmmac: fix gcc-10 -Wrestrict warning

gcc-10 and later warn about a theoretical array overrun when
accessing priv->int_name_rx_irq[i] with an out of bounds value
of 'i':

drivers/net/ethernet/stmicro/stmmac/stmmac_main.c: In function 'stmmac_request_irq_multi_msi':
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:3528:17: error: 'snprintf' argument 4 may overlap destination object 'dev' [-Werror=restrict]
 3528 |                 snprintf(int_name, int_name_len, "%s:%s-%d", dev->name, "tx", i);
      |                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:3404:60: note: destination object referenced by 'restrict'-qualified argument 1 was declared here
 3404 | static int stmmac_request_irq_multi_msi(struct net_device *dev)
      |                                         ~~~~~~~~~~~~~~~~~~~^~~

The warning is a bit strange since it's not actually about the array
bounds but rather about possible string operations with overlapping
arguments, but it's not technically wrong.

Avoid the warning by adding an extra bounds check.

Fixes: 8532f613bc78 ("net: stmmac: introduce MSI Interrupt routines for mac, safety, RX & TX")
Link: https://lore.kernel.org/all/20210421134743.3260921-1-arnd@kernel.org/
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ethernet: emac: utilize of_net's of_get_mac_address()
Christian Lamparter [Sun, 26 Sep 2021 09:56:48 +0000 (11:56 +0200)]
net: ethernet: emac: utilize of_net's of_get_mac_address()

of_get_mac_address() reads the same "local-mac-address" property.
... But goes above and beyond by checking the MAC value properly.

printk+message seems outdated too,
so let's put dev_err in the queue.

Signed-off-by: Christian Lamparter <chunkeey@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>