Merge tag 'mlx5-updates-2023-08-07' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2023-08-07

1) Few cleanups

2) Dynamic completion EQs

The driver creates completion EQs for all vectors directly on driver
load, even if those EQs will not be utilized later on.

To allow more flexibility in managing completion EQs and to reduce the
memory overhead of driver load, this series will adjust completion EQs
creation to be dynamic. Meaning, completion EQs will be created only
when needed.

Patch #1 introduces a counter for tracking the current number of
completion EQs.
Patches #2-6 refactor the existing infrastructure of managing completion
EQs and completion IRQs to be compatible with per-vector
allocation/release requests.
Patches #7-8 modify the CPU-to-IRQ affinity calculation to be resilient
in case the affinity is requested but completion IRQ is not allocated yet.
Patch #9 function rename.
Patch #10 handles the corner case of SF performing an IRQ request when no
SF IRQ pool is found, and no PF IRQ exists for the same vector.
Patch #11 modify driver to use dynamically allocate completion EQs.

* tag 'mlx5-updates-2023-08-07' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
  net/mlx5: Bridge, Only handle registered netdev bridge events
  net/mlx5: E-Switch, Remove redundant arg ignore_flow_lvl
  net/mlx5: Fix typo reminder -> remainder
  net/mlx5: remove many unnecessary NULL values
  net/mlx5: Allocate completion EQs dynamically
  net/mlx5: Handle SF IRQ request in the absence of SF IRQ pool
  net/mlx5: Rename mlx5_comp_vectors_count() to mlx5_comp_vectors_max()
  net/mlx5: Add IRQ vector to CPU lookup function
  net/mlx5: Introduce mlx5_cpumask_default_spread
  net/mlx5: Implement single completion EQ create/destroy methods
  net/mlx5: Use xarray to store and manage completion EQs
  net/mlx5: Refactor completion IRQ request/release handlers in EQ layer
  net/mlx5: Use xarray to store and manage completion IRQs
  net/mlx5: Refactor completion IRQ request/release API
  net/mlx5: Track the current number of completion EQs
====================

Link: https://lore.kernel.org/r/20230807175642.20834-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

docs: net: page_pool: de-duplicate the intro comment

In commit 82e896d992fa ("docs: net: page_pool: use kdoc to avoid
duplicating the information") I shied away from using the DOC:
comments when moving to kdoc for documenting page_pool API,
because I wasn't sure how familiar people are with it.

Turns out there is already a DOC: comment for the intro, which
is the same in both places, modulo what looks like minor rewording.
Use the version from Documentation/ but keep the contents with
the code.

Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>
Link: https://lore.kernel.org/r/20230807210051.1014580-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: Remove unused devlink_dpipe_table_resource_set() declaration

Commit f655dacb59ac ("net: devlink: remove unused locked functions")
removed this but leave the declaration.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230807143214.46648-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fq: Remove unused typedef fq_flow_get_default_t

Commitbf9009bf21b5 ("net/fq_impl: drop get_default_func, move default flow to fq_tin")
remove its last user, so can remove it.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230807142111.33524-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/tls: avoid TCP window full during ->read_sock()

When flushing the backlog after decoding a record we don't really
know how much data the caller want us to evaluate, so use INT_MAX
and 0 as arguments to tls_read_flush_backlog() to ensure we flush
at 128k of data. Otherwise we might be reading too much data and
trigger a TCP window full.

Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Link: https://lore.kernel.org/r/20230807071022.10091-1-hare@suse.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ipv6: exthdrs: Replace opencoded swap() implementation

Get a coccinelle warning as follows:
net/ipv6/exthdrs.c:800:29-30: WARNING opportunity for swap()

Use swap() to replace opencoded implementation.

Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230807020947.1991716-1-william.xuanziyang@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/ipv4: return the real errno instead of -EINVAL

For now, No matter what error pointer ip_neigh_for_gw() returns,
ip_finish_output2() always return -EINVAL, which may mislead the upper
users.

For exemple, an application uses sendto to send an UDP packet, but when the
neighbor table overflows, sendto() will get a value of -EINVAL, and it will
cause users to waste a lot of time checking parameters for errors.

Return the real errno instead of -EINVAL.

Signed-off-by: xu xin <xu.xin16@zte.com.cn>
Reviewed-by: Yang Yang <yang.yang29@zte.com.cn>
Cc: Si Hao <si.hao@zte.com.cn>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://lore.kernel.org/r/20230807015408.248237-1-xu.xin16@zte.com.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-renesas-rswitch-add-speed-change-support'

Yoshihiro Shimoda says:

====================
net: renesas: rswitch: Add speed change support

Add speed change support at runtime for the latest SoC version.
Also, add ethtool .[gs]et_link_ksettings.

v1: https://lore.kernel.org/all/20230803120621.1471440-1-yoshihiro.shimoda.uh@renesas.com/
====================

Link: https://lore.kernel.org/r/20230807003231.1552062-1-yoshihiro.shimoda.uh@renesas.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: renesas: rswitch: Add .[gs]et_link_ksettings support

Add .[gs]et_link_ksettings support by using
phy_ethtool_[gs]et_link_ksettings() functions.

Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230807003231.1552062-3-yoshihiro.shimoda.uh@renesas.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: renesas: rswitch: Add runtime speed change support

The latest SoC version can support runtime speed change. So,
add detect SoC version by using soc_device_match() and then
reconfigure the hardware of this and SerDes if needed.

Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230807003231.1552062-2-yoshihiro.shimoda.uh@renesas.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

rtnetlink: remove redundant checks for nlattr IFLA_BRIDGE_MODE

The commit d73ef2d69c0d ("rtnetlink: let rtnl_bridge_setlink checks
IFLA_BRIDGE_MODE length") added the nla_len check in rtnl_bridge_setlink,
which is the only caller for ndo_bridge_setlink handlers defined in
low-level driver codes. Hence, this patch cleanups the redundant checks in
each ndo_bridge_setlink handler function.

Suggested-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Lin Ma <linma@zju.edu.cn>
Acked-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230807091347.3804523-1-linma@zju.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'bnxt_en-fix-2-compile-warnings-in-bnxt_dcb-c'

Michael Chan says:

====================
bnxt_en: Fix 2 compile warnings in bnxt_dcb.c

Fix 2 similar warnings in bnxt_dcb.c related to the cos2bw interface
definitions in bnxt_hsi.h.
====================

Link: https://lore.kernel.org/r/20230807145720.159645-1-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

bnxt_en: Fix W=stringop-overflow warning in bnxt_dcb.c

Fix the following warning:

drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c: In function ‘bnxt_hwrm_queue_cos2bw_cfg’:
cc1: error: writing 12 bytes into a region of size 1 [-Werror=stringop-overflow ]
In file included from drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c:19:
drivers/net/ethernet/broadcom/bnxt/bnxt_hsi.h:6045:17: note: destination object ‘unused_0’ of size 1
 6045 |         u8      unused_0;

Fix it by modifying struct hwrm_queue_cos2bw_cfg_input to use an array
of sub struct similar to the previous patch.  This will eliminate the
pointer arithmetc to calculate the destination pointer passed to
memcpy().

Link: https://lore.kernel.org/netdev/CACKFLinikvXmKcxr4kjWO9TPYxTd2cb5agT1j=w9Qyj5-24s5A@mail.gmail.com/
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Link: https://lore.kernel.org/r/20230807145720.159645-3-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

bnxt_en: Fix W=1 warning in bnxt_dcb.c from fortify memcpy()

Fix the following warning:

inlined from ‘bnxt_hwrm_queue_cos2bw_qcfg’ at drivers/net/ethernet/broadcom/bnxt/bnxt_dcb.c:165:3,
./include/linux/fortify-string.h:592:4: error: call to ‘__read_overflow2_field’ declared with attribute warning: detected read beyond size of field (2nd parameter); maybe use struct_group()? [-Werror]
    __read_overflow2_field(q_size_field, size);
    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Modify the FW interface defintion of struct hwrm_queue_cos2bw_qcfg_output
to use an array of sub struct for the queue1 to queue7 fields.  Note that
the layout of the queue0 fields are different and these are not part of
the array.  This makes the code much cleaner by removing the pointer
arithmetic for memcpy().

Link: https://lore.kernel.org/netdev/20230727190726.1859515-2-kuba@kernel.org/
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Link: https://lore.kernel.org/r/20230807145720.159645-2-michael.chan@broadcom.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: bcmasp: Prevent array undereflow in bcmasp_netfilt_get_init()

The "loc" value comes from the user and it can be negative leading to an
an array underflow when we check "priv->net_filters[loc].claimed".  Fix
this by changing the type to u32.

Fixes: c5d511c49587 ("net: bcmasp: Add support for wake on net filters")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Justin Chen <justin.chen@broadcom.com>
Link: https://lore.kernel.org/r/b3b47b25-01fc-4d9f-a6c3-e037ad4d71d7@moroto.mountain
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'team-do-some-cleanups-in-team-driver'

Zhengchao Shao says:

====================
team: do some cleanups in team driver

Do some cleanups in team driver.
====================

Link: https://lore.kernel.org/r/20230807012556.3146071-1-shaozhengchao@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

team: remove unused input parameters in lb_htpm_select_tx_port and lb_hash_select_tx_port

The input parameters "lb_priv" and "skb" in lb_htpm_select_tx_port and
lb_hash_select_tx_port are unused, so remove them.

Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230807012556.3146071-6-shaozhengchao@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

team: change the getter function in the team_option structure to void

Because the getter function in the team_option structure always returns 0,
so change the getter function to void and remove redundant code.

Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230807012556.3146071-5-shaozhengchao@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

team: change the init function in the team_option structure to void

Because the init function in the team_option structure always returns 0,
so change the init function to void and remove redundant code.

Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230807012556.3146071-4-shaozhengchao@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

team: remove unreferenced header in broadcast and roundrobin files

Because linux/errno.h is unreferenced in broadcast and roundrobin
files, so remove it.

Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230807012556.3146071-3-shaozhengchao@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

team: add __exit modifier to team_nl_fini()

team_nl_fini is only called when the module exits, so add the __exit
modifier to it.

Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230807012556.3146071-2-shaozhengchao@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-fs_enet-driver-cleanup'

Christophe Leroy says:

====================
net: fs_enet: Driver cleanup

Over the years, platform and driver initialisation have evolved into
more generic ways, and driver or platform specific stuff has gone
away, leaving stale objects behind.

This series aims at cleaning all that up for fs_enet ethernet driver.
====================

Link: https://lore.kernel.org/r/cover.1691155346.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Use cpm_muram_xxx() functions instead of cpm_dpxxx() macros

cpm_dpxxx() macros are now always referring to cpm_muram_xxx() functions
directly since commit 3dd82a1ea724 ("[POWERPC] CPM: Always use new
binding.")

Use cpm_muram_xxx() functions directly so that the cpm_dpxxx() macros
can be removed in the near future.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/2400b3156891adb653dc387fff6393de10cf2b24.1691155347.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Remove linux/fs_enet_pd.h

linux/fs_enet_pd.h is not used anymore.

Remove it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/5be102791c987792ad127b15543ee6715394cf67.1691155347.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Don't include fs_enet_pd.h when not needed

Three platforms in arch/powerpc/platforms/8xx/ include fs_enet_pd.h
but don't use anything from it.

Remove the includes.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/de62ad1261a801c4a8ae4238bd4842ff278d2ddf.1691155347.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Move struct fs_platform_info into fs_enet.h

struct fs_platform_info is only used in fs_enet ethernet driver since
commit 3dd82a1ea724 ("[POWERPC] CPM: Always use new binding.").

Stale prototypes using fs_platform_info were left over in
arch/powerpc/sysdev/fsl_soc.c but they are now removed by
previous patch.

Move struct fs_platform_info into fs_enet.h

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/f882d6b0b7075d0d8435310634ceaa2cc8e9938f.1691155347.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Remove stale prototypes from fsl_soc.c

Commit 3dd82a1ea724 ("[POWERPC] CPM: Always use new binding.")
removed last use of init_fec_ioports() and init_fcc_ioports().

Remove stale prototypes then don't include anymore fs_enet_pd.h
which was only included to provide fs_platform_info structure
for the prototypes.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/f2f2db5dce3242eaa4a2aad6c226ba587fb0f513.1691155347.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Remove has_phy field in fs_platform_info struct

Since commit 3dd82a1ea724 ("[POWERPC] CPM: Always use new binding.")
has_phy field is never set.

Remove dead code and remove the field.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/bb5264e09e18f0ce8a0dbee399926a59f33cb248.1691155346.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Remove unused fields in fs_platform_info struct

Since commit 3dd82a1ea724 ("[POWERPC] CPM: Always use new binding.")
many fields of fs_platform_info structure are not used anymore.

Remove them.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/2e584fcd75e21a0f7e7d5f942eebdc067b2f82f9.1691155346.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Remove fs_get_id()

fs_get_id() hasn't been used since commit b219108cbace ("fs_enet:
Remove !CONFIG_PPC_CPM_NEW_BINDING code")

Remove it.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/7a53b88cc40302fcbea59554f5e7067e3594ad53.1691155346.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Fix address space and base types mismatches

CHECK   drivers/net/ethernet/freescale/fs_enet/mac-fcc.c
drivers/net/ethernet/freescale/fs_enet/mac-fcc.c:550:9: warning: cast removes address space '__iomem' of expression
drivers/net/ethernet/freescale/fs_enet/mac-fcc.c:550:9: error: subtraction of different types can't work (different address spaces)
  CC      drivers/net/ethernet/freescale/fs_enet/mii-bitbang.o
  CHECK   drivers/net/ethernet/freescale/fs_enet/mii-bitbang.c
drivers/net/ethernet/freescale/fs_enet/mii-bitbang.c:95:31: warning: incorrect type in argument 1 (different base types)
drivers/net/ethernet/freescale/fs_enet/mii-bitbang.c:95:31:    expected unsigned int [noderef] [usertype] __iomem *p
drivers/net/ethernet/freescale/fs_enet/mii-bitbang.c:95:31:    got restricted __be32 [noderef] [usertype] __iomem *dat
...
drivers/net/ethernet/freescale/fs_enet/mii-bitbang.c:63:31: warning: incorrect type in argument 1 (different base types)
drivers/net/ethernet/freescale/fs_enet/mii-bitbang.c:63:31:    expected unsigned int [noderef] [usertype] __iomem *p
drivers/net/ethernet/freescale/fs_enet/mii-bitbang.c:63:31:    got restricted __be32 [noderef] [usertype] __iomem *dir
...

Fix those address space and base type mismatches reported by sparse.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/25c7965e6aeeb6bbe1b6be5a3c2c7125182fcb02.1691155346.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fs_enet: Remove set but not used variable

CC      drivers/net/ethernet/freescale/fs_enet/fs_enet-main.o
drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c: In function 'fs_enet_interrupt':
drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c:321:40: warning: variable 'fpi' set but not used [-Wunused-but-set-variable]

Remove that variable.

Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/90b72c1708bb8ba2b7a1a688e8259e428968364d.1691155346.git.christophe.leroy@csgroup.eu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: hns: Remove unused function declaration mac_adjust_link()

Commit 511e6bc071db ("net: add Hisilicon Network Subsystem DSAF support")
declared but never implemented this, remove it.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Hao Lan <lanhao@huawei.com>
Link: https://lore.kernel.org/r/20230804130048.39808-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

i40e: Remove unused function declarations

Commit f62b5060d670 ("i40e: fix mac address checking") left behind
i40e_validate_mac_addr() declaration.
Also the other declarations are declared but never implemented in
commit 56a62fc86895 ("i40e: init code and hardware support").

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230804125525.20244-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ixgbe: Remove unused function declarations

Commit dc166e22ede5 ("ixgbe: DCB remove ixgbe_fcoe_getapp routine")
leave ixgbe_fcoe_getapp() unused.
Commit ffed21bcee7a ("ixgbe: Don't bother clearing buffer memory for descriptor rings")
leave ixgbe_unmap_and_free_tx_resource() declaration unused.
And commit 3b3bf3b92b31 ("ixgbe: remove unused fcoe.tc field and fcoe_setapp()")
removed the ixgbe_fcoe_setapp() implementation.

Commit c44ade9ef8ff ("ixgbe: update to latest common code module")
declared but never implemented ixgbe_init_ops_generic().

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230804125203.30924-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-remove-redundant-initialization-owner'

Li Zetao says:

====================
net: Remove redundant initialization owner

This patch set removes redundant initialization owner when register a
fsl_mc_driver driver
====================

Link: https://lore.kernel.org/r/20230804095946.99956-1-lizetao1@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dpaa2-switch: Remove redundant initialization owner in dpaa2_switch_drv

The fsl_mc_driver_register() will set "THIS_MODULE" to driver.owner when
register a fsl_mc_driver driver, so it is redundant initialization to set
driver.owner in dpaa2_switch_drv statement. Remove it for clean code.

Signed-off-by: Li Zetao <lizetao1@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230804095946.99956-3-lizetao1@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dpaa2-eth: Remove redundant initialization owner in dpaa2_eth_driver

The fsl_mc_driver_register() will set "THIS_MODULE" to driver.owner when
register a fsl_mc_driver driver, so it is redundant initialization to set
driver.owner in dpaa2_eth_driver statement. Remove it for clean code.

Signed-off-by: Li Zetao <lizetao1@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230804095946.99956-2-lizetao1@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'octeontx2-af-tc-flower-offload-changes'

Suman Ghosh says:

====================
octeontx2-af: TC flower offload changes

This patchset includes minor code restructuring related to TC
flower offload for outer vlan and adding support for TC inner
vlan offload.

Patch #1 Code restructure to handle TC flower outer vlan offload

Patch #2 Add TC flower offload support for inner vlan
====================

Link: https://lore.kernel.org/r/20230804045935.3010554-1-sumang@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

octeontx2-af: TC flower offload support for inner VLAN

Extend the current TC flower offload support to enable filters matching
inner VLAN, and support offload of those filters to hardware.

Signed-off-by: Suman Ghosh <sumang@marvell.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230804045935.3010554-3-sumang@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

octeontx2-af: Code restructure to handle TC outer VLAN offload

Moved the TC outer VLAN offload support to a separate function.
This change is done to handle all VLAN related changes cleanly from
a dedicated function.

Signed-off-by: Suman Ghosh <sumang@marvell.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230804045935.3010554-2-sumang@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'page_pool-a-couple-of-assorted-optimizations'

Alexander Lobakin says:

====================
page_pool: a couple of assorted optimizations

That initially was a spin-off of the IAVF PP series[0], but has grown
(and shrunk) since then a bunch. In fact, it consists of three
semi-independent blocks:

* #1-2: Compile-time optimization. Split page_pool.h into 2 headers to
  not overbloat the consumers not needing complex inline helpers and
  then stop including it in skbuff.h at all. The first patch is also
  prereq for the whole series.
* #3: Improve cacheline locality for users of the Page Pool frag API.
* #4-6: Use direct cache recycling more aggressively, when it is safe
  obviously. In addition, make sure nobody wants to use Page Pool API
  with disabled interrupts.

Patches #1 and #5 are authored by Yunsheng and Jakub respectively, with
small modifications from my side as per ML discussions.
For the perf numbers for #3-6, please see individual commit messages.

Also available on my GH with many more Page Pool goodies[1].

[0] https://lore.kernel.org/netdev/20230530150035.1943669-1-aleksander.lobakin@intel.com
[1] https://github.com/alobakin/linux/commits/iavf-pp-frag
====================

Link: https://lore.kernel.org/r/20230804180529.2483231-1-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: skbuff: always try to recycle PP pages directly when in softirq

Commit 8c48eea3adf3 ("page_pool: allow caching from safely localized
NAPI") allowed direct recycling of skb pages to their PP for some cases,
but unfortunately missed a couple of other majors.
For example, %XDP_DROP in skb mode. The netstack just calls kfree_skb(),
which unconditionally passes `false` as @napi_safe. Thus, all pages go
through ptr_ring and locks, although most of time we're actually inside
the NAPI polling this PP is linked with, so that it would be perfectly
safe to recycle pages directly.
Let's address such. If @napi_safe is true, we're fine, don't change
anything for this path. But if it's false, check whether we are in the
softirq context. It will most likely be so and then if ->list_owner
is our current CPU, we're good to use direct recycling, even though
@napi_safe is false -- concurrent access is excluded. in_softirq()
protection is needed mostly due to we can hit this place in the
process context (not the hardirq though).
For the mentioned xdp-drop-skb-mode case, the improvement I got is
3-4% in Mpps. As for page_pool stats, recycle_ring is now 0 and
alloc_slow counter doesn't change most of time, which means the
MM layer is not even called to allocate any new pages.

Suggested-by: Jakub Kicinski <kuba@kernel.org> # in_softirq()
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-7-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

page_pool: add a lockdep check for recycling in hardirq

Page pool use in hardirq is prohibited, add debug checks
to catch misuses. IIRC we previously discussed using
DEBUG_NET_WARN_ON_ONCE() for this, but there were concerns
that people will have DEBUG_NET enabled in perf testing.
I don't think anyone enables lockdep in perf testing,
so use lockdep to avoid pushback and arguing :)

Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-6-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: skbuff: avoid accessing page_pool if !napi_safe when returning page

Currently, pp->p.napi is always read, but the actual variable it gets
assigned to is read-only when @napi_safe is true. For the !napi_safe
cases, which yet is still a pack, it's an unneeded operation.
Moreover, it can lead to premature or even redundant page_pool
cacheline access. For example, when page_pool_is_last_frag() returns
false (with the recent frag improvements).
Thus, read it only when @napi_safe is true. This also allows moving
@napi inside the condition block itself. Constify it while we are
here, because why not.

Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-5-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

page_pool: place frag_* fields in one cacheline

On x86_64, frag_* fields of struct page_pool are scattered across two
cachelines despite the summary size of 24 bytes. All three fields are
used in pretty much the same places, but the last field, ::frag_users,
is pushed out to the next CL, provoking unwanted false-sharing on
hotpath (frags allocation code).
There are some holes and cold members to move around. Move frag_* one
block up, placing them right after &page_pool_params perfectly at the
beginning of CL2. This doesn't do any meaningful to the second block, as
those are some destroy-path cold structures, and doesn't do anything to
::alloc_stats, which still starts at 200-byte offset, 8 bytes after CL3
(still fitting into 1 cacheline).
On my setup, this yields 1-2% of Mpps when using PP frags actively.
When it comes to 32-bit architectures with 32-byte CL: &page_pool_params
plus ::pad is 44 bytes, the block taken care of is 16 bytes within one
CL, so there should be at least no regressions from the actual change.
::pages_state_hold_cnt is not related directly to that triple, but is
paired currently with ::frags_offset and decoupling them would mean
either two 4-byte holes or more invasive layout changes.

Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-4-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: skbuff: don't include <net/page_pool/types.h> to <linux/skbuff.h>

Currently, touching <net/page_pool/types.h> triggers a rebuild of more
than half of the kernel. That's because it's included in
<linux/skbuff.h>. And each new include to page_pool/types.h adds more
[useless] data for the toolchain to process per each source file from
that pile.

In commit 6a5bcd84e886 ("page_pool: Allow drivers to hint on SKB
recycling"), Matteo included it to be able to call a couple of functions
defined there. Then, in commit 57f05bc2ab24 ("page_pool: keep pp info as
long as page pool owns the page") one of the calls was removed, so only
one was left. It's the call to page_pool_return_skb_page() in
napi_frag_unref(). The function is external and doesn't have any
dependencies. Having very niche page_pool_types.h included only for that
looks like an overkill.

As %PP_SIGNATURE is not local to page_pool.c (was only in the
early submissions), nothing holds this function there. Teleport
page_pool_return_skb_page() to skbuff.c, just next to the main consumer,
skb_pp_recycle(), and rename it to napi_pp_put_page(), as it doesn't
work with skbs at all and the former name tells nothing. The #if guards
here are only to not compile and have it in the vmlinux when not needed
-- both call sites are already guarded.
Now, touching page_pool_types.h only triggers rebuilding of the drivers
using it and a couple of core networking files.

Suggested-by: Jakub Kicinski <kuba@kernel.org> # make skbuff.h less heavy
Suggested-by: Alexander Duyck <alexanderduyck@fb.com> # move to skbuff.c
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-3-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

page_pool: split types and declarations from page_pool.h

Split types and pure function declarations from page_pool.h
and add them in page_page/types.h, so that C sources can
include page_pool.h and headers should generally only include
page_pool/types.h as suggested by jakub.
Rename page_pool.h to page_pool/helpers.h to have both in
one place.

Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-2-aleksander.lobakin@intel.com
[Jakub: change microsoft/mana, fix kdoc paths in Documentation]
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'linux-can-next-for-6.6-20230807' of git://git./linux/kernel/git/mkl/linux-can-next

Marc Kleine-Budde says:

====================
pull-request: can-next 2023-08-07

The patch is from me and reverts the addition of the CAN controller
nodes in the allwinner d1 SoC.

* tag 'linux-can-next-for-6.6-20230807' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can-next:
  Revert "riscv: dts: allwinner: d1: Add CAN controller nodes"
====================

Link: https://lore.kernel.org/r/20230807074222.1576119-1-mkl@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-stmmac-correct-mac-propagation-delay'

Johannes Zink says:

====================
net: stmmac: correct MAC propagation delay

Changes in v3:
- work in Richard's review feedback. Thank you for reviewing my patch:
  - as some of the hardware may have no or invalid correction value
    registers: introduce feature switch which can be enabled in the glue
    code drivers depending on the actual hardware support
  - only enable the feature on the i.MX8MP for the time being, as the patch
    improves timing accuracy and is tested for this hardware
- Link to v2: https://lore.kernel.org/r/20230719-stmmac_correct_mac_delay-v2-1-3366f38ee9a6@pengutronix.de

Changes in v2:
- fix builds for 32bit, this was found by the kernel build bot
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202307200225.B8rmKQPN-lkp@intel.com/
- while at it also fix an overflow by shifting a u32 constant from macro by 10bits
  by casting the constant to u64
- Link to v1: https://lore.kernel.org/r/20230719-stmmac_correct_mac_delay-v1-1-768aa4d09334@pengutronix.de

Tested-by: Kurt Kanzenbach <kurt@linutronix.de> # imx8mp
====================

Link: https://lore.kernel.org/r/20230719-stmmac_correct_mac_delay-v3-0-61e63427735e@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: dwmac-imx: enable MAC propagation delay correction for i.MX8MP

As the i.MX8MP supports reading MAC propagation delay and correcting the
Hardware timestamp counter for additional delays [1], enable the feature
for this SoC.

This reduces phase error of the PPS output from the PTP Hardware Clock
from approx 150ns to 100ns.

[1] i.MX8MP Reference Manual, rev.1 Section 11.7.2.5.3 "Timestamp
correction"

Signed-off-by: Johannes Zink <j.zink@pengutronix.de>
Link: https://lore.kernel.org/r/20230719-stmmac_correct_mac_delay-v3-2-61e63427735e@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: correct MAC propagation delay

The IEEE1588 Standard specifies that the timestamps of Packets must be
captured when the PTP message timestamp point (leading edge of first
octet after the start of frame delimiter) crosses the boundary between
the node and the network. As the MAC latches the timestamp at an
internal point, the captured timestamp must be corrected for the
additional data transmission latency, as described in the publicly
available datasheet [1].

This patch only corrects for the MAC-Internal delay, which can be read
out from the MAC_Ingress_Timestamp_Latency register on DWMAC version 5,
since the Phy framework currently does not support querying the Phy
ingress and egress latency. The Closs Domain Crossing Circuits errors as
indicated in [1] are already being accounted in the
stmmac_get_tx_hwtstamp() function and are not corrected here.

As the Latency varies for different link speeds and MII
modes of operation, the correction value needs to be updated on each
link state change.

As the delay also causes a phase shift in the timestamp counter compared
to the rest of the network, this correction will also reduce phase error
when generating PPS outputs from the timestamp counter.

Since the correction registers may be unavailable on some hardware and
no feature bits are documented for dynamically detection of the MAC
propagation delay readout, introduce a feature bit to explicitely enable
MAC delay Correction in the gluecode driver.

[1] i.MX8MP Reference Manual, rev.1 Section 11.7.2.5.3 "Timestamp
correction"

Signed-off-by: Johannes Zink <j.zink@pengutronix.de>
Link: https://lore.kernel.org/r/20230719-stmmac_correct_mac_delay-v2-1-3366f38ee9a6@pengutronix.de
Link: https://lore.kernel.org/r/20230719-stmmac_correct_mac_delay-v3-1-61e63427735e@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Bridge, Only handle registered netdev bridge events

Don't handle bridge events for a netdev that doesn't belong to
an eswitch that registered for bridge events.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: E-Switch, Remove redundant arg ignore_flow_lvl

The arg is always passed as true and thus redundant.

Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Fix typo reminder -> remainder

Fix a typo in esw_qos_devlink_rate_to_mbps(): reminder -> remainder.

Signed-off-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Rahul Rameshbabu <rrameshbabu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: remove many unnecessary NULL values

There are many pointers assigned first, which need not to be
initialized, so remove the NULL assignments.

Signed-off-by: Ruan Jinjie <ruanjinjie@huawei.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Allocate completion EQs dynamically

This commit enables the dynamic allocation of EQs at runtime, allowing
for more flexibility in managing completion EQs and reducing the memory
overhead of driver load. Whenever a CQ is created for a given vector
index, the driver will lookup to see if there is an already mapped
completion EQ for that vector, if so, utilize it. Otherwise, allocate a
new EQ on demand and then utilize it for the CQ completion events.

Add a protection lock to the EQ table to protect from concurrent EQ
creation attempts.

While at it, replace mlx5_vector2irqn()/mlx5_vector2eqn() with
mlx5_comp_eqn_get() and mlx5_comp_irqn_get() which will allocate an
EQ on demand if no EQ is found for the given vector.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Handle SF IRQ request in the absence of SF IRQ pool

In case the SF IRQ pool is not available due to setup limitations,
SF currently relies on the already allocated PF IRQs to fulfill
its IRQ vector requests.

However, with the dynamic EQ allocation introduced in the next patch,
it is possible that not all IRQs of PF will be allocated after the driver
is loaded. In such case, if a SF requests a completion IRQ without having
its own independent IRQ pool, SF will lack a PF IRQ to utilize.

To address this scenario, allocate an IRQ for the SF from the PF's IRQ pool
on demand. The new IRQ will be shared between the SF and it's PF.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Rename mlx5_comp_vectors_count() to mlx5_comp_vectors_max()

To accurately represent its purpose, rename the function that retrieves
the value of maximum vectors from mlx5_comp_vectors_count() to
mlx5_comp_vectors_max().

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Add IRQ vector to CPU lookup function

Currently, once driver load completes, IRQ requests were performed for all
vectors. However, as we move to support dynamic creation of EQs, this will
not be the case as some IRQs will not exist at this stage. Thus, in such
case, use the default CPU to IRQ mapping which is the serial mapping based
on IRQ vector index. Meaning, the n'th vector gets mapped to the n'th CPU.

Introduce an API function mlx5_comp_vector_cpu() that takes an IRQ index and
provides the corresponding CPU mapping. It utilizes the existing IRQ
affinity if defined, or resorts to the default serialized CPU mapping
otherwise.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Introduce mlx5_cpumask_default_spread

For better code readability in the completion IRQ request code, define
the cpu lookup per completion vector logic in a separate function.

The new method mlx5_cpumask_default_spread() given a vector index 'n'
will return the 'nth' cpu. This new method will be used also in the next
patch.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Implement single completion EQ create/destroy methods

Currently, create_comp_eqs() function handles the creation of all
completion EQs for all the vectors on driver load. While on driver
unload, destroy_comp_eqs() performs the equivalent job.

In preparation for dynamic EQ creation, replace create_comp_eqs() /
destroy_comp_eqs() with  create_comp_eq() / destroy_comp_eq() functions
which will receive a vector index and allocate/destroy an EQ for that
specific vector. Thus, allowing more flexibility in the management
of completion EQs.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Use xarray to store and manage completion EQs

Use xarray to store the completion EQs instead of a linked list.
The xarray offers more scalability, reduced memory overhead, and
facilitates the lookup of a certain EQ given a vector index.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Refactor completion IRQ request/release handlers in EQ layer

Break the completion IRQ request/release functions into per-vector
handlers for both PCI devices and SFs in the EQ layer.

On EQ table creation, loop over all vectors and request an IRQ for each
one using the new per-vector functions. Perform the symmetrical change
when releasing IRQs on EQ table cleanup.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Use xarray to store and manage completion IRQs

Use xarray to store the completion IRQs instead of a fixed-size allocated
array as not all completion IRQs will be requested on driver load, but
rather on demand when an EQ is created. The xarray offers more scalability,
reduced memory overhead, and provides the ability to dynamically resize the
array when needed.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Refactor completion IRQ request/release API

Introduce a per-vector completion IRQ request API that requests a
single IRQ for a given vector index instead of multiple IRQs request API.
On driver load, loop over all completion vectors and request an IRQ for
each one via the newly introduced API.

Symmetrically, introduce an IRQ release API per vector. On driver
unload, loop over all vectors and release each completion IRQ via
the new per-vector API.

As IRQ vectors will be requested dynamically later in the patchset,
add a cpumask of the bounded CPUs to avoid the possible mapping of
two IRQs of the same device to the same cpu.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

net/mlx5: Track the current number of completion EQs

In preparation to allocate completion EQs, add a counter to track the
number of completion EQs currently allocated. Store the maximum number
of EQs in max_comp_eqs variable.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

udp/udplite: Remove unused function declarations udp{,lite}_get_port()

Commit 6ba5a3c52da0 ("[UDP]: Make full use of proto.h.udp_hash innovation.")
removed these implementations but leave declarations.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: sfp: Remove unused function declaration sfp_link_configure()

Commit ce0aa27ff3f6 ("sfp: add sfp-bus to bridge between network devices and sfp cages")
declared but never implemented it.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ndisc: Remove unused ndisc_ifinfo_sysctl_strategy() declaration

Commit f8572d8f2a2b ("sysctl net: Remove unused binary sysctl code")
left behind this declaration.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: pkt_cls: Remove unused inline helpers

Commit acb674428c3d ("net: sched: introduce per-block callbacks")
implemented these but never used it.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

neighbour: Remove unused function declaration pneigh_for_each()

pneigh_for_each() is never implemented since the beginning of git history.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/tls: Remove unused function declarations

Commit 3c4d7559159b ("tls: kernel TLS support") declared but never implemented
these functions.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Revert "riscv: dts: allwinner: d1: Add CAN controller nodes"

It turned out the dtsi changes were not quite ready, revert them for
now.

This reverts commit 6ea1ad888f5900953a21853e709fa499fdfcb317.

Link: https://lore.kernel.org/all/2690764.mvXUDI8C0e@jernej-laptop
Suggested-by: Jernej Škrabec <jernej.skrabec@gmail.com>
Link: https://lore.kernel.org/all/20230807-riscv-allwinner-d1-revert-can-controller-nodes-v1-1-eb3f70b435d9@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

net: omit ndo_hwtstamp_get() call when possible in dev_set_hwtstamp_phylib()

Setting dev->priv_flags & IFF_SEE_ALL_HWTSTAMP_REQUESTS is only legal
for drivers which were converted to ndo_hwtstamp_get() and
ndo_hwtstamp_set(), and it is only there that we call ndo_hwtstamp_set()
for a request that otherwise goes to phylib (for stuff like packet traps,
which need to be undone if phylib failed, hence the old_cfg logic).

The problem is that we end up calling ndo_hwtstamp_get() when we don't
need to (even if the SIOCSHWTSTAMP wasn't intended for phylib, or if it
was, but the driver didn't set IFF_SEE_ALL_HWTSTAMP_REQUESTS). For those
unnecessary conditions, we share a code path with virtual drivers (vlan,
macvlan, bonding) where ndo_hwtstamp_get() is implemented as
generic_hwtstamp_get_lower(), and may be resolved through
generic_hwtstamp_ioctl_lower() if the lower device is unconverted.

I.e. this situation:

$ ip link add link eno0 name eno0.100 type vlan id 100
$ hwstamp_ctl -i eno0.100 -t 1

We are unprepared to deal with this, because if ndo_hwtstamp_get() is
resolved through a legacy ndo_eth_ioctl(SIOCGHWTSTAMP) lower_dev
implementation, that needs a non-NULL old_cfg.ifr pointer, and we don't
have it.

But we don't even need to deal with it either. In the general case,
drivers may not even implement SIOCGHWTSTAMP handling, only SIOCSHWTSTAMP,
so it makes sense to completely avoid a SIOCGHWTSTAMP call if we can.

The solution is to split the single "if" condition into 3 smaller ones,
thus separating the decision to call ndo_hwtstamp_get() from the
decision to call ndo_hwtstamp_set(). The third "if" condition is
identical to the first one, and both are subsets of the second one.
Thus, the "cfg" argument of kernel_hwtstamp_config_changed() is always
valid.

Reported-by: Eric Dumazet <edumazet@google.com>
Closes: https://lore.kernel.org/netdev/CANn89iLOspJsvjPj+y8jikg7erXDomWe8sqHMdfL_2LQSFrPAg@mail.gmail.com/
Fixes: fd770e856e22 ("net: remove phy_has_hwtstamp() -> phy_mii_ioctl() decision from converted drivers")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ethernet: adi: adin1110: use eth_broadcast_addr() to assign broadcast address

Use eth_broadcast_addr() to assign broadcast address instead
of memset().

Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ibmvnic: remove unused rc variable

gcc with W=1 reports
drivers/net/ethernet/ibm/ibmvnic.c:194:13: warning: variable 'rc' set but not used [-Wunused-but-set-variable]
                                            ^
This variable is not used so remove it.

Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202308040609.zQsSXWXI-lkp@intel.com/
Signed-off-by: Yu Liao <liaoyu15@huawei.com>
Reviewed-by: Nick Child <nnac123@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: mana: Add page pool for RX buffers

Add page pool for RX buffers for faster buffer cycle and reduce CPU
usage.

The standard page pool API is used.

With iperf and 128 threads test, this patch improved the throughput
by 12-15%, and decreased the IRQ associated CPU's usage from 99-100% to
10-50%.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'gve-desc'

Rushil Gupta says:

====================
gve: Add QPL mode for DQO descriptor format

GVE supports QPL ("queue-page-list") mode where
all data is communicated through a set of pre-registered
pages. Adding this mode to DQO.
====================

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

gve: update gve.rst

Add a note about QPL and RDA mode

Signed-off-by: Rushil Gupta <rushilg@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Signed-off-by: Bailey Forrest <bcf@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gve: RX path for DQO-QPL

The RX path allocates the QPL page pool at queue creation, and
tries to reuse these pages through page recycling. This patch
ensures that on refill no non-QPL pages are posted to the device.

When the driver is running low on free buffers, an ondemand
allocation step kicks in that allocates a non-qpl page for
SKB business to free up the QPL page in use.

gve_try_recycle_buf was moved to gve_rx_append_frags so that driver does
not attempt to mark buffer as used if a non-qpl page was allocated
ondemand.

Signed-off-by: Rushil Gupta <rushilg@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Signed-off-by: Bailey Forrest <bcf@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gve: Tx path for DQO-QPL

Each QPL page is divided into GVE_TX_BUFS_PER_PAGE_DQO buffers.
When a packet needs to be transmitted, we break the packet into max
GVE_TX_BUF_SIZE_DQO sized chunks and transmit each chunk using a TX
descriptor.
We allocate the TX buffers from the free list in dqo_tx.
We store these TX buffer indices in an array in the pending_packet
structure.

The TX buffers are returned to the free list in dqo_compl after
receiving packet completion or when removing packets from miss
completions list.

Signed-off-by: Rushil Gupta <rushilg@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Signed-off-by: Bailey Forrest <bcf@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gve: Control path for DQO-QPL

GVE supports QPL ("queue-page-list") mode where
all data is communicated through a set of pre-registered
pages. Adding this mode to DQO descriptor format.

Add checks, abi-changes and device options to support
QPL mode for DQO in addition to GQI. Also, use
pages-per-qpl supplied by device-option to control the
size of the "queue-page-list".

Signed-off-by: Rushil Gupta <rushilg@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
Signed-off-by: Bailey Forrest <bcf@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'tcp-options-lockless'

Eric Dumazet says:

====================
tcp: set few options locklessly

This series is avoiding the socket lock for six TCP options.

They are not heavily used, but this exercise can give
ideas for other parts of TCP/IP stack :)
====================

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

tcp: set TCP_DEFER_ACCEPT locklessly

rskq_defer_accept field can be read/written without
the need of holding the socket lock.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: set TCP_LINGER2 locklessly

tp->linger2 can be set locklessly as long as readers
use READ_ONCE().

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: set TCP_KEEPCNT locklessly

tp->keepalive_probes can be set locklessly, readers
are already taking care of this field being potentially
set by other threads.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: set TCP_KEEPINTVL locklessly

tp->keepalive_intvl can be set locklessly, readers
are already taking care of this field being potentially
set by other threads.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: set TCP_USER_TIMEOUT locklessly

icsk->icsk_user_timeout can be set locklessly,
if all read sides use READ_ONCE().

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: set TCP_SYNCNT locklessly

icsk->icsk_syn_retries can safely be set without locking the socket.

We have to add READ_ONCE() annotations in tcp_fastopen_synack_timer()
and tcp_write_timeout().

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'wireless-next-2023-08-04' of git://git./linux/kernel/git/wireless/wireless-next

Kalle Valo says:

====================
wireless-next patches for v6.6

The first pull request for v6.6 and only driver patches this time.
Nothing special really standing out, it has been quiet most likely due
to vacations.

Major changes:

rtl8xxxu
 - enable AP mode for: RTL8192FU, RTL8710BU (RTL8188GU),
   RTL8192EU and RTL8723BU

mwifiex
 - allow moving to a different namespace

mt76
 - preparation for mt7925 support
 - mt7981 support

ath12k
 - Extremely High Throughput (EHT) PHY support for Wi-Fi 7

* tag 'wireless-next-2023-08-04' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next: (172 commits)
  wifi: rtw89: return failure if needed firmware elements are not recognized
  wifi: rtw89: add to parse firmware elements of BB and RF tables
  wifi: rtw89: introduce infrastructure of firmware elements
  wifi: rtw89: add firmware suit for BB MCU 0/1
  wifi: rtw89: add firmware parser for v1 format
  wifi: rtw89: introduce v1 format of firmware header
  wifi: rtw89: support firmware log with formatted text
  wifi: rtw89: recognize log format from firmware file
  wifi: ath12k: avoid deadlock by change ieee80211_queue_work for regd_update_work
  wifi: ath12k: add handler for scan event WMI_SCAN_EVENT_DEQUEUED
  wifi: ath12k: relax list iteration in ath12k_mac_vif_unref()
  wifi: ath12k: configure puncturing bitmap
  wifi: ath12k: parse WMI service ready ext2 event
  wifi: ath12k: add MLO header in peer association
  wifi: ath12k: peer assoc for 320 MHz
  wifi: ath12k: add WMI support for EHT peer
  wifi: ath12k: prepare EHT peer assoc parameters
  wifi: ath12k: add EHT PHY modes
  wifi: ath12k: propagate EHT capabilities to userspace
  wifi: ath12k: WMI support to process EHT capabilities
  ...
====================

Link: https://lore.kernel.org/r/87msz7j942.fsf@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'tcp-disable-header-prediction-for-md5'

Kuniyuki Iwashima says:

====================
tcp: Disable header prediction for MD5.

The 1st patch disable header prediction for MD5 flow and the 2nd
patch updates the stale comment in tcp_parse_options().
====================

Link: https://lore.kernel.org/r/20230803224552.69398-1-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: Update stale comment for MD5 in tcp_parse_options().

Since commit 9ea88a153001 ("tcp: md5: check md5 signature without socket
lock"), the MD5 option is checked in tcp_v[46]_rcv().

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230803224552.69398-3-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: Disable header prediction for MD5 flow.

TCP socket saves the minimum required header length in tcp_header_len
of struct tcp_sock, and later the value is used in __tcp_fast_path_on()
to generate a part of TCP header in tcp_sock(sk)->pred_flags.

In tcp_rcv_established(), if the incoming packet has the same pattern
with pred_flags, we enter the fast path and skip full option parsing.

The MD5 option is parsed in tcp_v[46]_rcv(), so we need not parse it
again later in tcp_rcv_established() unless other options exist.  We
add TCPOLEN_MD5SIG_ALIGNED to tcp_header_len in two paths to avoid the
slow path.

For passive open connections with MD5, we add TCPOLEN_MD5SIG_ALIGNED
to tcp_header_len in tcp_create_openreq_child() after 3WHS.

On the other hand, we do it in tcp_connect_init() for active open
connections.  However, the value is overwritten while processing
SYN+ACK or crossed SYN in tcp_rcv_synsent_state_process().

These two cases will have the wrong value in pred_flags and never go
into the fast path.

We could update tcp_header_len in tcp_rcv_synsent_state_process(), but
a test with slightly modified netperf which uses MD5 for each flow shows
that the slow path is actually a bit faster than the fast path.

  On c5.4xlarge EC2 instance (16 vCPU, 32 GiB mem)

  $ for i in {1..10}; do
  ./super_netperf $(nproc) -H localhost -l 10 -- -m 256 -M 256;
  done

  Avg of 10
  * 36e68eadd303  : 10.376 Gbps
  * all fast path : 10.374 Gbps (patch v2, See Link)
  * all slow path : 10.394 Gbps

The header prediction is not worth adding complexity for MD5, so let's
disable it for MD5.

Link: https://lore.kernel.org/netdev/20230803042214.38309-1-kuniyu@amazon.com/
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20230803224552.69398-2-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: phy: move marking PHY on SFP module into SFP code

Move marking the PHY as being on a SFP module into the SFP code between
getting the PHY device (and thus initialising the phy_device structure)
and registering the discovered device.

This means that PHY drivers can use phy_on_sfp() in their match and
get_features methods.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qRaga-001vKt-8X@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: spectrum: Remove unused function declarations

Commit c3d2ed93b14d ("mlxsw: Remove old parsing depth infrastructure")
left behind mlxsw_sp_nve_inc_parsing_depth_get()/mlxsw_sp_nve_inc_parsing_depth_put().
And commit 532b49e41e64 ("mlxsw: spectrum_span: Derive SBIB from maximum port speed & MTU")
remove mlxsw_sp_span_port_mtu_update()/mlxsw_sp_span_speed_update_work() but leave the
declarations.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://lore.kernel.org/r/20230803142047.42660-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ixgbevf: Remove unused function declarations

ixgbe_napi_add_all()/ixgbe_napi_del_all() are declared but never implemented in
commit 92915f71201b ("ixgbevf: Driver main and ethool interface module and main header")

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230803141904.15316-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

af_vsock: Remove unused declaration vsock_release_pending()/vsock_init_tap()

Commit d021c344051a ("VSOCK: Introduce VM Sockets") declared but never implemented
vsock_release_pending(). Also vsock_init_tap() never implemented since introduction
in commit 531b374834c8 ("VSOCK: Add vsockmon tap functions").

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://lore.kernel.org/r/20230803134507.22660-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: 802: Remove unused function declarations

Commit d8d9ba8dc9c7 ("net: 802: remove dead leftover after ipx driver removal")
remove these implementations but leave the declarations.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230803135424.41664-1-yuehaibing@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp_metrics: hash table allocation cleanup

After commit 098a697b497e ("tcp_metrics: Use a single hash table
for all network namespaces.") we can avoid calling tcp_net_metrics_init()
for each new netns.

Instead, rename tcp_net_metrics_init() to tcp_metrics_hash_alloc(),
and move it to __init section.

Also move tcpmhash_entries to __initdata section.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230803135417.2716879-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>