GRO queue more pkts is enabled in bus vote low, when
bus vote back to idle, these frames may not flushed to
stack, use force flush to reduce latency.
Change-Id: I22b5cf994fe02b7e3156a845ee13a5acbf791b22
CRs-Fixed: 2774662
Currently, sta_ctx->hdd_reassoc_scenario is set when ROAM_FT_START
and cleared when ROAM_SET_KEY_COMPLETE. But for some roaming case,
no KEY exchange happen after roaming complete. Then this flag
is not cleared, which cause failure to get station stats in
wlan_hdd_get_sta_stats, such as RSSI info. To address this, clear
sta_ctx->hdd_reassoc_scenario when handling
ROAM_ASSOCIATION_COMPLETION.
Change-Id: I0b96246afe227023fd52df93f27bed23ffd7cb1c
CRs-Fixed: 2763178
The array size of 'channels' in struct channel_list_info
is CFG_VALID_CHANNEL_LIST_LEN, which may smaller than
NUM_CHANNELS, function 'iw_get_channel_list()' may attempt
to access element CFG_VALID_CHANNEL_LIST_LEN..NUM_CHANNELS
of array 'channels'.
To make align, fix array size of 'channels' to NUM_CHANNELS,
and refine 'iw_get_channel_list_with_cc()'.
CRs-Fixed: 2777780
Change-Id: Ie064322f14186310a769db2642e69c8a520f1b26
Enable low_tput GRO skip logic in 5.4 kernel based GRO.
It is required to always flush listified normal Rx frames at end of
context. Skip flush for GRO rx at low TPUT.
So it requires flush to be aware of low TPUT, hence pass the flush code
to DP GRO flush.
Change-Id: I20c1efcc7bbe0aa50f350ebc31eaefbbefd69616
CRs-Fixed: 2745763
To send the vdev delete to firmware, all the references needs to
be released, so on receiving the iff_down from the userspace,driver
is currently draining the peer management frames which are queued
to the firmware, if the firmware is active it can access these
drained frames resulting in system instability.
Drain the management frames only when the driver is recovering.
Change-Id: Ia7a83088cde857c7111f8dff9e18422257fc8cb8
CRs-Fixed: 2778328
In the process of rso update config, need to send roam
scan_mode/deauth/idle/trigger commands when rso enabled.
Another fix is to call csr_cm_roam_scan_offload_fill_rso_configs()
with ROAM_SCAN_OFFLOAD_STOP in wlan_cm_roam_fill_stop_req().
Change-Id: Ibcc2ab5d7da1bf838b386e0e1e486cd43072225a
CRs-Fixed: 2779549
Supplicant disabled flag is used to skip the RSO stop command
sending to FW as RSO stop aborts the roaming in FW. Do not reset
the flag in the reassoc call from user space to complete the
roaming in FW by not sending the RSO stop command.
Change-Id: I1cda88d24942d60ca98b497ddbbbea3a5fc52a56
CRs-Fixed: 2771009
While initial connection with SPMK AP1 supplicant generates
PMK1, and DUT updates global mlme single pmk entry and
csr session pmk.
Now DUT roam to non-SPMK AP2 and generates PMK2, Global mlme
single pmk has SPMK AP1 PMK1 and csr_roam_session has non-SPMK
AP2 since FULL SAE happen.
Now Roam back to SPMK AP1, since the Full SAE does not happen
and due to check roam_synch_data->authStatus ==
CSR_ROAM_AUTH_STATUS_AUTHENTICATED in
csr_process_roam_synch_callback(). The lookup and update
csr_roam_session code is not hit and driver don’t update
csr session pmk after roaming to SPMK AP. So after roam sync complete,
csr_check_and_set_sae_single_pmk_cap() updates the global mlme
single pmk with session->psk_pmk. This results DUT uses PMK generated
from a Non-Single PMK supported AP in the ReAssoc Request frame
while Roaming to Single PMK supported AP.
Fix is to allow driver to update session->psk_pmk even in case
Full SAE does not happen.
Change-Id: Ie4f06cfcb066ae245de024b62da586aade783aec
CRs-Fixed: 2761771
Enable beaconing for 160 MHz BW when all the channels
are non-DFS. Addition of new 5.9 GHz channels 169, 173,
and 177 has created a new 160 MHz BW channel bonding
pair (149-177) and beacons can be sent out from SAP
for this pair.
Change-Id: Ifd222a01a5058cf02d6e1d819da9d213eb0fd9ca
CRs-Fixed: 2778707
Check for the Moselle bus type whether the selfrecovery ini is enabled,
if the ini is disabled let the system enter to bad state and debug
the issue.
Change-Id: Ib6e720bbc8d1421706a5631ecc7d11989538e2d5
CRs-Fixed: 2780882
As part of WOW feature in WCN6750, send HOST wake up
to FW over QMI instead of WMI. To send HOST wake, add
PLD layer and call platform driver API to send QMI
EXIT_POWER_SAVE to FW.
When FW receive EXIT_POWER_SAVE QMI, it will implicitly
consider that it also received HOST_WAKEUP_FROM_SLEEP
and send ACK for it through CE2.
Change-Id: Ia0a3f4c15662f9ee58649f5c6de965f65aeafe32
CRs-Fixed: 2779370
The 11ax cap and oper IE presence debug printing in the following
functions are not bringing any useful info for debugging:
sir_convert_probe_req_frame2_struct
sir_convert_probe_frame2_struct
sir_convert_assoc_resp_frame2_struct
so remove them to reduce driver log verbosity.
Change-Id: I2afc96e39dbf5ae30ca1c6a3d803064ee018a8aa
CRs-Fixed: 2781096
During probe IPA SMMU S1 enable check will pass only if IPA
registration for ready callback is successful and IPA ready
callback is invoked.
This ensures that IPA APIs are not invoked unless IPA ready
callback is invoked and ipa ready flag is set in the host
driver.
Change-Id: Ie157a2f851bb1e914a098b62ede1280aa0243d19
CRs-Fixed: 2778504
Broadcast TWT is disabled by overwriting the user
configuration. Remove the overwrite to enable the
broadcast TWT.
Change-Id: I79742eaaa61236f92526435ec4a59e75829cf0fb
CRs-Fixed: 2780168
In csr_issue_set_context_req_helper() profile is dereferenced
to get the encryption type.
Add null check before accessing the profile pointer.
Change-Id: I40c148648130918f9da0194ff6e132d55b113008
CRs-Fixed: 2777778
The current contract between the host and the firmware is that
0 means 1x1 nss
1 means 2x2 nss
with the new nl80211 interface the driver is sending 1 for 1x1
and 2 for 2x2 nss, so the command is failing for the 2x2 nss.
Send the nss value as per the current expectation so that rate
setting would take place.
Change-Id: I3cb376704e7e3430fbce0015183812484127cbb8
CRs-Fixed: 2772297
Free keys for vdev when disconnect from AP in STA mode,
otherwise, the keys may be used in the next connection, which will
result in connection failure.
CRs-Fixed: 2773040
Change-Id: I6d3e858cdc5b38b25a4431b9aaf5f2fc3e1019c7
wlan_reg_set_channel_params_for_freq API will check the bonded
channels are valid or not. If invalid ch width is returned, that
means no suitable channel bonding in current regdomain for
requested channel frequency, then abort the vdev start request.
Change-Id: Ibca5ef7af72f1fd45855d64f380abaa49c0f54c4
CRs-Fixed: 2777637
Issue: Currently, host calculates he_caps based on assoc resp
he caps. In case assoc resp doesn't have he_caps host will
set peer mode vht and vdev mode he which may cause issue due
to different peer and vdev modes.
Fix: Consider beacon he caps if he caps not present in
assoc resp to keep vdev and he param same.
Change-Id: Ia47b3d9c30b5ebc59a398a5ece2263ea65b03550
CRs-Fixed: 2777916
Host received mixed vlan header frames and non-vlan header frames,
and these frames is belong to same UDP flow. currently wlan host
will assume same UDP flow frames with same L2 header length which
then get incorrect IP/UDP header for non-vlan frames.
Determine FISA continuation by RX frame self L2/L3 header length,
update head skb L2/L3 header length in fisa flow when head skb
changed.
Change-Id: Id61b856773b8dd7f0199055b1e0bf9a2bd98e401
CRs-Fixed: 2778403
In some cases, max_2_4_g_power.len/max_5_g_power.len value
could be 0 and driver tries to allocate 0 bytes through
qdf_mem_alloc. qdf_mem_alloc has a check for 0 size which
logs the failure and returns an error code. But this error
log could cause some delay in the roaming process
unnecessarily as this is not an error case from roaming
perspective. It's better to avoid calling qdf_mem_alloc if
the length is 0.
Add validation for cfg_length in csr_get_cfg_max_tx_power,
if cfg_length is 0 return default maxTxPwr.
Change-Id: Ifd5d90186605e141ed2c107b4170a1d2c82bee0e
CRs-Fixed: 2768190
