wfx_tx_queue_get_num_queued() can take advantage of BIT() instead of
maintaining one variable for a counter and another for a mask.
In add, wfx_tx_queue_get_num_queued() has no real reason to return a
size_t instead of an int.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-64-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since enum action has now only two cases, it can be dropped. Then
hif_handle_tx_data() can be simplified.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-63-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Obviously, the value of wvif->state as no reason to be wrong. At least,
if it the case, dropping the frame is probably not the bast thing to do.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-62-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In current code, packet_id is derivated from mac802.11 packet sequence
number, but this number is only unique for a station. It is not
sufficient.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-60-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When mac80211 ask for a frame to be sent after a DTIM, driver should:
1. Update TIM with multicast bit set (using update_ie). This function
can be called whenever.
2. Keep buffered all frames marked "after dtim"
3. When it receive a suspend_resume indication (see
wfx_suspend_resume_mc()), send all the buffered frames. This
indication is sent by the firmware 4ms before the dtim.
4. If one of the frames returns status "REQUEUE", it means that the
DTIM period was ended before to be able to send the frame.
5. When all the buffered frames were sent or if DTIM period was ended,
driver should update the TIM with multicast bit reset.
All the mess with the asynchronous works can be dropped.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-58-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Indeed, it is not necessary to pass whole hif_ind_suspend_resume_tx to
wfx_suspend_resume_mc().
In add, the structure hif_ind_suspend_resume_tx come from hardware API.
It is not intended to be manipulated in upper layers of the driver.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-57-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
At the beginning, firmware could send suspend_resume indication to
notify that a station wake up or sleep down. However, mac80211 already
handles power save status of stations and this behavior has been removed
from the firmware. So now, when suspend_resume indication is received,
it is always to notify that a DTIM is about to be sent.
So, it is possible to simply wfx_suspend_resume().
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-56-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wfx_ps_notify() is called for any changes in the TIM. However,
association ID 0 is a very special case that should be handled
independently.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-55-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
pspoll_mask is here to send data buffered in driver. But since station
is marked buffered, TIM for this station is 1 and mac80211 will call
sta_notify when a ps-poll is received. So pspoll_mask is useless and
sta_alseep_mask is sufficient.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-54-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wfx_tx_get_tid() was used as a wrapper around ieee80211_get_tid(). It
did sometime return WFX_MAX_TID to ask to upper layers to not include
the frame in "buffered" counter. The objective of this behavior is not
clear, but tests has shown that wfx_tx_get_tid() can be replaced by
ieee80211_get_tid() without any regressions.
BTW, it is not necessary to save the tid in tx_rpiv since it can be
retrieved from the 802.11 header.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-53-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The "link-id" is a slot number provided to the chip. A link-id is
allocated to every station associated with the chip (mainly when the
chip is in AP mode). It is more or less the same thing than the
association ID, but it is limited to 14 values.
Firmware uses the link-id to track the power save status of the
stations.
The current code try to associate a link-id as soon as data are
exchanged with station. It is far easier to rely on sta_add() and
sta_remove().
Until now the value WFX_LINK_ID_NO_ASSOC, was only used when no more
link-id was available. Now, we also use this value for
not-yet-associated stations (that was its primary behavior).
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-51-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Argument provided to wfx_set_tim_impl() is always wvif->aid0_bit_set and
there is no reason to provide another argument.
Also rename wfx_set_tim_impl() into wfx_update_tim() to reflect the new
behavior.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-50-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wfx_tx_confirm_cb() retrieves the station associated with a frame using
the MAC address from the 802.11 header. In the other side wfx_tx()
retrieves the station using sta field from the ieee80211_tx_control
argument.
In wfx_cqm_bssloss_sm(), wfx_tx() was called directly without valid sta
field, but with a valid MAC address in 802.11 header. So there the
processing of this packet was unbalanced and may produce weird bugs.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-48-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It simplify the code if field buffered is hosted in the struct sta_priv
instead of in the struct wfx_link_entry. More globally, struct
wfx_link_entry has no real reasons to exist and should be dropped soon.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-47-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It seems that purpose of "early_data" was to prevent sending data to
mac80211 before station was completely associated. It is a useless
precaution.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-46-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
It is recommended to use function time_*() to compare jiffies.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-45-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wfx_conf_tx() is called for each queue. On every call, the function
updates UAPSD mask and PM mode for all queues. It is a pity since the
UAPSD configuration very rarely changes and it makes exchanges between
the host and the chip more difficult to track.
This patch avoid to update UAPSD and Power Mode in most usual cases.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-43-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When multiple vif are in use (the only supported configuration is one
station and one AP), the driver force power save flag on station.
This behavior allows the station to leave the station channel and make
its business on AP channel.
However, this has a big impact on station performances (especially since
only legacy PS is supported).
When both vifs use the same channel, it is not necessary to keep this
restriction. This greatly improve station performances.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-42-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mac80211 and the device are both able to control power-save per vif.
But, the current code retrieve power-save from wfx_config(). So, it does
not allow to setup power-save independently for each vif. Driver just
has to rely on wfx_bss_info_changed() instead of wfx_config().
wfx_config() has nothing to do anymore, but we keep it since it is
mandatory for mac80211.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-41-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wfx_scan_complete() do nothing with argument hif_ind_scan_cmpl. In add,
hif_ind_scan_cmpl come from hardware API and is not expected to be used
with upper layers of the driver.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-40-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wfx_update_filtering() has no reason to instantiate a struct
hif_mib_bcn_filter_enable. Drop it and simplify the code.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-39-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
HIF functions return only serious errors (OOM or device freeze). The
current handling of errors in wfx_set_mcast_filter() does not bring
anything. Finally it may disturb the developer more than it helps.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-38-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The structure hif_mib_config_data_filter come from hardware
API. It is not intended to be manipulated in upper layers of the driver.
In add, current code for hif_set_config_data_filter() is too dumb. It
should pack data with hardware representation instead of leaving all
work to the caller.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-37-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The structure hif_mib_mac_addr_data_frame_condition come from hardware
API. It is not intended to be manipulated in upper layers of the driver.
In add, current code for hif_set_mac_addr_condition() is too dumb. It
should pack data with hardware representation instead of leaving all
work to the caller.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-36-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The structure hif_mib_set_data_filtering come from hardware API. It is
not intended to be manipulated in upper layers of the driver.
In add, current code for hif_set_data_filtering() is too dumb. It should
pack data with hardware representation instead of leaving all work to
the caller.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-35-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The field "default_filter" was not obvious.
In add, explicitly declare that fields default_filter and enable are
booleans.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-34-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The structure hif_mib_uc_mc_bc_data_frame_condition come from hardware
API. It is not intended to be manipulated in upper layers of the driver.
In add, current code for hif_set_uc_mc_bc_condition() is too dumb. It
should pack data with hardware representation instead of leaving all
work to the caller.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-32-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The structure hif_mib_set_association_mode come from hardware API. It is
not intended to be manipulated in upper layers of the driver.
In add, current code for hif_set_association_mode() is too dumb. It
should pack data with hardware representation instead of leaving all
work to the caller.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-31-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The structure hif_req_join come from hardware API. It is not intended
to be manipulated in upper layers of the driver.
In add, current code for hif_join() is too dumb. It should pack data
with hardware representation instead of leaving all work to the caller.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-30-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
hif_update_ie() is only called to change the beacon template. So,
specialize this function and simplify the way to call it.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-29-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wfx_bss_info_changed() is not called from atomic contexts. So, it is not
necessary to raise an asynchronous work to change ERP. Thus
wfx_set_cts_work() become useless.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-27-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
setbssparams_done was here to ensure that the firmware does not enable
powersave before to get the first beacon. However, mac80211 already
ensures it gets a beacon before to associate to the BSS. And even, if
it won't, the firmware wake up at least on every DTIM, which is
sufficient to finalize the association.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-26-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Current code keeps a copy of bss_conf->cqm_rssi_thold in wfx_vif. There
is no sane reason for that.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-25-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It seems that current code try to save calls to hif_beacon_transmit() by
keeping a copy of the previous value of bss_conf->enable_beacon.
However, hif_beacon_transmit() does not cost so much and mac80211
already take care to not send useless events.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-24-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It is not necessary to keep a copy of dtim_period in wfx_vif. Prefer to
just rely on bss_conf->dtim_period.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-23-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Current code parse the TIM and retrieve the DTIM period. It is far
easier to rely on bss_info_changed() for this job.
It is no more necessary to run task asynchronously. So
set_beacon_wakeup_period_work is now useless.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-22-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Current code of wfx_flush() force to drop packets in some contexts.
However, there is no obvious reasons to do that. It looks like a
workaround for a bug with the old implementation of __wfx_flush().
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-21-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
wfx_tx_queues_clear() only clear not yet sent requests. So, it always
necessary to wait for tx_queue_stats.wait_link_id_empty whatever the
value of "drop" argument.
In add, it is not necessary to return with tx queue locked since all
calls to __wfx_flush() unlock the tx queue just after the call to
__wfx_tx_flush().
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-20-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This function built probe response from data retrieved in beacon. Yet,
this job can be done with ieee80211_proberesp_get(). So, we can simplify
that code (and fix bugs like inclusion of TIM in probe responses).
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-18-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In fact, wfx_upload_beacon() uploads beacon and probe response. So,
rename it in wfx_upload_ap_templates().
The call to wfx_fwd_probe_req() has nothing to do with template
uploading, so relocate it.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200115135338.14374-17-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Jérôme Pouiller