Since channel active time is not available from MIB registers, emulate it
using the kernel boot time (in us)
Signed-off-by: Felix Fietkau <[email protected]>
Using the NAPI or netdev frag cache along with other drivers can lead to
32 KiB pages being held for a long time, despite only being used for
very few page fragments.
This can happen if the driver grabs one or two fragments for rx ring
refill, while other drivers use (and free up) the remaining fragments.
The 32 KiB higher-order page can only be freed once all users have freed
their fragments.
Depending on the traffic patterns, this can waste a lot of memory and
look a lot like a memory leak.
Signed-off-by: Felix Fietkau <[email protected]>
When switching between low gain (high RSSI) and high gain settings, it
can take a few seconds to adjust to the current environment.
This can lead to short periods of time with extreme packet loss.
When switching from low_gain=1 to low_gain=2, start with the same gain
adjustment value instead of the lowest to avoid spikes of huge numbers
of false CCA events
Also avoid resetting adjustment values on switching between low_gain
values 0 and 1, since it affects only the upper limit of vga adjustment
Signed-off-by: Felix Fietkau <[email protected]>
The range should only be limited to 4 for really weak signals, for all
other gain settings the range is 16.
Signed-off-by: Felix Fietkau <[email protected]>
usleep_range can cause excessive latency on channel change if waiting
for the MAC to stop fails. It will be forced to stop by the code
following that loop anyway.
Signed-off-by: Felix Fietkau <[email protected]>
During scans, mac80211 frequently switches back to the home channel to
minimize interruption of ongoing traffic. Keep regular tx queues active
during that time.
Signed-off-by: Felix Fietkau <[email protected]>
The beacon timer drifts by 1 microsecond every TBTT. After 20 minutes
with a beacon interval of 100, the drift will be almost 12 ms, enough to
cause weird issues for devices in powersave mode.
Since the beacon timer is configured in units of 1/16 TU (64 us), we
need to adjust it once every 64 beacons and only for one beacon.
Signed-off-by: Felix Fietkau <[email protected]>
If the radio was previously in running state, it can receive some
packets before it is able to process them.
This can lead to a crash if the channel is not initialized yet.
Discard all rx packets until start() is called
Signed-off-by: Felix Fietkau <[email protected]>
For encryption to work properly, the BSS index needs to be initialized
for the WCID entry used for the interface.
Signed-off-by: Felix Fietkau <[email protected]>
Add a spinlock in mt76_rx_complete. Without this, multiple stats updates
could happen in parallel, which can lead to deadlocks. There are
probably more corner cases fixed by this change.
Signed-off-by: Felix Fietkau <[email protected]>
The check within mt76_txq_send_burst is not enough, as it happens after
a first frame has already been queued up
Signed-off-by: Felix Fietkau <[email protected]>
- discard packets that have no rx group 3 (signal/rate info missing)
- discard packets marked as VHT (no hardware support)
- discard packets with invalid HT rate values
Signed-off-by: Felix Fietkau <[email protected]>
When doing rate probing, multiple frames might end up getting sent with
the probe rate, because of the way that rate probing is set up.
This race window is most visible with MCS rates and aggregation.
To improve rate control stats accuracy, pull the final tx rate from the
descriptor
Signed-off-by: Felix Fietkau <[email protected]>
Tx status reporting cannot account for the tx attempt starting at a rate
further down the chain, so without this flag it leads to inaccurate
results.
Signed-off-by: Felix Fietkau <[email protected]>
There were a number of issues in the existing tx power handling code:
1. The EEPROM target power for chain 0 refers to the actual output power
after the channel and bandwidth delta have been added to the initial
channel gain. This means the delta values should not be added/subtracted
for the rate power calculation
2. When power is reduced significantly, the initial channel gain
underflows very quickly, while the per-rate power offsets are high.
This miscalculation causes effective tx power to be increased when it
should actually be lowered.
Fix this by trying to adjust the channel gain to the lowest power from
the rate table. In case of under- or overflow, compensate by adding an
appropriate delta to the rate power values.
This makes power configuration more accurate on a wider range of
configurable values
Signed-off-by: Felix Fietkau <[email protected]>
Allows the driver to control whether to send a BlockAckReq after waking
up. MT76x2 needs this set to true, for MT7603 it needs to be false
Signed-off-by: Felix Fietkau <[email protected]>
With software A-MPDU reordering in place, frames that notify mac80211 of
powersave changes are reordered as well, which can cause connection
stalls. Fix this by implementing powersave state processing in the
driver.
Signed-off-by: Felix Fietkau <[email protected]>
Apparently hardware does not perform CCMP PN validation in hardware, so
we need to take care of this in the driver.
Signed-off-by: Felix Fietkau <[email protected]>
Like beacons, probe responses need a hardware-generated TSF value. Set
the flag that causes the hw to generate it
Signed-off-by: Felix Fietkau <[email protected]>
Sending frames to mac80211 needs time, which could allow for more rx
packets to end up in the DMA ring. Retry polling until there are no more
frames left. Improves rx latency under load.
Signed-off-by: Felix Fietkau <[email protected]>
Due to an unrelated issue, the MT_RX_FILTR_CFG_PROMISC flag is currently
unset, which means that monitor mode is unconditionally enabled.
Toggle this flag based on the mac80211 monitor mode setting instead
Signed-off-by: Felix Fietkau <[email protected]>
The hardware has a separate set of registers to configure a
per-interface MAC address.
Signed-off-by: Lorenzo Bianconi <[email protected]>
Signed-off-by: Felix Fietkau <[email protected]>
It can happen during normal operation and it confuses users more than it
actually helps. workaround trigger counts are still available via
debugfs
Signed-off-by: Felix Fietkau <[email protected]>
Hardware encryption seems to break encrypted unicast mgmt tx.
Unfortunately the hardware TXWI header does not have a bit to indicate
that a frame is software encrypted, so sw-encrypted frames need to use a
different WCID. For that to work, the CCMP PN needs to be generated in
software, which makes things a bit slower, so only do it for keys that
also need to tx management frames.
Signed-off-by: Felix Fietkau <[email protected]>
Felix Fietkau