Skip to content

L
linux
Switch branch/tag
  1. 09 Sep, 2015 2 commits
    • Vasanthakumar Thiagarajan's avatar
      ath10k: fix DMA alloc failure for target requested memory chunks · a925a376
      Vasanthakumar Thiagarajan authored 
      During long hours of stress testing like AP interface up/down along
with continuous ping flood from a station doing connect/disconnect,
it is observed that the system is not able to allocate DMA consistent
memory of size > 512KB chunks as requested by firmware in WMI_SERVICE_EVENTID.
With the system memory getting fragmented during the run based on the
size of the memory requested, the failure to return physically continguous
memory of high order can happen. Once the system gets to this situation,
bringing up the wifi interface will fail and a system reboot may be needed
to make it work again. This problem is obseved with QCA99X0.

To fix this issue, allocate the DMA memory requested by firmware during
device probe time and keep it during the life time of the device. WMI service
ready event handler is changed to allocate the memory chunks if it is
not already allocated or if the memory allocated for the previous ready
event is not same as the current requested ones. After this patch the
memory usage when wifi is inactive will be inceased by few 100KB to
3MB based on the target type.

Failure happens with the following stack trace

[29557.488773] kworker/u4:1: page allocation failure: order:8, mode:0xd0
[29557.494297] CPU: 0 PID: 8402 Comm: kworker/u4:1 Not tainted 3.14.43 #7
[29557.500793] Workqueue: ath10k_aux_wq ath10k_wmi_event_service_ready_work [ath10k_core]
[29557.508602] [<c021e9b0>] (unwind_backtrace) from [<c021ba90>] (show_stack+0x10/0x14)
[29557.516580] [<c021ba90>] (show_stack) from [<c03bdddc>] (dump_stack+0x88/0xcc)
[29557.523612] [<c03bdddc>] (dump_stack) from [<c0290e34>] (warn_alloc_failed+0xdc/0x108)
[29557.531515] [<c0290e34>] (warn_alloc_failed) from [<c0292d88>] (__alloc_pages_nodemask+0x4f0/0x654)
[29557.540485] [<c0292d88>] (__alloc_pages_nodemask) from [<c0222b48>] (__dma_alloc_buffer.isra.20+0x2c/0x104)
[29557.550260] [<c0222b48>] (__dma_alloc_buffer.isra.20) from [<c0222c34>] (__alloc_remap_buffer.isra.23+0x14/0xb8)
[29557.560413] [<c0222c34>] (__alloc_remap_buffer.isra.23) from [<c022305c>] (__dma_alloc+0x224/0x2b8)
[29557.569490] [<c022305c>] (__dma_alloc) from [<c0223208>] (arm_dma_alloc+0x84/0x90)
[29557.577010] [<c0223208>] (arm_dma_alloc) from [<bf5159d0>] (ath10k_wmi_event_service_ready_work+0x2f8/0x420 [ath10k_core])
[29557.588055] [<bf5159d0>] (ath10k_wmi_event_service_ready_work [ath10k_core]) from [<c024260c>] (process_one_work+0x20c/0x328)
[29557.599305] [<c024260c>] (process_one_work) from [<c02432d0>] (worker_thread+0x228/0x360)
[29557.607470] [<c02432d0>] (worker_thread) from [<c0247f88>] (kthread+0xd8/0xec)
[29557.614750] [<c0247f88>] (kthread) from [<c0208d18>] (ret_from_fork+0x14/0x3c)
[29557.712751] Normal: 696*4kB (UEMR) 512*8kB (UEMR) 367*16kB (UEMR) 404*32kB (UEMR) 455*64kB (UEMR) 424*128kB (UEMR) 379*256kB (UMR) 327*512kB (UMR) 1*1024kB (R) 0*2048kB 0*4096kB = 374544kB
Signed-off-by: default avatarVasanthakumar Thiagarajan <vthiagar@qti.qualcomm.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      a925a376
    • Michal Kazior's avatar
      ath10k: fix mu-mimo rx status reporting · 6986fdd6
      Michal Kazior authored 
      MU-MIMO Rx involves different interpretation of
the VHT-SIG-A compared to SU-MIMO.

The incorrect interpretation led ath10k to report
VHT MCS values greater than 9 which subsequently
prompted mac80211 to drop such frames. This
effectively broke Rx with MU-MIMO in many cases
and manifested with a kernel warning in the log
which looked like this:

  [   14.552520] WARNING: CPU: 2 PID: 0 at net/mac80211/rx.c:3578 ieee80211_rx+0x26c/0x940 [mac80211]()
  [   14.552522] Rate marked as a VHT rate but data is invalid: MCS: 10, NSS: 2
  ... call trace follows ...
Signed-off-by: default avatarMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      6986fdd6
  3. 26 Aug, 2015 5 commits
    • Raja Mani's avatar
      ath10k: fix compilation warnings in wmi phyerr pull function · ee92a209
      Raja Mani authored 
      Below compilation warnings are observed in gcc version 4.8.2.
Even though it's not seen in bit older gcc versions (for ex, 4.7.3),
It's good to fix it by changing format specifier from %d to
%zd in wmi pull phyerr functions.

wmi.c: In function 'ath10k_wmi_op_pull_phyerr_ev':
wmi.c:3567:8: warning: format '%d' expects argument of type 'int',
              but argument 4 has type 'long unsigned int' [-Wformat=]
              left_len, sizeof(*phyerr));
                        ^
wmi.c: In function 'ath10k_wmi_10_4_op_pull_phyerr_ev':
wmi.c:3612:8: warning: format '%d' expects argument of type 'int',
	      but argument 4 has type 'long unsigned int' [-Wformat=]
              left_len, sizeof(*phyerr));
                        ^
Fixes: 991adf71 ("ath10k: refactor phyerr event handlers")
Fixes: 2b0a2e0d ("ath10k: handle 10.4 firmware phyerr event")
Signed-off-by: default avatarRaja Mani <rmani@qti.qualcomm.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      ee92a209
    • Michal Kazior's avatar
      ath10k: add qca6164 support · 36582e5d
      Michal Kazior authored 
      This adds additional 0x0041 PCI Device ID
definition to ath10k for QCA6164 which is a 1
spatial stream sibling of the QCA6174 (which is 2
spatial stream chip).

The QCA6164 needs a dedicated board.bin file which
is different than the one used for QCA6174. If the
board.bin is wrong the device will crash early
while trying to boot firmware. The register dump
will look like this:

 ath10k_pci 0000:02:00.0: firmware register dump:
 ath10k_pci 0000:02:00.0: [00]: 0x05010000 0x000015B3 0x000A012D 0x00955B31
 ...

Note the value 0x000A012D.

Special credit goes to Alan Liu
<alanliu@qca.qualcomm.com> for providing support
help which enabled me to come up with this patch.
Signed-off-by: default avatarMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      36582e5d
    • Raja Mani's avatar
      ath10k: add spectral scan support for 10.4 fw · 4535edbd
      Raja Mani authored 
      To enable/configure spectral scan parameters in 10.4 firmware, existing
wmi spectral related functions can be reused. Link those functions in
10.4 wmi ops table.

In addition, adjust bin size (only when size is 68 bytes) before reporting
bin samples to user space. The background for this adjustment is that
qca99x0 reports bin size as 68 bytes (64 bytes + 4 bytes) in report
mode 2. First 64 bytes carries in-band tones (-32 to +31) and last 4 byte
carries band edge detection data (+32) mainly used in radar detection
purpose. Additional last 4 bytes are stripped to make bin size valid one.

This bin size adjustment will happen only for qca99x0, all other chipsets
will report proper bin sizes (64/128) without extra 4 bytes being added
at the end. The changes are validated in qca99x0 using 10.4 firmware.
Signed-off-by: default avatarRaja Mani <rmani@qti.qualcomm.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      4535edbd
    • Michal Kazior's avatar
      ath10k: fix dma_mapping_error() handling · 5e55e3cb
      Michal Kazior authored 
      The function returns 1 when DMA mapping fails. The
driver would return bogus values and could
possibly confuse itself if DMA failed.

Fixes: 767d34fc ("ath10k: remove DMA mapping wrappers")
Reported-by: default avatarDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: default avatarMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      5e55e3cb
    • Michal Kazior's avatar
      ath10k: add missing mutex unlock on failpath · 503422d9
      Michal Kazior authored 
      Kernel would complain about leaving a held lock
after going back to userspace and would
subsequently deadlock.

Fixes: e04cafbc ("ath10k: fix peer limit enforcement")
Reported-by: default avatarDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: default avatarMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      503422d9
  5. 19 Aug, 2015 2 commits
  7. 17 Aug, 2015 10 commits
    • Vasanthakumar Thiagarajan's avatar
      ath10k: fill in wmi 10.4 command handlers for addba/delba debug commands · b2887410
      Vasanthakumar Thiagarajan authored 
      WMI 10.4 uses the same command interface as QCA988X for addba/delba
debug wmi commands. Fill wmi_10_4_ops table with the functions used
for QCA988X for these commands.

With this change, the following debugfs entries can be used to
configure the aggregation mode and to send addba request,
addba response and delba respectively in manual aggregation mode
for QCA99X0 chip.

/sys/kernel/debug/ieee80211/phyX/netdev:wlanX/stations/XX:XX:XX:XX:XX:XX/aggr_mode
/sys/kernel/debug/ieee80211/phyX/netdev:wlanX/stations/XX:XX:XX:XX:XX:XX/addba
/sys/kernel/debug/ieee80211/phyX/netdev:wlanX/stations/XX:XX:XX:XX:XX:XX/addba_resp
/sys/kernel/debug/ieee80211/phyX/netdev:wlanX/stations/XX:XX:XX:XX:XX:XX/delba
Signed-off-by: default avatarVasanthakumar Thiagarajan <vthiagar@qti.qualcomm.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      b2887410
    • Raja Mani's avatar
      ath10k: handle 10.4 firmware phyerr event · 2b0a2e0d
      Raja Mani authored 
      Header format of 10.4 firmware phyerr event is not alligned
with pre 10.4 firmware. Introduce new wmi handlers to parse
10.4 firmware specific phyerror event header.

With changes covered in this patch, radar detection works on
qca9x0 hw 2.0 which uses 10.4 firmware.
Signed-off-by: default avatarRaja Mani <rmani@qti.qualcomm.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      2b0a2e0d
    • Raja Mani's avatar
      ath10k: refactor phyerr event handlers · 991adf71
      Raja Mani authored 
      Existing phyerr event handlers directly uses phyerr header format
(ie, struct wmi_phyerr and struct wmi_phyerr_event) in the code
exactly on how firmware packs it. This is the problem in 10.4 fw
specific phyerr event handling where it uses different phyerror
header format. Before adding 10.4 specific handler, little bit of
refactor is done in existing phyerr handlers.

Two new abstracted structures (struct wmi_phyerr_ev_hdr_arg and
struct wmi_phyerr_ev_arg) are introduced to remove dependency of using
firmware specific header format in the code. So that firmware specific
phyerror handlers can populate values to abstracted structures and
the following code can use abstracted struct for further operation.

.pull_phyerr_hdr is added newly to pull common phyerr header info
like tsf, buf_len, number of phyerr packed. Existing .pull_phyerr
handler is changed and called to parse every sub phyerrs in the event.

Validated these refactoring on qca988x hw2.0 using fw 10.2.4 version.
Signed-off-by: default avatarRaja Mani <rmani@qti.qualcomm.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      991adf71
    • Nicholas Mc Guire's avatar
      wil6210: match wait_for_completion_timeout return type · f4bbb829
      Nicholas Mc Guire authored 
      Return type of wait_for_completion_timeout is unsigned long not int.
As remain is exclusively used for wait_for_completion_timeout here its
type is simply changed to unsigned long.

API conformance testing for completions with coccinelle spatches are being
used to locate API usage inconsistencies:
./drivers/net/wireless/ath/wil6210/wmi.c:827
	int return assigned to unsigned long

Patch was compile tested with x86_64_defconfig + CONFIG_ATH_CARDS=m,
CONFIG_WIL6210=m

Patch is against 4.1-rc3 (localversion-next is -next-20150514)
Signed-off-by: default avatarNicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      f4bbb829
    • Michal Kazior's avatar
      ath10k: split ap/ibss wep key install process · 8674d909
      Michal Kazior authored 
      Apparently it's not safe to install both pairwise
and groupwise keys on AP vdevs as it can cause
traffic to stop working in some multi-vif
(WPA+WEP) cases.

Fixes: ce90b271 ("ath10k: fix multiple key static wep with ibss")
Signed-off-by: default avatarMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      8674d909
    • Vasanthakumar Thiagarajan's avatar
      ath10k: add cycle/rx_clear counters frequency to hw_params · 9c8fb548
      Vasanthakumar Thiagarajan authored 
      The frequency at which cycle/rx_clear counters are running might
change from one target type to another. QCA99X0 is running the
counters at 150Mhz while QCA9888X and QCA6174 are running at 88Mhz.
Add a new entry to hw_params to store the target specific frequency
and use it in msecs conversion. This change fixes inconsistent
channel active/busy time.
Signed-off-by: default avatarVasanthakumar Thiagarajan <vthiagar@qti.qualcomm.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      9c8fb548
    • Vasanthakumar Thiagarajan's avatar
      ath10k: fix invalid survey reporting for QCA99X0 · 3d2a2e29
      Vasanthakumar Thiagarajan authored 
      There are three WMI_CHAN_INFO events reported per channel
in QCA99X0 firmware. First one is a notification at the begining
of the channel dwell time with cmd_flag as CHAN_INFO_START(cmd_flag = 0),
second one is a notification at the end of the dwell time with cmd_flag
CHAN_INFO_PRE_COMPLETE (cmd_flag = 2) and the third is the indication
with CHAN_INFO_COMPLETE (cmd_flag = 1) which is the last indication for
the channel. Since there is a new state before the completion, the handler
is to fixed so that the counts are deducted from the ones reported with
CHAN_INFO_START rather than the ones reported with CHAN_INFO_PRE_COMPLETE.
Without this fix there will be lots of 0 msecs reported as active
and busy time.
Signed-off-by: default avatarVasanthakumar Thiagarajan <vthiagar@qti.qualcomm.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      3d2a2e29
    • Michal Kazior's avatar
      ath10k: wake up queue upon vif creation · 6d2d51ec
      Michal Kazior authored 
      Vif's vdev_id is used as queue number. However due
to the tx pausing design in ath10k it was possible
for a new interface to be created with its tx
queue stopped (via ieee80211_stop_queues). This
could in turn leave the interface inoperable until
ath10k_mac_tx_unlock() was called.

This problem only affected multi-vif scenarios when
new interfaces were created some time later after
other interfaces have been running for some time
and had Tx queue full at some point prior.

Possible manifestation of the bug was
authentication timeout for a client vif.

Fixes: 96d828d4 ("ath10k: rework tx queue locking")
Signed-off-by: default avatarMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      6d2d51ec
    • Michal Kazior's avatar
      ath10k: wake up offchannel queue properly · 3a73d1a6
      Michal Kazior authored 
      Once HTT Tx queue got full offchannel queue was
stopped and never woken up again. This broke, e.g.
P2P. This could be reproduced after running a lot
of traffic enough to saturate 100% of the driver
Tx queue and then trying to send offchannel
traffic.

Fixes: 96d828d4 ("ath10k: rework tx queue locking")
Signed-off-by: default avatarMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: default avatarKalle Valo <kvalo@qca.qualcomm.com>
      3a73d1a6
    • Kalle Valo's avatar
      Merge ath-next from ath.git. Major changes in ath10k: · 94e92a7b
      Kalle Valo authored 
      * add support for qca99x0 family of devices
* improve performance of tx_lock
* add support for raw mode (802.11 frame format) and software crypto
  engine enabled via a module parameter

wil6210:

* implement TSO support
* support bootloader v1 and onwards
      94e92a7b
  9. 13 Aug, 2015 21 commits
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7