- 16 Jan, 2020 40 commits
-
-
Jérôme Pouiller authored
wfx_ps_notify_sta() is too complex compared to the task it do. Signed-off-by:
Jérôme Pouiller <jerome.pouiller@silabs.com> Link: https://lore.kernel.org/r/20200115135338.14374-59-Jerome.Pouiller@silabs.comSigned-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
Jérôme Pouiller authored
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 authored
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 authored
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 authored
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 authored
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 authored
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 authored
Just for sanity. Signed-off-by:
-
Jérôme Pouiller authored
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 authored
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 authored
Indeed, sta was used after call to rcu_unlock() Signed-off-by:
-
Jérôme Pouiller authored
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 authored
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 authored
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 authored
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 authored
If RTS threshold is 0, it currently disables RTS. It should mean "enabled for every frames". Signed-off-by:
-
Jérôme Pouiller authored
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 authored
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 authored
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 authored
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 authored
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 authored
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 authored
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 authored
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 authored
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 authored
The field "default_filter" was not obvious. In add, explicitly declare that fields default_filter and enable are booleans. Signed-off-by:
-
Jérôme Pouiller authored
The current API defines bitfields. It is not very convenient. Prefer to use bitmasks. Signed-off-by:
-
Jérôme Pouiller authored
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 authored
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 authored
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 authored
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 authored
SSID is hidden in beacon but firmware has to know to which probe requests it has to answer. Signed-off-by:
-
Jérôme Pouiller authored
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 authored
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 authored
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 authored
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 authored
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 authored
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 authored
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 authored
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:
-
