GitHub/LineageOS/android_kernel_motorola_exynos9610.git
10 years agoath10k: add modpram 'skip_otp' to ignore empty otp error during BMI
Rajkumar Manoharan [Mon, 17 Nov 2014 14:44:14 +0000 (16:44 +0200)]
ath10k: add modpram 'skip_otp' to ignore empty otp error during BMI

This patch would help bring up wifi interface with default board
data in case of failures in otp download. It is useful for initial
calibration.

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: handle ieee80211 header and payload tracing separately
Rajkumar Manoharan [Wed, 5 Nov 2014 13:44:31 +0000 (19:14 +0530)]
ath10k: handle ieee80211 header and payload tracing separately

For packet log, the transmitted frame 802.11 header alone is sufficient.
Recording entire packet is also consuming lot of disk space. To optimize
this, tx and rx data tracepoints are splitted into header and payload
tracepoints.

To record tx ieee80211 headers

     trace-cmd record -e ath10k_tx_hdr

To record complete packets

     trace-cmd record -e ath10k_tx_hdr -e ath10k_tx_payload

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>
10 years agoath10k: fix pm resume after suspend
Bartosz Markowski [Fri, 31 Oct 2014 08:03:43 +0000 (09:03 +0100)]
ath10k: fix pm resume after suspend

Firmware was crashing when we were trying to warm reset it
after suspend. This was due to the fact that target registeres
can be accessed only if the hardware is awaken.

This patch makes sure to awake the device also on the hif up,
not only in case of probe call.

Signed-off-by: Bartosz Markowski <bartosz.markowski@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath6kl: remove incorrect reset_resume handler
Alexey Khoroshilov [Fri, 31 Oct 2014 00:45:37 +0000 (02:45 +0200)]
ath6kl: remove incorrect reset_resume handler

Existing implementation of reset_resume handler just calls
ath6kl_usb_remove() that deallocates all resources.
It can lead to double free, etc. on disconnect.

The patch removes reset_resume handler,
so usb core could conservatively reset the driver.

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: speed up hw recovery
Michal Kazior [Tue, 28 Oct 2014 09:34:38 +0000 (10:34 +0100)]
ath10k: speed up hw recovery

In some cases hw recovery was taking an absurdly
long time due to ath10k waiting for things that
would never really complete.

Instead of waiting for inevitable timeouts poke
all completions and wakequeues and check if it's
still worth waiting.

Reading/writing ar->state requires conf_mutex.
Since waiters might be holding it introduce a new
flag CRASH_FLUSH so it's possible to tell waiters
to abort whatever they were waiting for.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: expose hw restart via debugfs
Michal Kazior [Tue, 28 Oct 2014 09:34:37 +0000 (10:34 +0100)]
ath10k: expose hw restart via debugfs

Until now it was possible to simulate soft and
hard fw crashes but it wasn't possible to trigger
an immediately hw restart itself (without the fw
crash).

This can be useful when stress testing hw
restarting stability, e.g. during heavy tx/rx
traffic.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix possible bmi crash
Michal Kazior [Tue, 28 Oct 2014 09:34:36 +0000 (10:34 +0100)]
ath10k: fix possible bmi crash

While testing other things I've found that CE
items aren't cleared properly. This could lead to
null dereferences in BMI.

To prevent that make sure CE revoking clears the
nbytes value (which is used as a buffer completion
indication) and memset the entire CE ring data
shared between host and target when
(re)initializing.

Also make sure to check BMI xfer pointer and print
a splat instead of crashing the kernel.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: don't reset chip on power_down
Michal Kazior [Tue, 28 Oct 2014 09:32:08 +0000 (10:32 +0100)]
ath10k: don't reset chip on power_down

Currently hif_power_up performs effectively a
reset and hif_stop resets the chip as well so
there's no point in resetting here.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: split reset logic from power up
Michal Kazior [Tue, 28 Oct 2014 09:32:07 +0000 (10:32 +0100)]
ath10k: split reset logic from power up

The power up procedure was overly complex due to
warm/cold reset workarounds and issues.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: make warm reset a bit safer and faster
Michal Kazior [Tue, 28 Oct 2014 09:32:06 +0000 (10:32 +0100)]
ath10k: make warm reset a bit safer and faster

One of the problems with warm reset I've found is
that it must be guaranteed that copy engine
registers are not being accessed while being
reset. Otherwise in worst case scenario the host
may lock up.

Instead of using sleeps and hoping the device is
operational in some arbitrary timeframes use
firmware indication register.

As a side effect this makes driver
boot/stop/recovery faster.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: change ce ring cleanup logic
Michal Kazior [Tue, 28 Oct 2014 09:32:05 +0000 (10:32 +0100)]
ath10k: change ce ring cleanup logic

Make ath10k_pci_init_pipes() effectively only
alter shared target-host data.

The per_transfer_context is a host-only thing.
It is necessary to preserve it's contents for a
more robust ring cleanup.

This is required for future warm reset fixes.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: avoid possible deadlock with scan timeout
Michal Kazior [Tue, 28 Oct 2014 09:23:09 +0000 (10:23 +0100)]
ath10k: avoid possible deadlock with scan timeout

This should prevent deadlock predicted by the
following splat:

 ======================================================
 [ INFO: possible circular locking dependency detected ]
 3.17.0-wl-ath+ #67 Not tainted
 -------------------------------------------------------
 kworker/u32:1/7230 is trying to acquire lock:
  (&ar->conf_mutex){+.+.+.}, at: [<ffffffffa040a57d>] ath10k_scan_timeout_work+0x2d/0x50 [ath10k_core]

 but task is already holding lock:
  ((&(&ar->scan.timeout)->work)){+.+...}, at: [<ffffffff8106dae1>] process_one_work+0x151/0x470

 which lock already depends on the new lock.

 the existing dependency chain (in reverse order) is:

 -> #1 ((&(&ar->scan.timeout)->work)){+.+...}:
        [<ffffffff810a12e5>] lock_acquire+0x85/0x100
        [<ffffffff8106cb4d>] flush_work+0x3d/0x270
        [<ffffffff8106e49d>] __cancel_work_timer+0x7d/0x110
        [<ffffffff8106e543>] cancel_delayed_work_sync+0x13/0x20
        [<ffffffffa0409f16>] ath10k_cancel_remain_on_channel+0x36/0x60 [ath10k_core]
        [<ffffffffa028c75c>] ieee80211_cancel_roc+0x1cc/0x2f0 [mac80211]
        [<ffffffffa028c8a2>] ieee80211_mgmt_tx_cancel_wait+0x22/0x30 [mac80211]
        [<ffffffffa0132288>] nl80211_tx_mgmt_cancel_wait+0xa8/0x130 [cfg80211]
        [<ffffffff816654a5>] genl_family_rcv_msg+0x1a5/0x3c0
        [<ffffffff81665749>] genl_rcv_msg+0x89/0xc0
        [<ffffffff81664e91>] netlink_rcv_skb+0xb1/0xc0
        [<ffffffff816650bc>] genl_rcv+0x2c/0x40
        [<ffffffff8166474d>] netlink_unicast+0x18d/0x200
        [<ffffffff81664add>] netlink_sendmsg+0x31d/0x430
        [<ffffffff8161a9ac>] sock_sendmsg+0x9c/0xd0
        [<ffffffff8161b469>] ___sys_sendmsg+0x389/0x3a0
        [<ffffffff8161bed9>] __sys_sendmsg+0x49/0x90
        [<ffffffff8161bf32>] SyS_sendmsg+0x12/0x20
        [<ffffffff8174c456>] system_call_fastpath+0x1a/0x1f

 -> #0 (&ar->conf_mutex){+.+.+.}:
        [<ffffffff810a0bde>] __lock_acquire+0x1b6e/0x1ce0
        [<ffffffff810a12e5>] lock_acquire+0x85/0x100
        [<ffffffff817491eb>] mutex_lock_nested+0x4b/0x370
        [<ffffffffa040a57d>] ath10k_scan_timeout_work+0x2d/0x50 [ath10k_core]
        [<ffffffff8106db41>] process_one_work+0x1b1/0x470
        [<ffffffff8106df63>] worker_thread+0x123/0x460
        [<ffffffff81073f34>] kthread+0xe4/0x100
        [<ffffffff8174c3ac>] ret_from_fork+0x7c/0xb0

 other info that might help us debug this:

  Possible unsafe locking scenario:

        CPU0                    CPU1
        ----                    ----
   lock((&(&ar->scan.timeout)->work));
                                lock(&ar->conf_mutex);
                                lock((&(&ar->scan.timeout)->work));
   lock(&ar->conf_mutex);

  *** DEADLOCK ***

Reported-by: Marek Puzyniak <marek.puzyniak@tieto.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: call correct function for frag threshold
Michal Kazior [Thu, 23 Oct 2014 14:04:29 +0000 (17:04 +0300)]
ath10k: call correct function for frag threshold

Rts threshold was being configured instead of
fragmentation threshold.

Keep in mind available firmware binaries don't
seem to support fragmentation anyway so this
doesn't fix fragmentation threshold per se.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add SURVEY_INFO_IN_USE for current channel on survey
Felix Fietkau [Thu, 23 Oct 2014 14:04:28 +0000 (17:04 +0300)]
ath10k: add SURVEY_INFO_IN_USE for current channel on survey

When user space requests survey info, it is useful to know which of the survey
data refers to the channel that is currently actively being used.  One of the
use cases is getting the current channel noise for status output.  Without this
flag you would have to look up the channel separately and then compare it
against the frequency in the survey output in user space.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: re-use trace class
Michal Kazior [Thu, 23 Oct 2014 14:04:28 +0000 (17:04 +0300)]
ath10k: re-use trace class

Instead of defining a completely new tracepoint
use an existing tracepoint class.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: remove tsf argument from rx_desc tracing
Michal Kazior [Thu, 23 Oct 2014 14:04:27 +0000 (17:04 +0300)]
ath10k: remove tsf argument from rx_desc tracing

Fundamentally this was wrong. Tsf is only valid
in last MPDU of a PPDU. This means tsf value was
wrong most of the time during heavy traffic.

Also I don't see much point in exposing a
redundant (and broken) tsf value. Userspace can
already read it from the dumped rx descriptor
buffer.

Cc: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: clear htt->rx_confused on load
Michal Kazior [Thu, 23 Oct 2014 14:04:24 +0000 (17:04 +0300)]
ath10k: clear htt->rx_confused on load

Once driver entered the rx_confused state it would
refuse to rx even after firmware is restarted.
Make sure to clear it so that rx works after, e.g.
hw restart or after all interfaces are stopped.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: don't forget to replenish after fragmented Rx
Michal Kazior [Thu, 23 Oct 2014 14:04:24 +0000 (17:04 +0300)]
ath10k: don't forget to replenish after fragmented Rx

In theory it was possible to drain entire HTT Rx
ring via fragmented Rx leading to Rx lockup.

In practice non-data traffic would always trigger
replenishment via the regular Rx handler.

For correctness sake make sure to replenish the
ring on fragmented Rx.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add extra sanity check when popping amsdu
Michal Kazior [Thu, 23 Oct 2014 14:04:24 +0000 (17:04 +0300)]
ath10k: add extra sanity check when popping amsdu

The netbuf pop can return NULL. Make sure to check
for that. It shouldn't happen but better safe than
sorry.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: don't drop frames aggressively
Michal Kazior [Thu, 23 Oct 2014 14:04:23 +0000 (17:04 +0300)]
ath10k: don't drop frames aggressively

There's little point in dropping, e.g. frames with
FCS error early in ath10k.

This simplifies amsdu_allowed() and gets rid of
htt_rx_mpdu_status usage finally.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: deduplicate htt rx dma unmapping
Michal Kazior [Thu, 23 Oct 2014 14:04:23 +0000 (17:04 +0300)]
ath10k: deduplicate htt rx dma unmapping

Treat non-chained and chained popping the same
way. Also this makes netbuf pop fully symmetrical
to (re)filling.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix rx buffer tracing
Michal Kazior [Thu, 23 Oct 2014 14:04:23 +0000 (17:04 +0300)]
ath10k: fix rx buffer tracing

Tracing function was called before buffers were
unmapped from DMA.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: use ieee80211 defines for crypto param lengths
Michal Kazior [Thu, 23 Oct 2014 14:04:22 +0000 (17:04 +0300)]
ath10k: use ieee80211 defines for crypto param lengths

Use the globally defined ieee80211 values instead
of re-defining them in the driver again.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: remove unused variable
Michal Kazior [Thu, 23 Oct 2014 14:04:22 +0000 (17:04 +0300)]
ath10k: remove unused variable

The rx descriptor variable was no longer used in
the rx handler.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: don't drop control and null func Rx
Michal Kazior [Thu, 23 Oct 2014 14:04:22 +0000 (17:04 +0300)]
ath10k: don't drop control and null func Rx

HTT_RX_IND_MPDU_STATUS_MGMT_CTRL was pretty greedy
and because of that ath10k ended up dropping
Control Frames as well as Null Func frames.

Reported-by: Okhwan Lee <ohlee@mwnl.snu.ac.kr>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: split ce pipe init/alloc further
Michal Kazior [Mon, 20 Oct 2014 12:14:39 +0000 (14:14 +0200)]
ath10k: split ce pipe init/alloc further

Calling init to reinit ce pipe state would also
re-set all static structure links and setting
(which don't change over driver lifecycle).

Make it so alloc links structures and initializes
static data and init part to setup state
variables and clear stuff.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: mask/unmask msi fw irq
Michal Kazior [Mon, 20 Oct 2014 12:14:38 +0000 (14:14 +0200)]
ath10k: mask/unmask msi fw irq

This was the final missing bit to making sure the
device doesn't assert interrupts to host.

This should fix possible race when target crashes
during driver teardown.

This also removes an early warm reset workaround
during pci probing.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: re-disable interrupts after target init
Michal Kazior [Mon, 20 Oct 2014 12:14:37 +0000 (14:14 +0200)]
ath10k: re-disable interrupts after target init

If MSI isn't configured device ROM program expects
legacy interrupts to be enabled before it can
fully boot. Don't forget to disable legacy
interrupts after that.

While at it re-use the legacy irq enabling helper
instead of calling ath10k_pci_write32().

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: use bss_info as txpower source
Michal Kazior [Tue, 21 Oct 2014 07:40:15 +0000 (10:40 +0300)]
ath10k: use bss_info as txpower source

This simply changes the source for txpower setup.
It does not make ath10k use different txpower
values for different vifs.

This will make it easier to implement chanctx in
ath10k in the future.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: simplify computation of mgmt rx band
Michal Kazior [Tue, 21 Oct 2014 07:10:30 +0000 (10:10 +0300)]
ath10k: simplify computation of mgmt rx band

Using global channel won't work with chanctx. Try
to determine the channel from the information
provided in the wmi event itself alone. This
should be sufficient.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: clean up sta auth/assoc code
Michal Kazior [Tue, 21 Oct 2014 07:10:29 +0000 (10:10 +0300)]
ath10k: clean up sta auth/assoc code

The code can be symmetrical so make it so. This
makes it easier to understand and work with.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: skip some commands on reassoc
Michal Kazior [Tue, 21 Oct 2014 07:10:29 +0000 (10:10 +0300)]
ath10k: skip some commands on reassoc

It doesn't make much sense to reconfigure peer
completely upon reassociation. This will make it
easier to have a more uniform association code
across different modes.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: clean up assoc code
Michal Kazior [Tue, 21 Oct 2014 07:10:29 +0000 (10:10 +0300)]
ath10k: clean up assoc code

There's no need to pass bss_conf explicitly as it
is accessible via vif pointer. This requires
slight changes in function prototypes. While at it
clean up listen interval workaround/command.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: retrieve calibration data from file
Kalle Valo [Mon, 13 Oct 2014 06:40:59 +0000 (09:40 +0300)]
ath10k: retrieve calibration data from file

A frequent request have been to be able to provide calibration data from a
file as some of the AP devices store the calibration data on an MTD partition.
This patchset adds support for that and also makes it easier to add Device Tree
support later on.

The calibration data is found by using the id string provided by dev_name()
using this format:

cal-<bus>-<id>.bin

With PCI the id string contains bus, slot and func values. For example for a
PCI device in bus 2 slot 0, ath10k will try to retrieve a calibration data from
a file:

/lib/firmware/ath10k/cal-pci-0000:02:00.0.bin

The calibration data sequence is:

1. Check with request_firmware() if there's a calibration file
   ("cal-<bus>-<id>.bin") on the filesystem for this device. If yes, use that. If
   not, goto 2

2. Check if otp.bin is able to successfully load the calibration data
   from OTP. If yes, use that. If not, goto 3.

4. Print an error message that no calibration data found and stop driver
   initialization for this device.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: refactor ath10k_init_download_firmware()
Kalle Valo [Mon, 13 Oct 2014 06:40:53 +0000 (09:40 +0300)]
ath10k: refactor ath10k_init_download_firmware()

This is preparation for being able to download calibration data from a file.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add back enum ath10k_bus
Kalle Valo [Mon, 13 Oct 2014 06:40:47 +0000 (09:40 +0300)]
ath10k: add back enum ath10k_bus

Commit 3a0861fffd223 ("ath10k: remove ath10k_bus") removed enum ath10k_bus
because it was not used for anything at the time. But now it's needed for for
retrieving the right calibration data file so add it back. Only new addition is
ath10k_bus_str().

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix kernel panic while shutting down AP
Rajkumar Manoharan [Fri, 10 Oct 2014 12:08:27 +0000 (17:38 +0530)]
ath10k: fix kernel panic while shutting down AP

Based on GFP flag given to DMA coherent allocation, the behaviour of
dma_free_coherent is changed. This behavioural diffrence is noticeable
in ARM platform. If DMA memory is allocated with GFP_KERNEL, free
coherent can not be called inside spin lock. This is causing kernel
crash in ARM platforms. Fix this by changing GFP flag to atomic.

This is most likely a regression from commit 64badcb6d645 ("ath10k: workaround
fw beaconing bug").

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>
10 years agoath10k: add tracing for frame transmission
Rajkumar Manoharan [Mon, 6 Oct 2014 17:09:06 +0000 (22:39 +0530)]
ath10k: add tracing for frame transmission

Add tracing support to forward management and data frames to
user space for packet inspection.

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix WMI scan command length
Janusz Dziedzic [Tue, 7 Oct 2014 10:26:43 +0000 (12:26 +0200)]
ath10k: fix WMI scan command length

Fix WMI scan command length we setup when scan request.

This fix issue with 636 firmware when scan always failed
with message:

ath10k_pci 0000:02:00.0: wmi start scan
ath10k_pci 0000:02:00.0: wmi stop scan reqid 1 req_type 0 vdev/scan_id 0
ath10k_pci 0000:02:00.0: failed to stop wmi scan: -11
ath10k_pci 0000:02:00.0: failed to stop scan: -11
ath10k_pci 0000:02:00.0: failed to start hw scan: -110

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: advertise all possible firmware(-api) files
Bartosz Markowski [Mon, 6 Oct 2014 12:16:41 +0000 (14:16 +0200)]
ath10k: advertise all possible firmware(-api) files

This is required if we take into account possibility to load the driver
from initrd (RAM disk), so in other words: very early in the boot process,
before the file system is visible.

In such case we need to have the firmware files accessible from ram disk too,
and this patch guarantee this.

Signed-off-by: Bartosz Markowski <bartosz.markowski@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: don't create bssid peer for ibss
Janusz Dziedzic [Thu, 2 Oct 2014 11:56:40 +0000 (13:56 +0200)]
ath10k: don't create bssid peer for ibss

It's not really necessary to create bssid peer for
bssid. Self-address peer is sufficient.

This prevents some firmware revisions from crashing.

Signed-off-by: Janusz Dziedzic <janusz.dziedzic@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: warn on unhandled htt events
Michal Kazior [Thu, 2 Oct 2014 11:32:55 +0000 (13:32 +0200)]
ath10k: warn on unhandled htt events

It makes a lot more sense to print these kinds of
problems as a warning instead of a debug.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add tracing for tx info
Rajkumar Manoharan [Fri, 3 Oct 2014 05:02:54 +0000 (08:02 +0300)]
ath10k: add tracing for tx info

The tx info such as msdu_id, frame len, vdev id and tid are reported
to user space by tracepoint. This is useful for collecting tx
statistics.

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add tracing for rx descriptor
Rajkumar Manoharan [Fri, 3 Oct 2014 05:02:47 +0000 (08:02 +0300)]
ath10k: add tracing for rx descriptor

Upon the reception of frame, the descriptor status are reported
to user space by tracepoint. This is useful for collecting rx
statistics.

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add tracing for ath10k_htt_pktlog
Rajkumar Manoharan [Fri, 3 Oct 2014 05:02:40 +0000 (08:02 +0300)]
ath10k: add tracing for ath10k_htt_pktlog

This is useful for collecting pktlog statistics of tx, rx
and rate information, so add tracing for the API call.

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add support to configure pktlog filter
Rajkumar Manoharan [Fri, 3 Oct 2014 05:02:33 +0000 (08:02 +0300)]
ath10k: add support to configure pktlog filter

Add support to configure packet log filters (tx, rx, rate control)
via debugfs. To disable htt pktlog events set the filters to 0.

ex:

To enable pktlog for all filters

   echo 0x1f > /sys/kernel/debug/ieee80211/phy*/ath10k/pktlog_filter

To disable pktlog

   echo 0 > /sys/kernel/debug/ieee80211/phy*/ath10k/pktlog_filter

Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: use 64-bit vdev map
Ben Greear [Tue, 23 Sep 2014 21:17:16 +0000 (14:17 -0700)]
ath10k: use 64-bit vdev map

This can allow more than 32 stations to be supported
without over-running the bitmap.

Signed-off-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: support ethtool stats
Ben Greear [Mon, 29 Sep 2014 11:41:46 +0000 (14:41 +0300)]
ath10k: support ethtool stats

Add support for reading firmware stats through the ethtool
API.  This may be easier for applications to manipulate
compared to parsing a text based debugfs file.

kvalo: remove unneeded ifdefs, call ath10k_debug_fw_stats_request() and added
simple error handling

Signed-off-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add firmware crash counters
Ben Greear [Mon, 29 Sep 2014 11:41:46 +0000 (14:41 +0300)]
ath10k: add firmware crash counters

Add three counters related to firmware crashes or resets.

Usage:

# cat /sys/kernel/debug/ieee80211/phy0/ath10k/fw_reset_stats
fw_crash_counter                2
fw_warm_reset_counter           43
fw_cold_reset_counter           0
#

kvalo: split into it's own patch, add debugfs file and add locking

Signed-off-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix fw stats processing
Michal Kazior [Thu, 25 Sep 2014 10:33:50 +0000 (12:33 +0200)]
ath10k: fix fw stats processing

If stat data exceeds wmi-htc buffer limits
firmware splits it into many wmi stats update
events which are delivered in a ping-pong fashion
triggered by wmi stats request command.

Since there's only an implicit start-of-data and
no end-of-data indications the driver has to
perform some trickery to get complete stat data.

kvalo: use %zu to fix a compiler warning and fix a typo in a comment

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: request fw_stats once on open
Michal Kazior [Thu, 25 Sep 2014 10:33:49 +0000 (12:33 +0200)]
ath10k: request fw_stats once on open

Stats were requested and processed for each read
call. This caused inconsistent readings.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: rename fw_stats related stuff
Michal Kazior [Thu, 25 Sep 2014 10:33:48 +0000 (12:33 +0200)]
ath10k: rename fw_stats related stuff

The naming was a bit inconsistent.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: split wmi stats parsing
Michal Kazior [Thu, 25 Sep 2014 10:33:47 +0000 (12:33 +0200)]
ath10k: split wmi stats parsing

The parsing function was rather complex. Simplify
by splitting it up into firmware branch specific
implementations.

While at it move the parsing code into wmi.c where
it belongs.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath: change logging functions to return void
Joe Perches [Mon, 22 Sep 2014 17:35:34 +0000 (10:35 -0700)]
ath: change logging functions to return void

The return values are not used by callers of these functions
so change the functions to return void.

Other miscellanea:

o add __printf verification to wil6210 logging functions
  No format/argument mismatches found

Acked-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add cal_data debugfs file
Kalle Valo [Wed, 24 Sep 2014 11:16:58 +0000 (14:16 +0300)]
ath10k: add cal_data debugfs file

Provide calibration data used by the firmware to user space via a debugfs file.
This makes it easier to debug calibration related problems.

Example:

sudo cp /sys/kernel/debug/ieee80211/phy0/ath10k/cal_data 1.cal

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add diag_read() to hif ops
Kalle Valo [Wed, 24 Sep 2014 11:16:52 +0000 (14:16 +0300)]
ath10k: add diag_read() to hif ops

diag_read() is used for reading from firmware memory via the diagnose window.
First user will be cal_data debugfs file.

To serialise diagnostic window access and make it safe to use while firmware is
running take ce_lock both in ath10k_pci_diag_write_mem() and
ath10k_pci_diag_read_mem(). Because of that all the CE calls had to be changed
to _nolock variants.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: don't enable interrupts for the diagnostic window
Kalle Valo [Wed, 24 Sep 2014 11:16:46 +0000 (14:16 +0300)]
ath10k: don't enable interrupts for the diagnostic window

The diagnostic window (CE7) uses polling and is not initiliased to retrieve
interrupts so disable interrupts altogether for CE7. Otherwise ath10k crashes
when using the diagnostic window while the firmware is running due to NULL
dereference and polling reads timeout.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add debug dump for pci rx
Michal Kazior [Tue, 23 Sep 2014 08:22:54 +0000 (10:22 +0200)]
ath10k: add debug dump for pci rx

This makes it easier to debug the device-target
communication at a very low level.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: dump hex bytes with dev string prefix
Michal Kazior [Tue, 23 Sep 2014 08:22:53 +0000 (10:22 +0200)]
ath10k: dump hex bytes with dev string prefix

This makes it easier to debug hex dumps on systems
with more than a single ath10k device.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: print wmi version info
Michal Kazior [Tue, 23 Sep 2014 08:22:52 +0000 (10:22 +0200)]
ath10k: print wmi version info

HTT version is already printed so print WMI
version as well for consistency.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: re-work scan start command building
Michal Kazior [Thu, 18 Sep 2014 13:21:27 +0000 (15:21 +0200)]
ath10k: re-work scan start command building

This gets rid of the ugly scan structure building
and uses a saner way to do it.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: unify wmi event function names
Michal Kazior [Thu, 18 Sep 2014 13:21:26 +0000 (15:21 +0200)]
ath10k: unify wmi event function names

Make all wmi event functions match the same naming
style, i.e. ath10k_wmi_event_<name>.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: clean up phyerr code
Michal Kazior [Thu, 18 Sep 2014 13:21:25 +0000 (15:21 +0200)]
ath10k: clean up phyerr code

Make the phyerr structures more compact and easier
to understand. Also add constness.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: deduplicate wmi service ready logic
Michal Kazior [Thu, 18 Sep 2014 13:21:24 +0000 (15:21 +0200)]
ath10k: deduplicate wmi service ready logic

The logic responsible for processing the event is
no different across different firmware binaries.
The difference that needs to be dealt with is the
ABI of data structures.

The intermediate structure uses __le32 to avoid
extra memory allocations to byteswap
variable-length substructures (i.e. host mem
chunks).

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: relocate wmi attach/deatch functions
Michal Kazior [Thu, 18 Sep 2014 13:21:23 +0000 (15:21 +0200)]
ath10k: relocate wmi attach/deatch functions

Init functions should be placed at the end of
files in most cases to avoid forward declarations
for static functions.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: deduplicate host mem chunk code
Michal Kazior [Thu, 18 Sep 2014 13:21:22 +0000 (15:21 +0200)]
ath10k: deduplicate host mem chunk code

Simplify the code by deduplicating structure
definitions and code.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: deduplicate wmi_channel code
Michal Kazior [Thu, 18 Sep 2014 13:21:21 +0000 (15:21 +0200)]
ath10k: deduplicate wmi_channel code

The structure is being set up in 2 places.
Deduplicate the code by creating a helper.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: remove unused pdev_set_channel command
Michal Kazior [Thu, 18 Sep 2014 13:21:20 +0000 (15:21 +0200)]
ath10k: remove unused pdev_set_channel command

This command is not used anymore and most firmware
revisions do not seem to handle it well. Channel
switching is done via vdev restarting.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix tx/rx chainmask init
Michal Kazior [Thu, 18 Sep 2014 13:21:19 +0000 (15:21 +0200)]
ath10k: fix tx/rx chainmask init

Firmware reports the number of RF chains so use
that for initialization of supp_{tx,rx}_chainmask
instead of using a macro for 3x3 chips.

This should make tx/rx chainmask reports correct
for chips other than 3x3.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: workaround fw beaconing bug
Michal Kazior [Thu, 18 Sep 2014 08:18:02 +0000 (11:18 +0300)]
ath10k: workaround fw beaconing bug

Some firmware revisions don't wait for beacon tx
completion before sending another SWBA event. This
could lead to hardware using old (freed) beacon
data in some cases, e.g. tx credit starvation
combined with missed TBTT. This is very very rare.

On non-IOMMU-enabled hosts this could be a
possible security issue because hw could beacon
some random data on the air.  On IOMMU-enabled
hosts DMAR faults would occur in most cases and
target device would crash.

Since there are no beacon tx completions (implicit
nor explicit) propagated to host the only
workaround for this is to allocate a DMA-coherent
buffer for a lifetime of a vif and use it for all
beacon tx commands. Worst case for this approach
is some beacons may become corrupted, e.g. garbled
IEs or out-of-date TIM bitmap.

Keep the original beacon-related code as-is in
case future firmware revisions solve this problem
so that the old path can be easily re-enabled with
a fw_feature flag.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: use ether_addr_copy()
Kalle Valo [Sun, 14 Sep 2014 09:50:49 +0000 (12:50 +0300)]
ath10k: use ether_addr_copy()

As suggeested by checkpatch:

WARNING: Prefer ether_addr_copy() over memcpy() if the Ethernet addresses are __aligned(2)

In wmi.c I had to change due to sparse warnings copying of struct wmi_mac_addr
from form &cmd->peer_macaddr.addr to cmd->peer_macaddr.addr. In
ath10k_wmi_set_ap_ps_param() I also added the missing ".addr" to the copy
command.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: reformat help text in ath10k_read_simulate_fw_crash()
Kalle Valo [Sun, 14 Sep 2014 09:50:44 +0000 (12:50 +0300)]
ath10k: reformat help text in ath10k_read_simulate_fw_crash()

Makes it more readable and fixes checkpatch warnings:

drivers/net/wireless/ath/ath10k/debug.c:593: WARNING: quoted string split across lines
drivers/net/wireless/ath/ath10k/debug.c:594: WARNING: quoted string split across lines
drivers/net/wireless/ath/ath10k/debug.c:595: WARNING: quoted string split across lines
drivers/net/wireless/ath/ath10k/debug.c:596: WARNING: quoted string split across lines
drivers/net/wireless/ath/ath10k/debug.c:597: WARNING: quoted string split across lines
drivers/net/wireless/ath/ath10k/debug.c:599: WARNING: quoted string split across lines
drivers/net/wireless/ath/ath10k/debug.c:600: WARNING: quoted string split across lines

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: miscellaneous checkpatch fixes
Kalle Valo [Sun, 14 Sep 2014 09:50:39 +0000 (12:50 +0300)]
ath10k: miscellaneous checkpatch fixes

Fixes checkpatch warnings:

ath10k/htc.c:49: WARNING: Possible unnecessary 'out of memory' message
ath10k/htc.c:810: WARNING: Possible unnecessary 'out of memory' message
ath10k/htt.h:1034: CHECK: Please use a blank line after function/struct/union/enum declarations
ath10k/htt_rx.c:135: CHECK: Unnecessary parentheses around htt->rx_ring.alloc_idx.vaddr
ath10k/htt_rx.c:173: CHECK: Unnecessary parentheses around htt->rx_ring.alloc_idx.vaddr
ath10k/pci.c:633: WARNING: macros should not use a trailing semicolon
ath10k/wmi.c:3594: WARNING: quoted string split across lines

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: else is not generally useful after a break or return
Kalle Valo [Sun, 14 Sep 2014 09:50:33 +0000 (12:50 +0300)]
ath10k: else is not generally useful after a break or return

Fixes checkpatch warnings:

WARNING: else is not generally useful after a break or return

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: don't use return on void functions
Kalle Valo [Sun, 14 Sep 2014 09:50:28 +0000 (12:50 +0300)]
ath10k: don't use return on void functions

Fixes a checkpatch warning:

WARNING: void function return statements are not generally useful

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix space after a cast style errors
Kalle Valo [Sun, 14 Sep 2014 09:50:23 +0000 (12:50 +0300)]
ath10k: fix space after a cast style errors

Fixes checkpatch warnings:

CHECK: No space is necessary after a cast

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix missing a blank line after declarations
Kalle Valo [Sun, 14 Sep 2014 09:50:17 +0000 (12:50 +0300)]
ath10k: fix missing a blank line after declarations

Fixes checkpatch warnings:

WARNING: Missing a blank line after declarations

Please note that some of the cases I fixed by moving the variable declarations
to the beginning of the function, which is the preferred style in ath10k.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix use of multiple blank lines
Kalle Valo [Sun, 14 Sep 2014 09:50:11 +0000 (12:50 +0300)]
ath10k: fix use of multiple blank lines

Fixes checkpatch warnings:

CHECK: Please don't use multiple blank lines

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix checkpatch warnings about parenthesis alignment
Kalle Valo [Sun, 14 Sep 2014 09:50:06 +0000 (12:50 +0300)]
ath10k: fix checkpatch warnings about parenthesis alignment

CHECK: Alignment should match open parenthesis

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix parenthesis alignment warning in ath10k_htt_rx_alloc()
Kalle Valo [Sun, 14 Sep 2014 09:50:00 +0000 (12:50 +0300)]
ath10k: fix parenthesis alignment warning in ath10k_htt_rx_alloc()

Fixes checkpatch warning:

ath10k/htt_rx.c:519: CHECK: Alignment should match open parenthesis

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix debugfs_create_dir() checking
Michal Kazior [Thu, 4 Sep 2014 10:36:45 +0000 (12:36 +0200)]
ath10k: fix debugfs_create_dir() checking

The function may return an -ENODEV if debugfs is
disabled in kernel. This should originally be
guarded by ath10k's Kconfig but it still makes
sense to check for the non-NULL errno return
value.

Reported-by: Matteo Croce <matteo@openwrt.org>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: support firmware crash-by-assert
Ben Greear [Wed, 10 Sep 2014 15:59:28 +0000 (18:59 +0300)]
ath10k: support firmware crash-by-assert

10.1 firmware does not have an official way to
cause assert on purpose, but it can be done with
carefully crafted WMI command.  This is a different
kind of crash from the 'hard' crash, which is
a bad memory dereference.

Different crashes decode in different manners, so
this will help the crash-report testing as well as
offer better ways to test firmware failure and
recovery.

kvalo: move the wmi command creation to debug.c, modify
the info print

Signed-off-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add testmode
Kalle Valo [Wed, 10 Sep 2014 15:23:30 +0000 (18:23 +0300)]
ath10k: add testmode

Add testmode interface for starting and using UTF firmware which is used to run
factory tests. This is implemented by adding new state ATH10K_STATE_UTF and user
space can enable this state with ATH10K_TM_CMD_UTF_START command. To go back to
normal mode user space can send ATH10K_TM_CMD_UTF_STOP.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: make ath10k_wmi_cmd_send() public
Kalle Valo [Wed, 10 Sep 2014 15:23:23 +0000 (18:23 +0300)]
ath10k: make ath10k_wmi_cmd_send() public

We need this function to send wmi packets from testmode.c.

Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: ATH10K_DEBUGFS depends on DEBUG_FS
Matteo Croce [Thu, 4 Sep 2014 12:51:09 +0000 (14:51 +0200)]
ath10k: ATH10K_DEBUGFS depends on DEBUG_FS

ATH10K_DEBUGFS must depend on DEBUG_FS, otherwise
ath10k will generate an invalid pointer on module load.

Signed-off-by: Matteo Croce <matteo@openwrt.org>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: use proper service bitmap size
Michal Kazior [Thu, 4 Sep 2014 08:18:32 +0000 (10:18 +0200)]
ath10k: use proper service bitmap size

On 32bit systems the bitmap was too small and it
was overwritten partially by the stat completion
structure. This was visible with 10.2 firmware
only due to it using a few of the last service
ids.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: move fw_crash_dump allocation
Michal Kazior [Thu, 4 Sep 2014 07:13:08 +0000 (09:13 +0200)]
ath10k: move fw_crash_dump allocation

The fw_crash_data was allocated too late. Upon
early firmware crash, before registering to
mac80211, it was possible to crash the whole
system:

 ath10k_pci 0000:00:05.0: device has crashed during init
 BUG: unable to handle kernel NULL pointer dereference at           (null)
 IP: [<ffffffffa0058005>] ath10k_debug_get_new_fw_crash_data+0x15/0x30 [ath10k_core]
 PGD 0
 Oops: 0002 [#1] SMP
 Modules linked in: ath10k_pci(O) ath10k_core(O) ath [last unloaded: ath]
 CPU: 3 PID: 29 Comm: kworker/u8:1 Tainted: G           O   3.17.0-rc2-wl-ath+ #447
 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
 Workqueue: ath10k_wq ath10k_core_register_work [ath10k_core]
 task: ffff88001eb01ad0 ti: ffff88001eb60000 task.ti: ffff88001eb60000
 RIP: 0010:[<ffffffffa0058005>]  [<ffffffffa0058005>] ath10k_debug_get_new_fw_crash_data+0x15/0x30 [ath10k_core]
 RSP: 0018:ffff88001eb63ce8  EFLAGS: 00010246
 RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
 RDX: 0000000000000000 RSI: ffffc90001a09030 RDI: 0000000000000001
 RBP: ffff88001eb63cf0 R08: 0000000000000000 R09: ffff8800000bb200
 R10: 00000000000001e2 R11: ffff88001eb638de R12: ffff88001d7459a0
 R13: ffff88001d746ab0 R14: 00000000fffe14d4 R15: ffff88001d747c60
 FS:  0000000000000000(0000) GS:ffff88001fd80000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
 CR2: 0000000000000000 CR3: 000000001df34000 CR4: 00000000000006e0
 Stack:
  ffff88001d7459a0 ffff88001eb63d58 ffffffffa0083bbe ffff880000000010
  ffff88001eb63d68 ffff88001eb63d18 0000000000000002 0000000000059010
  ffffffffa0086fef 00000000deadbeef ffff88001d747a28 ffff88001d7459a0
 Call Trace:
  [<ffffffffa0083bbe>] ath10k_pci_fw_crashed_dump+0x2e/0xd0 [ath10k_pci]
  [<ffffffffa0085410>] __ath10k_pci_hif_power_up+0x5f0/0x700 [ath10k_pci]
  [<ffffffffa0085550>] ath10k_pci_hif_power_up+0x30/0xe0 [ath10k_pci]
  [<ffffffffa005bc7b>] ath10k_core_register_work+0x2b/0x520 [ath10k_core]
  [<ffffffff810689cc>] process_one_work+0x18c/0x3f0
  [<ffffffff81069011>] worker_thread+0x121/0x4a0
  [<ffffffff81068ef0>] ? rescuer_thread+0x2c0/0x2c0
  [<ffffffff8106daf2>] kthread+0xd2/0xf0
  [<ffffffff8106da20>] ? kthread_create_on_node+0x170/0x170
  [<ffffffff81857cfc>] ret_from_fork+0x7c/0xb0
  [<ffffffff8106da20>] ? kthread_create_on_node+0x170/0x170
 Code: 8b 40 38 48 c7 80 00 01 00 00 00 00 00 00 5b 5d c3 0f 1f 44 00 00 0f 1f 44 00 00 55 48 89 e5 53 48 8b 9f 90 1d 00 00 48 8d 7b 01 <c6> 03 01 e8 e3 ec 2b e1 48 8d 7b 18 e8 6a 4f 05 e1 48 89 d8 5b
 RIP  [<ffffffffa0058005>] ath10k_debug_get_new_fw_crash_data+0x15/0x30 [ath10k_core]
  RSP <ffff88001eb63ce8>
 CR2: 0000000000000000
 ---[ end trace 5d0ed15b050bcc1f ]---
 Kernel panic - not syncing: Fatal exception in interrupt
 Kernel Offset: 0x0 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffff9fffffff)
 ---[ end Kernel panic - not syncing: Fatal exception in interrupt

To prevent that split debug functions and allocate
fw_crash_data earlier.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: don't access tx_info while overwriting it
Michal Kazior [Tue, 2 Sep 2014 08:00:22 +0000 (11:00 +0300)]
ath10k: don't access tx_info while overwriting it

Nothing important was being overwritten so it
didn't yield any bugs yet.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: add device/driver strings to tracepoints
Michal Kazior [Tue, 2 Sep 2014 08:00:21 +0000 (11:00 +0300)]
ath10k: add device/driver strings to tracepoints

This makes it easier to log and debug via tracing
with more than 1 ath10k device on a system.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: remove diag_*_access functions
Michal Kazior [Tue, 2 Sep 2014 08:00:21 +0000 (11:00 +0300)]
ath10k: remove diag_*_access functions

Remove the ugly _access functions. Being explicit
is a good thing.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: stop monitor vdev for sta assoc
Michal Kazior [Thu, 28 Aug 2014 10:58:17 +0000 (12:58 +0200)]
ath10k: stop monitor vdev for sta assoc

This prevents some fw revisions from crashing in
many cases when user is trying to run a
promiscuous station interface (e.g. sniffing,
4addr bridge).

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix monitor start/stop sequences
Michal Kazior [Thu, 28 Aug 2014 10:58:16 +0000 (12:58 +0200)]
ath10k: fix monitor start/stop sequences

Fix some cases where monitor start failure left
the driver in a confused state.

This also makes the monitor code simpler.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: kill tasklets after free_irq
Michal Kazior [Thu, 28 Aug 2014 08:24:40 +0000 (10:24 +0200)]
ath10k: kill tasklets after free_irq

Commit 5c771e7454d148af35e8b4297d00f880de79ea49
introduced a regression. On some systems spurious
interrupts could schedule a tasklet while tearing
down leading to, e.g.:

 BUG: unable to handle kernel paging request at fe589030
 IP: [<c1316fb0>] ioread32+0x30/0x40
 ...
 Call Trace:
  [<fe576c1b>] ath10k_pci_tasklet+0x1b/0x60 [ath10k_pci]
  [<c1053fbe>] tasklet_action+0x9e/0xb0
  [<c10534f1>] __do_softirq+0xf1/0x3f0
  [<c1053400>] ? ftrace_raw_event_irq_handler_entry+0xa0/0xa0
  [<c1004999>] do_softirq_own_stack+0x29/0x40
  <IRQ>
  [<c1053a76>] irq_exit+0x86/0xb0
 ...
  [<c132d522>] do_pci_disable_device+0x52/0x60
  [<c132d57f>] pci_disable_device+0x4f/0xb0
  [<c132a961>] ? __pci_set_master+0x51/0x80
  [<fe5740b3>] ath10k_pci_release+0x33/0x40 [ath10k_pci]
  [<fe575d4b>] ath10k_pci_remove+0x7b/0x90 [ath10k_pci]

Reported-by: Kalle Valo <kvalo@qca.qualcomm.com>
Tested-by: Kalle Valo <kvalo@qca.qualcomm.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: fix num_legacy_stations tracking
Michal Kazior [Thu, 28 Aug 2014 07:59:39 +0000 (09:59 +0200)]
ath10k: fix num_legacy_stations tracking

If a station was reassociated, i.e. due to change
of supported rates update via sta_rc_update() the
num_legacy_stations would be (incorrectly) bumped
up leading to unbalanced usage of the var. This in
turn could lock rtscts protection up as enabled.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agoath10k: re-enable interrupts properly in hw recovery
Michal Kazior [Thu, 28 Aug 2014 19:14:16 +0000 (22:14 +0300)]
ath10k: re-enable interrupts properly in hw recovery

Recent changes done to start/restart sequences
broke hw recovery in some hw configurations. The
pci transport was stopped twice however due to a
workaround in the pci disabling code the
disable/enable for first msi interrupt was not
balanced. This ended up with irqs not being
properly re-enabled and the following print out
during recovery:

 ath10k: failed to receive control response completion, polling..
 ath10k: Service connect timeout: -110
 ath10k: Could not init core: -110

Legacy interrupt mode was unaffected while msi
ranged mode would be partially crippled (it would
miss fw indication interrupts but otherwise it
worked fine).

This fixes completely broken fw recovery for a
single msi interrupt mode and fixes subsequent fw
crash reports for msi range interrupt mode.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
10 years agocarl9170: tx: Replace rcu_assign_pointer() with RCU_INIT_POINTER()
Andreea-Cristina Bernat [Wed, 27 Aug 2014 13:27:30 +0000 (16:27 +0300)]
carl9170: tx: Replace rcu_assign_pointer() with RCU_INIT_POINTER()

According to RCU_INIT_POINTER()'s block comment 3.a, it can be used if
"3.   The referenced data structure has already been exposed to readers either
at compile time or via rcu_assign_pointer() -and-
 a.   You have not made -any- reader-visible changes to this structure since
then".

This case fulfills the conditions above because between the rcu_dereference()
call (cvif = rcu_dereference(ar->beacon_iter);) and the rcu_assign_pointer()
call there is no update of the "cvif" variable.
Therefore, this patch makes the replacement.

The following Coccinelle semantic patch was used:
@@
identifier v;
@@

v = rcu_dereference(...);
... when != rcu_dereference(...);
    when != v = ...;
    when != (<+...v...+>)++;
    when != \(memcpy\|memset\)(...);
(
- rcu_assign_pointer
+ RCU_INIT_POINTER
  (..., v);
|
 if(...) {
... when != v = ...;
- rcu_assign_pointer
+ RCU_INIT_POINTER
  (..., v);
... when any
 }
)

Because there are cases where between a “rcu_dereference()” call and a
“rcu_assign_pointer()” call might be updates of the value that interests us,
the Coccinelle semantic patch ignores them and replaces with
"RCU_INIT_POINTER()" only when the update is not happening.

Signed-off-by: Andreea-Cristina Bernat <bernat.ada@gmail.com>
Acked-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
10 years agoath9k: Fix channel context timer
Sujith Manoharan [Wed, 27 Aug 2014 06:37:25 +0000 (12:07 +0530)]
ath9k: Fix channel context timer

Setup the channel context correctly. Not doing this was
causing beacon loss in a P2P-GO/STA concurrent setup.

Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
10 years agoath9k: Add more debug statements for channel context
Sujith Manoharan [Wed, 27 Aug 2014 06:37:24 +0000 (12:07 +0530)]
ath9k: Add more debug statements for channel context

Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
10 years agoath9k: Disable fastcc for channel context mode
Sujith Manoharan [Wed, 27 Aug 2014 06:37:23 +0000 (12:07 +0530)]
ath9k: Disable fastcc for channel context mode

Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
10 years agort2x00: support Ralink 5362.
Canek Peláez Valdés [Mon, 25 Aug 2014 00:06:11 +0000 (19:06 -0500)]
rt2x00: support Ralink 5362.

Signed-off-by: Canek Peláez Valdés <canek@ciencias.unam.mx>
Signed-off-by: John W. Linville <linville@tuxdriver.com>