- 16 Jan, 2020 40 commits
-
-
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 <jerome.pouiller@silabs.com> Link: https://lore.kernel.org/r/20200115135338.14374-46-Jerome.Pouiller@silabs.comSigned-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
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:
-
Jérôme Pouiller authored
Remove most of indentation of wfx_update_beaconing() by reworking the error handling. Signed-off-by:
-
Jérôme Pouiller authored
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 authored
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 authored
Now that wfx_config() only handles IEEE80211_CONF_CHANGE_PS, it can be simplified. Signed-off-by:
-
Jérôme Pouiller authored
mac80211 and the device are both able to control tx power per vif. But, the current code retrieve tx power from wfx_config(). So, it does not allow to setup the tx power independently for each vif. Driver just has to rely on bss_conf->txpower to get the correct information. In add, it is no more necessary to protect access to wdev->output_power with scan_lock. Signed-off-by:
-
Jérôme Pouiller authored
This struct is no more used. Signed-off-by:
-
Jérôme Pouiller authored
wvif->ht_info contains useless copies of sta->ht_cap and bss_conf->ht_operation_mode. Prefer to retrieve information from the original structs instead of rely on wvif->ht_info. Signed-off-by:
-
Jérôme Pouiller authored
wvif->ht_info.ht_cap is a useless copy of sta->ht_cap. It makes no sense to rely on it. Signed-off-by:
-
Jérôme Pouiller authored
Most of the commands that are sent to device should take struct in argument. In the current code, when this struct is binary compatible with a __le32, the driver use a __le32. This behavior is error prone. This patch fixes that and uses the specialized structs instead. Signed-off-by:
-
Jérôme Pouiller authored
The structure hif_req_start come from hardware API. It is not intended to be manipulated in upper layers of the driver. In add, current code for hif_start() 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_arp_ip_addr_table come from hardware API. It is not intended to be manipulated in upper layers of the driver. In add, current code for hif_set_arp_ipv4_filter() is too dumb. It should pack data using the hardware representation instead of leaving all work to the caller. Signed-off-by:
-
Jérôme Pouiller authored
The structure hif_mib_rcpi_rssi_threshold come from hardware API. It is not intended to be manipulated in upper layers of the driver. In add, current code for hif_set_rcpi_rssi_threshold() is dumb. It should pack data using the hardware representation instead of leaving all work to the caller. Signed-off-by:
-
Jérôme Pouiller authored
Hardware API use 10th of dBm for output power unit. Upper layers should use same units than mac80211 and the conversion should be done by low level layer of the driver (hif_set_output_power()) In add, current code of hif_set_output_power() use a __le32 while the device API specify a specific structure for this. Signed-off-by:
-
