openwrt/staging/blogic.git
9 years agoath10k: implement more versatile set_bitrate_mask
Michal Kazior [Tue, 21 Apr 2015 17:42:58 +0000 (20:42 +0300)]
ath10k: implement more versatile set_bitrate_mask

Until now only a single fixed tx rate or nss was
allowed to be set.

The patch attempts to improve this by allowing
most bitrate masks. The limitation is VHT MCS
rates cannot be expressed separately using
existing firmware interfaces and only the
following VHT MCS ranges are supported: none, 0-7,
0-8, and 0-9.

This keeps the old behaviour when requesting
single tx rate or single nss. The new bitrate mask
logic is only applied to other cases that would
return -EINVAL until now.

Depending on firmware revisions some combinations
may crash firmware so use with care, please.

This depends on "ath10k: don't use reassoc flag".
Without it key cache would effectively be
invalidated upon bitrate change leading to
communication being no longer possible.

These work:

  iw wlan0 set bitrates legacy-5 6 12 ht-mcs-5 1 2 3
  iw wlan0 set bitrates legacy-5 ht-mcs-5 7 8 9
  iw wlan0 set bitrates legacy-5 24 ht-mcs-5 vht-mcs-5 1:0-9

These won't work:

  iw wlan0 set bitrates legacy-5 ht-mcs-5 vht-mcs-5 1:0-5
  iw wlan0 set bitrates vht-mcs-5 2:7-9

(note the invalid VHT MCS ranges)

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix qca61x4 hw2.1 support
Michal Kazior [Mon, 20 Apr 2015 09:20:41 +0000 (09:20 +0000)]
ath10k: fix qca61x4 hw2.1 support

During initialization firmware does some sort of
memory switch between DRAM and IRAM. If
configuration value for bank switching isn't
correct device crashes during init.

The new value prevents firmware 11.0.0.302 (and
possibly others) for qca61x4 hw2.1 from crashing
during init.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: allow loading device specific board files
Michal Kazior [Fri, 17 Apr 2015 09:19:17 +0000 (09:19 +0000)]
ath10k: allow loading device specific board files

Some devices differ slightly and require different
board files. If wrong board data is used they
crash or behave incorrectly.

These devices can be differentiated by looking at
PCI subsystem device id. That is the case for
qca61x4 devices at least.

The board specific filename is constructed as:

 board-<bus>-<id>.bin

For PCI in particular it is:

 board-pci-<vendor>:<dev>:<subsys_vendor>:<subsys_dev>.bin

These files are looked in device/hw specific
directories. Hence for Killer 1525 (qca6174 hw2.1)
ath10k will request:

  /lib/firmware/ath10k/QCA6174/hw2.1/board-pci-168c:003e:1a56:1525.bin

To not break any existing setups (e.g. in case
some devices in the wild already have subsys ids)
if a board specific file isn't found a generic one
is used which is the one which would be used until
now. This guarantees that after upgrading a driver
device will not suddenly stop working due to
now-missing specific board file. If this is the
case a "fallback" string is appended to the info
string when driver boots.

Keep in mind this is distinct from cal-pci-*.bin
files which contain full calibration data and MAC
address. Cal data is aimed at systems where
calibration data is stored out of band, e.g. on
nand flash instead of device EEPROM - an approach
taken by some AP/router vendors.

Board files are more of a template and needs some
bits to be filled in by the OTP program using
device EEPROM contents.

One could argue to map subsystem ids to some board
design codename strings instead of using raw ids
when building the board filename. Using a mapping
however would make it a lot more cumbersome and
time consuming (due to how patches propagate over
various kernel trees) to add support for some new
device board designs. Adding a board file is a lot
quicker and doesn't require recompilation.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: share board file loading code across FW APIs
Michal Kazior [Fri, 17 Apr 2015 09:19:16 +0000 (09:19 +0000)]
ath10k: share board file loading code across FW APIs

There's no need to implement the same thing twice.
Reduce code duplication.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: enable the BT-coex feature
Yanbo Li [Wed, 15 Apr 2015 12:28:08 +0000 (15:28 +0300)]
ath10k: enable the BT-coex feature

The new 10.2.4 FW support the BT-coex feature with external BT module,
The external BT modules can communicate with it via the GPIO. This patch
check the BT-coex capability of the FW and enable it by default.

Signed-off-by: Yanbo Li <yanbol@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add the WMI Air Time Fairness service define
Yanbo Li [Wed, 15 Apr 2015 12:28:07 +0000 (15:28 +0300)]
ath10k: add the WMI Air Time Fairness service define

Add the FW ATF (Air Time Fairness) service define to keep consistent
with 10.2 FW capability.

Signed-off-by: Yanbo Li <yanbol@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: use MAC clock rate same as FW uses for survey accuracy
Peter Oh [Mon, 13 Apr 2015 18:01:20 +0000 (11:01 -0700)]
ath10k: use MAC clock rate same as FW uses for survey accuracy

WLAN survey data provides several informative values on each channels
such as noise, active time, transmit time, and etc.
Regarding the channel activity time data, it is reported from firmware
in different form which is number of cycles instead of time itself.
Hence host driver converts it to time unit by dividing it by clock rate
that is used at baseband MAC.

Using difference clock rate from that actual HW is using will result in
inaccurate survey data. For instance, channel active time can be reported
with bigger than 150ms even though we set Dwell time to 150ms.
Therefore set the clock rate to 88000 (88 MHz) which is ath10k baseband
MAC is using.

Signed-off-by: Peter Oh <poh@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: enable ibss-rsn
Janusz Dziedzic [Fri, 10 Apr 2015 13:23:23 +0000 (13:23 +0000)]
ath10k: enable ibss-rsn

With latest additions to the driver it seems
viable to enable support for IBSS-RSN.

It seems to work on QCA988X and 999.999.0.636 but
is a bit slow to exchange RSN keys for some
reason. This may be a firmware quirk or ath10k is
missing something. Nevertheless it makes sense to
finally enable IBSS-RSN in ath10k even if somewhat
handicapped.

QCA6174 firmware doesn't seem to be able to Tx
EAPOL frames at all now (they get stuck in hw
queues for some reason) so it never gets to set
the keys in driver. It's fairly safe to assume that
once this is fixed IBSS-RSN will work with QCA6174
firmware without any additional changes. Hence no
special handling for advertising
IEEE80211_HW_SUPPORTS_PER_STA_GTK and
WIPHY_FLAG_IBSS_RSN is done now.

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: set def key idx for ibss
Michal Kazior [Fri, 10 Apr 2015 13:23:22 +0000 (13:23 +0000)]
ath10k: set def key idx for ibss

Some time ago there was a weird issue with AP
using wrong multicast keys and generating
corrupted traffic on 10.1 firmware.

Apparently a very similar problem applies for
IBSS-RSN on 999.999.0.636.

ath10k doesn't have IBSS-RSN enabled yet. This
patch is a prerequisite to support it.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix multiple key static wep with ibss
Michal Kazior [Fri, 10 Apr 2015 13:23:21 +0000 (13:23 +0000)]
ath10k: fix multiple key static wep with ibss

Apparently firmware requires both pairwise and
groupwise keys to be installed per-peer for static
WEP in IBSS. This wasn't necessary for AP mode
(and installing both doesn't seem to break AP
mode thus there's no special handling).

Also there seems to be some kind of issue with
mapping tx/rx keys in firmware properly which
resulted in wrong keys being used and broken
communication between devices.

It can be argued the vdev param part is more of a
workaround than a real fix. However I couldn't
figure out how to fix this differently. It works
and isn't super ugly.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: don't use reassoc flag
Michal Kazior [Fri, 10 Apr 2015 13:23:20 +0000 (13:23 +0000)]
ath10k: don't use reassoc flag

Firmware actually re-creates peer entry when
reassoc flag is set. This is undesired and could
cause trouble with IBSS crypto-wise. This is also
important for upcomming bitrate mask improvement.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: change static wep key handling for client mode
Michal Kazior [Fri, 10 Apr 2015 13:05:58 +0000 (13:05 +0000)]
ath10k: change static wep key handling for client mode

mac80211 sets static WEP keys as groupwise while
firmware requires them to be installed twice as
both pairwise and groupwise.

Until now these keys were installed once as
pairwise only and, due to that special handling,
needed additional tricks to support 802.1x as
well.

Without this patch in some cases (when AP and
ath10k client use different default tx keys)
multicast communication was broken.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add extra check for frame tracing
Michal Kazior [Fri, 10 Apr 2015 13:02:28 +0000 (13:02 +0000)]
ath10k: add extra check for frame tracing

Frames are logged via tracing in two slices:
header and payload, separately. This is done for
performance reasons when one wants to, e.g.
analyse metadata only of frames only.

If for some reason device delivered a frame buffer
which was sized below what 802.11 header implied
tracing logic would blow doing an invalid memory
accesses.

I've hit this problem when running IBSS on QCA988X
with 999.999.0.636 and tracing at the same time.

Fixes: 5ce8e7fdcc7a ("ath10k: handle ieee80211 header and payload tracing separately")
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix error handling in pci_probe
Michal Kazior [Fri, 10 Apr 2015 13:01:27 +0000 (13:01 +0000)]
ath10k: fix error handling in pci_probe

If chip_id wasn't recognized clean up code wasn't
executed properly. It would skip freeing memory
causing a leak and irqs causing possibly MSI
warning splats later or even kernel crashes.

Fixes: 1a7fecb766c8 ("ath10k: reset chip before reading chip_id in probe")
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath: lower JP W53 band DFS detection threshold around 30%
Peter Oh [Tue, 31 Mar 2015 23:44:43 +0000 (16:44 -0700)]
ath: lower JP W53 band DFS detection threshold around 30%

Japan's W53 band requires 50% data traffic during its DFS test,
but WLAN baseband used by ath9k and ath10k is not able to achieve
current threshold rate, 50%, under the data traffic rate.
In other words, HW occasionally fails detecting radar pulses,
so that SW cannot get enough radar reports to achieve the rate.

Signed-off-by: Peter Oh <poh@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath: define JP DFS patterns separated from FCC
Peter Oh [Tue, 31 Mar 2015 23:44:42 +0000 (16:44 -0700)]
ath: define JP DFS patterns separated from FCC

Separate Japan's DFS pattern from FCC to control PPB threshold.

Currently all the radar detectors use the same threshold rate at
50%, but it's not able to achieve if data traffic rate is higher
than 40% because WLAN baseband used by ath9k and ath10k often fails
detecting radar pulses, so that SW cannot get enough radar reports
to achieve the rate.

Since Japan's W53 band requires 50% data traffic during its DFS
test we need to apply different threshold rate than others on it.
Hence define its own pattern to give flexibility to threshold rate.

Signed-off-by: Peter Oh <poh@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: use scan burst for hw roc
Michal Kazior [Tue, 31 Mar 2015 11:03:48 +0000 (11:03 +0000)]
ath10k: use scan burst for hw roc

This improves chances of getting onto a foreign
channel and thus makes P2P a bit more reliable.

Without scan burst it was possible for firmware to
not switch to foreign channel resulting in "failed
to switch to channel for roc scan" warning. This
would also effectively fail some offchan tx
requests and lead to P2P find/connect taking
longer. This could be observed when other vifs
were running/busy, e.g. with P2P GO.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix hw roc expiration
Michal Kazior [Tue, 31 Mar 2015 11:03:47 +0000 (11:03 +0000)]
ath10k: fix hw roc expiration

It is not guaranteed firmware will switch to
foreign channel immediately after starting scan
sequence. To account for that don't use duration
parameter for scan time. Instead request insanely
long scan and use timeout worker to cancel it from
driver.

This should improve P2P reliability a bit.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath: support new FCC DFS Radar Type 1
Peter Oh [Thu, 9 Apr 2015 15:50:10 +0000 (18:50 +0300)]
ath: support new FCC DFS Radar Type 1

Add support for new FCC DFS rules released on August 14, 2014.
FCC has added a new radar type named Radar Type 1 and original
Radar Type 1 is renamed to Radar Type 0 in consequence.

During the certificate test, Type 1 PRI values are randomly selected
within the range of 518 and 3066 and we divide it to 3 groups based on
practical test result data collected for more than a year.

For about Radar type ID, it does nothing to functionalities.
In other words, even if we re-order the IDs, DFS detection will
work as well, but we give the ID with matching to FCC doc.

By adding this support, the drivers using this DFS function are
able to support both of old and new FCC DFS rules simultaneously
without any other changes.

Signed-off-by: Peter Oh <poh@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix no-ack frame status
Sujith Manoharan [Wed, 1 Apr 2015 19:53:21 +0000 (22:53 +0300)]
ath10k: fix no-ack frame status

Use the new IEEE80211_TX_STAT_NOACK_TRANSMITTED flag
to indicate successful transmission of no-ack frames.
This fixes multicast frame accounting.

Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: change the fw stats access mask
Yanbo Li [Wed, 1 Apr 2015 19:53:21 +0000 (22:53 +0300)]
ath10k: change the fw stats access mask

This fix the regerssion introduced after "ath10k_wmi_requests_stats()
call to use more stat bits", both the 10.2.X FW and community FW 10.1.X
will return time out when access the fw_stats with the STAT_VDEV and
STAT_PDEV. Add the condition to use different fw_stats mask for different
FW versions.

Fixes: 7777d8c7ef6f ("ath10k: implement fw stats for wmi-tlv")
Signed-off-by: Yanbo Li <yanbol@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix static wep with more than 1 key in client mode
Michal Kazior [Wed, 1 Apr 2015 19:53:20 +0000 (22:53 +0300)]
ath10k: fix static wep with more than 1 key in client mode

The default keyidx callback may be called after
more than 1 key is installed. This led to only 1
WEP key being reinstalled only. This caused Rxed
traffic encrypted with other WEP keys to be
dropped in client mode.

Tested-by: Bartosz Markowski <bartosz.markowski@tieto.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix arvif->wep_keys clearing
Michal Kazior [Wed, 1 Apr 2015 19:53:19 +0000 (22:53 +0300)]
ath10k: fix arvif->wep_keys clearing

The variable should be cleared regardless of
whether there's a peer associated with the key or
not.

This fixes case when user first associates with 2
WEP keys and then disconnects and connects with 1
WEP key. This resulted in WEP key count being 2 in
the driver leading to default keyidx fixup
failure.

Tested-by: Bartosz Markowski <bartosz.markowski@tieto.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: replenish htc tx credits always
Michal Kazior [Wed, 1 Apr 2015 19:53:19 +0000 (22:53 +0300)]
ath10k: replenish htc tx credits always

There's always at most 2 credits and it makes
little sense to set the
ATH10K_HTC_FLAG_NEED_CREDIT_UPDATE flag
conditionally.

This seems to fix some random issues with tx
credit starvation on WLAN.RM.2.0-00073 I've been
seeing. Note: this isn't related to wmi mgmt tx.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix insufficient tracing buffer size
Michal Kazior [Wed, 1 Apr 2015 19:53:19 +0000 (22:53 +0300)]
ath10k: fix insufficient tracing buffer size

Some trace messages were truncated and a kernel
splat could be seen in the log:

  WARNING: CPU: 3 PID: 0 at /devel/src/linux/drivers/net/wireless/ath/ath10k/./trace.h:114 ftrace_raw_event_ath10k_log_dbg+0x20e/0x220 [ath10k_core]()
  Modules linked in: ath10k_pci(O) ath10k_core(O) ath iwldvm iwlwifi [last unloaded: iwlwifi]
  CPU: 3 PID: 0 Comm: swapper/3 Tainted: G        W  O    4.0.0-rc3-wl-ath+ #703
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
   ffffffffa01d4cb0 ffff88001fd83998 ffffffff8191b86c ffffffff81e3b718
   0000000000000000 ffff88001fd839d8 ffffffff8105573a ffff88001c0a5528
   ffff88001bea9ae0 ffff88001c3dd940 00000000000d0018 ffff88001fd83a80
  Call Trace:
   <IRQ>  [<ffffffff8191b86c>] dump_stack+0x45/0x57
   [<ffffffff8105573a>] warn_slowpath_common+0x8a/0xc0
   [<ffffffff8105582a>] warn_slowpath_null+0x1a/0x20
   [<ffffffffa01c61fe>] ftrace_raw_event_ath10k_log_dbg+0x20e/0x220 [ath10k_core]
   [<ffffffffa01aaa6b>] ath10k_dbg+0xbb/0xd0 [ath10k_core]
   [<ffffffff810eed89>] ? trace_clock_local+0x9/0x10
   [<ffffffffa01bd069>] ath10k_wmi_event_service_ready+0x479/0x520 [ath10k_core]
   [<ffffffff810fdf90>] ? trace_buffer_unlock_commit+0x50/0x60
   [<ffffffffa01c2113>] ath10k_wmi_tlv_op_rx+0x6b3/0x8b0 [ath10k_core]

This could be reproduced with:

  trace-cmd record -e ath10k
  ifconfig wlan0 down
  ifconfig wlan0 up

Fixes: 5c01aa3de918 ("ath10k: deduplicate wmi service ready logic")
Fixes: ca996ec56608 ("ath10k: implement wmi-tlv backend")
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: disable burst mode
Janusz Dziedzic [Wed, 1 Apr 2015 19:53:18 +0000 (22:53 +0300)]
ath10k: disable burst mode

Firmwares that indicate BURST_SERVICE as enabled
could overwrite some wmm parameters (eg. txop).
This could lead IOT and balancing issues.
Hence disable this by default.

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix IBSS scanning on wmi-tlv and qca6174
Michal Kazior [Tue, 24 Mar 2015 13:14:03 +0000 (13:14 +0000)]
ath10k: fix IBSS scanning on wmi-tlv and qca6174

WLAN.RM.2.0-00073 firmware requires self-peer to
be created prior to issuing scan command. Without
this wmi-tlv with qca6174 firmware crashes after
submitting a scan request.

Creating the peer as soon as add_interface()
shouldn't be a problem.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: clean up set_bitrate_mask handling
Michal Kazior [Tue, 31 Mar 2015 10:49:20 +0000 (10:49 +0000)]
ath10k: clean up set_bitrate_mask handling

The code was a bit convoluted. Clean it up and
prepare for future changes.

While at it this fixes incorrect verification of
'single nss' case when ss2 rates were missing
while ss1 and ss3 were requested resulting in
nss=3 being set:

  iw wlan1 set bitrates legacy-5 ht-mcs-5 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23 vht-mcs-5 1:0-9 3:0-9

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: allocate fw resources for iface combinations
Michal Kazior [Tue, 31 Mar 2015 10:26:26 +0000 (10:26 +0000)]
ath10k: allocate fw resources for iface combinations

The number of 3 vdevs wasn't enough to handle the
worst case for interface combinations in practice.

wpa_supplicant may need up to 4 vifs to have 2
vifs actually connected, i.e. p2pdev + client + 2x
p2p (either p2p client or p2p go).

This fixes worst case warning:

  Free vdev map is empty, no more interfaces allowed.

This keeps the ability to associate 32 stations in
AP mode at the cost of not being able to guarantee
that under all circumstances, i.e. some
combinations may consume additional fw peer
entries for internal purposes leaving less
resource for stations in AP mode.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: enable multi-channel on supported devices
Michal Kazior [Tue, 31 Mar 2015 10:26:25 +0000 (10:26 +0000)]
ath10k: enable multi-channel on supported devices

This effectively enables multi-channel operation
on qca6174 WLAN.RM.2.0-00073 (and possibly any
newer firmware release for qca6174).

This adds appopriate interface combinations and
initializes firmware channel scheduler.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: implement tx pause wmi event
Michal Kazior [Tue, 31 Mar 2015 10:26:24 +0000 (10:26 +0000)]
ath10k: implement tx pause wmi event

qca6174 wmi-tlv firmware defines a new wmi event
for host tx pausing (i.e. stop/wake tx queues).

Map these events to ath10k/mac80211 tx queue
control.

This is important for multi-channel throughput
performance.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: rework tx queue locking
Michal Kazior [Tue, 31 Mar 2015 10:26:23 +0000 (10:26 +0000)]
ath10k: rework tx queue locking

Tx queue locking was very simple until now.
Multi-channel support will require a more flexible
and fine grained control.

This introduces a per-hw and per-vif (each with a
bitmask of reasons) tx queue locking.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: implement adaptive qcs command
Michal Kazior [Tue, 31 Mar 2015 10:26:22 +0000 (10:26 +0000)]
ath10k: implement adaptive qcs command

This command will be used to configure
multi-channel scheduler in firmware.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: implement chanctx API
Michal Kazior [Tue, 31 Mar 2015 10:26:21 +0000 (10:26 +0000)]
ath10k: implement chanctx API

The chanctx API will allow ath10k to support
multi-channel operation.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: allow empty ssid vdev config
Michal Kazior [Tue, 31 Mar 2015 10:26:20 +0000 (10:26 +0000)]
ath10k: allow empty ssid vdev config

It doesn't make much sense to reject a valid
firmware configuration combination.

Since SSID isn't known early on it might make
sense to allow driver to start vdev without SSID
and restart it later.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: don't call quiet mode if it's not implemented
Michal Kazior [Tue, 24 Mar 2015 12:40:17 +0000 (12:40 +0000)]
ath10k: don't call quiet mode if it's not implemented

qca6174 and wmi-tlv doesn't have quiet mode
implemented. Don't even attempt to call it. This
fixes a warning when bringing first interface up:

  failed to set quiet mode period 100 duarion 0 enabled 0 ret -95

Fixes: 8515b5c79a54 ("ath10k: configure thermal throttle while powering up")
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix survey reporting
Michal Kazior [Tue, 24 Mar 2015 12:38:11 +0000 (12:38 +0000)]
ath10k: fix survey reporting

Number of channels is stored in a separate macro
in a header file and channel list is constructed
independently. The macro is used to define survey
array.

This fixes a recent regression introduced after
adding support for 144 channel. The regression
would lead to a warning and incomplete survey data
on channel 165:

  chan info: invalid frequency 5825 (idx 38 out of bounds)

Also make sure to enforce the sizes and avoid this
kind of problem in the future.

Fixes: 4a7898fed5f3 ("ath10k: enable channel 144 on 5GHz band")
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: harmonize error case handling in ath10k_core_start
Nicholas Mc Guire [Mon, 30 Mar 2015 12:39:21 +0000 (15:39 +0300)]
ath10k: harmonize error case handling in ath10k_core_start

All of the bringup/init functions called in ath10k_core_start return 0
on success and != 0 on failure. ath10k_wmi_wait_for_service_ready(),
ath10k_wmi_wait_for_unified_ready() and their call sites were adjusted
to fit this model.
The return type of wait_for_completion_timeout is unsigned long not int so
ath10k_wmi_wait_for_service_ready() and ath10k_wmi_wait_for_unified_ready()
were fixed up accordingly.

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: core: match wait_for_completion_timeout return type
Nicholas Mc Guire [Mon, 30 Mar 2015 12:39:21 +0000 (15:39 +0300)]
ath10k: core: match wait_for_completion_timeout return type

Return type of wait_for_completion_timeout is unsigned long not int.
An appropriately named unsigned long is added and the assignments fixed up.

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: thermal: match wait_for_completion_timeout return type
Nicholas Mc Guire [Mon, 30 Mar 2015 12:39:20 +0000 (15:39 +0300)]
ath10k: thermal: match wait_for_completion_timeout return type

Return type of wait_for_completion_timeout is unsigned long not int.
An appropriately named unsigned long is added and the assignments fixed up.

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: debug: match wait_for_completion_timeout return type
Nicholas Mc Guire [Mon, 30 Mar 2015 12:39:20 +0000 (15:39 +0300)]
ath10k: debug: match wait_for_completion_timeout return type

Return type of wait_for_completion_timeout is unsigned long not int.
An appropriately named unsigned long is added and the assignments fixed up.
Missing spaces 1*HZ -> 1 * HZ were also added along the way.

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: htc: match wait_for_completion_timeout return type
Nicholas Mc Guire [Mon, 30 Mar 2015 12:39:20 +0000 (15:39 +0300)]
ath10k: htc: match wait_for_completion_timeout return type

Return type of wait_for_completion_timeout is unsigned long not int.
An appropriately named unsigned long is added and the assignments fixed up.

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: mac: match wait_for_completion_timeout return type
Nicholas Mc Guire [Mon, 30 Mar 2015 12:39:19 +0000 (15:39 +0300)]
ath10k: mac: match wait_for_completion_timeout return type

Return type of wait_for_completion_timeout is unsigned long not int.
An appropriately named unsigned long is added, respectively 'ret'
renamed, and the assignments fixed up.

Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix aid setup in station mode
Michal Kazior [Mon, 30 Mar 2015 06:51:57 +0000 (09:51 +0300)]
ath10k: fix aid setup in station mode

While debugging something else I noticed AID was
set to 0. This could lead to powersave issues in
station mode. Maybe this isn't really necessary
but set it properly just to be sure.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: document ofdm/5ghz rate offset with a macro
Michal Kazior [Mon, 30 Mar 2015 06:51:57 +0000 (09:51 +0300)]
ath10k: document ofdm/5ghz rate offset with a macro

Don't use literal values for offsets. While at it
rename a function so it is more clear what it
checks for.

This finally takes care of the last magic
5GHz/OFDM offset.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: deduplicate bitrate to rate idx conversion
Michal Kazior [Mon, 30 Mar 2015 06:51:56 +0000 (09:51 +0300)]
ath10k: deduplicate bitrate to rate idx conversion

It's possible to derive rate index from bitrate
without any additional mapping structures/logic.

This should have little to none impact on
performance since this is only done for management
frames and the previous approach wasn't
particularly optimized.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: rework legacy rx rate decoding
Michal Kazior [Mon, 30 Mar 2015 06:51:56 +0000 (09:51 +0300)]
ath10k: rework legacy rx rate decoding

Instead of using a hacky table and magic values
use supported band information advertised to
mac80211.

This may impact performance a little when dealing
with legacy rx rates depending on system
architecture. It's probably negligible.

This also fixes a highly theoretical corner case
when HT/VHT rates weren't reported correctly if
channel frequency wasn't known.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: simplify fixed rate selection
Michal Kazior [Mon, 30 Mar 2015 06:51:56 +0000 (09:51 +0300)]
ath10k: simplify fixed rate selection

Use the existing ieee80211_rate array instead of
definining separate one. This gets rid of the ugly
4-index offset when dealing with 5GHz band.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: use hw rate definitions for fixed rate
Michal Kazior [Mon, 30 Mar 2015 06:51:55 +0000 (09:51 +0300)]
ath10k: use hw rate definitions for fixed rate

Using raw values is discouraged. Having
defines/enums is preferred.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add hw rate definitions
Michal Kazior [Mon, 30 Mar 2015 06:51:55 +0000 (09:51 +0300)]
ath10k: add hw rate definitions

Prepare defines for future use.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: drop RATETAB_ENT macro
Michal Kazior [Mon, 30 Mar 2015 06:51:54 +0000 (09:51 +0300)]
ath10k: drop RATETAB_ENT macro

It was superfluous and confusing. It's better to
define the structure explicitly.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: derive rate from bitrate
Michal Kazior [Mon, 30 Mar 2015 06:51:54 +0000 (09:51 +0300)]
ath10k: derive rate from bitrate

There's no need to store rate values in hw_value.
This frees up the hw_value for better future use.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: move rate definitions to file start
Michal Kazior [Mon, 30 Mar 2015 06:51:54 +0000 (09:51 +0300)]
ath10k: move rate definitions to file start

Prepare the code for future changes so that new
code can refer to rate-related stuff without
forward declarations.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: introduce basic tdls functionality
Marek Puzyniak [Mon, 30 Mar 2015 06:51:53 +0000 (09:51 +0300)]
ath10k: introduce basic tdls functionality

This patch introduces tdls without tdls peer uapsd and tdls channel switching.
Transmitting tdls data frames works only for ethernet type frames, that's why
data addressed to tdls sta is in ethernet format.

This patch depends on commit c23e31cf7b55 ("mac80211: initialize rate control
earlier for tdls station").

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Marek Kwaczynski <marek.kwaczynski@tieto.com>
Signed-off-by: Marek Puzyniak <marek.puzyniak@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add wmi support for tdls
Marek Puzyniak [Mon, 30 Mar 2015 06:51:53 +0000 (09:51 +0300)]
ath10k: add wmi support for tdls

As a part of tdls implementation introduce
tdls related wmi data structures, constant
values and functions.

Signed-off-by: Marek Puzyniak <marek.puzyniak@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: update station counting
Marek Puzyniak [Mon, 30 Mar 2015 06:51:52 +0000 (09:51 +0300)]
ath10k: update station counting

Currently station counting functions
(inc_num_stations/dec_num_stations) does not handle
tdls type of stations. Tdls station should be counted
because it consumes peer in firmware. Only not tdls
stations are excluded from this counting.

Signed-off-by: Marek Puzyniak <marek.puzyniak@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: store max tdls vdevs that fw can handle
Marek Puzyniak [Mon, 30 Mar 2015 06:51:52 +0000 (09:51 +0300)]
ath10k: store max tdls vdevs that fw can handle

Currently number of tdls vdevs supported by firmware
is hardcoded. For future usage it is stored in ath10k
structure based on defined value.

Signed-off-by: Marek Puzyniak <marek.puzyniak@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: make peer type configurable
Marek Puzyniak [Mon, 30 Mar 2015 06:51:52 +0000 (09:51 +0300)]
ath10k: make peer type configurable

Peer type was hardcoded to default value.
For future implementation it is required
to make is configurable.

Signed-off-by: Marek Puzyniak <marek.puzyniak@tieto.com>
Signed-off-by: Marek Kwaczynski <marek.kwaczynski@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: unify tx mode and dispatch
Michal Kazior [Mon, 30 Mar 2015 06:51:51 +0000 (09:51 +0300)]
ath10k: unify tx mode and dispatch

There are a few different tx paths depending on
firmware and frame itself.

Creating a uniform decision will make it possible
to switch between different txmode easier, both
for testing and for future features as well.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Marek Puzyniak <marek.puzyniak@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: clear htt.freq
Michal Kazior [Mon, 30 Mar 2015 06:51:51 +0000 (09:51 +0300)]
ath10k: clear htt.freq

If htt.freq isn't cleared and contains garbage fw
may discard tx packets. Prevent this from
happening by clearing htt.freq properly.

Possible manifestation of the problem could be not
being able to send auth request/response frames on
firmware with HTT >= 3.4 (when freq param was
introduced), e.g. on qca6174.

Fixes: 8d6d36243610 ("ath10k: fix offchan reliability")
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix HTT op backwards compatibility for QCA6174
Kalle Valo [Mon, 30 Mar 2015 11:14:28 +0000 (14:14 +0300)]
ath10k: fix HTT op backwards compatibility for QCA6174

Commit 8348db298496 ("ath10k: add ATH10K_FW_IE_HTT_OP_VERSION") broke QCA6174
receive side as the HTT ids from ATH10K_FW_HTT_OP_VERSION_MAIN don't have
HTT_TLV_T2H_MSG_TYPE_RX_IN_ORD_PADDR_IND set. Fix this by adding a backwards
compatibility code which sets HTT op version if it's not set by the firmware
image.

Fixes: 8348db298496 ("ath10k: add ATH10K_FW_IE_HTT_OP_VERSION")
Reported-by: Michal Kazior <michal.kazior@tieto.com>
Tested-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoMerge mac80211-next into ath-next
Kalle Valo [Mon, 30 Mar 2015 07:07:12 +0000 (10:07 +0300)]
Merge mac80211-next into ath-next

Patch "ath10k: introduce basic tdls functionality" depends on this mac80211
patch:

c23e31cf7b55 mac80211: initialize rate control earlier for tdls station

9 years agoath10k: bump up FW API to 5
Kalle Valo [Wed, 25 Mar 2015 11:12:42 +0000 (13:12 +0200)]
ath10k: bump up FW API to 5

Firmware 10.2.4.48-3 now supports management frames over HTT feature and has
ATH10K_FW_FEATURE_HAS_WMI_MGMT_TX. But as 10.2.4 branch has conflicting HTT ids
patch "ath10k: add ATH10K_FW_IE_HTT_OP_VERSION" is needed to fix the issue.
Older ath10k versions don't have support that support and to maintain backwards
compatibility we need bump up the FW API to 5 not break older versions.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add ATH10K_FW_IE_HTT_OP_VERSION
Rajkumar Manoharan [Wed, 25 Mar 2015 11:12:27 +0000 (13:12 +0200)]
ath10k: add ATH10K_FW_IE_HTT_OP_VERSION

Target to host HTT messages are conflicting between 10.x and other
firmware revisions. By maintaining separate HTT T2H tables for each
firmware revisions (main, 10x and tlv) similar to WMI abstraction,
solves the conflicts. Add ATH10K_FW_IE_HTT_OP_VERSION so that the firmware can
advertise the HTT interface to ath10k.

This fix is needed to get management frames over HTT (ie.
ATH10K_FW_FEATURE_HAS_WMI_MGMT_TX disabled) working with 10.2.4.48-2 firmware.
Otherwise there will be unknown htt events and nothing works:

[30087.438343] ath10k_pci 0000:02:00.0: htt event (19) not handled
[30087.448691] ath10k_pci 0000:02:00.0: htt event (19) not handled
[30149.032974] ath10k_pci 0000:02:00.0: htt event (19) not handled

If the firmware does not have ATH10K_FW_IE_HTT_OP_VERSION use the main HTT
interface. That way old firmware images will still work.

Cc: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add WOW patterns support
Janusz Dziedzic [Mon, 23 Mar 2015 15:32:54 +0000 (17:32 +0200)]
ath10k: add WOW patterns support

Add patterns support for WOW.

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add WMI support for WOW patterns
Janusz Dziedzic [Mon, 23 Mar 2015 15:32:54 +0000 (17:32 +0200)]
ath10k: add WMI support for WOW patterns

Add WMI support for WOW patterns.

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add WOW disconnect/magic-packet support
Janusz Dziedzic [Mon, 23 Mar 2015 15:32:53 +0000 (17:32 +0200)]
ath10k: add WOW disconnect/magic-packet support

Add support for WOW disconnect and magic-packet.

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add WMI support for WOW
Janusz Dziedzic [Mon, 23 Mar 2015 15:32:53 +0000 (17:32 +0200)]
ath10k: add WMI support for WOW

Add WMI support for WOW like enable,
wakeup events and host wakeup indication.

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: allow user to toggle ani_enable via debugfs
Ashok Raj Nagarajan [Thu, 19 Mar 2015 11:08:00 +0000 (16:38 +0530)]
ath10k: allow user to toggle ani_enable via debugfs

Now that ANI is enabled by default, allow user to disable or enable ANI feature
from debugfs

echo 0|1 > /sys/kernel/debug/ieee80211/phyX/ath10k/ani_enable

Signed-off-by: Ashok Raj Nagarajan <arnagara@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: enable Adaptive Noise Immunity (ANI) by default
Ashok Raj Nagarajan [Thu, 19 Mar 2015 11:07:59 +0000 (16:37 +0530)]
ath10k: enable Adaptive Noise Immunity (ANI) by default

ANI helps to improve connectvity and performance in a noisy environment.
Enabling this feature would help the user experience a better and stable
wireless connection in a noisy environmnet. This feature is currently not
enabled for ath10k. Enable this feature by default.

Signed-off-by: Ashok Raj Nagarajan <arnagara@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: enable channel 144 on 5GHz band
Peter Oh [Wed, 18 Mar 2015 18:39:18 +0000 (11:39 -0700)]
ath10k: enable channel 144 on 5GHz band

Enable channel 144 on 5GHz band since 802.11ac introduced it.

Signed-off-by: Peter Oh <poh@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: move driver state check before setting throttle
Rajkumar Manoharan [Sun, 15 Mar 2015 15:06:25 +0000 (20:36 +0530)]
ath10k: move driver state check before setting throttle

Since thermal daemon is unaware of the device state, it might
try to adjust the throttle state when the device is powered down.
So the driver caches the value and will configure it while
powering up the target. The cached value will be programed later
once the device is brought up. In such case, returning error
status is confusing and misleading the user application. Hence
moving the driver state check before sending wmi command to target.

Reported-by: Matthias Kaehlcke <mka@google.com>
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: cache throttle state when device is down
Rajkumar Manoharan [Sun, 15 Mar 2015 15:06:24 +0000 (20:36 +0530)]
ath10k: cache throttle state when device is down

Allow driver to cache the throttle state when the devie is not
yet started. Configure the cached throttle state while powering
up the device. Since thermal daemon is unaware of the up/down cycle,
it assumes that device is throttled.

Reported-by: Matthias Kaehlcke <mka@google.com>
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: do not restrict thermal throttling to ap mode
Rajkumar Manoharan [Sun, 15 Mar 2015 15:06:23 +0000 (20:36 +0530)]
ath10k: do not restrict thermal throttling to ap mode

Recently thermal mitigation is validated in station mode as well.
Hence allowing thermal throttling for all interfaces. This enables
user to validate thermal mitigation with different modes.

Reported-by: Matthias Kaehlcke <mka@google.com>
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: configure thermal throttle while powering up
Rajkumar Manoharan [Sun, 15 Mar 2015 15:06:22 +0000 (20:36 +0530)]
ath10k: configure thermal throttle while powering up

Thermal throttling is not handled in software restart and device
bootup. Also it needs to be configured whenever quiet period got
updated. Fix that.

Reported-by: Matthias Kaehlcke <mka@google.com>
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix interpretation of cooling device state
Rajkumar Manoharan [Sun, 15 Mar 2015 15:06:21 +0000 (20:36 +0530)]
ath10k: fix interpretation of cooling device state

Setting the sysfs attribute ends up configuring the duty cycle,
but the interface through which the attribute is exposed
(cooling_device) is for setting the throttle/cooling state. This
is confusing the user. Hence renaming the cooling device interfaces
for better readability.

Cc: Matthias Kaehlcke <mka@google.com>
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add debugfs entry to configure quiet period
Rajkumar Manoharan [Sun, 15 Mar 2015 15:06:20 +0000 (20:36 +0530)]
ath10k: add debugfs entry to configure quiet period

Add support to configure quiet period (in milliseconds) via debugfs.
This is useful to experiment different quiet period values along with
different throttle ratio.

echo 100 > /sys/kernel/debug/ieee80211/phyX/ath10k/quiet_period

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: increase station kickout threshold
Rajkumar Manoharan [Thu, 19 Mar 2015 14:03:29 +0000 (16:03 +0200)]
ath10k: increase station kickout threshold

Since the station kickout threshold is also counting software
reries (Default sw count in firmware is 16), increasing the threshold
to try with atleast 20 data frames before kicking out the station.

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix rts profile for second rate series
Rajkumar Manoharan [Thu, 19 Mar 2015 14:03:29 +0000 (16:03 +0200)]
ath10k: fix rts profile for second rate series

By default rts protection is enabled in firmware for the second
rateset. Currently ath10k selects RTS profile (only for software
retries), when legacy stations are associated or asked by mac80211.
On congested environment, when AP is running in HT/VHT mode and
there are no legacy clients associated, this will impact the
robustness. Also enabling RTS protection only for second rateset will
not impact performance on clear environment. Fix that.

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agocfg80211: add vlan to station add/change tracing
Johannes Berg [Fri, 20 Mar 2015 18:56:41 +0000 (19:56 +0100)]
cfg80211: add vlan to station add/change tracing

This helps debug issues with VLAN modifications that are otherwise
not really visible in any tracing/debugging.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: avoid duplicate TX path station lookup
Johannes Berg [Fri, 20 Mar 2015 13:18:27 +0000 (14:18 +0100)]
mac80211: avoid duplicate TX path station lookup

Instead of looking up the destination station twice in the TX path
(first to build the header, and then for control processing), save
it when building the header and use it later in the TX path.

To avoid having to look up the station in the many callers, allow
those to pass %NULL which keeps the existing lookup.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: mesh: avoid pointless station lookup
Johannes Berg [Fri, 20 Mar 2015 15:01:52 +0000 (16:01 +0100)]
mac80211: mesh: avoid pointless station lookup

In ieee80211_build_hdr(), the station is looked up to build the
header correctly (QoS field) and to check for authorization. For
mesh, authorization isn't checked here, and QoS capability is
mandatory, so the station lookup can be avoided.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: drop 4-addr VLAN frames earlier if not connected
Johannes Berg [Fri, 20 Mar 2015 13:05:02 +0000 (14:05 +0100)]
mac80211: drop 4-addr VLAN frames earlier if not connected

If there's no station on the 4-addr VLAN interface, then frames
cannot be transmitted. Drop such frames earlier, before setting
up all the information for them.

We should keep the old check though since that code might be used
for other internally-generated frames.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: don't look up destination station twice
Johannes Berg [Fri, 20 Mar 2015 12:29:29 +0000 (13:29 +0100)]
mac80211: don't look up destination station twice

There's no need to look up the destination station twice while
building the 802.11 header for a given frame if the frame will
actually be transmitted to the station we initially looked up.

This happens for 4-addr VLAN interfaces and TDLS connections, which
both directly send the frame to the station they looked up, though
in the case of TDLS some station conditions need to be checked.

To avoid that, add a variable indicating that we've looked up the
station that the frame is going to be transmitted to, and avoid the
lookup/flag checking if it already has been done.

In the TDLS case, also move the authorized/wme_sta flag assignment
to the correct place, i.e. only when that station is really used.
Before this change, the new lookup should always have succeeded so
that the potentially erroneous data would be overwritten.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: remove drop_unencrypted code
Johannes Berg [Fri, 20 Mar 2015 10:37:36 +0000 (11:37 +0100)]
mac80211: remove drop_unencrypted code

This mechanism was historic, and only ever used by IBSS, which
also doesn't need to have it as it properly manages station's
802.1X PAE state (or, with WEP, always has a key.)

Remove the mechanism to clean up the code.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agoath10k: setup maximum number of supported stations
Janusz.Dziedzic@tieto.com [Thu, 12 Mar 2015 12:11:41 +0000 (13:11 +0100)]
ath10k: setup maximum number of supported stations

Setup maximum number of associated stations supported in
AP/P2P_GO mode.

This allow hostapd/wpa_supplicant to play with that limit
and set more information (eg. P2P group limit bit for P2P_GO).

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: protect driver from unsolicited htc ep0 events
Michal Kazior [Wed, 11 Mar 2015 13:37:11 +0000 (14:37 +0100)]
ath10k: protect driver from unsolicited htc ep0 events

Some firmware revisions (tested with qca6174
rm2.0-00088) deliver unsolicited unknown (kind of
garbled) HTC ep0 event to host in some cases.

This issue was mainly observed with both qca988x
and qca6174 being installed on a single host
system. During driver probing if qca6174 booting
sequences were somehow deferred (e.g. by qca988x
implicitly making some resources busy presumably)
the unsolicited event would came around 1s after
ATH10K_HTC_MSG_READY_ID was delivered to host for
qca6174.

The unsolicited event would confuse driver and
cause HTT initialization (and subsequently
driver probing) to fail.

Make the ep0 event processing more robust. The
event will still be caught but instead will
only generate a warning now.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix wrong symlink name on error path
Rajkumar Manoharan [Thu, 12 Mar 2015 17:32:00 +0000 (19:32 +0200)]
ath10k: fix wrong symlink name on error path

Wrong symlink name is used on error path of thermal registration
and also correcting the error message.

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: strip qos data bit always
Michal Kazior [Wed, 11 Mar 2015 13:25:26 +0000 (14:25 +0100)]
ath10k: strip qos data bit always

NativeWifi tx mode expects QoS Data frames to be
delivered as Data frames with QoS part (e.g. tid)
being delievered out-of-band in fw tx command.

The QoS bit wasn't stripped before submitting to
firmware.

Stripping fixes two known problems:

 * qca6174 IOT with some APs, e.g.
   Cisco AIR-AP 1252 (which would crash after
   ath10k association). Some ath9k APs would
   crash as well.

 * sniffing own tx frames via radiotap because,
   e.g. wireshark was seeing QoS bit set but
   since QoS Control was stripped in ath10k it
   would parse beginning of LLC/SNAP

>From debugability point of view this removes the
ability to distinguish QoS from non-QoS frames
when sniffing own tx via radiotap. On the other
hand frames can be now parsed correctly without
special software modification.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agomac80211: Get IV len from key conf and not cipher scheme
Cedric Izoard [Tue, 17 Mar 2015 10:47:33 +0000 (10:47 +0000)]
mac80211: Get IV len from key conf and not cipher scheme

When a key is installed using a cipher scheme, set a new
internal key flag (KEY_FLAG_CIPHER_SCHEME) on it, to allow
distinguishing such keys more easily.

In particular, use this flag on the TX path instead of
testing the sta->cipher_scheme pointer, as the station is
NULL for broad-/multicast message, and use the key's iv_len
instead of the cipher scheme information.

Signed-off-by: Cedric Izoard <cedric.izoard@ceva-dsp.com>
[add missing documentation, rewrite commit message]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: IBSS: refactor ieee80211_rx_bss_info
Janusz.Dziedzic@tieto.com [Mon, 9 Mar 2015 06:58:15 +0000 (07:58 +0100)]
mac80211: IBSS: refactor ieee80211_rx_bss_info

Put station specific code in ieee80211_update_sta_info
function.

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: minstrel_ht: fix rounding issue in MCS duration calculation
Felix Fietkau [Fri, 13 Mar 2015 09:54:44 +0000 (10:54 +0100)]
mac80211: minstrel_ht: fix rounding issue in MCS duration calculation

On very high MCS bitrates, the calculated duration of rates that are
next to each other can be very imprecise, due to the small packet size
used as reference (1200 bytes).
This is most visible in VHT80 nss=2 MCS8/9, for which minstrel shows the
same throughput when the probability is also the same. This leads to a
bad rate selection for such rates.

Fix this issue by introducing an average A-MPDU size factor into the
calculation.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agocfg80211: Process all pending regulatory requests/hints
Ben [Thu, 12 Mar 2015 13:37:34 +0000 (09:37 -0400)]
cfg80211: Process all pending regulatory requests/hints

It is possible that there are several regulatory requests
pending, but the processing of the last one does not call
CRDA, and thus the other requests are not handled.

Fix this by rescheduling the work until all requests have
been processed.

Signed-off-by: Ben Rosenfeld <ben.rosenfeld@intel.com>
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: initialize rate control earlier for tdls station
Marek Puzyniak [Sun, 8 Mar 2015 16:04:22 +0000 (18:04 +0200)]
mac80211: initialize rate control earlier for tdls station

Currently when TDLS station in driver goes from authenticated
to associated state it can not use rate control parameters
because rate control is not initialized yet. Some drivers
require parameters already initialized by rate control when
entering associated state. It can be done by initializing
rate control after station transition to associated state but
before notifying driver about that.

Signed-off-by: Marek Puzyniak <marek.puzyniak@tieto.com>
Signed-off-by: Arik Nemtsov <arikx.nemtsov@intel.com>
[fix comment to say 'associated' instead of 'authorized']
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: refactor drop connection/unlock in CSA processing
Johannes Berg [Thu, 12 Mar 2015 06:53:25 +0000 (08:53 +0200)]
mac80211: refactor drop connection/unlock in CSA processing

The schedule_work()/mutex unlocking code is duplicated many times,
refactor that to a common place in the function.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: allow to get wireless_dev structure from ieee80211_vif
Emmanuel Grumbach [Thu, 12 Mar 2015 06:53:24 +0000 (08:53 +0200)]
mac80211: allow to get wireless_dev structure from ieee80211_vif

This will allow mac80211 drivers to call cfg80211 APIs with
the right handle.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: add comment for rx_path_lock
Johannes Berg [Mon, 16 Mar 2015 08:08:20 +0000 (09:08 +0100)]
mac80211: add comment for rx_path_lock

Add a comment explaining how the RX path lock is used.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agomac80211: move netdev stats to common function
Johannes Berg [Mon, 16 Mar 2015 08:05:23 +0000 (09:05 +0100)]
mac80211: move netdev stats to common function

Move the netdev stats accounting into the common function
ieee80211_deliver_skb() that is called in both places.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
9 years agoath10k: fix PMF by using AES-CMAC/IGTK software crypto
Bartosz Markowski [Tue, 10 Mar 2015 13:32:19 +0000 (14:32 +0100)]
ath10k: fix PMF by using AES-CMAC/IGTK software crypto

While testing with older supplicant, .drv_set_key() was failing due to
higher than ath10k firmware could handle key_index (WMI_MAX_KEY_INDEX == 3).

--
wpa_driver_nl80211_set_key: ifindex=15 alg=4 addr=0x7f02b129fbe3 key_idx=4 set_tx=0 seq_len=6 key_len=16
    broadcast key
nl80211: set_key failed; err=-22 Invalid argument)
wlan0: WPA: Failed to configure IGTK to the driver
wlan0: RSN: Failed to configure IGTK
--

In order to fix this case (PMF: AES-CMAC/IGTK) force the AES_CMAC cipher to
be handled by software.

Signed-off-by: Bartosz Markowski <bartosz.markowski@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add hw connection monitor support
Michal Kazior [Tue, 10 Mar 2015 14:22:01 +0000 (16:22 +0200)]
ath10k: add hw connection monitor support

Some firmware revisions (e.g. qca6174 with fw73)
don't deliver beacons to host reliably. This
causes random disconnects even in perfect
conditions. This is most visible with
multi-channel operation.

All available firmware revisions seem to support
beacon miss offloading so there shouldn't be any
problems.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>