platform/kernel/linux-rpi.git
18 months agonetfilter: nf_tables: merge nft_rules_old structure and end of ruleblob marker
Florian Westphal [Tue, 11 Apr 2023 14:29:45 +0000 (16:29 +0200)]
netfilter: nf_tables: merge nft_rules_old structure and end of ruleblob marker

In order to free the rules in a chain via call_rcu, the rule array used
to stash a rcu_head and space for a pointer at the end of the rule array.

When the current nft_rule_dp blob format got added in
2c865a8a28a1 ("netfilter: nf_tables: add rule blob layout"), this results
in a double-trailer:

  size (unsigned long)
  struct nft_rule_dp
    struct nft_expr
         ...
    struct nft_rule_dp
     struct nft_expr
         ...
    struct nft_rule_dp (is_last=1) // Trailer

The trailer, struct nft_rule_dp (is_last=1), is not accounted for in size,
so it can be located via start_addr + size.

Because the rcu_head is stored after 'start+size' as well this means the
is_last trailer is *aliased* to the rcu_head (struct nft_rules_old).

This is harmless, because at this time the nft_do_chain function never
evaluates/accesses the trailer, it only checks the address boundary:

        for (; rule < last_rule; rule = nft_rule_next(rule)) {
...

But this way the last_rule address has to be stashed in the jump
structure to restore it after returning from a chain.

nft_do_chain stack usage has become way too big, so put it on a diet.

Without this patch is impossible to use
        for (; !rule->is_last; rule = nft_rule_next(rule)) {

... because on free, the needed update of the rcu_head will clobber the
nft_rule_dp is_last bit.

Furthermore, also stash the chain pointer in the trailer, this allows
to recover the original chain structure from nf_tables_trace infra
without a need to place them in the jump struct.

After this patch it is trivial to diet the jump stack structure,
done in the next two patches.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
18 months agoMerge tag 'wireless-next-2023-04-21' of git://git.kernel.org/pub/scm/linux/kernel...
Jakub Kicinski [Fri, 21 Apr 2023 14:35:51 +0000 (07:35 -0700)]
Merge tag 'wireless-next-2023-04-21' of git://git./linux/kernel/git/wireless/wireless-next

Kalle Valo says:

====================
wireless-next patches for v6.4

Most likely the last -next pull request for v6.4. We have changes all
over. rtw88 now supports SDIO bus and iwlwifi continues to work on
Wi-Fi 7 support. Not much stack changes this time.

Major changes:

cfg80211/mac80211
 - fix some Fine Time Measurement (FTM) frames not being bufferable
 - flush frames before key removal to avoid potential unencrypted
   transmission depending on the hardware design

iwlwifi
 - preparation for Wi-Fi 7 EHT and multi-link support

rtw88
 - SDIO bus support
 - RTL8822BS, RTL8822CS and RTL8821CS SDIO chipset support

rtw89
 - framework firmware backwards compatibility

brcmfmac
 - Cypress 43439 SDIO support

mt76
 - mt7921 P2P support
 - mt7996 mesh A-MSDU support
 - mt7996 EHT support
 - mt7996 coredump support

wcn36xx
 - support for pronto v3 hardware

ath11k
 - PCIe DeviceTree bindings
 - WCN6750: enable SAR support

ath10k
 - convert DeviceTree bindings to YAML

* tag 'wireless-next-2023-04-21' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next: (261 commits)
  wifi: rtw88: Update spelling in main.h
  wifi: airo: remove ISA_DMA_API dependency
  wifi: rtl8xxxu: Simplify setting the initial gain
  wifi: rtl8xxxu: Add rtl8xxxu_write{8,16,32}_{set,clear}
  wifi: rtl8xxxu: Don't print the vendor/product/serial
  wifi: rtw88: Fix memory leak in rtw88_usb
  wifi: rtw88: call rtw8821c_switch_rf_set() according to chip variant
  wifi: rtw88: set pkg_type correctly for specific rtw8821c variants
  wifi: rtw88: rtw8821c: Fix rfe_option field width
  wifi: rtw88: usb: fix priority queue to endpoint mapping
  wifi: rtw88: 8822c: add iface combination
  wifi: rtw88: handle station mode concurrent scan with AP mode
  wifi: rtw88: prevent scan abort with other VIFs
  wifi: rtw88: refine reserved page flow for AP mode
  wifi: rtw88: disallow PS during AP mode
  wifi: rtw88: 8822c: extend reserved page number
  wifi: rtw88: add port switch for AP mode
  wifi: rtw88: add bitmap for dynamic port settings
  wifi: rtw89: mac: use regular int as return type of DLE buffer request
  wifi: mac80211: remove return value check of debugfs_create_dir()
  ...
====================

Link: https://lore.kernel.org/r/20230421104726.800BCC433D2@smtp.kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet/packet: support mergeable feature of virtio
Jianfeng Tan [Wed, 19 Apr 2023 07:24:16 +0000 (15:24 +0800)]
net/packet: support mergeable feature of virtio

Packet sockets, like tap, can be used as the backend for kernel vhost.
In packet sockets, virtio net header size is currently hardcoded to be
the size of struct virtio_net_hdr, which is 10 bytes; however, it is not
always the case: some virtio features, such as mrg_rxbuf, need virtio
net header to be 12-byte long.

Mergeable buffers, as a virtio feature, is worthy of supporting: packets
that are larger than one-mbuf size will be dropped in vhost worker's
handle_rx if mrg_rxbuf feature is not used, but large packets
cannot be avoided and increasing mbuf's size is not economical.

With this virtio feature enabled by virtio-user, packet sockets with
hardcoded 10-byte virtio net header will parse mac head incorrectly in
packet_snd by taking the last two bytes of virtio net header as part of
mac header.
This incorrect mac header parsing will cause packet to be dropped due to
invalid ether head checking in later under-layer device packet receiving.

By adding extra field vnet_hdr_sz with utilizing holes in struct
packet_sock to record currently used virtio net header size and supporting
extra sockopt PACKET_VNET_HDR_SZ to set specified vnet_hdr_sz, packet
sockets can know the exact length of virtio net header that virtio user
gives.
In packet_snd, tpacket_snd and packet_recvmsg, instead of using
hardcoded virtio net header size, it can get the exact vnet_hdr_sz from
corresponding packet_sock, and parse mac header correctly based on this
information to avoid the packets being mistakenly dropped.

Signed-off-by: Jianfeng Tan <henry.tjf@antgroup.com>
Co-developed-by: Anqi Shen <amy.saq@antgroup.com>
Signed-off-by: Anqi Shen <amy.saq@antgroup.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'mlx5-ipsec-fixes'
David S. Miller [Fri, 21 Apr 2023 10:49:47 +0000 (11:49 +0100)]
Merge branch 'mlx5-ipsec-fixes'

Leon Romanovsky says:

====================
Fixes to mlx5 IPsec implementation

This small patchset includes various fixes and one refactoring patch
which I collected for the features sent in this cycle, with one exception -
first patch.

First patch fixes code which was introduced in previous cycle, however I
was able to trigger FW error only in custom debug code, so don't see a
need to send it to net-rc.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet/mlx5e: Refactor duplicated code in mlx5e_ipsec_init_macs
Leon Romanovsky [Thu, 20 Apr 2023 08:02:51 +0000 (11:02 +0300)]
net/mlx5e: Refactor duplicated code in mlx5e_ipsec_init_macs

ARP discovery code has same logic for RX and TX flows, but with
different source and destination fields. Instead of duplicating
same code in mlx5e_ipsec_init_macs, let's refactor.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet/mlx5e: Properly release work data structure
Leon Romanovsky [Thu, 20 Apr 2023 08:02:50 +0000 (11:02 +0300)]
net/mlx5e: Properly release work data structure

There are some flows in which work structure is not allocated at all
and it is needed to be checked prior release of data structure.

 general protection fault, probably for non-canonical address 0xdffffc000000000a: 0000 [#1] SMP KASAN
 KASAN: null-ptr-deref in range [0x0000000000000050-0x0000000000000057]
 CPU: 6 PID: 3486 Comm: kworker/6:0 Not tainted 6.3.0-rc5_for_upstream_debug_2023_04_06_11_01 #1
 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
 Workqueue: events xfrm_state_gc_task
 RIP: 0010:mlx5e_xfrm_free_state+0x177/0x260 [mlx5_core]
 Code: c1 ea 03 80 3c 02 00 0f 85 f5 00 00 00 4c 8b a5 08 01 00 00 48 b8 00 00 00 00 00 fc ff df 49 8d 7c 24 50 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 b7 00 00 00 49 8b 7c 24 50 e8 85 7c 09 e0 4c 89
 RSP: 0018:ffff888137a8fc50 EFLAGS: 00010206
 RAX: dffffc0000000000 RBX: ffff888180398000 RCX: 0000000000000000
 RDX: 000000000000000a RSI: ffffffffa1878227 RDI: 0000000000000050
 RBP: ffff88812a0c8000 R08: ffff888137a8fb60 R09: 0000000000000000
 R10: fffffbfff09aba0c R11: 0000000000000001 R12: 0000000000000000
 R13: ffff88812a0c8108 R14: ffffffff84c63480 R15: ffff8881acb63118
 FS:  0000000000000000(0000) GS:ffff88881eb00000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 00007f667e8bc000 CR3: 0000000004693006 CR4: 0000000000370ea0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 Call Trace:

  ___xfrm_state_destroy+0x3c8/0x5e0
  xfrm_state_gc_task+0xf6/0x140
  ? ___xfrm_state_destroy+0x5e0/0x5e0
  process_one_work+0x7c2/0x1340
  ? lockdep_hardirqs_on_prepare+0x3f0/0x3f0
  ? pwq_dec_nr_in_flight+0x230/0x230
  ? spin_bug+0x1d0/0x1d0
  worker_thread+0x59d/0xec0
  ? __kthread_parkme+0xd9/0x1d0
  ? process_one_work+0x1340/0x1340
  kthread+0x28f/0x330
  ? kthread_complete_and_exit+0x20/0x20
  ret_from_fork+0x1f/0x30

 Modules linked in: sch_ingress openvswitch nsh mlx5_vdpa vringh vhost_iotlb vdpa mlx5_ib mlx5_core xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink xt_addrtype iptable_nat nf_nat br_netfilter rpcrdma rdma_ucm ib_iser libiscsi scsi_transport_iscsi rdma_cm iw_cm ib_umad ib_ipoib ib_cm ib_uverbs ib_core vfio_pci vfio_pci_core vfio_iommu_type1 vfio cuse overlay zram zsmalloc fuse [last unloaded: mlx5_core]
 ---[ end trace 0000000000000000 ]---

Fixes: 4562116f8a56 ("net/mlx5e: Generalize IPsec work structs")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet/mlx5e: Compare all fields in IPv6 address
Leon Romanovsky [Thu, 20 Apr 2023 08:02:49 +0000 (11:02 +0300)]
net/mlx5e: Compare all fields in IPv6 address

Fix size argument in memcmp to compare whole IPv6 address.

Fixes: b3beba1fb404 ("net/mlx5e: Allow policies with reqid 0, to support IKE policy holes")
Reviewed-by: Raed Salem <raeds@nvidia.com>
Reviewed-by: Emeel Hakim <ehakim@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet/mlx5e: Don't overwrite extack message returned from IPsec SA validator
Leon Romanovsky [Thu, 20 Apr 2023 08:02:48 +0000 (11:02 +0300)]
net/mlx5e: Don't overwrite extack message returned from IPsec SA validator

Addition of new err_xfrm label caused to error messages be overwritten.
Fix it by using proper NL_SET_ERR_MSG_WEAK_MOD macro together with change
in a default message.

Fixes: aa8bd0c9518c ("net/mlx5e: Support IPsec acquire default SA")
Reviewed-by: Raed Salem <raeds@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet/mlx5e: Fix FW error while setting IPsec policy block action
Leon Romanovsky [Thu, 20 Apr 2023 08:02:47 +0000 (11:02 +0300)]
net/mlx5e: Fix FW error while setting IPsec policy block action

When trying to set IPsec policy block action the following error is
generated:

 mlx5_cmd_out_err:803:(pid 3426): SET_FLOW_TABLE_ENTRY(0x936) op_mod(0x0) failed,
status bad parameter(0x3), syndrome (0x8708c3), err(-22)

This error means that drop action is not allowed when modify action is
set, so update the code to skip modify header for XFRM_POLICY_BLOCK action.

Fixes: 6721239672fe ("net/mlx5e: Skip IPsec encryption for TX path without matching policy")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet: stmmac:fix system hang when setting up tag_8021q VLAN for DSA ports
Yan Wang [Wed, 19 Apr 2023 14:13:46 +0000 (22:13 +0800)]
net: stmmac:fix system hang when setting up tag_8021q VLAN for DSA ports

The system hang because of dsa_tag_8021q_port_setup()->
stmmac_vlan_rx_add_vid().

I found in stmmac_drv_probe() that cailing pm_runtime_put()
disabled the clock.

First, when the kernel is compiled with CONFIG_PM=y,The stmmac's
resume/suspend is active.

Secondly,stmmac as DSA master,the dsa_tag_8021q_port_setup() function
will callback stmmac_vlan_rx_add_vid when DSA dirver starts. However,
The system is hanged for the stmmac_vlan_rx_add_vid() accesses its
registers after stmmac's clock is closed.

I would suggest adding the pm_runtime_resume_and_get() to the
stmmac_vlan_rx_add_vid().This guarantees that resuming clock output
while in use.

Fixes: b3dcb3127786 ("net: stmmac: correct clocks enabled in stmmac_vlan_rx_kill_vid()")
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Yan Wang <rk.code@outlook.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'pds_core'
David S. Miller [Fri, 21 Apr 2023 07:29:14 +0000 (08:29 +0100)]
Merge branch 'pds_core'

Shannon Nelson says:

====================
pds_core driver

Summary:
--------
This patchset implements a new driver for use with the AMD/Pensando
Distributed Services Card (DSC), intended to provide core configuration
services through the auxiliary_bus and through a couple of EXPORTed
functions for use initially in VFio and vDPA feature specific drivers.

To keep this patchset to a manageable size, the pds_vdpa and pds_vfio
drivers have been split out into their own patchsets to be reviewed
separately.

Detail:
-------
AMD/Pensando is making available a new set of devices for supporting vDPA,
VFio, and potentially other features in the Distributed Services Card
(DSC).  These features are implemented through a PF that serves as a Core
device for controlling and configuring its VF devices.  These VF devices
have separate drivers that use the auxiliary_bus to work through the Core
device as the control path.

Currently, the DSC supports standard ethernet operations using the
ionic driver.  This is not replaced by the Core-based devices - these
new devices are in addition to the existing Ethernet device.  Typical DSC
configurations will include both PDS devices and Ionic Eth devices.
However, there is a potential future path for ethernet services to come
through this device as well.

The Core device is a new PCI PF/VF device managed by a new driver
'pds_core'.  The PF device has access to an admin queue for configuring
the services used by the VFs, and sets up auxiliary_bus devices for each
vDPA VF for communicating with the drivers for the vDPA devices.  The VFs
may be for VFio or vDPA, and other services in the future; these VF types
are selected as part of the DSC internal FW configurations, which is out
of the scope of this patchset.

When the vDPA support set is enabled in the core PF through its devlink
param, auxiliary_bus devices are created for each VF that supports the
feature.  The vDPA driver then connects to and uses this auxiliary_device
to do control path configuration through the PF device.  This can then be
used with the vdpa kernel module to provide devices for virtio_vdpa kernel
module for host interfaces, or vhost_vdpa kernel module for interfaces
exported into your favorite VM.

A cheap ASCII diagram of a vDPA instance looks something like this:

                                ,----------.
                                |   vdpa   |
                                '----------'
                                  |     ||
                                 ctl   data
                                  |     ||
                          .----------.  ||
                          | pds_vdpa |  ||
                          '----------'  ||
                               |        ||
                       pds_core.vDPA.1  ||
                               |        ||
                    .---------------.   ||
                    |   pds_core    |   ||
                    '---------------'   ||
                        ||         ||   ||
                      09:00.0      09:00.1
        == PCI ============================================
                        ||            ||
                   .----------.   .----------.
            ,------|    PF    |---|    VF    |-------,
            |      '----------'   '----------'       |
            |                  DSC                   |
            |                                        |
            ------------------------------------------

Changes:
  v11:
 - change strncpy to strscpy
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202304181137.WaZTYyAa-lkp@intel.com/
  v10:
Link: https://lore.kernel.org/netdev/20230418003228.28234-1-shannon.nelson@amd.com/
 - remove CONFIG_DEBUG_FS guard static inline stuff
 - remove unnecessary 0 and null initializations
 - verify in driver load that PDS_CORE_DRV_NAME matches KBUILD_MODNAME
 - remove debugfs irqs_show(), redundant with /proc
 - return -ENOMEM if intr_info = kcalloc() fails
 - move the status code enum into pds_core_if.h as part of API definition
 - fix up one place in pdsc_devcmd_wait() we're using the status codes where we could use the errno
 - remove redundant calls to flush_workqueue()
 - grab config_lock before testing state bits in pdsc_fw_reporter_diagnose()
 - change pdsc_color_match() to return bool
 - remove useless VIF setup loop and just setup vDPA services for now
 - remove pf pointer from struct padev and have clients use pci_physfn()
 - drop use of "vf" in auxdev.c function names, make more generic
 - remove last of client ops struct and simply export the functions
 - drop drivers@pensando.io from MAINTAINERS and add new include dir
 - include dynamic_debug.h in adminq.c to protect dynamic_hex_dump()
 - fixed fw_slot type from u8 to int for handling error returns
 - fixed comment spelling
 - changed void arg in pdsc_adminq_post() to struct pdsc *

  v9:
Link: https://lore.kernel.org/netdev/20230406234143.11318-1-shannon.nelson@amd.com/
 - change pdsc field name id to uid to clarify the unique id used for aux device
 - remove unnecessary pf->state and other checks in aux device creation
 - hardcode fw slotnames for devlink info, don't use strings from FW
 - handle errors from PDS_CORE_CMD_INIT devcmd call
 - tighten up health thread use of config_lock
 - remove pdsc_queue_health_check() layer over queuing health check
 - start pds_core.rst file in first patch, add to it incrementally
 - give more user interaction info in commit messages
 - removed a few more extraneous includes

  v8:
Link: https://lore.kernel.org/netdev/20230330234628.14627-1-shannon.nelson@amd.com/
 - fixed deadlock problem, use devl_health_reporter_destroy() when devlink is locked
 - don't clear client_id until after auxiliary_device_uninit()

  v7:
Link: https://lore.kernel.org/netdev/20230330192313.62018-1-shannon.nelson@amd.com/
 - use explicit devlink locking and devl_* APIs
 - move some of devlink setup logic into probe and remove
 - use debugfs_create_u{type}() for state and queue head and tail
 - add include for linux/vmalloc.h
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202303260420.Tgq0qobF-lkp@intel.com/
  v6:
Link: https://lore.kernel.org/netdev/20230324190243.27722-1-shannon.nelson@amd.com/
 - removed version.h include noticed by kernel test robot's version check
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202303230742.pX3ply0t-lkp@intel.com/
 - fixed up the more egregious checkpatch line length complaints
 - make sure pdsc_auxbus_dev_register() checks padev pointer errcode

  v5:
Link: https://lore.kernel.org/netdev/20230322185626.38758-1-shannon.nelson@amd.com/
 - added devlink health reporter for FW issues
 - removed asic_type, asic_rev, serial_num, fw_version from debugfs as
   they are available through other means
 - trimed OS info in pdsc_identify(), we don't need to send that much info to the FW
 - removed reg/unreg from auxbus client API, they are now in the core when VF
   is started
 - removed need for pdsc definition in client by simplifying the padev to only carry
   struct pci_dev pointers rather than full struct pdsc to the pf and vf
 - removed the unused pdsc argument in pdsc_notify()
 - moved include/linux/pds/pds_core.h to driver/../pds_core/core.h
 - restored a few pds_core_if.h interface values and structs that are shared
   with FW source
 - moved final config_lock unlock to before tear down of timer and workqueue
   to be sure there are no deadlocks while waiting for any stragglers
 - changed use of PAGE_SIZE to local PDS_PAGE_SIZE to keep with FW layout needs
   without regard to kernel PAGE_SIZE configuration
 - removed the redundant *adminqcq argument from pdsc_adminq_post()

  v4:
Link: https://lore.kernel.org/netdev/20230308051310.12544-1-shannon.nelson@amd.com/
 - reworked to attach to both Core PF and vDPA VF PCI devices
 - now creates auxiliary_device as part of each VF PCI probe, removes them on PCI remove
 - auxiliary devices now use simple unique id rather than PCI address for identifier
 - replaced home-grown event publishing with kernel-based notifier service
 - dropped live_migration parameter, not needed when not creating aux device for it
 - replaced devm_* functions with traditional interfaces
 - added MAINTAINERS entry
 - removed lingering traces of set/get_vf attribute adminq commands
 - trimmed some include lists
 - cleaned a kernel test robot complaint about a stray unused variable
Link: https://lore.kernel.org/oe-kbuild-all/202302181049.yeUQMeWY-lkp@intel.com/
  v3:
Link: https://lore.kernel.org/netdev/20230217225558.19837-1-shannon.nelson@amd.com/
 - changed names from "pensando" to "amd" and updated copyright strings
 - dropped the DEVLINK_PARAM_GENERIC_ID_FW_BANK for future development
 - changed the auxiliary device creation to be triggered by the
   PCI bus event BOUND_DRIVER, and torn down at UNBIND_DRIVER in order
   to properly handle users using the sysfs bind/unbind functions
 - dropped some noisy log messages
 - rebased to current net-next

  RFC to v2:
Link: https://lore.kernel.org/netdev/20221207004443.33779-1-shannon.nelson@amd.com/
 - added separate devlink param patches for DEVLINK_PARAM_GENERIC_ID_ENABLE_MIGRATION
   and DEVLINK_PARAM_GENERIC_ID_FW_BANK, and dropped the driver specific implementations
 - updated descriptions for the new devlink parameters
 - dropped netdev support
 - dropped vDPA patches, will followup later
 - separated fw update and fw bank select into their own patches

  RFC:
Link: https://lore.kernel.org/netdev/20221118225656.48309-1-snelson@pensando.io/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: Kconfig and pds_core.rst
Shannon Nelson [Wed, 19 Apr 2023 17:04:27 +0000 (10:04 -0700)]
pds_core: Kconfig and pds_core.rst

Remaining documentation and Kconfig hook for building the driver.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: publish events to the clients
Shannon Nelson [Wed, 19 Apr 2023 17:04:26 +0000 (10:04 -0700)]
pds_core: publish events to the clients

When the Core device gets an event from the device, or notices
the device FW to be up or down, it needs to send those events
on to the clients that have an event handler.  Add the code to
pass along the events to the clients.

The entry points pdsc_register_notify() and pdsc_unregister_notify()
are EXPORTed for other drivers that want to listen for these events.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: add the aux client API
Shannon Nelson [Wed, 19 Apr 2023 17:04:25 +0000 (10:04 -0700)]
pds_core: add the aux client API

Add the client API operations for running adminq commands.
The core registers the client with the FW, then the client
has a context for requesting adminq services.  We expect
to add additional operations for other clients, including
requesting additional private adminqs and IRQs, but don't have
the need yet.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: devlink params for enabling VIF support
Shannon Nelson [Wed, 19 Apr 2023 17:04:24 +0000 (10:04 -0700)]
pds_core: devlink params for enabling VIF support

Add the devlink parameter switches so the user can enable
the features supported by the VFs.  The only feature supported
at the moment is vDPA.

Example:
    devlink dev param set pci/0000:2b:00.0 \
    name enable_vnet cmode runtime value true

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: add auxiliary_bus devices
Shannon Nelson [Wed, 19 Apr 2023 17:04:23 +0000 (10:04 -0700)]
pds_core: add auxiliary_bus devices

An auxiliary_bus device is created for each vDPA type VF at VF
probe and destroyed at VF remove.  The aux device name comes
from the driver name + VIF type + the unique id assigned at PCI
probe.  The VFs are always removed on PF remove, so there should
be no issues with VFs trying to access missing PF structures.

The auxiliary_device names will look like "pds_core.vDPA.nn"
where 'nn' is the VF's uid.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: add initial VF device handling
Shannon Nelson [Wed, 19 Apr 2023 17:04:22 +0000 (10:04 -0700)]
pds_core: add initial VF device handling

This is the initial VF PCI driver framework for the new
pds_vdpa VF device, which will work in conjunction with an
auxiliary_bus client of the pds_core driver.  This does the
very basics of registering for the new VF device, setting
up debugfs entries, and registering with devlink.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: set up the VIF definitions and defaults
Shannon Nelson [Wed, 19 Apr 2023 17:04:21 +0000 (10:04 -0700)]
pds_core: set up the VIF definitions and defaults

The Virtual Interfaces (VIFs) supported by the DSC's
configuration (vDPA, Eth, RDMA, etc) are reported in the
dev_ident struct and made visible in debugfs.  At this point
only vDPA is supported in this driver so we only setup
devices for that feature.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: add FW update feature to devlink
Shannon Nelson [Wed, 19 Apr 2023 17:04:20 +0000 (10:04 -0700)]
pds_core: add FW update feature to devlink

Add in the support for doing firmware updates.  Of the two
main banks available, a and b, this updates the one not in
use and then selects it for the next boot.

Example:
    devlink dev flash pci/0000:b2:00.0 \
    file pensando/dsc_fw_1.63.0-22.tar

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: Add adminq processing and commands
Shannon Nelson [Wed, 19 Apr 2023 17:04:19 +0000 (10:04 -0700)]
pds_core: Add adminq processing and commands

Add the service routines for submitting and processing
the adminq messages and for handling notifyq events.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: set up device and adminq
Shannon Nelson [Wed, 19 Apr 2023 17:04:18 +0000 (10:04 -0700)]
pds_core: set up device and adminq

Set up the basic adminq and notifyq queue structures.  These are
used mostly by the client drivers for feature configuration.
These are essentially the same adminq and notifyq as in the
ionic driver.

Part of this includes querying for device identity and FW
information, so we can make that available to devlink dev info.

  $ devlink dev info pci/0000:b5:00.0
  pci/0000:b5:00.0:
    driver pds_core
    serial_number FLM18420073
    versions:
        fixed:
          asic.id 0x0
          asic.rev 0x0
        running:
          fw 1.51.0-73
        stored:
          fw.goldfw 1.15.9-C-22
          fw.mainfwa 1.60.0-73
          fw.mainfwb 1.60.0-57

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: add devlink health facilities
Shannon Nelson [Wed, 19 Apr 2023 17:04:17 +0000 (10:04 -0700)]
pds_core: add devlink health facilities

Add devlink health reporting on top of our fw watchdog.

Example:
  # devlink health show pci/0000:2b:00.0 reporter fw
  pci/0000:2b:00.0:
    reporter fw
      state healthy error 0 recover 0
  # devlink health diagnose pci/0000:2b:00.0 reporter fw
   Status: healthy State: 1 Generation: 0 Recoveries: 0

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: health timer and workqueue
Shannon Nelson [Wed, 19 Apr 2023 17:04:16 +0000 (10:04 -0700)]
pds_core: health timer and workqueue

Add in the periodic health check and the related workqueue,
as well as the handlers for when a FW reset is seen.

The firmware is polled every 5 seconds to be sure that it is
still alive and that the FW generation didn't change.

The alive check looks to see that the PCI bus is still readable
and the fw_status still has the RUNNING bit on.  If not alive,
the driver stops activity and tears things down.  When the FW
recovers and the alive check again succeeds, the driver sets
back up for activity.

The generation check looks at the fw_generation to see if it
has changed, which can happen if the FW crashed and recovered
or was updated in between health checks.  If changed, the
driver counts that as though the alive test failed and forces
the fw_down/fw_up cycle.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: add devcmd device interfaces
Shannon Nelson [Wed, 19 Apr 2023 17:04:15 +0000 (10:04 -0700)]
pds_core: add devcmd device interfaces

The devcmd interface is the basic connection to the device through the
PCI BAR for low level identification and command services.  This does
the early device initialization and finds the identity data, and adds
devcmd routines to be used by later driver bits.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agopds_core: initial framework for pds_core PF driver
Shannon Nelson [Wed, 19 Apr 2023 17:04:14 +0000 (10:04 -0700)]
pds_core: initial framework for pds_core PF driver

This is the initial PCI driver framework for the new pds_core device
driver and its family of devices.  This does the very basics of
registering for the new PF PCI device 1dd8:100c, setting up debugfs
entries, and registering with devlink.

Signed-off-by: Shannon Nelson <shannon.nelson@amd.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'bridge-neigh-suppression'
David S. Miller [Fri, 21 Apr 2023 07:25:51 +0000 (08:25 +0100)]
Merge branch 'bridge-neigh-suppression'

Ido Schimmel says:

====================
bridge: Add per-{Port, VLAN} neighbor suppression

Background
==========

In order to minimize the flooding of ARP and ND messages in the VXLAN
network, EVPN includes provisions [1] that allow participating VTEPs to
suppress such messages in case they know the MAC-IP binding and can
reply on behalf of the remote host. In Linux, the above is implemented
in the bridge driver using a per-port option called "neigh_suppress"
that was added in kernel version 4.15 [2].

Motivation
==========

Some applications use ARP messages as keepalives between the application
nodes in the network. This works perfectly well when two nodes are
connected to the same VTEP. When a node goes down it will stop
responding to ARP requests and the other node will notice it
immediately.

However, when the two nodes are connected to different VTEPs and
neighbor suppression is enabled, the local VTEP will reply to ARP
requests even after the remote node went down, until certain timers
expire and the EVPN control plane decides to withdraw the MAC/IP
Advertisement route for the address. Therefore, some users would like to
be able to disable neighbor suppression on VLANs where such applications
reside and keep it enabled on the rest.

Implementation
==============

The proposed solution is to allow user space to control neighbor
suppression on a per-{Port, VLAN} basis, in a similar fashion to other
per-port options that gained per-{Port, VLAN} counterparts such as
"mcast_router". This allows users to benefit from the operational
simplicity and scalability associated with shared VXLAN devices (i.e.,
external / collect-metadata mode), while still allowing for per-VLAN/VNI
neighbor suppression control.

The user interface is extended with a new "neigh_vlan_suppress" bridge
port option that allows user space to enable per-{Port, VLAN} neighbor
suppression on the bridge port. When enabled, the existing
"neigh_suppress" option has no effect and neighbor suppression is
controlled using a new "neigh_suppress" VLAN option. Example usage:

 # bridge link set dev vxlan0 neigh_vlan_suppress on
 # bridge vlan add vid 10 dev vxlan0
 # bridge vlan set vid 10 dev vxlan0 neigh_suppress on

Testing
=======

Tested using existing bridge selftests. Added a dedicated selftest in
the last patch.

Patchset overview
=================

Patches #1-#5 are preparations.

Patch #6 adds per-{Port, VLAN} neighbor suppression support to the
bridge's data path.

Patches #7-#8 add the required netlink attributes to enable the feature.

Patch #9 adds a selftest.

iproute2 patches can be found here [3].

Changelog
=========

Since RFC [4]:

No changes.

[1] https://www.rfc-editor.org/rfc/rfc7432#section-10
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=a42317785c898c0ed46db45a33b0cc71b671bf29
[3] https://github.com/idosch/iproute2/tree/submit/neigh_suppress_v1
[4] https://lore.kernel.org/netdev/20230413095830.2182382-1-idosch@nvidia.com/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoselftests: net: Add bridge neighbor suppression test
Ido Schimmel [Wed, 19 Apr 2023 15:35:00 +0000 (18:35 +0300)]
selftests: net: Add bridge neighbor suppression test

Add test cases for bridge neighbor suppression, testing both per-port
and per-{Port, VLAN} neighbor suppression with both ARP and NS packets.

Example truncated output:

 # ./test_bridge_neigh_suppress.sh
 [...]
 Tests passed: 148
 Tests failed:   0

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agobridge: Allow setting per-{Port, VLAN} neighbor suppression state
Ido Schimmel [Wed, 19 Apr 2023 15:34:59 +0000 (18:34 +0300)]
bridge: Allow setting per-{Port, VLAN} neighbor suppression state

Add a new bridge port attribute that allows user space to enable
per-{Port, VLAN} neighbor suppression. Example:

 # bridge -d -j -p link show dev swp1 | jq '.[]["neigh_vlan_suppress"]'
 false
 # bridge link set dev swp1 neigh_vlan_suppress on
 # bridge -d -j -p link show dev swp1 | jq '.[]["neigh_vlan_suppress"]'
 true
 # bridge link set dev swp1 neigh_vlan_suppress off
 # bridge -d -j -p link show dev swp1 | jq '.[]["neigh_vlan_suppress"]'
 false

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agobridge: vlan: Allow setting VLAN neighbor suppression state
Ido Schimmel [Wed, 19 Apr 2023 15:34:58 +0000 (18:34 +0300)]
bridge: vlan: Allow setting VLAN neighbor suppression state

Add a new VLAN attribute that allows user space to set the neighbor
suppression state of the port VLAN. Example:

 # bridge -d -j -p vlan show dev swp1 vid 10 | jq '.[]["vlans"][]["neigh_suppress"]'
 false
 # bridge vlan set vid 10 dev swp1 neigh_suppress on
 # bridge -d -j -p vlan show dev swp1 vid 10 | jq '.[]["vlans"][]["neigh_suppress"]'
 true
 # bridge vlan set vid 10 dev swp1 neigh_suppress off
 # bridge -d -j -p vlan show dev swp1 vid 10 | jq '.[]["vlans"][]["neigh_suppress"]'
 false

 # bridge vlan set vid 10 dev br0 neigh_suppress on
 Error: bridge: Can't set neigh_suppress for non-port vlans.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agobridge: Add per-{Port, VLAN} neighbor suppression data path support
Ido Schimmel [Wed, 19 Apr 2023 15:34:57 +0000 (18:34 +0300)]
bridge: Add per-{Port, VLAN} neighbor suppression data path support

When the bridge is not VLAN-aware (i.e., VLAN ID is 0), determine if
neighbor suppression is enabled on a given bridge port solely based on
the existing 'BR_NEIGH_SUPPRESS' flag.

Otherwise, if the bridge is VLAN-aware, first check if per-{Port, VLAN}
neighbor suppression is enabled on the given bridge port using the
'BR_NEIGH_VLAN_SUPPRESS' flag. If so, look up the VLAN and check whether
it has neighbor suppression enabled based on the per-VLAN
'BR_VLFLAG_NEIGH_SUPPRESS_ENABLED' flag.

If the bridge is VLAN-aware, but the bridge port does not have
per-{Port, VLAN} neighbor suppression enabled, then fallback to
determine neighbor suppression based on the 'BR_NEIGH_SUPPRESS' flag.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agobridge: Encapsulate data path neighbor suppression logic
Ido Schimmel [Wed, 19 Apr 2023 15:34:56 +0000 (18:34 +0300)]
bridge: Encapsulate data path neighbor suppression logic

Currently, there are various places in the bridge data path that check
whether neighbor suppression is enabled on a given bridge port.

As a preparation for per-{Port, VLAN} neighbor suppression, encapsulate
this logic in a function and pass the VLAN ID of the packet as an
argument.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agobridge: Take per-{Port, VLAN} neighbor suppression into account
Ido Schimmel [Wed, 19 Apr 2023 15:34:55 +0000 (18:34 +0300)]
bridge: Take per-{Port, VLAN} neighbor suppression into account

The bridge driver gates the neighbor suppression code behind an internal
per-bridge flag called 'BROPT_NEIGH_SUPPRESS_ENABLED'. The flag is set
when at least one bridge port has neighbor suppression enabled.

As a preparation for per-{Port, VLAN} neighbor suppression, make sure
the global flag is also set if per-{Port, VLAN} neighbor suppression is
enabled. That is, when the 'BR_NEIGH_VLAN_SUPPRESS' flag is set on at
least one bridge port.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agobridge: Add internal flags for per-{Port, VLAN} neighbor suppression
Ido Schimmel [Wed, 19 Apr 2023 15:34:54 +0000 (18:34 +0300)]
bridge: Add internal flags for per-{Port, VLAN} neighbor suppression

Add two internal flags that will be used to enable / disable per-{Port,
VLAN} neighbor suppression:

1. 'BR_NEIGH_VLAN_SUPPRESS': A per-port flag used to indicate that
per-{Port, VLAN} neighbor suppression is enabled on the bridge port.
When set, 'BR_NEIGH_SUPPRESS' has no effect.

2. 'BR_VLFLAG_NEIGH_SUPPRESS_ENABLED': A per-VLAN flag used to indicate
that neighbor suppression is enabled on the given VLAN.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agobridge: Pass VLAN ID to br_flood()
Ido Schimmel [Wed, 19 Apr 2023 15:34:53 +0000 (18:34 +0300)]
bridge: Pass VLAN ID to br_flood()

Subsequent patches are going to add per-{Port, VLAN} neighbor
suppression, which will require br_flood() to potentially suppress ARP /
NS packets on a per-{Port, VLAN} basis.

As a preparation, pass the VLAN ID of the packet as another argument to
br_flood().

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agobridge: Reorder neighbor suppression check when flooding
Ido Schimmel [Wed, 19 Apr 2023 15:34:52 +0000 (18:34 +0300)]
bridge: Reorder neighbor suppression check when flooding

The bridge does not flood ARP / NS packets for which a reply was sent to
bridge ports that have neighbor suppression enabled.

Subsequent patches are going to add per-{Port, VLAN} neighbor
suppression, which is going to make it more expensive to check whether
neighbor suppression is enabled since a VLAN lookup will be required.

Therefore, instead of unnecessarily performing this lookup for every
packet, only perform it for ARP / NS packets for which a reply was sent.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'macsec-vlan'
David S. Miller [Fri, 21 Apr 2023 07:22:15 +0000 (08:22 +0100)]
Merge branch 'macsec-vlan'

Emeel Hakim says:

====================
Support MACsec VLAN

This patch series introduces support for hardware (HW) offload MACsec
devices with VLAN configuration. The patches address both scenarios
where the VLAN header is both the inner and outer header for MACsec.

The changes include:

1. Adding MACsec offload operation for VLAN.
2. Considering VLAN when accessing MACsec net device.
3. Currently offloading MACsec when it's configured over VLAN with
current MACsec TX steering rules would wrongly insert the MACsec sec tag
after inserting the VLAN header. This resulted in an ETHERNET | SECTAG |
VLAN packet when ETHERNET | VLAN | SECTAG is configured. The patche
handles this issue when configuring steering rules.
4. Adding MACsec rx_handler change support in case of a marked skb and a
mismatch on the dst MAC address.

Please review these changes and let me know if you have any feedback or
concerns.

Updates since v1:
- Consult vlan_features when adding NETIF_F_HW_MACSEC.
- Allow grep for the functions.
- Add helper function to get the macsec operation to allow the compiler
  to make some choice.

Updates since v2:
- Don't use macros to allow direct navigattion from mdo functions to its
  implementation.
- Make the vlan_get_macsec_ops argument a const.
- Check if the specific mdo function is available before calling it.
- Enable NETIF_F_HW_MACSEC by default when the lower device has it enabled
  and in case the lower device currently has NETIF_F_HW_MACSEC but disabled
  let the new vlan device also have it disabled.

Updates since v3:
- Split patch ("vlan: Add MACsec offload operations for VLAN interface")
  to prevent mixing generic vlan code changes with driver changes.
- Add mdo_open, stop and stats to support drivers which have those.
- Don't fail if macsec offload operations are available but a specific
  function is not, to support drivers which does not implement all
  macsec offload operations.
- Don't call find_rx_sc twice in the same loop, instead save the result
  in a parameter and re-use it.
- Completely remove _BUILD_VLAN_MACSEC_MDO macro, to prevent returning
  from a macro.
- Reorder the functions inside struct macsec_ops to match the struct
  decleration.

 Updates since v4:
 - Change subject line of ("macsec: Add MACsec rx_handler change support") and adapt commit message.
 - Don't separate the new check in patch ("macsec: Add MACsec rx_handler change support")
   from the previous if/else if.
 - Drop"_found" from the parameter naming "rx_sc_found" and move the definition to
   the relevant block.
 - Remove "{}" since not needed around a single line.

 Updates since v5:
 - Consider promiscuous mode case.

 Updates since v6:
 - Use IS_ENABLED instead of checking for ifdef.
 - Don't add inline keywork in c files, let the compiler make its own decisions.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agomacsec: Don't rely solely on the dst MAC address to identify destination MACsec device
Emeel Hakim [Wed, 19 Apr 2023 14:21:26 +0000 (17:21 +0300)]
macsec: Don't rely solely on the dst MAC address to identify destination MACsec device

Offloading device drivers will mark offloaded MACsec SKBs with the
corresponding SCI in the skb_metadata_dst so the macsec rx handler will
know to which interface to divert those skbs, in case of a marked skb
and a mismatch on the dst MAC address, divert the skb to the macsec
net_device where the macsec rx_handler will be called to consider cases
where relying solely on the dst MAC address is insufficient.

One such instance is when using MACsec with a VLAN as an inner
header, where the packet structure is ETHERNET | SECTAG | VLAN.
In such a scenario, the dst MAC address in the ethernet header
will correspond to the VLAN MAC address, resulting in a mismatch.

Signed-off-by: Emeel Hakim <ehakim@nvidia.com>
Reviewed-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet/mlx5: Consider VLAN interface in MACsec TX steering rules
Emeel Hakim [Wed, 19 Apr 2023 14:21:25 +0000 (17:21 +0300)]
net/mlx5: Consider VLAN interface in MACsec TX steering rules

Offloading MACsec when its configured over VLAN with current MACsec
TX steering rules will wrongly insert MACsec sec tag after inserting
the VLAN header leading to a ETHERNET | SECTAG | VLAN packet when
ETHERNET | VLAN | SECTAG is configured.

The above issue is due to adding the SECTAG by HW which is a later
stage compared to the VLAN header insertion stage.

Detect such a case and adjust TX steering rules to insert the
SECTAG in the correct place by using reformat_param_0 field in
the packet reformat to indicate the offset of SECTAG from end of
the MAC header to account for VLANs in granularity of 4Bytes.

Signed-off-by: Emeel Hakim <ehakim@nvidia.com>
Reviewed-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet/mlx5: Support MACsec over VLAN
Emeel Hakim [Wed, 19 Apr 2023 14:21:24 +0000 (17:21 +0300)]
net/mlx5: Support MACsec over VLAN

MACsec device may have a VLAN device on top of it.
Detect MACsec state correctly under this condition,
and return the correct net device accordingly.

Signed-off-by: Emeel Hakim <ehakim@nvidia.com>
Reviewed-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agonet/mlx5: Enable MACsec offload feature for VLAN interface
Emeel Hakim [Wed, 19 Apr 2023 14:21:23 +0000 (17:21 +0300)]
net/mlx5: Enable MACsec offload feature for VLAN interface

Enable MACsec offload feature over VLAN by adding NETIF_F_HW_MACSEC
to the device vlan_features.

Signed-off-by: Emeel Hakim <ehakim@nvidia.com>
Reviewed-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agovlan: Add MACsec offload operations for VLAN interface
Emeel Hakim [Wed, 19 Apr 2023 14:21:22 +0000 (17:21 +0300)]
vlan: Add MACsec offload operations for VLAN interface

Add support for MACsec offload operations for VLAN driver
to allow offloading MACsec when VLAN's real device supports
Macsec offload by forwarding the offload request to it.

Signed-off-by: Emeel Hakim <ehakim@nvidia.com>
Reviewed-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'sctp-nested-flex-arrays'
David S. Miller [Fri, 21 Apr 2023 07:19:30 +0000 (08:19 +0100)]
Merge branch 'sctp-nested-flex-arrays'

Xin Long says:

====================
sctp: fix a plenty of flexible-array-nested warnings

Paolo noticed a compile warning in SCTP,

../net/sctp/stream_sched_fc.c: note: in included file (through ../include/net/sctp/sctp.h):
../include/net/sctp/structs.h:335:41: warning: array of flexible structures

But not only this, there are actually quite a lot of such warnings in
some SCTP structs. This patchset fixes most of warnings by deleting
these nested flexible array members.

After this patchset, there are still some warnings left:

  # make C=2 CF="-Wflexible-array-nested" M=./net/sctp/
  ./include/net/sctp/structs.h:1145:41: warning: nested flexible array
  ./include/uapi/linux/sctp.h:641:34: warning: nested flexible array
  ./include/uapi/linux/sctp.h:643:34: warning: nested flexible array
  ./include/uapi/linux/sctp.h:644:33: warning: nested flexible array
  ./include/uapi/linux/sctp.h:650:40: warning: nested flexible array
  ./include/uapi/linux/sctp.h:653:39: warning: nested flexible array

the 1st is caused by __data[] in struct ip_options, not in SCTP;
the others are in uapi, and we should not touch them.

Note that instead of completely deleting it, we just leave it as a
comment in the struct, signalling to the reader that we do expect
such variable parameters over there, as Marcelo suggested.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agosctp: delete the nested flexible array payload
Xin Long [Wed, 19 Apr 2023 15:16:33 +0000 (11:16 -0400)]
sctp: delete the nested flexible array payload

This patch deletes the flexible-array payload[] from the structure
sctp_datahdr to avoid some sparse warnings:

  # make C=2 CF="-Wflexible-array-nested" M=./net/sctp/
  net/sctp/socket.c: note: in included file (through include/net/sctp/structs.h, include/net/sctp/sctp.h):
  ./include/linux/sctp.h:230:29: warning: nested flexible array

This member is not even used anywhere.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agosctp: delete the nested flexible array hmac
Xin Long [Wed, 19 Apr 2023 15:16:32 +0000 (11:16 -0400)]
sctp: delete the nested flexible array hmac

This patch deletes the flexible-array hmac[] from the structure
sctp_authhdr to avoid some sparse warnings:

  # make C=2 CF="-Wflexible-array-nested" M=./net/sctp/
  net/sctp/auth.c: note: in included file (through include/net/sctp/structs.h, include/net/sctp/sctp.h):
  ./include/linux/sctp.h:735:29: warning: nested flexible array

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agosctp: delete the nested flexible array peer_init
Xin Long [Wed, 19 Apr 2023 15:16:31 +0000 (11:16 -0400)]
sctp: delete the nested flexible array peer_init

This patch deletes the flexible-array peer_init[] from the structure
sctp_cookie to avoid some sparse warnings:

  # make C=2 CF="-Wflexible-array-nested" M=./net/sctp/
  net/sctp/sm_make_chunk.c: note: in included file (through include/net/sctp/sctp.h):
  ./include/net/sctp/structs.h:1588:28: warning: nested flexible array
  ./include/net/sctp/structs.h:343:28: warning: nested flexible array

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agosctp: delete the nested flexible array variable
Xin Long [Wed, 19 Apr 2023 15:16:30 +0000 (11:16 -0400)]
sctp: delete the nested flexible array variable

This patch deletes the flexible-array variable[] from the structure
sctp_sackhdr and sctp_errhdr to avoid some sparse warnings:

  # make C=2 CF="-Wflexible-array-nested" M=./net/sctp/
  net/sctp/sm_statefuns.c: note: in included file (through include/net/sctp/structs.h, include/net/sctp/sctp.h):
  ./include/linux/sctp.h:451:28: warning: nested flexible array
  ./include/linux/sctp.h:393:29: warning: nested flexible array

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agosctp: delete the nested flexible array skip
Xin Long [Wed, 19 Apr 2023 15:16:29 +0000 (11:16 -0400)]
sctp: delete the nested flexible array skip

This patch deletes the flexible-array skip[] from the structure
sctp_ifwdtsn/fwdtsn_hdr to avoid some sparse warnings:

  # make C=2 CF="-Wflexible-array-nested" M=./net/sctp/
  net/sctp/stream_interleave.c: note: in included file (through include/net/sctp/structs.h, include/net/sctp/sctp.h):
  ./include/linux/sctp.h:611:32: warning: nested flexible array
  ./include/linux/sctp.h:628:33: warning: nested flexible array

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agosctp: delete the nested flexible array params
Xin Long [Wed, 19 Apr 2023 15:16:28 +0000 (11:16 -0400)]
sctp: delete the nested flexible array params

This patch deletes the flexible-array params[] from the structure
sctp_inithdr, sctp_addiphdr and sctp_reconf_chunk to avoid some
sparse warnings:

  # make C=2 CF="-Wflexible-array-nested" M=./net/sctp/
  net/sctp/input.c: note: in included file (through include/net/sctp/structs.h, include/net/sctp/sctp.h):
  ./include/linux/sctp.h:278:29: warning: nested flexible array
  ./include/linux/sctp.h:675:30: warning: nested flexible array

This warning is reported if a structure having a flexible array
member is included by other structures.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 months agoMerge branch 'net-extend-drop-reasons'
Jakub Kicinski [Fri, 21 Apr 2023 03:20:51 +0000 (20:20 -0700)]
Merge branch 'net-extend-drop-reasons'

Johannes Berg says:

====================
net: extend drop reasons

Here's v4 of the extended drop reasons, with fixes to kernel-doc
and checkpatch.
====================

Link: https://lore.kernel.org/r/20230419125254.20789-1-johannes@sipsolutions.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agomac80211: use the new drop reasons infrastructure
Johannes Berg [Wed, 19 Apr 2023 12:52:54 +0000 (14:52 +0200)]
mac80211: use the new drop reasons infrastructure

It can be really hard to analyse or debug why packets are
going missing in mac80211, so add the needed infrastructure
to use use the new per-subsystem drop reasons.

We actually use two drop reason subsystems here because of
the different handling of frames that are dropped but still
go to monitor for old versions of hostapd, and those that
are just completely unusable (e.g. crypto failed.)

Annotate a few reasons here just to illustrate this, we'll
need to go through and annotate more of them later.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: extend drop reasons for multiple subsystems
Johannes Berg [Wed, 19 Apr 2023 12:52:53 +0000 (14:52 +0200)]
net: extend drop reasons for multiple subsystems

Extend drop reasons to make them usable by subsystems
other than core by reserving the high 16 bits for a
new subsystem ID, of which 0 of course is used for the
existing reasons immediately.

To still be able to have string reasons, restructure
that code a bit to make the loopup under RCU, the only
user of this (right now) is drop_monitor.

Link: https://lore.kernel.org/netdev/00659771ed54353f92027702c5bbb84702da62ce.camel@sipsolutions.net
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: move dropreason.h to dropreason-core.h
Johannes Berg [Wed, 19 Apr 2023 12:52:52 +0000 (14:52 +0200)]
net: move dropreason.h to dropreason-core.h

This will, after the next patch, hold only the core
drop reasons and minimal infrastructure. Fix a small
kernel-doc issue while at it, to avoid the move
triggering a checker.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoipv6: add icmpv6_error_anycast_as_unicast for ICMPv6
Mahesh Bandewar [Wed, 19 Apr 2023 01:32:38 +0000 (18:32 -0700)]
ipv6: add icmpv6_error_anycast_as_unicast for ICMPv6

ICMPv6 error packets are not sent to the anycast destinations and this
prevents things like traceroute from working. So create a setting similar
to ECHO when dealing with Anycast sources (icmpv6_echo_ignore_anycast).

Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Maciej Żenczykowski <maze@google.com>
Link: https://lore.kernel.org/r/20230419013238.2691167-1-maheshb@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoMerge branch 'ethtool-mm-api-consolidation'
Jakub Kicinski [Fri, 21 Apr 2023 03:03:24 +0000 (20:03 -0700)]
Merge branch 'ethtool-mm-api-consolidation'

Vladimir Oltean says:

====================
ethtool mm API consolidation

This series consolidates the behavior of the 2 drivers that implement
the ethtool MAC Merge layer by making NXP ENETC commit its preemptible
traffic classes to hardware only when MM TX is active (same as Ocelot).

Then, after resolving an issue with the ENETC driver, it restricts user
space from entering 2 states which don't make sense:

- pmac-enabled off tx-enabled on  verify-enabled *
- pmac-enabled *   tx-enabled off verify-enabled on

Then, it introduces a selftest (ethtool_mm.sh) which puts everything
together and tests all valid configurations known to me.

This is simultaneously the v2 of "[PATCH net-next 0/2] ethtool mm API
improvements":
https://lore.kernel.org/netdev/20230415173454.3970647-1-vladimir.oltean@nxp.com/
which had caused some problems to openlldp. Those were solved in the
meantime, see:
https://github.com/intel/openlldp/commit/11171b474f6f3cbccac5d608b7f26b32ff72c651

and of "[RFC PATCH net-next] selftests: forwarding: add a test for MAC
Merge layer":
https://lore.kernel.org/netdev/20230210221243.228932-1-vladimir.oltean@nxp.com/
====================

Link: https://lore.kernel.org/r/20230418111459.811553-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoselftests: forwarding: add a test for MAC Merge layer
Vladimir Oltean [Tue, 18 Apr 2023 11:14:59 +0000 (14:14 +0300)]
selftests: forwarding: add a test for MAC Merge layer

The MAC Merge layer (IEEE 802.3-2018 clause 99) does all the heavy
lifting for Frame Preemption (IEEE 802.1Q-2018 clause 6.7.2), a TSN
feature for minimizing latency.

Preemptible traffic is different on the wire from normal traffic in
incompatible ways. If we send a preemptible packet and the link partner
doesn't support preemption, it will drop it as an error frame and we
will never know. The MAC Merge layer has a control plane of its own,
which can be manipulated (using ethtool) in order to negotiate this
capability with the link partner (through LLDP).

Actually the TLV format for LLDP solves this problem only partly,
because both partners only advertise:
- if they support preemption (RX and TX)
- if they have enabled preemption (TX)
so we cannot tell the link partner what to do - we cannot force it to
enable reception of our preemptible packets.

That is fully solved by the verification feature, where the local device
generates some small probe frames which look like preemptible frames
with no useful content, and the link partner is obliged to respond to
them if it supports the standard. If the verification times out, we know
that preemption isn't active in our TX direction on the link.

Having clarified the definition, this selftest exercises the manual
(ethtool) configuration path of 2 link partners (with and without
verification), and the LLDP code path, using the openlldp project.

The test also verifies the TX activity of the MAC Merge layer by
sending traffic through a traffic class configured as preemptible
(using mqprio). There isn't a good way to make this really portable
(user space cannot find out how many traffic classes there are for
a device), but I chose num_tc 4 here, that should work reasonably well.
I also know that some devices (stmmac) only permit TXQ0 to be
preemptible, so this is why PREEMPTIBLE_PRIO was strategically chosen
as 0. Even if other hardware is more configurable, this test should
cover the baseline.

This is not really a "forwarding" selftest, but I put it near the other
"ethtool" selftests.

$ ./ethtool_mm.sh eno0 swp0
TEST: Manual configuration with verification: eno0 to swp0          [ OK ]
TEST: Manual configuration with verification: swp0 to eno0          [ OK ]
TEST: Manual configuration without verification: eno0 to swp0       [ OK ]
TEST: Manual configuration without verification: swp0 to eno0       [ OK ]
TEST: Manual configuration with failed verification: eno0 to swp0   [ OK ]
TEST: Manual configuration with failed verification: swp0 to eno0   [ OK ]
TEST: LLDP                                                          [ OK ]

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoselftests: forwarding: introduce helper for standard ethtool counters
Vladimir Oltean [Tue, 18 Apr 2023 11:14:58 +0000 (14:14 +0300)]
selftests: forwarding: introduce helper for standard ethtool counters

Counters for the MAC Merge layer and preemptible MAC have standardized
so far on using structured ethtool stats as opposed to the driver
specific names and meanings.

Benefit from that rare opportunity and introduce a helper to lib.sh for
querying standardized counters, in the hope that these will take off for
other uses as well.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoselftests: forwarding: generalize bail_on_lldpad from mlxsw
Petr Machata [Tue, 18 Apr 2023 11:14:57 +0000 (14:14 +0300)]
selftests: forwarding: generalize bail_on_lldpad from mlxsw

mlxsw selftests often invoke a bail_on_lldpad() helper to make sure LLDPAD
is not running, to prevent conflicts between the QoS configuration applied
through TC or DCB command line tool, and the DCB configuration that LLDPAD
might apply. This helper might be useful to others. Move the function to
lib.sh, and parameterize to make reusable in other contexts.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoselftests: forwarding: sch_tbf_*: Add a pre-run hook
Petr Machata [Tue, 18 Apr 2023 11:14:56 +0000 (14:14 +0300)]
selftests: forwarding: sch_tbf_*: Add a pre-run hook

The driver-specific wrappers of these selftests invoke bail_on_lldpad to
make sure that LLDPAD doesn't trample the configuration. The function
bail_on_lldpad is going to move to lib.sh in the next patch. With that, it
won't be visible for the wrappers before sourcing the framework script. And
after sourcing it, it is too late: the selftest will have run by then.

One option might be to source NUM_NETIFS=0 lib.sh from the wrapper, but
even if that worked (it might, it might not), that seems cumbersome. lib.sh
is doing fair amount of stuff, and even if it works today, it does not look
particularly solid as a solution.

Instead, introduce a hook, sch_tbf_pre_hook(), that when available, gets
invoked. Move the bail to the hook.

Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: ethtool: mm: sanitize some UAPI configurations
Vladimir Oltean [Tue, 18 Apr 2023 11:14:55 +0000 (14:14 +0300)]
net: ethtool: mm: sanitize some UAPI configurations

The verify-enabled boolean (ETHTOOL_A_MM_VERIFY_ENABLED) was intended to
be a sub-setting of tx-enabled (ETHTOOL_A_MM_TX_ENABLED). IOW, MAC Merge
TX can be enabled with or without verification, but verification with TX
disabled makes no sense.

The pmac-enabled boolean (ETHTOOL_A_MM_PMAC_ENABLED) was intended to be
a global toggle from an API perspective, whereas tx-enabled just handles
the TX direction. IOW, the pMAC can be enabled with or without TX, but
it doesn't make sense to enable TX if the pMAC is not enabled.

Add two checks which sanitize and reject these invalid cases.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: enetc: include MAC Merge / FP registers in register dump
Vladimir Oltean [Tue, 18 Apr 2023 11:14:54 +0000 (14:14 +0300)]
net: enetc: include MAC Merge / FP registers in register dump

These have been useful in debugging various problems related to frame
preemption, so make them available through ethtool --register-dump for
later too.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: enetc: only commit preemptible TCs to hardware when MM TX is active
Vladimir Oltean [Tue, 18 Apr 2023 11:14:53 +0000 (14:14 +0300)]
net: enetc: only commit preemptible TCs to hardware when MM TX is active

This was left as TODO in commit 01e23b2b3bad ("net: enetc: add support
for preemptible traffic classes") since it's relatively complicated.

Where this makes a difference is with a configuration as follows:

ethtool --set-mm eno0 pmac-enabled on tx-enabled on verify-enabled on

Preemptible packets should only be sent when the MAC Merge TX direction
becomes active (i.o.w. when the verification process succeeds, aka when
the link partner confirms it can process preemptible traffic). But the
tc qdisc with the preemptible traffic classes is offloaded completely
asynchronously w.r.t. the MM becoming active.

The ENETC manual does suggest that this should be handled in the driver:
"On startup, software should wait for the verification process to
complete (MMCSR[VSTS]=011) before initiating traffic".

Adding the necessary logic allows future selftests to uphold the claim
that an inactive or disabled MAC Merge layer should never send data
packets through the pMAC.

This change moves enetc_set_ptcfpr() from enetc.c to enetc_ethtool.c,
where its only caller is now - enetc_mm_commit_preemptible_tcs().

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: enetc: report mm tx-active based on tx-enabled and verify-status
Vladimir Oltean [Tue, 18 Apr 2023 11:14:52 +0000 (14:14 +0300)]
net: enetc: report mm tx-active based on tx-enabled and verify-status

The MMCSR register contains 2 fields with overlapping meaning:

- LPA (Local preemption active):
This read-only status bit indicates whether preemption is active for
this port. This bit will be set if preemption is both enabled and has
completed the verification process.
- TXSTS (Merge status):
This read-only status field provides the state of the MAC Merge sublayer
transmit status as defined in IEEE Std 802.3-2018 Clause 99.
00 Transmit preemption is inactive
01 Transmit preemption is active
10 Reserved
11 Reserved

However none of these 2 fields offer reliable reporting to software.

When connecting ENETC to a link partner which is not capable of Frame
Preemption, the expectation is that ENETC's verification should fail
(VSTS=4) and its MM TX direction should be inactive (LPA=0, TXSTS=00)
even though the MM TX is enabled (ME=1). But surprise, the LPA bit of
MMCSR stays set even if VSTS=4 and ME=1.

OTOH, the TXSTS field has the opposite problem. I cannot get its value
to change from 0, even when connecting to a link partner capable of
frame preemption, which does respond to its verification frames (ME=1
and VSTS=3, "SUCCEEDED").

The only option with such buggy hardware seems to be to reimplement the
formula for calculating tx-active in software, which is for tx-enabled
to be true, and for the verify-status to be either SUCCEEDED, or
DISABLED.

Without reliable tx-active reporting, we have no good indication when
to commit the preemptible traffic classes to hardware, which makes it
possible (but not desirable) to send preemptible traffic to a link
partner incapable of receiving it. However, currently we do not have the
logic to wait for TX to be active yet, so the impact is limited.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: enetc: fix MAC Merge layer remaining enabled until a link down event
Vladimir Oltean [Tue, 18 Apr 2023 11:14:51 +0000 (14:14 +0300)]
net: enetc: fix MAC Merge layer remaining enabled until a link down event

Current enetc_set_mm() is designed to set the priv->active_offloads bit
ENETC_F_QBU for enetc_mm_link_state_update() to act on, but if the link
is already up, it modifies the ENETC_MMCSR_ME ("Merge Enable") bit
directly.

The problem is that it only *sets* ENETC_MMCSR_ME if the link is up, it
doesn't *clear* it if needed. So subsequent enetc_get_mm() calls still
see tx-enabled as true, up until a link down event, which is when
enetc_mm_link_state_update() will get called.

This is not a functional issue as far as I can assess. It has only come
up because I'd like to uphold a simple API rule in core ethtool code:
the pMAC cannot be disabled if TX is going to be enabled. Currently,
the fact that TX remains enabled for longer than expected (after the
enetc_set_mm() call that disables it) is going to violate that rule,
which is how it was caught.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agowwan: core: add print for wwan port attach/disconnect
Slark Xiao [Thu, 20 Apr 2023 02:36:17 +0000 (10:36 +0800)]
wwan: core: add print for wwan port attach/disconnect

Refer to USB serial device or net device, there is a notice to
let end user know the status of device, like attached or
disconnected. Add attach/disconnect print for wwan device as
well.

Signed-off-by: Slark Xiao <slark_xiao@163.com>
Reviewed-by: Loic Poulain <loic.poulain@linaro.org>
Link: https://lore.kernel.org/r/20230420023617.3919569-1-slark_xiao@163.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: skbuff: update and rename __kfree_skb_defer()
Jakub Kicinski [Thu, 20 Apr 2023 02:00:05 +0000 (19:00 -0700)]
net: skbuff: update and rename __kfree_skb_defer()

__kfree_skb_defer() uses the old naming where "defer" meant
slab bulk free/alloc APIs. In the meantime we also made
__kfree_skb_defer() feed the per-NAPI skb cache, which
implies bulk APIs. So take away the 'defer' and add 'napi'.

While at it add a drop reason. This only matters on the
tx_action path, if the skb has a frag_list. But getting
rid of a SKB_DROP_REASON_NOT_SPECIFIED seems like a net
benefit so why not.

Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Link: https://lore.kernel.org/r/20230420020005.815854-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoeth: mlx5: avoid iterator use outside of a loop
Jakub Kicinski [Thu, 20 Apr 2023 01:58:02 +0000 (18:58 -0700)]
eth: mlx5: avoid iterator use outside of a loop

Fix the following warning about risky iterator use:

drivers/net/ethernet/mellanox/mlx5/core/eq.c:1010 mlx5_comp_irq_get_affinity_mask() warn: iterator used outside loop: 'eq'

Acked-by: Saeed Mahameed <saeed@kernel.org>
Link: https://lore.kernel.org/r/20230420015802.815362-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoflow_dissector: Address kdoc warnings
Simon Horman [Wed, 19 Apr 2023 11:34:35 +0000 (13:34 +0200)]
flow_dissector: Address kdoc warnings

Address a number of warnings flagged by
./scripts/kernel-doc -none include/net/flow_dissector.h

 include/net/flow_dissector.h:23: warning: Function parameter or member 'addr_type' not described in 'flow_dissector_key_control'
 include/net/flow_dissector.h:23: warning: Function parameter or member 'flags' not described in 'flow_dissector_key_control'
 include/net/flow_dissector.h:46: warning: Function parameter or member 'padding' not described in 'flow_dissector_key_basic'
 include/net/flow_dissector.h:145: warning: Function parameter or member 'tipckey' not described in 'flow_dissector_key_addrs'
 include/net/flow_dissector.h:157: warning: cannot understand function prototype: 'struct flow_dissector_key_arp '
 include/net/flow_dissector.h:171: warning: cannot understand function prototype: 'struct flow_dissector_key_ports '
 include/net/flow_dissector.h:203: warning: cannot understand function prototype: 'struct flow_dissector_key_icmp '

Also improve indentation on adjacent lines to those changed
to address the above.

No functional changes intended.

Signed-off-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230419-flow-dissector-kdoc-v1-1-1aa0cca1118b@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agopage_pool: unlink from napi during destroy
Jakub Kicinski [Wed, 19 Apr 2023 18:20:06 +0000 (11:20 -0700)]
page_pool: unlink from napi during destroy

Jesper points out that we must prevent recycling into cache
after page_pool_destroy() is called, because page_pool_destroy()
is not synchronized with recycling (some pages may still be
outstanding when destroy() gets called).

I assumed this will not happen because NAPI can't be scheduled
if its page pool is being destroyed. But I missed the fact that
NAPI may get reused. For instance when user changes ring configuration
driver may allocate a new page pool, stop NAPI, swap, start NAPI,
and then destroy the old pool. The NAPI is running so old page
pool will think it can recycle to the cache, but the consumer
at that point is the destroy() path, not NAPI.

To avoid extra synchronization let the drivers do "unlinking"
during the "swap" stage while NAPI is indeed disabled.

Fixes: 8c48eea3adf3 ("page_pool: allow caching from safely localized NAPI")
Reported-by: Jesper Dangaard Brouer <jbrouer@redhat.com>
Link: https://lore.kernel.org/all/e8df2654-6a5b-3c92-489d-2fe5e444135f@redhat.com/
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Link: https://lore.kernel.org/r/20230419182006.719923-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agonet: phy: fix circular LEDS_CLASS dependencies
Arnd Bergmann [Thu, 20 Apr 2023 08:45:51 +0000 (10:45 +0200)]
net: phy: fix circular LEDS_CLASS dependencies

The CONFIG_PHYLIB symbol is selected by a number of device drivers that
need PHY support, but it now has a dependency on CONFIG_LEDS_CLASS,
which may not be enabled, causing build failures.

Avoid the risk of missing and circular dependencies by guarding the
phylib LED support itself in another Kconfig symbol that can only be
enabled if the dependency is met.

This could be made a hidden symbol and always enabled when both CONFIG_OF
and CONFIG_LEDS_CLASS are reachable from the phylib, but there may be an
advantage in having users see this option when they have a misconfigured
kernel without built-in LED support.

Fixes: 01e5b728e9e4 ("net: phy: Add a binding for PHY LEDs")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/20230420084624.3005701-1-arnd@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Jakub Kicinski [Thu, 20 Apr 2023 23:27:33 +0000 (16:27 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Adjacent changes:

net/mptcp/protocol.h
  63740448a32e ("mptcp: fix accept vs worker race")
  2a6a870e44dd ("mptcp: stops worker on unaccepted sockets at listener close")
  ddb1a072f858 ("mptcp: move first subflow allocation at mpc access time")

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
18 months agowifi: ath9k: Don't mark channelmap stack variable read-only in ath9k_mci_update_wlan_...
Toke Høiland-Jørgensen [Thu, 13 Apr 2023 21:41:18 +0000 (23:41 +0200)]
wifi: ath9k: Don't mark channelmap stack variable read-only in ath9k_mci_update_wlan_channels()

This partially reverts commit e161d4b60ae3a5356e07202e0bfedb5fad82c6aa.

Turns out the channelmap variable is not actually read-only, it's modified
through the MCI_GPM_CLR_CHANNEL_BIT() macro further down in the function,
so making it read-only causes page faults when that code is hit.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=217183
Link: https://lore.kernel.org/r/20230413214118.153781-1-toke@toke.dk
Fixes: e161d4b60ae3 ("wifi: ath9k: Make arrays prof_prio and channelmap static const")
Cc: stable@vger.kernel.org
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agoMerge tag 'rust-fixes-6.3' of https://github.com/Rust-for-Linux/linux
Linus Torvalds [Thu, 20 Apr 2023 19:46:18 +0000 (12:46 -0700)]
Merge tag 'rust-fixes-6.3' of https://github.com/Rust-for-Linux/linux

Pull Rust fixes from Miguel Ojeda:
 "Most of these are straightforward.

  The last one is more complex, but it only touches Rust + GCC builds
  which are for the moment best-effort.

   - Code: Missing 'extern "C"' fix.

   - Scripts: 'is_rust_module.sh' and 'generate_rust_analyzer.py' fixes.

   - A couple trivial fixes

   - Build: Rust + GCC build fix and 'grep' warning fix"

* tag 'rust-fixes-6.3' of https://github.com/Rust-for-Linux/linux:
  rust: allow to use INIT_STACK_ALL_ZERO
  rust: fix regexp in scripts/is_rust_module.sh
  rust: build: Fix grep warning
  scripts: generate_rust_analyzer: Handle sub-modules with no Makefile
  rust: kernel: Mark rust_fmt_argument as extern "C"
  rust: sort uml documentation arch support table
  rust: str: fix requierments->requirements typo

18 months agoMerge tag 'net-6.3-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 20 Apr 2023 18:03:51 +0000 (11:03 -0700)]
Merge tag 'net-6.3-rc8' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Including fixes from netfilter and bpf.

  There are a few fixes for new code bugs, including the Mellanox one
  noted in the last networking pull. No known regressions outstanding.

  Current release - regressions:

   - sched: clear actions pointer in miss cookie init fail

   - mptcp: fix accept vs worker race

   - bpf: fix bpf_arch_text_poke() with new_addr == NULL on s390

   - eth: bnxt_en: fix a possible NULL pointer dereference in unload
     path

   - eth: veth: take into account peer device for
     NETDEV_XDP_ACT_NDO_XMIT xdp_features flag

  Current release - new code bugs:

   - eth: revert "net/mlx5: Enable management PF initialization"

  Previous releases - regressions:

   - netfilter: fix recent physdev match breakage

   - bpf: fix incorrect verifier pruning due to missing register
     precision taints

   - eth: virtio_net: fix overflow inside xdp_linearize_page()

   - eth: cxgb4: fix use after free bugs caused by circular dependency
     problem

   - eth: mlxsw: pci: fix possible crash during initialization

  Previous releases - always broken:

   - sched: sch_qfq: prevent slab-out-of-bounds in qfq_activate_agg

   - netfilter: validate catch-all set elements

   - bridge: don't notify FDB entries with "master dynamic"

   - eth: bonding: fix memory leak when changing bond type to ethernet

   - eth: i40e: fix accessing vsi->active_filters without holding lock

  Misc:

   - Mat is back as MPTCP co-maintainer"

* tag 'net-6.3-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (33 commits)
  net: bridge: switchdev: don't notify FDB entries with "master dynamic"
  Revert "net/mlx5: Enable management PF initialization"
  MAINTAINERS: Resume MPTCP co-maintainer role
  mailmap: add entries for Mat Martineau
  e1000e: Disable TSO on i219-LM card to increase speed
  bnxt_en: fix free-runnig PHC mode
  net: dsa: microchip: ksz8795: Correctly handle huge frame configuration
  bpf: Fix incorrect verifier pruning due to missing register precision taints
  hamradio: drop ISA_DMA_API dependency
  mlxsw: pci: Fix possible crash during initialization
  mptcp: fix accept vs worker race
  mptcp: stops worker on unaccepted sockets at listener close
  net: rpl: fix rpl header size calculation
  net: vmxnet3: Fix NULL pointer dereference in vmxnet3_rq_rx_complete()
  bonding: Fix memory leak when changing bond type to Ethernet
  veth: take into account peer device for NETDEV_XDP_ACT_NDO_XMIT xdp_features flag
  mlxfw: fix null-ptr-deref in mlxfw_mfa2_tlv_next()
  bnxt_en: Fix a possible NULL pointer dereference in unload path
  bnxt_en: Do not initialize PTP on older P3/P4 chips
  netfilter: nf_tables: tighten netlink attribute requirements for catch-all elements
  ...

18 months agoMerge ath-next from git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/ath.git
Kalle Valo [Thu, 20 Apr 2023 16:43:45 +0000 (19:43 +0300)]
Merge ath-next from git://git./linux/kernel/git/kvalo/ath.git

ath.git patches for v6.4. Major changes:

wcn36xx

* support for pronto v3 hardware

ath11k

* PCIe DeviceTree bindings

* WCN6750: enable SAR support

ath10k

* convert DeviceTree bindings to YAML

18 months agonet: libwx: fix memory leak in wx_setup_rx_resources
Zhengchao Shao [Tue, 18 Apr 2023 06:54:50 +0000 (14:54 +0800)]
net: libwx: fix memory leak in wx_setup_rx_resources

When wx_alloc_page_pool() failed in wx_setup_rx_resources(), it doesn't
release DMA buffer. Add dma_free_coherent() in the error path to release
the DMA buffer.

Fixes: 850b971110b2 ("net: libwx: Allocate Rx and Tx resources")
Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230418065450.2268522-1-shaozhengchao@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
18 months agoMerge tag 'mt76-for-kvalo-2023-04-18' of https://github.com/nbd168/wireless
Kalle Valo [Thu, 20 Apr 2023 12:53:58 +0000 (15:53 +0300)]
Merge tag 'mt76-for-kvalo-2023-04-18' of https://github.com/nbd168/wireless

mt76 patches for 6.4

- fixes
- connac code unification
- mt7921 p2p support
- mt7996 mesh a-msdu support
- mt7996 eht support
- mt7996 coredump support

18 months agowifi: rtw88: Update spelling in main.h
Simon Horman [Tue, 18 Apr 2023 11:29:22 +0000 (13:29 +0200)]
wifi: rtw88: Update spelling in main.h

Update spelling in comments in main.h

Found by inspection.

Signed-off-by: Simon Horman <horms@kernel.org>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230418-rtw88-starspell-v1-1-70e52a23979b@kernel.org
18 months agowifi: airo: remove ISA_DMA_API dependency
Arnd Bergmann [Mon, 17 Apr 2023 20:51:24 +0000 (22:51 +0200)]
wifi: airo: remove ISA_DMA_API dependency

This driver does not actually use the ISA DMA API, it is purely
PIO based, so remove the dependency.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230417205131.1560074-1-arnd@kernel.org
18 months agowifi: rtl8xxxu: Simplify setting the initial gain
Bitterblue Smith [Mon, 17 Apr 2023 17:08:20 +0000 (20:08 +0300)]
wifi: rtl8xxxu: Simplify setting the initial gain

The goal of writing 0x6954341e / 0x6955341e to REG_OFDM0_XA_AGC_CORE1
appears to be setting the initial gain, which is stored in bits 0..6.
Bits 7..31 are the same as what the phy init tables write.

Modify only bits 0..6 so that we don't have to care about the values
of the others. This way we don't have to add another "else if" for the
RTL8192FU.

Why we need to change the initial gain from the default 0x20 to 0x1e?
Not sure. Some of the vendor drivers change it to 0x1e before scanning
and then restore it to the original value after.

Signed-off-by: Bitterblue Smith <rtl8821cerfe2@gmail.com>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/cf91ca69-70e3-4c20-c0b1-e59d452356a1@gmail.com
18 months agowifi: rtl8xxxu: Add rtl8xxxu_write{8,16,32}_{set,clear}
Bitterblue Smith [Mon, 17 Apr 2023 17:07:09 +0000 (20:07 +0300)]
wifi: rtl8xxxu: Add rtl8xxxu_write{8,16,32}_{set,clear}

Also add rtl8xxxu_write32_mask, rtl8xxxu_write_rfreg_mask.

These helper functions make it easier to modify only parts of a register
by eliminating the call to the register reading function and the bit
manipulations.

Signed-off-by: Bitterblue Smith <rtl8821cerfe2@gmail.com>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/9430b841-1048-b27c-14ec-fca447dc32af@gmail.com
18 months agowifi: rtl8xxxu: Don't print the vendor/product/serial
Bitterblue Smith [Mon, 17 Apr 2023 17:05:43 +0000 (20:05 +0300)]
wifi: rtl8xxxu: Don't print the vendor/product/serial

Most devices have a vendor name, product name, and serial number in the
efuse, but it's pretty useless. It duplicates the information already
printed by the USB subsystem:

   usb 1-4: New USB device found, idVendor=0bda, idProduct=8178, bcdDevice= 2.00
   usb 1-4: New USB device strings: Mfr=1, Product=2, SerialNumber=3
   usb 1-4: Product: 802.11n WLAN Adapter
   usb 1-4: Manufacturer: Realtek
   usb 1-4: SerialNumber: 00e04c000001
-> usb 1-4: Vendor: Realtek
-> usb 1-4: Product: 802.11n WLAN Adapter

   usb 1-4: New USB device found, idVendor=0bda, idProduct=818b, bcdDevice= 2.00
   usb 1-4: New USB device strings: Mfr=1, Product=2, SerialNumber=3
   usb 1-4: Product: 802.11n NIC
   usb 1-4: Manufacturer: Realtek
   usb 1-4: SerialNumber: 00e04c000001
-> usb 1-4: Vendor: Realtek
-> usb 1-4: Product: 802.11n NIC
-> usb 1-4: Serial not available.

   usb 1-4: New USB device found, idVendor=0bda, idProduct=f179, bcdDevice= 0.00
   usb 1-4: New USB device strings: Mfr=1, Product=2, SerialNumber=3
   usb 1-4: Product: 802.11n
   usb 1-4: Manufacturer: Realtek
   usb 1-4: SerialNumber: 002E2DC0041F
-> usb 1-4: Vendor: Realtek
-> usb 1-4: Product: 802.11n

   usb 1-4: New USB device found, idVendor=0bda, idProduct=8179, bcdDevice= 0.00
   usb 1-4: New USB device strings: Mfr=1, Product=2, SerialNumber=3
   usb 1-4: Product: 802.11n NIC
   usb 1-4: Manufacturer: Realtek
   usb 1-4: SerialNumber: 00E04C0001
-> usb 1-4: Vendor: Realtek
-> usb 1-4: Product: 802.11n NIC
-> usb 1-4: Serial: 00E04C0001

Also, that data is not interpreted correctly in all cases:

usb 3-1.1.2: New USB device found, idVendor=0bda, idProduct=8179, bcdDevice= 0.00
usb 3-1.1.2: New USB device strings: Mfr=1, Product=2, SerialNumber=3
usb 3-1.1.2: Product: 802.11n NIC
usb 3-1.1.2: Manufacturer: Realtek
usb 3-1.1.2: Vendor: Realtek
usb 3-1.1.2: Product: \x03802.11n NI
usb 3-1.1.2: Serial: \xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff

Link: https://bugzilla.kernel.org/show_bug.cgi?id=217231
Signed-off-by: Bitterblue Smith <rtl8821cerfe2@gmail.com>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/a2a7d9df-0529-7890-3522-48dce613753f@gmail.com
18 months agowifi: rtw88: Fix memory leak in rtw88_usb
Larry Finger [Mon, 17 Apr 2023 16:03:31 +0000 (11:03 -0500)]
wifi: rtw88: Fix memory leak in rtw88_usb

Kmemleak shows the following leak arising from routine in the usb
probe routine:

unreferenced object 0xffff895cb29bba00 (size 512):
  comm "(udev-worker)", pid 534, jiffies 4294903932 (age 102751.088s)
  hex dump (first 32 bytes):
    77 30 30 30 00 00 00 00 02 2f 2d 2b 30 00 00 00  w000...../-+0...
    02 00 2a 28 00 00 00 00 ff 55 ff ff ff 00 00 00  ..*(.....U......
  backtrace:
    [<ffffffff9265fa36>] kmalloc_trace+0x26/0x90
    [<ffffffffc17eec41>] rtw_usb_probe+0x2f1/0x680 [rtw_usb]
    [<ffffffffc03e19fd>] usb_probe_interface+0xdd/0x2e0 [usbcore]
    [<ffffffff92b4f2fe>] really_probe+0x18e/0x3d0
    [<ffffffff92b4f5b8>] __driver_probe_device+0x78/0x160
    [<ffffffff92b4f6bf>] driver_probe_device+0x1f/0x90
    [<ffffffff92b4f8df>] __driver_attach+0xbf/0x1b0
    [<ffffffff92b4d350>] bus_for_each_dev+0x70/0xc0
    [<ffffffff92b4e51e>] bus_add_driver+0x10e/0x210
    [<ffffffff92b50935>] driver_register+0x55/0xf0
    [<ffffffffc03e0708>] usb_register_driver+0x88/0x140 [usbcore]
    [<ffffffff92401153>] do_one_initcall+0x43/0x210
    [<ffffffff9254f42a>] do_init_module+0x4a/0x200
    [<ffffffff92551d1c>] __do_sys_finit_module+0xac/0x120
    [<ffffffff92ee6626>] do_syscall_64+0x56/0x80
    [<ffffffff9300006a>] entry_SYSCALL_64_after_hwframe+0x46/0xb0

The leak was verified to be real by unloading the driver, which resulted
in a dangling pointer to the allocation.

The allocated memory is freed in rtw_usb_intf_deinit().

Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Cc: Sascha Hauer <s.hauer@pengutronix.de>
Cc: Ping-Ke Shih <pkshih@realtek.com>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230417160331.23071-1-Larry.Finger@lwfinger.net
18 months agowifi: rtw88: call rtw8821c_switch_rf_set() according to chip variant
Sascha Hauer [Mon, 17 Apr 2023 14:03:58 +0000 (16:03 +0200)]
wifi: rtw88: call rtw8821c_switch_rf_set() according to chip variant

We have to call rtw8821c_switch_rf_set() with SWITCH_TO_WLG or
SWITCH_TO_BTG according to the chip variant as denoted in rfe_option.
The information which argument to use for which variant has been
taken from the vendor driver.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230417140358.2240429-5-s.hauer@pengutronix.de
18 months agowifi: rtw88: set pkg_type correctly for specific rtw8821c variants
Sascha Hauer [Mon, 17 Apr 2023 14:03:57 +0000 (16:03 +0200)]
wifi: rtw88: set pkg_type correctly for specific rtw8821c variants

According to the vendor driver the pkg_type has to be set to '1'
for some rtw8821c variants. As the pkg_type has been hardcoded to
'0', add a field for it in struct rtw_hal and set this correctly
in the rtw8821c part.
With this parsing of a rtw_table is influenced and check_positive()
in phy.c returns true for some cases here. The same is done in the
vendor driver. However, this has no visible effect on the driver
here.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230417140358.2240429-4-s.hauer@pengutronix.de
18 months agowifi: rtw88: rtw8821c: Fix rfe_option field width
Sascha Hauer [Mon, 17 Apr 2023 14:03:56 +0000 (16:03 +0200)]
wifi: rtw88: rtw8821c: Fix rfe_option field width

On my RTW8821CU chipset rfe_option reads as 0x22. Looking at the
vendor driver suggests that the field width of rfe_option is 5 bit,
so rfe_option should be masked with 0x1f.

Without this the rfe_option comparisons with 2 further down the
driver evaluate as false when they should really evaluate as true.
The effect is that 2G channels do not work.

rfe_option is also used as an array index into rtw8821c_rfe_defs[].
rtw8821c_rfe_defs[34] (0x22) was added as part of adding USB support,
likely because rfe_option reads as 0x22. As this now becomes 0x2,
rtw8821c_rfe_defs[34] is no longer used and can be removed.

Note that this might not be the whole truth. In the vendor driver
there are indeed places where the unmasked rfe_option value is used.
However, the driver has several places where rfe_option is tested
with the pattern if (rfe_option == 2 || rfe_option == 0x22) or
if (rfe_option == 4 || rfe_option == 0x24), so that rfe_option BIT(5)
has no influence on the code path taken. We therefore mask BIT(5)
out from rfe_option entirely until this assumption is proved wrong
by some chip variant we do not know yet.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Tested-by: Alexandru gagniuc <mr.nuke.me@gmail.com>
Tested-by: Larry Finger <Larry.Finger@lwfinger.net>
Tested-by: ValdikSS <iam@valdikss.org.ru>
Cc: stable@vger.kernel.org
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230417140358.2240429-3-s.hauer@pengutronix.de
18 months agowifi: rtw88: usb: fix priority queue to endpoint mapping
Sascha Hauer [Mon, 17 Apr 2023 14:03:55 +0000 (16:03 +0200)]
wifi: rtw88: usb: fix priority queue to endpoint mapping

The RTW88 chipsets have four different priority queues in hardware. For
the USB type chipsets the packets destined for a specific priority queue
must be sent through the endpoint corresponding to the queue. This was
not fully understood when porting from the RTW88 USB out of tree driver
and thus violated.

This patch implements the qsel to endpoint mapping as in
get_usb_bulkout_id_88xx() in the downstream driver.

Without this the driver often issues "timed out to flush queue 3"
warnings and often TX stalls completely.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Tested-by: ValdikSS <iam@valdikss.org.ru>
Tested-by: Alexandru gagniuc <mr.nuke.me@gmail.com>
Tested-by: Larry Finger <Larry.Finger@lwfinger.net>
Cc: stable@vger.kernel.org
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230417140358.2240429-2-s.hauer@pengutronix.de
18 months agowifi: rtw88: 8822c: add iface combination
Po-Hao Huang [Fri, 14 Apr 2023 12:13:31 +0000 (20:13 +0800)]
wifi: rtw88: 8822c: add iface combination

Allow 8822c to operate two interface concurrently, only 1 AP mode plus
1 station mode under same frequency is supported. Combination of other
types will not be added until requested.

Signed-off-by: Po-Hao Huang <phhuang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414121331.18062-1-pkshih@realtek.com
18 months agowifi: rtw88: handle station mode concurrent scan with AP mode
Po-Hao Huang [Fri, 14 Apr 2023 12:13:23 +0000 (20:13 +0800)]
wifi: rtw88: handle station mode concurrent scan with AP mode

This patch allows vifs sharing same hardware with the AP mode vif to
do scan, do note that this could lead to packet loss or disconnection
of the AP's clients. Since we don't have chanctx, update scan info
upon set channel so bandwidth changes won't go unnoticed and get
misconfigured after scan. Download beacon just before scan starts to
allow hardware to get proper content to do beaconing. Last, beacons
should only be transmitted in AP's operating channel. Turn related
beacon functions off while we're in other channels so the receiving
stations won't get confused.

Signed-off-by: Po-Hao Huang <phhuang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414121323.18008-1-pkshih@realtek.com
18 months agowifi: rtw88: prevent scan abort with other VIFs
Po-Hao Huang [Fri, 14 Apr 2023 12:13:12 +0000 (20:13 +0800)]
wifi: rtw88: prevent scan abort with other VIFs

Only abort scan with current scanning VIF. If we have more than one
interface, we could call rtw_hw_scan_abort() with the wrong VIF as
input. This avoids potential null pointer being accessed when actually
the other VIF is scanning.

Signed-off-by: Po-Hao Huang <phhuang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414121312.17954-1-pkshih@realtek.com
18 months agowifi: rtw88: refine reserved page flow for AP mode
Po-Hao Huang [Fri, 14 Apr 2023 12:13:00 +0000 (20:13 +0800)]
wifi: rtw88: refine reserved page flow for AP mode

Only gather reserved pages from AP interface after it has started. Or
else ieee80211_beacon_get_*() returns NULL and causes other VIFs'
reserved pages fail to download. Update location of current reserved page
after beacon renews so offsets changed by beacon can be recognized.

Signed-off-by: Po-Hao Huang <phhuang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414121300.17900-1-pkshih@realtek.com
18 months agowifi: rtw88: disallow PS during AP mode
Po-Hao Huang [Fri, 14 Apr 2023 12:11:31 +0000 (20:11 +0800)]
wifi: rtw88: disallow PS during AP mode

Firmware can't support PS mode during AP mode, so disallow this case.

Signed-off-by: Po-Hao Huang <phhuang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414121135.17828-5-pkshih@realtek.com
18 months agowifi: rtw88: 8822c: extend reserved page number
Po-Hao Huang [Fri, 14 Apr 2023 12:11:30 +0000 (20:11 +0800)]
wifi: rtw88: 8822c: extend reserved page number

Extend 8822c's reserved page number to accommodate additional required
pages. Reserved page is an area of memory in the FIFO dedicated for
special purposes. Previously only one interface is supported so 8 pages
should suffice, extend it so we can support 2 interfaces concurrently.

Signed-off-by: Po-Hao Huang <phhuang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414121135.17828-4-pkshih@realtek.com
18 months agowifi: rtw88: add port switch for AP mode
Po-Hao Huang [Fri, 14 Apr 2023 12:11:29 +0000 (20:11 +0800)]
wifi: rtw88: add port switch for AP mode

Switch port settings if AP mode does not start on port 0 because of
hardware limitation. For some ICs, beacons on ports other than zero
could misbehave and do not issue properly, to fix this we change AP
VIFs to port zero when multiple interfaces is active.

Signed-off-by: Po-Hao Huang <phhuang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414121135.17828-3-pkshih@realtek.com
18 months agowifi: rtw88: add bitmap for dynamic port settings
Po-Hao Huang [Fri, 14 Apr 2023 12:11:28 +0000 (20:11 +0800)]
wifi: rtw88: add bitmap for dynamic port settings

In order to support multiple interfaces, multiple port settings will
be required. Current code always uses port 0 and should be changed.
Declare a bitmap with size equal to hardware port number to record
the current usage.

Signed-off-by: Po-Hao Huang <phhuang@realtek.com>
Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414121135.17828-2-pkshih@realtek.com
18 months agowifi: rtw89: mac: use regular int as return type of DLE buffer request
Ping-Ke Shih [Fri, 14 Apr 2023 08:22:28 +0000 (16:22 +0800)]
wifi: rtw89: mac: use regular int as return type of DLE buffer request

The function to request DLE (data link engine) buffer uses 'u16' as return
value that mixes error code, so change it to 'int' as regular error code.
Also, treat invalid register value (0xfff) as an error.

Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/r/20230414082228.30766-1-pkshih@realtek.com
19 months agowifi: mac80211: remove return value check of debugfs_create_dir()
Yingsha Xu [Wed, 19 Apr 2023 10:45:47 +0000 (18:45 +0800)]
wifi: mac80211: remove return value check of debugfs_create_dir()

Smatch complains that:
debugfs_hw_add() warn: 'statsd' is an error pointer or valid

Debugfs checks are generally not supposed to be checked for errors
and it is not necessary here.

Just delete the dead code.

Signed-off-by: Yingsha Xu <ysxu@hust.edu.cn>
Reviewed-by: Dongliang Mu <dzm91@hust.edu.cn>
Link: https://lore.kernel.org/r/20230419104548.30124-1-ysxu@hust.edu.cn
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
19 months agowifi: iwlwifi: mvm: fix RFKILL report when driver is going down
Avraham Stern [Tue, 18 Apr 2023 09:28:14 +0000 (12:28 +0300)]
wifi: iwlwifi: mvm: fix RFKILL report when driver is going down

When CSME takes ownership, the driver sets RFKILL on, and this
triggers driver unload and sending the confirmation SAP message.
However, when IWL_MVM_MEI_REPORT_RFKILL is set, RFKILL was not
reported and as a result, the driver did not confirm the ownership
transition. Fix it.

Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Link: https://lore.kernel.org/r/20230418122405.29ac3cd3df73.I96b32bc274bfe1e3871e54d3fa29c7ac4f40446f@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
19 months agowifi: iwlwifi: mei: re-ask for ownership after it was taken by CSME
Avraham Stern [Tue, 18 Apr 2023 09:28:13 +0000 (12:28 +0300)]
wifi: iwlwifi: mei: re-ask for ownership after it was taken by CSME

When the host disconnects from the AP CSME takes ownership right away.
Since the driver never asks for ownership again wifi is left in rfkill
until CSME releases the NIC, although in many cases the host could
re-connect shortly after the disconnection. To allow the host to
recover from occasional disconnection, re-ask for ownership to let
the host connect again.
Allow one minute before re-asking for ownership to avoid too frequent
ownership transitions.

Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Link: https://lore.kernel.org/r/20230418122405.a6c6ebc48f2d.I8a17003b86e71b3567521cc69864b9cbe9553ea9@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
19 months agowifi: iwlwifi: mei: make mei filtered scan more aggressive
Avraham Stern [Tue, 18 Apr 2023 09:28:12 +0000 (12:28 +0300)]
wifi: iwlwifi: mei: make mei filtered scan more aggressive

When mei filtered scan is performed, it must find the AP on the first
scan, otherwise CSME will take the ownership of the NIC.
Make this scan more aggressive by scanning the channel the AP is
supposed to be on (as reported by CSME) several times.

Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Link: https://lore.kernel.org/r/20230418122405.47e383b10b18.I14340a118acdb19ecb7214e7ff413054c77bd99c@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
19 months agowifi: iwlwifi: modify scan request and results when in link protection
Avraham Stern [Tue, 18 Apr 2023 09:28:11 +0000 (12:28 +0300)]
wifi: iwlwifi: modify scan request and results when in link protection

When CSME is connected and has link protection set, the driver must
connect to the same AP CSME is connected to.
When in link protection, modify scan request parameters to include
only the channel of the AP CSME is connected to and scan for the
same SSID. In addition, filter the scan results to include only
results from the same AP. This will make sure the driver will connect
to the same AP and will do it fast enough to keep the session alive.

Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Link: https://lore.kernel.org/r/20230418122405.c1b55de3d704.I3895eebe18b3b672607695c887d728e113fc85ec@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>