Align the mac80211 implementation with P802.11be_D1.5.
Signed-off-by: NIlan Peer <ilan.peer@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

There are a few places that check ps_sdata and/or the dynamic
PS timeout, but they're erroneous in case SUPPORTS_DYNAMIC_PS
is set by the driver.

Skip the entire recalculation in this case so we cannot get
into those paths elsewhere, and so we simplify this for the
purpose of implementing MLO.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

If we don't really have multiple links, omit the link ID from
link debug prints, otherwise we change the format for all of
the existing drivers (most of which might never support MLO),
and also have extra noise in the logs.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Since links can be added and removed dynamically, we need to
somehow protect the sdata->link[] and vif->link_conf[] array
pointers from disappearing when accessing them without locks.
RCU-ify the pointers to achieve this, which requires quite a
bit of rework.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Implement callbacks for cfg80211 add_link_station, mod_link_station,
and del_link_station API.
Signed-off-by: NShaul Triebitz <shaul.triebitz@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Put the link_station_parameters structure in the station_parameters
structure (and remove the station_parameters fields already existing
in link_station_parameters).
Now, for an MLD station, the default link is added together with
the station.
Signed-off-by: NShaul Triebitz <shaul.triebitz@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Similar to ieee80211_get_sband but get the sband of the link_conf.
Signed-off-by: NShaul Triebitz <shaul.triebitz@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

AP SMPS was removed and not needed anymore. Remove the leftovers.
Signed-off-by: NAndrei Otcheretianski <andrei.otcheretianski@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Check all the MLO links to decide whether offchannel TX is needed.
Signed-off-by: NAndrei Otcheretianski <andrei.otcheretianski@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

When checking whether or not to accept a frame in RX,
take into account the configured link addresses. Also
look up the link station correctly.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Instead of open-coding it, use sdata_dereference().
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

ieee80211s_update_metric function uses sta_set_rate_info_tx
function to get struct rate_info data from ieee80211_tx_rate
struct, present in ieee80211_sta->deflink.tx_stats. However,
drivers can skip tx rate calculation by setting rate idx as
-1. Such drivers provides rate_info directly and hence
ieee80211s metric is updated incorrectly since ieee80211_tx_rate
has inconsistent data.

Add fix to use rate_info directly if present instead of
sta_set_rate_info_tx for updating ieee80211s metric.
Signed-off-by: NAditya Kumar Singh <quic_adisi@quicinc.com>
Link: https://lore.kernel.org/r/20220701133611.544-1-quic_adisi@quicinc.comSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

WDS needs 4addr packets to trigger AP for wlan0.sta creation.
However, the 4addr null frame is sent at a high rate so that
sometimes the AP can't receive it. Switch to using min rate.
Signed-off-by: NLian Chen <lian.chen@mediatek.com>
Link: https://lore.kernel.org/r/20220714091636.59107-1-lian.chen@mediatek.comSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

In AP mode, multicast traffic is handled very differently from normal traffic,
especially if at least one client is in powersave mode.
This means that multicast packets can be buffered a lot longer than normal
unicast packets, and can eat up the AQL budget very quickly because of the low
data rate.
Along with the recent change to maintain a global PHY AQL limit, this can lead
to significant latency spikes for unicast traffic.

Since queueing multicast to hardware is currently not constrained by AQL limits
anyway, let's just exclude it from the AQL pending airtime calculation entirely.

Fixes: 8e4bac06 ("wifi: mac80211: add a per-PHY AQL limit to improve fairness")
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Link: https://lore.kernel.org/r/20220713083444.86129-1-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

When using iTXQ, the code assumes that there is only one vif queue for
broadcast packets, using the BE queue. Allowing non-BE queue marking
violates that assumption and txq->ac == skb_queue_mapping is no longer
guaranteed. This can cause issues with queue handling in the driver and
also causes issues with the recent ATF change, resulting in an AQL
underflow warning.

Cc: stable@vger.kernel.org
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Link: https://lore.kernel.org/r/20220702145227.39356-1-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

Use bitmap_zalloc()/bitmap_free() instead of hand-writing them.

It is less verbose and it improves the semantic.
Signed-off-by: NChristophe JAILLET <christophe.jaillet@wanadoo.fr>
Link: https://lore.kernel.org/r/dfb438a6a199ee4c95081fa01bd758fd30e50931.1656962156.git.christophe.jaillet@wanadoo.frSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

When mac80211 downgrades working bandwidth, the
center_freq and center_freq1 need to be recalculated.
There is a typo in the case of downgrading bandwidth from
320MHz to 160MHz which would cause a wrong frequency value.
Reviewed-by: NMoney Wang <Money.Wang@mediatek.com>
Signed-off-by: NMeiChia Chiu <MeiChia.Chiu@mediatek.com>
Link: https://lore.kernel.org/r/20220708095823.12959-1-MeiChia.Chiu@mediatek.comSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

If the interface isn't (yet) added to the driver, skip the
link info update. This was previously done for the BSS info
changes, but I forgot to copy the same check here.

Fixes: 7b7090b4 ("wifi: mac80211: split bss_info_changed method")
Reported-by: syzbot+bce2ca140cc00578ed07@syzkaller.appspotmail.com
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

When a client does not generate any local tx activity, accumulating airtime
deficit for the round-robin scheduler can be harmful. If this goes on for too
long, the deficit could grow quite large, which might cause unreasonable
initial latency once the client becomes active
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Acked-by: NToke Høiland-Jørgensen <toke@toke.dk>
Link: https://lore.kernel.org/r/20220625212411.36675-7-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

Now that the global pending airtime is more relevant for airtime fairness,
it makes sense to make it accessible via debugfs for debugging
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Acked-by: NToke Høiland-Jørgensen <toke@toke.dk>
Link: https://lore.kernel.org/r/20220625212411.36675-6-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

In order to maintain fairness, the amount of queueing needs to be limited
beyond the simple per-station AQL budget, otherwise the driver can simply
repeatedly do scheduling rounds until all queues that have not used their
AQL budget become eligble.

To be conservative, use the high AQL limit for the first txq and add half
of the low AQL for each subsequent queue.
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Acked-by: NToke Høiland-Jørgensen <toke@toke.dk>
Link: https://lore.kernel.org/r/20220625212411.36675-5-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

This allows proper deficit accounting to ensure that they don't carry their
deficit until the next time they become active
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Acked-by: NToke Høiland-Jørgensen <toke@toke.dk>
Link: https://lore.kernel.org/r/20220625212411.36675-4-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

When queueing packets for a station, deficit only gets added once the packets
have been transmitted, which could be much later. During that time, a lot of
temporary unfairness could happen, which could lead to bursty behavior.
Fix this by subtracting the aql_tx_pending when checking the deficit in tx
scheduling.
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Acked-by: NToke Høiland-Jørgensen <toke@toke.dk>
Link: https://lore.kernel.org/r/20220625212411.36675-3-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

32 bit is more than enough range for the airtime deficit
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Acked-by: NToke Høiland-Jørgensen <toke@toke.dk>
Link: https://lore.kernel.org/r/20220625212411.36675-2-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

This reverts commits 6a789ba6 and
2433647b.

The virtual time scheduler code has a number of issues:
- queues slowed down by hardware/firmware powersave handling were not properly
  handled.
- on ath10k in push-pull mode, tx queues that the driver tries to pull from
  were starved, causing excessive latency
- delay between tx enqueue and reported airtime use were causing excessively
  bursty tx behavior

The bursty behavior may also be present on the round-robin scheduler, but there
it is much easier to fix without introducing additional regressions
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Acked-by: NToke Høiland-Jørgensen <toke@toke.dk>
Link: https://lore.kernel.org/r/20220625212411.36675-1-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

struct link_sta_info has now a cur_max_bandwidth data:

	net/mac80211/sta_info.h:569: warning: Function parameter or member 'cur_max_bandwidth' not described in 'link_sta_info'

Copy the meaning from struct sta_info, documenting it.
Signed-off-by: NMauro Carvalho Chehab <mchehab@kernel.org>
Link: https://lore.kernel.org/r/37d898634bb30776442a33833c48cbb21c90ecc6.1656409369.git.mchehab@kernel.orgSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

When a vif is being removed and sdata->bss is cleared, __ieee80211_wake_txqs
can still be called on it, which crashes as soon as sdata->bss is being
dereferenced.
To fix this properly, check for SDATA_STATE_RUNNING before waking queues,
and take the fq lock when setting it (to ensure that __ieee80211_wake_txqs
observes the change when running on a different CPU)
Signed-off-by: NFelix Fietkau <nbd@nbd.name>
Acked-by: NToke Høiland-Jørgensen <toke@kernel.org>
Link: https://lore.kernel.org/r/20220531190824.60019-1-nbd@nbd.nameSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

Add a missing skb_shared check into 802.3 path to prevent potential
use-after-free from happening. This also uses skb_share_check()
instead of open-coding in tx path.
Signed-off-by: NRyder Lee <ryder.lee@mediatek.com>
Link: https://lore.kernel.org/r/e7a73aaf7742b17e43421c56625646dfc5c4d2cb.1653571902.git.ryder.lee@mediatek.comSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

Introduce the capability to specify gfp_t parameter to
ieeee80211_obss_color_collision_notify routine since it runs in
interrupt context in ieee80211_rx_check_bss_color_collision().

Fixes: 6d945a33 ("mac80211: introduce BSS color collision detection")
Co-developed-by: NRyder Lee <ryder.lee@mediatek.com>
Signed-off-by: NRyder Lee <ryder.lee@mediatek.com>
Signed-off-by: NLorenzo Bianconi <lorenzo@kernel.org>
Link: https://lore.kernel.org/r/02c990fb3fbd929c8548a656477d20d6c0427a13.1655419135.git.lorenzo@kernel.orgSigned-off-by: NJohannes Berg <johannes.berg@intel.com>

We should set the STA deflink addresses in case no
link is really added.

Fixes: 046d2e7c ("mac80211: prepare sta handling for MLO support")
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Maintain a hash table of link-sta addresses so we can find
them for management frames etc. where addresses haven't
been replaced by the drivers to the MLD address yet.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

We need to be able to access these in a race-free way under
traffic while adding/removing them, so RCU-ify the pointers.
This requires passing a link_sta to a lot of functions so
we don't have to do the RCU handling everywhere.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Hardcoding link[0] is more confusing than using deflink,
so use deflink for now, which requires making the macro
a bit safer.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

This is only used in rx.c, so move it into the file.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Pass the link id through to the get_beacon and return
the beacon for a specific link id.
Signed-off-by: NShaul Triebitz <shaul.triebitz@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

In start_ap and stop_ap mac80211 callbacks pass the link_id
to the drivers.
Signed-off-by: NShaul Triebitz <shaul.triebitz@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Use link and link_conf according to the link_id
provided by cfg in start_ap/stop_ap and change_beacon.
Also use them in the functions called by them.
Note that for a non MLD device, the link_id is 0,
and link[0] and link_conf[0] equal to deflink and
bss_conf respectively (what was there before).

Also, call vif_info_change for BSS related changes (SSID), and
link_info_change for LINK related changes (instead of the
legacy bss_info_change).
Signed-off-by: NShaul Triebitz <shaul.triebitz@intel.com>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

When a link is added or removed via nl80211, these are
called. Implement them so we don't have to check in all
the different per-link commands whether we've already
created the necessary datastructures.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Add the necessary infrastructure, including a new driver
method, to add/remove links to/from a station. To do this,
refactor the link alloc/free a bit, splitting that so we
can do it without linking them, to handle failures better.

Note that a station entry must be created representing an
MLD or a non-MLD STA, it cannot change between the two.
When representing an MLD, the 'deflink' is used for the
first link, which might be removed later, in which case
the memory isn't reused.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

Take a few bits out of the control.flags to add the link ID
to TX frame metadata, so drivers don't need to look it up
by the address themselves. Implement that lookup where it's
needed, for internal frame TX, and set it to "unspecified"
for data transmissions.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>