Some kernel-doc is missing here, add it.

Link: https://msgid.link/20240206134555.eb95c1dfc1f0.Ibaf8b3249d9de59358bf6503fe4a186d9ac6544d@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

This was added earlier, add kernel-doc for it.

Link: https://msgid.link/20240206134555.6354b0ac8610.Ib90d3651834c556b73697388f59bd396a1f6f9b0@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

The vif pointer at least looks like it can actually be NULL
in some cases such as the monitor-mode vif, causing static
checkers to complain with the immediate derefence. In these
cases the sta pointer will also be NULL, but clarify it in
the code anyway.

Link: https://msgid.link/20240131164910.60066625a239.Idfb6a5a9876f9f631eae760055e1c4018259a971@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

This could be used e.g. for temporarily sending quiet
(mode=1 in CSA/ECSA), or updating bandwidth. This is
also useful for testing, since it's something that an
AP may do and the client needs to be prepared. Simply
allow it.

Link: https://msgid.link/20240129203544.ef7258d5790d.Idafe22e41621757458d4960659b9621853f7104d@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Convert ieee80211_ie_build_eht_cap() to the SKB-put function
style, renaming it to ieee80211_put_eht_cap().

Link: https://msgid.link/20240129202041.ece9769e3c94.Ibd17bea6311f0c7ba56f6c1803fa3208abaaebb9@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Convert ieee80211_ie_build_he_cap() to the SKB-put function
style, renaming it to ieee80211_put_he_cap().

Link: https://msgid.link/20240129202041.e6ef888980d9.Ied9e014314b5d27611e693e3d4cb63bdc8d7de17@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

This code got really big, move it to a new file.
Pure code move.

Link: https://msgid.link/20240129202041.7f27f7c895e4.I0adfc28bd656a4d44c2bf47966277eecf56cbaa0@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

The building of elements is really mess, and really the only
reason we're not doing it in SKBs in the first place is that
the scan code in ieee80211_build_preq_ies() doesn't.

Convert ieee80211_build_preq_ies() to use an SKB internally
so that we can gradually convert other things to ..._put_*()
style interfaces.

Link: https://msgid.link/20240129202041.c3a8e3c2cc99.I9d9920858c30ae5154719783933de0d7bc2a2cb9@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Make this a new-style "put" function, and change the
parameters to pass more information directly, this
makes it usable also for the MLME code.

Link: https://msgid.link/20240129202041.f604a03bd728.I8c798ea45b8479ac9982e77d0378af11a09ccdaf@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

We don't need to use the write function here since we already
have an SKB, so use ieee80211_put_he_6ghz_cap() with the SMPS
mode taken from the link we're using.

Link: https://msgid.link/20240129202041.6454ac78ff8c.I7152e3c27645105478c68d40ca493feb27cac6bf@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

The term 'IE' isn't really in use in the spec, and I want
to rework all of this to use SKBs as the primary method
for building elements. Rename this one already.

Link: https://msgid.link/20240129202041.b8064a4e73b5.I8d2f4526562029107c6414c6cda378b300b1b0b0@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

When limiting a station's supported bandwidth while connecting,
also limit various other HE capabilities according to the bandwidth
needed for them.

Link: https://msgid.link/20240129202041.34be99efca25.I02a695961bc6aadd37768b17c50fcdec4427d460@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

If intending to associate with a lower bandwidth, remove capabilities
related to 320 MHz from the EHT capabilities element. Also change the
EHT MCS-NSS set accordingly: if just reducing 320->160 or similar the
format doesn't change, just cut off the last bytes. If changing from
higher bandwidth to 20 MHz only EHT STA, adjust the format.

Note that this also requires adjusting the caller in mlme.c since the
data written can now be shorter than it determined. We need to clean
all that up. Since the other callers pass NULL for the conn limit, we
don't need to change things there.

Link: https://msgid.link/20240129202041.b5f6df108c77.I0d8ea04079c61cb3744cc88625eeaf0d4776dc2b@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

We don't need to pass the iftype there, we already have it
in the sdata. Simplify this code.

Link: https://msgid.link/20240129202041.5890eb1d4184.Ibce7e5abcc7887630da03ac2263d8004ec541418@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

There really shouldn't be a basic multi-link element in any
per-STA profile in an association response, it's not clear
what that would really mean. Refuse connecting in this case
since the AP isn't following the spec.

Link: https://msgid.link/20240129200652.23f1e3b337f1.Idd2e43cdbfe3ba15b3e9b8aeb54c8115587177a0@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Let the element parsing function return what kind of error
was encountered, as a bitmap, even if nothing currently
checks for which specific error it was, we'll use it later.

Link: https://msgid.link/20240129200652.1a69f2a31ec7.I55b86561d64e7ef1504c73f6f2813c33030c8136@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

If the vif is an MLD then it may receive multicast from
different links, and should drop those frames according
to the SN. Implement that.

Link: https://msgid.link/20240129200456.693b77d14b44.I491846f2bea0058c14eab6422962c10bfae9b675@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

This will also be useful for MLO duplicate multicast
detection, but add it already here and use it in one
place that trivially converts.

Link: https://msgid.link/20240129200456.f0ff49c80006.I850d2785ab1640e56e262d3ad7343b87f6962552@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Aloka originally suggested that puncturing should be part of
the chandef, so that it's treated correctly. At the time, I
disagreed and it ended up not part of the chandef, but I've
now realized that this was wrong. Even for clients, the RX,
and perhaps more importantly, CCA configuration needs to take
puncturing into account.

Move puncturing into the chandef, and adjust all the code
accordingly. Also add a few tests for puncturing in chandef
compatibility checking.

Link: https://lore.kernel.org/linux-wireless/20220214223051.3610-1-quic_alokad@quicinc.com/Suggested-by: Aloka Dixit <quic_alokad@quicinc.com>
Link: https://msgid.link/20240129194108.307183a5d2e5.I4d7fe2f126b2366c1312010e2900dfb2abffa0f6@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Instead of calculating the new primary 40/80/160 MHz
center frequency here, use the new helper function from
cfg80211.

Link: https://msgid.link/20240129194108.eb59d6433d18.I74b745f0d1a32e779fb25d50c56407be7c35b840@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Simplify cfg80211_chandef_compatible() a bit by switching
c1 and c2 around so that c1 is always the narrower one
(once they're not identical or narrow/S1G). Then we can
just check the various primary channels and exit with the
wider one (c2), or NULL.

Also refactor the primary 40/80/160 function to not have
all the calculations hard-coded, and use a wrapper around
it to check primary 40/80/160 compatibility.

While at it, add some kunit tests for this functionality.

Also expose the new cfg80211_chandef_primary_freq() to
drivers, mac80211 will use it.

Link: https://msgid.link/20240129194108.be3e6eccaba3.I8399c2ff1435d7378e5837794cb5aa6dd2ee1416@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

It doesn't look like we can get into this code, but make it
more robust and declare two S1G chandefs to be incompatible
unless they're identical.

Link: https://msgid.link/20240129194108.b28fb0644a8c.I9297ada5cf1baf00dbbdf8fcffd1806883489fc9@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Add a new inline helper function to ieee80211.h to
extract the disabled subchannels bitmap from an EHT
operation element, and use that in mac80211 where
we do that.

Link: https://msgid.link/20240129194108.d9f50dcec8d0.I8b08cbc2490a734fafcce0fa0fc328211ba6f10b@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Upcoming patches will move the puncturing bitmap into the
chandef, so chandef validation will need to check for correct
puncturing. Purely move the code first so later changes are
easier to review.

Link: https://msgid.link/20240129194108.1ca184427c76.I077deb8d52c4648eac145b63f88b6c5a3b920ddc@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Due to the earlier restructuring we now mostly ignore
the channel configuration in the association response,
apart from the HT/VHT checks we had.

Don't do that, but parse it and update, also dropping the
association if the AP changed its mode in the response.

Link: https://msgid.link/20240129194108.b3efa5eae60c.I1b70c9fd56781b22cdfdca55d34d69f7d0733e31@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

EHT requires that stations are able to participate in
wider bandwidth OFDMA, i.e. parse downlink OFDMA and
uplink OFDMA triggers when they're not capable of (or
not connected at) the (wider) bandwidth that the AP
is using. This requires hardware configuration, since
the entity responsible for parsing (possibly hardware)
needs to know the AP bandwidth.

To support this, change the channel request to have
the AP's bandwidth for clients, and track that in the
channel context in mac80211. This means that the same
chandef might need to be split up into two different
contexts, if the APs are different. Interfaces other
than client are not participating in OFDMA the same
way, so they don't request any AP setting.

Note that this doesn't introduce any API to split a
channel context, so that there are cases where this
might lead to a disconnect, e.g. if there are two
client interfaces using the same channel context, e.g.
both 160 MHz connected to different 320 MHz APs, and
one of the APs switches to 160 MHz.

Note also there are possible cases where this can be
optimised, e.g. when using the upper or lower 160 Mhz,
but I haven't been able to really fully understand the
spec and/or hardware limitations.

If, for some reason, there are no hardware limits on
this because the OFDMA (downlink/trigger) parsing is
done in firmware and can take the transmitter into
account, then drivers can set the new flag
IEEE80211_VIF_IGNORE_OFDMA_WIDER_BW on interfaces to
not have them request any AP bandwidth in the channel
context and ignore this issue entirely. The bss_conf
still contains the AP configuration (if any, i.e. EHT)
in the chanreq.

Link: https://msgid.link/20240129194108.d3d5b35dd783.I939d04674f4ff06f39934b1591c8d36a30ce74c2@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

In the channel context code we have quite a few instances
of nested loops iterating the interfaces and then links.
Add a new for_each_sdata_link() macro and use it. Also,
since it's easier, convert all the loops and a few other
places away from RCU as we now hold the wiphy mutex
everywhere anyway.

This does cause a little bit more work (such as checking
interface types for each link of an interface rather than
not iterating links in some cases), but that's not a huge
issue and seems like an acceptable trade-off, readability
is important too.

Link: https://msgid.link/20240129194108.7240829bd96d.I5ccbb8dd019cbcb5326c85d76121359225d6541a@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

For channel contexts, mac80211 currently uses the cfg80211
chandef struct (control channel, center freq(s), width) to
define towards drivers and internally how these behave. In
fact, there are _two_ such structs used, where the min_def
can reduce bandwidth according to the stations connected.

Unfortunately,  with EHT this is longer be sufficient,  at
least not for all hardware.  EHT requires that non-AP STAs
that are connected to an AP with a lower bandwidth than it
(the AP) advertises (e.g. 160 MHz STA connected to 320 MHz
AP) still be able to receive downlink OFDMA and respond to
trigger frames for uplink OFDMA  that specify the position
and bandwidth  for the non-AP STA  relative to the channel
the AP is using.  Therefore, they need to be aware of this,
and at least for some hardware (e.g. Intel) this awareness
is in the hardware. As a result, use of the "same" channel
may need to be split over  two channel contexts where they
differ by the AP being used.

As a first step,  introduce a concept of a channel request
('chanreq') for each interface,  to control the context it
requests.   This step does nothing but reorganise the code,
so that later the AP's chandef can be added to the request
in order to handle the EHT case described above.

Link: https://msgid.link/20240129194108.2e88e48bd2e9.I4256183debe975c5ed71621611206fdbb69ba330@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

The last caller of this with a NULL argument was related to
the non-chanctx code, so we can now remove this odd logic.
Reviewed-by: Miriam Rachel Korenblit <miriam.rachel.korenblit@intel.com>
Link: https://msgid.link/20240129194108.bad8ec1e76c8.I12287452f42c54baf75821e75491cf6d021af20a@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

There are still surprisingly many non-chanctx drivers, but in
mac80211 that code is a bit awkward. Simplify this by having
those drivers assign 'emulated' ops, so that the mac80211 code
can be more unified between non-chanctx/chanctx drivers. This
cuts the number of places caring about it by about 15, which
are scattered across - now they're fewer and no longer in the
channel context handling.

Link: https://msgid.link/20240129194108.6d0ead50f5cf.I60d093b2fc81ca1853925a4d0ac3a2337d5baa5b@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

In the code we currently check for support 80+80, 160
and 320 channel widths, but really the way this should
be (and is otherwise) handled is that we compute the
highest channel bandwidth given there, and then cut it
down to what we support. This is also needed for wider
bandwidth OFDMA support.

Change the code to remove this limitation and always
parse the highest possible channel width.
Reviewed-by: Miriam Rachel Korenblit <miriam.rachel.korenblit@intel.com>
Link: https://msgid.link/20240129194108.d06f85082e29.I47e68ed3d97b0a2f4ee61e5d8abfcefc8a5b9c08@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Rewrite the station-side connection handling. The connection
flags (IEEE80211_DISABLE_*) are rather confusing, and they're
not always maintained well. Additionally, for wider-bandwidth
OFDMA support we need to know the precise bandwidth of the AP,
which is currently somewhat difficult.

Rewrite this to have a 'mode' (S1G/legacy/HT/...) and a limit
on the bandwidth. This is not entirely clean because some of
those modes aren't completely sequenced (as this assumes in
some places), e.g. VHT doesn't exist on 2.4 GHz, but HE does.
However, it still simplifies things and gives us a good idea
what we're operating as, so we can parse elements accordingly
etc.

This leaves a FIXME for puncturing, this is addressed in a
later patch.
Reviewed-by: Ilan Peer <ilan.peer@intel.com>
Reviewed-by: Miriam Rachel Korenblit <miriam.rachel.korenblit@intel.com>
Link: https://msgid.link/20240129194108.9451722c0110.I3e61f4cfe9da89008e1854160093c76a1e69dc2a@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Most devices now do duration calculations, so we don't hit
this code at all any more. Clearly the approach of warning
at compile time here when new bands are added didn't work,
the new bands were just added with "TODO". Clean it up, it
won't matter for new bands since they'll just not have any
need to calculate durations in software.

While at it, also clean up and unify the code a bit.

Link: https://msgid.link/20240129194108.70a97bd69265.Icdd8b0ac60a382244466510090eb0f5868151f39@changeidSigned-off-by: Johannes Berg <johannes.berg@intel.com>

Not sure how this happened or how nothing complained, but
this variable already exists in the outer function scope
with the same value (and the SKB isn't changed either.)
Remove the extra one.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>

This got unused when the tracing was converted to dynamic
strings, so the define can be removed.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>

There are some changes coming to wireless-next that will
otherwise cause conflicts, pull wireless in first to be
able to resolve that when applying the individual changes
rather than having to do merge resolution later.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>

Until now, rtl8xxxu_watchdog_callback() only fetches RSSI and updates
the rate mask in station mode. This means, in AP mode only the default
rate mask is used.

In order to have the rate mask reflect the actual connection quality,
extend rtl8xxxu_watchdog_callback() to iterate over every sta. Like in
the rtw88 driver, add a function to collect all currently present stas
and then iterate over a list of copies to ensure no RCU lock problems
for register access via USB. Remove the existing RCU lock in
rtl8xxxu_refresh_rate_mask().

Since the currently used ieee80211_ave_rssi() is only for 'vif', add
driver-level tracking of RSSI per sta.
Signed-off-by: Martin Kaistra <martin.kaistra@linutronix.de>
Reviewed-by: Ping-Ke Shih <pkshih@realtek.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://msgid.link/20240205093040.1941140-1-martin.kaistra@linutronix.de

Now that the driver core can properly handle constant struct bus_type,
move the bcma_bus_type variable to be a constant structure as well,
placing it into read-only memory which can not be modified at runtime.

Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Suggested-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ricardo B. Marliere <ricardo@marliere.net>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://msgid.link/20240204-bus_cleanup-bcma-v1-1-0d881c793190@marliere.net

Now that the driver core can properly handle constant struct bus_type,
move the ssb_bustype variable to be a constant structure as well,
placing it into read-only memory which can not be modified at runtime.

Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Suggested-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ricardo B. Marliere <ricardo@marliere.net>
Acked-by: Michael Büsch <m@bues.ch>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://msgid.link/20240204-bus_cleanup-ssb-v1-1-511026cd5f3c@marliere.net

Kmemleak reported this error:

    unreferenced object 0xd73d1180 (size 184):
      comm "wpa_supplicant", pid 1559, jiffies 13006305 (age 964.245s)
      hex dump (first 32 bytes):
        00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
        00 00 00 00 00 00 00 00 1e 00 01 00 00 00 00 00  ................
      backtrace:
        [<5ca11420>] kmem_cache_alloc+0x20c/0x5ac
        [<127bdd74>] __alloc_skb+0x144/0x170
        [<fb8a5e38>] __netdev_alloc_skb+0x50/0x180
        [<0f9fa1d5>] __ieee80211_beacon_get+0x290/0x4d4 [mac80211]
        [<7accd02d>] ieee80211_beacon_get_tim+0x54/0x18c [mac80211]
        [<41e25cc3>] wfx_start_ap+0xc8/0x234 [wfx]
        [<93a70356>] ieee80211_start_ap+0x404/0x6b4 [mac80211]
        [<a4a661cd>] nl80211_start_ap+0x76c/0x9e0 [cfg80211]
        [<47bd8b68>] genl_rcv_msg+0x198/0x378
        [<453ef796>] netlink_rcv_skb+0xd0/0x130
        [<6b7c977a>] genl_rcv+0x34/0x44
        [<66b2d04d>] netlink_unicast+0x1b4/0x258
        [<f965b9b6>] netlink_sendmsg+0x1e8/0x428
        [<aadb8231>] ____sys_sendmsg+0x1e0/0x274
        [<d2b5212d>] ___sys_sendmsg+0x80/0xb4
        [<69954f45>] __sys_sendmsg+0x64/0xa8
    unreferenced object 0xce087000 (size 1024):
      comm "wpa_supplicant", pid 1559, jiffies 13006305 (age 964.246s)
      hex dump (first 32 bytes):
        00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
        10 00 07 40 00 00 00 00 00 00 00 00 00 00 00 00  ...@............
      backtrace:
        [<9a993714>] __kmalloc_track_caller+0x230/0x600
        [<f83ea192>] kmalloc_reserve.constprop.0+0x30/0x74
        [<a2c61343>] __alloc_skb+0xa0/0x170
        [<fb8a5e38>] __netdev_alloc_skb+0x50/0x180
        [<0f9fa1d5>] __ieee80211_beacon_get+0x290/0x4d4 [mac80211]
        [<7accd02d>] ieee80211_beacon_get_tim+0x54/0x18c [mac80211]
        [<41e25cc3>] wfx_start_ap+0xc8/0x234 [wfx]
        [<93a70356>] ieee80211_start_ap+0x404/0x6b4 [mac80211]
        [<a4a661cd>] nl80211_start_ap+0x76c/0x9e0 [cfg80211]
        [<47bd8b68>] genl_rcv_msg+0x198/0x378
        [<453ef796>] netlink_rcv_skb+0xd0/0x130
        [<6b7c977a>] genl_rcv+0x34/0x44
        [<66b2d04d>] netlink_unicast+0x1b4/0x258
        [<f965b9b6>] netlink_sendmsg+0x1e8/0x428
        [<aadb8231>] ____sys_sendmsg+0x1e0/0x274
        [<d2b5212d>] ___sys_sendmsg+0x80/0xb4

However, since the kernel is build optimized, it seems the stack is not
accurate. It appears the issue is related to wfx_set_mfp_ap(). The issue
is obvious in this function: memory allocated by ieee80211_beacon_get()
is never released. Fixing this leak makes kmemleak happy.
Reported-by: Ulrich Mohr <u.mohr@semex-engcon.com>
Co-developed-by: Ulrich Mohr <u.mohr@semex-engcon.com>
Signed-off-by: Ulrich Mohr <u.mohr@semex-engcon.com>
Fixes: 268bceec ("staging: wfx: fix BA when device is AP and MFP is enabled")
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Signed-off-by: Kalle Valo <kvalo@kernel.org>
Link: https://msgid.link/20240202164213.1606145-1-jerome.pouiller@silabs.com