openwrt/staging/blogic.git
14 years agoath9k_hw: remove ATH9K_CAP_CIPHER
Felix Fietkau [Sat, 12 Jun 2010 15:22:28 +0000 (17:22 +0200)]
ath9k_hw: remove ATH9K_CAP_CIPHER

All of the ciphers that are tested for are always supported

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: enable AR9003 PCI IDs
Luis R. Rodriguez [Sat, 12 Jun 2010 04:34:02 +0000 (00:34 -0400)]
ath9k: enable AR9003 PCI IDs

All AR9003 features are now complete so enable AR9003
support.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: implement PA predistortion support
Felix Fietkau [Sat, 12 Jun 2010 04:34:01 +0000 (00:34 -0400)]
ath9k: implement PA predistortion support

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: add functions for controlling PA predistortion calibration
Felix Fietkau [Sat, 12 Jun 2010 04:34:00 +0000 (00:34 -0400)]
ath9k_hw: add functions for controlling PA predistortion calibration

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: add support for parsing PA predistortion related EEPROM fields
Felix Fietkau [Sat, 12 Jun 2010 04:33:59 +0000 (00:33 -0400)]
ath9k_hw: add support for parsing PA predistortion related EEPROM fields

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: add register definitions related to PA predistortion
Felix Fietkau [Sat, 12 Jun 2010 04:33:58 +0000 (00:33 -0400)]
ath9k_hw: add register definitions related to PA predistortion

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: remove a useless function for setting the mac address
Felix Fietkau [Sat, 12 Jun 2010 04:33:57 +0000 (00:33 -0400)]
ath9k_hw: remove a useless function for setting the mac address

ath9k_hw_setmac() only copies the mac address it is called with into
common->macaddr, yet in all call sites, the supplied mac address pointer
is already common->macaddr.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: add a debugfs entry for ignoring CCA on the extension channel in HT40
Felix Fietkau [Sat, 12 Jun 2010 04:33:56 +0000 (00:33 -0400)]
ath9k: add a debugfs entry for ignoring CCA on the extension channel in HT40

Debugfs requires a u32 for bool knobs though so we turn the
ath9k_hw knob into a u32 as well.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: fix queue stopping threshold
Felix Fietkau [Sat, 12 Jun 2010 04:33:55 +0000 (00:33 -0400)]
ath9k: fix queue stopping threshold

ath9k tries to prevent WMM queue tx buffer starvation caused by
traffic on different queues by limiting the number of pending frames
in a tx queue (tracked in the ath_buf structure). This had a leak
issue, because the a skb can be reassigned to a different ath_buf
in the tx path, causing the pending frame counter to become inaccurate.

To fix this, track the number of frames in an array in the softc,
using the mac80211 queue mapping as index.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: fix extending the rx timestamp with the hardware TSF
Felix Fietkau [Sat, 12 Jun 2010 04:33:54 +0000 (00:33 -0400)]
ath9k: fix extending the rx timestamp with the hardware TSF

AR5416 and all newer chipsets use a 32 bit rx timestamp, so there
is no need to keep the 15 bit timestamp extending logic around.

This patch removes ath9k_hw_extend_tsf (replaced by a call to
ath9k_hw_gettsf64), and reduces the frequency of TSF reads, which
can improve performance in some cases.

This change also has the side effect of making rx timestamps
more accurate.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: make ath_get_hal_qnum static
Felix Fietkau [Sat, 12 Jun 2010 04:33:53 +0000 (00:33 -0400)]
ath9k: make ath_get_hal_qnum static

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: remove declarations of some nonexistant functions
Felix Fietkau [Sat, 12 Jun 2010 04:33:52 +0000 (00:33 -0400)]
ath9k: remove declarations of some nonexistant functions

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: remove duplicate WMM AC definitions
Felix Fietkau [Sat, 12 Jun 2010 04:33:51 +0000 (00:33 -0400)]
ath9k: remove duplicate WMM AC definitions

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_htc: use common WMM AC definitions instead of ath9k ones
Felix Fietkau [Sat, 12 Jun 2010 04:33:50 +0000 (00:33 -0400)]
ath9k_htc: use common WMM AC definitions instead of ath9k ones

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: fix mac80211 queue lookup for waking up queues
Felix Fietkau [Sat, 12 Jun 2010 04:33:49 +0000 (00:33 -0400)]
ath9k: fix mac80211 queue lookup for waking up queues

ath_get_mac80211_qnum() expects the queue 'subtype'
(internal ID for the WMM AC) as argument when looking up
the mac80211 queue, however ath_wake_mac80211_queue provides
txq->axq_qnum instead, which contains the hardware queue
number. Fix this by keeping track of the WMM class ID in
the txq data structure.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Cc: stable@kernel.org
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: update 5 GHz tx gain tables for femless and high power PA
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:48 +0000 (00:33 -0400)]
ath9k_hw: update 5 GHz tx gain tables for femless and high power PA

This updates the initvals for AR9003 to adjust the 5 GHz tx gain
tables for femless and high power PA.

References:

Osprey 2.0 header file ver 72
Osprey 2.2 header file ver 20

Checksums:

$ ./initvals -f ar9003-2p0
0x00000000c2bfa7d5        ar9300_2p0_radio_postamble
0x00000000ada2b114        ar9300Modes_lowest_ob_db_tx_gain_table_2p0
0x00000000e0bc2c84        ar9300Modes_fast_clock_2p0
0x00000000056eaf74        ar9300_2p0_radio_core
0x0000000000000000        ar9300Common_rx_gain_table_merlin_2p0
0x0000000078658fb5        ar9300_2p0_mac_postamble
0x0000000023235333        ar9300_2p0_soc_postamble
0x0000000054d41904        ar9200_merlin_2p0_radio_core
0x00000000748572cf        ar9300_2p0_baseband_postamble
0x000000009aa5a0a4        ar9300_2p0_baseband_core
0x000000003dffa526        ar9300Modes_high_power_tx_gain_table_2p0
0x000000001cfda724        ar9300Modes_high_ob_db_tx_gain_table_2p0
0x0000000011302700        ar9300Common_rx_gain_table_2p0
0x00000000e3eab114        ar9300Modes_low_ob_db_tx_gain_table_2p0
0x00000000c9d66d40        ar9300_2p0_mac_core
0x000000001e1d0800        ar9300Common_wo_xlna_rx_gain_table_2p0
0x00000000a0c54980        ar9300_2p0_soc_preamble
0x00000000292e2544        ar9300PciePhy_pll_on_clkreq_disable_L1_2p0
0x000000002d3e2544        ar9300PciePhy_clkreq_enable_L1_2p0
0x00000000293e2544        ar9300PciePhy_clkreq_disable_L1_2p0

$ ./initvals -f ar9003-2p2
0x00000000c2bfa7d5        ar9300_2p2_radio_postamble
0x00000000ada2b114        ar9300Modes_lowest_ob_db_tx_gain_table_2p2
0x00000000e0bc2c84        ar9300Modes_fast_clock_2p2
0x00000000056eaf74        ar9300_2p2_radio_core
0x0000000000000000        ar9300Common_rx_gain_table_merlin_2p2
0x0000000078658fb5        ar9300_2p2_mac_postamble
0x0000000023235333        ar9300_2p2_soc_postamble
0x0000000054d41904        ar9200_merlin_2p2_radio_core
0x000000008475a084        ar9300_2p2_baseband_postamble
0x000000009aaafd90        ar9300_2p2_baseband_core
0x000000003dffa526        ar9300Modes_high_power_tx_gain_table_2p2
0x000000001cfda724        ar9300Modes_high_ob_db_tx_gain_table_2p2
0x0000000011302700        ar9300Common_rx_gain_table_2p2
0x00000000a9a2b114        ar9300Modes_low_ob_db_tx_gain_table_2p2
0x00000000a9d66d40        ar9300_2p2_mac_core
0x000000001e1d0800        ar9300Common_wo_xlna_rx_gain_table_2p2
0x00000000a0c531c8        ar9300_2p2_soc_preamble
0x00000000292e2544        ar9300PciePhy_pll_on_clkreq_disable_L1_2p2
0x000000002d3e2544        ar9300PciePhy_clkreq_enable_L1_2p2
0x00000000293e2544        ar9300PciePhy_clkreq_disable_L1_2p2

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: reduce delay on programming INI on AR9003
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:47 +0000 (00:33 -0400)]
ath9k_hw: reduce delay on programming INI on AR9003

All AR9003 devices are PCI-E only, the extra delay here
is not required and only reduces the delay for loading
the initial register values by at least 14ms.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: enable ANI for AR9003
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:46 +0000 (00:33 -0400)]
ath9k_hw: enable ANI for AR9003

AR9003 has been tested with the new ANI implementation
and so ANI can now be enabled for that family.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k: add new ANI implementation for AR9003
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:45 +0000 (00:33 -0400)]
ath9k: add new ANI implementation for AR9003

This adds support for ANI for AR9003. The implementation for
ANI for AR9003 is slightly different than the one used for
the older chipset families. It can technically be used for
the older families as well but this is not yet fully tested
so we only enable the new ANI for the AR5008, AR9001 and AR9002
families with a module parameter, force_new_ani.

The old ANI implementation is left intact.

Details of the new ANI implemention:

  * ANI adjustment logic is now table driven so that each ANI level
    setting is parameterized. This makes adjustments much more
    deterministic than the old procedure based logic and allows
    adjustments to be made incrementally to several parameters per
    level.

  * ANI register settings are now relative to INI values; so ANI
    param zero level == INI value. Appropriate floor and ceiling
    values are obeyed when adjustments are combined with INI values.

  * ANI processing is done once per second rather that every 100ms.
    The poll interval is now a set upon hardware initialization and
    can be picked up by the core driver.

  * OFDM error and CCK error processing are made in a round robin
    fashion rather than allowing all OFDM adjustments to be made
    before CCK adjustments.

  * ANI adjusts MRC CCK off in the presence of high CCK errors

  * When adjusting spur immunity (SI) and OFDM weak signal detection,
    ANI now sets register values for the extension channel too

  * When adjusting FIR step (ST), ANI now sets register for FIR step
    low too

  * FIR step adjustments now allow for an extra level of immunity for
    extremely noisy environments

  * The old Noise immunity setting (NI), which changes coarse low, size
    desired, etc have been removed. Changing these settings could affect
    up RIFS RX as well.

  * CCK weak signal adjustment is no longer used

  * ANI no longer enables phy error interrupts; in all cases phy hw
    counting registers are used instead

  * The phy error count (overflow) interrupts are also no longer used
    for ANI adjustments. All ANI adjustments are made via the polling
    routine and no adjustments are possible in the ISR context anymore

  * A history settings buffer is now correctly used for each channel;
    channel settings are initialized with the defaults but later
    changes are restored when returning back to that channel

  * When scanning, ANI is disabled settings are returned to (INI) defaults.

  * OFDM phy error thresholds are now 400 & 1000 (errors/second units) for
    low/high water marks, providing increased stability/hysteresis when
    changing levels.

  * Similarly CCK phy error thresholds are now 300 & 600 (errors/second)

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: inform ANI calibration when scanning
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:44 +0000 (00:33 -0400)]
ath9k_hw: inform ANI calibration when scanning

The new ANI implementation will use this to skip ANI
calibration upon a scan. This cannot be ported to the
older ANI implementation unless default ANI values from
the ANI are also used upon a scan. This is essentially
what one of the things thenew ANI does.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: add register definitions for the new ANI
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:43 +0000 (00:33 -0400)]
ath9k_hw: add register definitions for the new ANI

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: allow for spliting up ANI operations by family
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:42 +0000 (00:33 -0400)]
ath9k_hw: allow for spliting up ANI operations by family

The AR9003 hardware family will use a slightly modified ANI
implementation which has not yet been tested on the other hardware
families. To allow for this new ANI implementation a few ANI
calls need to be abstracted away. This patch just allows for
each hardware family to declare their own ANI ops and annotates
the current ANI implementation as old.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: clear MIB interrupt causes when skipping ANI adjustments
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:41 +0000 (00:33 -0400)]
ath9k_hw: clear MIB interrupt causes when skipping ANI adjustments

We get an MIB interrupt when we hit certain PHY error counter
thresholds. If ANI is disabled but the MIB interrupt is enabled
we'll keep around the old MIB interrupt causes.

Since ath9k disables the MIB interrupt when ANI is disabled
this is not a fix, but more of a sanity fix in case we ever
need the MIB interrupt enabled but disabling ANI.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: fix clock rate calculations for ANI
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:40 +0000 (00:33 -0400)]
ath9k_hw: fix clock rate calculations for ANI

The clock rate was assumed to be static but it actually
changes depending on the mode of operation, correct this
to help improve the calcuation of the listenTime for ANI.
This change will help adjust ANI more accurately on different
PHY thresholds.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath9k_hw: move clock definitions from hw.c to hw.h
Luis R. Rodriguez [Sat, 12 Jun 2010 04:33:39 +0000 (00:33 -0400)]
ath9k_hw: move clock definitions from hw.c to hw.h

These will be used by the ANI code next.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: bracket driver tracing
Johannes Berg [Thu, 10 Jun 2010 08:56:20 +0000 (10:56 +0200)]
mac80211: bracket driver tracing

Currently, driver tracing is sometimes invoked
after and sometimes before the actual driver
callback. This is fine as long as the driver
has no tracing itself, but as soon as it does
it gets confusing.

To make traces containing such information
easier to read, introduce a return tracer in
mac80211 that essentially brackets any driver
tracing, and invoke the real trace before the
driver's callback, only showing the return
value, if any, afterwards.

Since tracing records the process, there's no
problem with overlapping calls if that should
happen.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: fix mgmt frame accounting
Johannes Berg [Thu, 10 Jun 2010 08:21:51 +0000 (10:21 +0200)]
mac80211: fix mgmt frame accounting

The recent change to processing action frames from
the management frame queue had already broken action
frame accounting, and my rework didn't help either.
So add back accounting and simplify the code with a
label rather than duplicating it, and also add
accounting for management frames.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: update aggregation documentation
Johannes Berg [Thu, 10 Jun 2010 08:21:50 +0000 (10:21 +0200)]
mac80211: update aggregation documentation

Even before the recent changes, the documentation
for TX aggregation was somewhat out of date. Update
it and also add documentation for the RX side.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: allow drivers to sleep in ampdu_action
Johannes Berg [Thu, 10 Jun 2010 08:21:49 +0000 (10:21 +0200)]
mac80211: allow drivers to sleep in ampdu_action

Allow drivers to sleep, and indicate this in
the documentation. ath9k has some locking I
don't understand, so keep it safe and disable
BHs in it, all other drivers look fine with
the context change.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: change TX aggregation locking
Johannes Berg [Thu, 10 Jun 2010 08:21:48 +0000 (10:21 +0200)]
mac80211: change TX aggregation locking

To prepare for allowing drivers to sleep in
ampdu_action, change the locking in the TX
aggregation code to use the mutex the RX part
already uses. The spinlock is still necessary
around some code to avoid races with TX, but
now we can also synchronize_net() to avoid
getting an inconsistent sequence number.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: defer TX agg session teardown to work
Johannes Berg [Thu, 10 Jun 2010 08:21:47 +0000 (10:21 +0200)]
mac80211: defer TX agg session teardown to work

Since we want the code to be able to sleep
in the future, it must not be called from
the timer directly. To achieve that, simply
call the function drivers would call, and
also use RCU in the timer to get the struct
so we don't need to rely on the spinlock in
the future.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: change RX aggregation locking
Johannes Berg [Thu, 10 Jun 2010 08:21:46 +0000 (10:21 +0200)]
mac80211: change RX aggregation locking

To prepare for allowing drivers to sleep in
ampdu_action, change the locking in the RX
aggregation code to use a mutex, so that it
would already allow drivers to sleep. But
explicitly disable BHs around the callback
for now since the TX part cannot yet sleep,
and drivers' locking might require it.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: fix RX aggregation timer
Johannes Berg [Thu, 10 Jun 2010 08:21:45 +0000 (10:21 +0200)]
mac80211: fix RX aggregation timer

I noticed that when there was _no_ traffic at
all on a given aggregation session, it would
never time out. This won't happen unless you
forced creating a session, but fix it anyway.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: defer RX agg session teardown to work
Johannes Berg [Thu, 10 Jun 2010 08:21:44 +0000 (10:21 +0200)]
mac80211: defer RX agg session teardown to work

Since we want the code to be able to sleep
in the future, it must not be called from
the timer directly. To prepare, move it out
into the aggregation work.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: move BA session work
Johannes Berg [Thu, 10 Jun 2010 08:21:43 +0000 (10:21 +0200)]
mac80211: move BA session work

Move the block-ack session works into common
code, since it will be needed for RX agg too
in the next patches.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: make TX aggregation start/stop request async
Johannes Berg [Thu, 10 Jun 2010 08:21:42 +0000 (10:21 +0200)]
mac80211: make TX aggregation start/stop request async

When the driver or rate control requests starting
or stopping an aggregation session, that currently
causes a direct callback into the driver, which
could potentially cause locking problems. Also,
the functions need to be callable from contexts
that cannot sleep, and thus will interfere with
making the ampdu_action callback sleeping.

To address these issues, add a new work item for
each station that will process any start or stop
requests out of line.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: refcount aggregation queue stop
Johannes Berg [Thu, 10 Jun 2010 08:21:41 +0000 (10:21 +0200)]
mac80211: refcount aggregation queue stop

mac80211 currently maintains the ampdu_lock to
avoid starting a queue due to one aggregation
session while another aggregation session needs
the queue stopped.

We can do better, however, and instead refcount
the queue stops for this particular purpose,
thus removing the need for the lock. This will
help making ampdu_action able to sleep.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: remove non-irqsafe aggregation callbacks
Johannes Berg [Thu, 10 Jun 2010 08:21:40 +0000 (10:21 +0200)]
mac80211: remove non-irqsafe aggregation callbacks

The non-irqsafe aggregation start/stop done
callbacks are currently only used by ath9k_htc,
and can cause callbacks into the driver again.
This might lead to locking issues, which will
only get worse as we modify locking. To avoid
trouble, remove the non-irqsafe versions and
change ath9k_htc to use those instead.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: use RCU for TX aggregation
Johannes Berg [Thu, 10 Jun 2010 08:21:39 +0000 (10:21 +0200)]
mac80211: use RCU for TX aggregation

Currently we allocate some memory for each TX
aggregation session and additionally keep a
state bitmap indicating the state it is in.
By using RCU to protect the pointer, moving
the state into the structure and some locking
trickery we can avoid locking when the TX agg
session is fully operational.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: use RCU for RX aggregation
Johannes Berg [Thu, 10 Jun 2010 08:21:38 +0000 (10:21 +0200)]
mac80211: use RCU for RX aggregation

Currently we allocate some memory for each RX
aggregation session and additionally keep a
flag indicating whether or not it is valid.
By using RCU to protect the pointer and making
sure that the memory is fully set up before it
becomes visible to the RX path, we can remove
the need for the bool that indicates validity,
as well as for locking on the RX path since it
is always synchronised against itself, and we
can guarantee that all other modifications are
done when the structure is not visible to the
RX path.

The net result is that since we remove locking
requirements from the RX path, we can in the
future use any kind of lock for the setup and
teardown code paths.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: move aggregation callback processing
Johannes Berg [Thu, 10 Jun 2010 08:21:37 +0000 (10:21 +0200)]
mac80211: move aggregation callback processing

This moves the aggregation callback processing
to the per-sdata skb queue and a work function
rather than the tasklet.

Unfortunately, this means that it extends the
pkt_type hack to that skb queue. However, it
will enable making ampdu_action API changes
gradually, my current plan is to get rid of
this again by forcing drivers to only return
from ampdu_action() when everything is done,
thus removing the callbacks completely.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: move blockack stop due to fragmentation
Johannes Berg [Thu, 10 Jun 2010 08:21:36 +0000 (10:21 +0200)]
mac80211: move blockack stop due to fragmentation

There's a corner case where we receive a fragmented
frame during a blockack session, in which case we
will terminate that session. To simplify future work
in this area that will culminate in allowing the
driver callbacks for aggregation to sleep, move the
processing of this case out of the RX path into the
interface work.

This will simplify future work because the new place
for this code doesn't require that the function will
always be atomic, which the RX path needs.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: always process blockack action from workqueue
Johannes Berg [Thu, 10 Jun 2010 08:21:35 +0000 (10:21 +0200)]
mac80211: always process blockack action from workqueue

To prepare for making the ampdu_action callback
sleep, make mac80211 always process blockack
action frames from the skb queue. This gets rid
of the current special case for managed mode
interfaces as well.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: pull mgmt frame rx into rx handler
Johannes Berg [Thu, 10 Jun 2010 08:21:34 +0000 (10:21 +0200)]
mac80211: pull mgmt frame rx into rx handler

Some code is duplicated between ibss, mesh and
managed mode regarding the queueing of management
frames. Since all modes now use a common skb
queue and a common work function, we can pull
the queueing code into the rx handler directly
and remove the duplicated length checks etc.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: common work skb freeing
Johannes Berg [Thu, 10 Jun 2010 08:21:33 +0000 (10:21 +0200)]
mac80211: common work skb freeing

All the management processing functions free the
skb after they are done, so this can be done in
the new common code instead.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: use common work function
Johannes Berg [Thu, 10 Jun 2010 08:21:32 +0000 (10:21 +0200)]
mac80211: use common work function

Even with the previous patch, IBSS, managed
and mesh modes all attach their own work
function to the shared work struct, which
means some duplicated code. Change that to
only have a frame processing function and a
further work function for each of them and
share some common code.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: use common work struct
Johannes Berg [Thu, 10 Jun 2010 08:21:31 +0000 (10:21 +0200)]
mac80211: use common work struct

IBSS, managed and mesh modes all have their
own work struct, and in the future we want
to also use it in other modes to process
frames from the now common skb queue.

This also makes the skb queue and work safe
to use from other interface types.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: use common skb queue
Johannes Berg [Thu, 10 Jun 2010 08:21:30 +0000 (10:21 +0200)]
mac80211: use common skb queue

IBSS, managed and mesh modes all have an
skb queue, and in the future we want to
also use it in other modes, so make them
all use a common skb queue already.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: simplify station/aggregation code
Johannes Berg [Thu, 10 Jun 2010 08:21:29 +0000 (10:21 +0200)]
mac80211: simplify station/aggregation code

A number of places use RCU locking for accessing
the station list, even though they do not need
to. Use mutex locking instead to prepare for the
locking changes I want to make. The mlme code is
also using a WLAN_STA_DISASSOC flag that has the
same meaning as WLAN_STA_BLOCK_BA, so use that.

While doing so, combine places where we loop
over stations twice, and optimise away some of
the loops by checking if the hardware supports
aggregation at all first.

Also fix a more theoretical race condition: right
now we could resume, set up an aggregation session,
and right after tear it down again due to the code
that is needed for hardware reconfiguration here.
Also mark add a comment to that code marking it as
a workaround.

Finally, remove a pointless aggregation disabling
loop when an interface is stopped, directly after
that we remove all stations from it which will also
disable all aggregation sessions that may still be
active, and does so in a race-free way unlike the
current loop that doesn't block new sessions.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agocfg80211/mac80211: allow action frame TX/RX in IBSS
Johannes Berg [Wed, 9 Jun 2010 15:20:33 +0000 (17:20 +0200)]
cfg80211/mac80211: allow action frame TX/RX in IBSS

When in IBSS mode, currently action frame TX and RX
cannot be used. Allow using it to talk to any peer,
or for public action frames. Also, while at it,
restructure the code in mac80211 to make it easier
to add this for other interface types in the future.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoiwlwifi: print warning about disconnected antennas
Johannes Berg [Wed, 9 Jun 2010 08:46:32 +0000 (01:46 -0700)]
iwlwifi: print warning about disconnected antennas

When we detect that not all antennas are
properly connected, we simply disable the
associated chains, but never notify the
user at all. Print out a warning so it is
obvious that happened and we know where
to start looking for related issues.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: rename iwl4965_rx_mpdu_res_start
Emmanuel Grumbach [Mon, 7 Jun 2010 20:21:47 +0000 (13:21 -0700)]
iwlwifi: rename iwl4965_rx_mpdu_res_start

iwl4965_rx_mpdu_res_start is not 4695 specific, so rename it to more
general name iwl_rx_mpdu_res_start.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: rename rxq->dma_addr
Emmanuel Grumbach [Mon, 7 Jun 2010 20:21:46 +0000 (13:21 -0700)]
iwlwifi: rename rxq->dma_addr

Rename rxq->dma_addr to rxq->bd_dma to better emphasize that the
physical address stands for the receive buffer descriptor's address.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: cancel run time calibration work when going down
Wey-Yi Guy [Thu, 3 Jun 2010 17:14:01 +0000 (10:14 -0700)]
iwlwifi: cancel run time calibration work when going down

Cancel scheduled run time calibration work when interface is going
down.

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: Fix null pointer referencing in iwl_dbgfs_rx_queue_read.
Dor Shaish [Tue, 1 Jun 2010 07:04:08 +0000 (00:04 -0700)]
iwlwifi: Fix null pointer referencing in iwl_dbgfs_rx_queue_read.

Test for null pointer prior to access.
Print "Not Allocated" if null pointer.

Signed-off-by: Dor Shaish <dor.shaish@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: trace full RX
Johannes Berg [Fri, 28 May 2010 11:08:30 +0000 (04:08 -0700)]
iwlwifi: trace full RX

The length contained in the status word doesn't
include the status word's length itself, so we
need to account for that for tracing.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agowl1251: fix ELP_CTRL register reads
Grazvydas Ignotas [Tue, 8 Jun 2010 11:33:31 +0000 (14:33 +0300)]
wl1251: fix ELP_CTRL register reads

Reading the ELP_CTRL register with sdio_readb causes problems because
hardware seems to be performing a write using stuff bits in the request
(those bits contain write data in write request). This indicates that it
actually expects RAW (read after write) type of request, so perform that
when reading ELP_CTRL instead. Also cache last written value so we know
what to write when doing RAW request.

Because of the above it was not possible to wake the chip from ELP power
saving mode, PM had to be disabled to have the driver usable in SDIO
mode. After this patch PM is functional.

For backporting to 2.6.34 or earlier, this patch depends on
6c1f716e8154ee9315534782b9b1eedea0559a24, which adds the
required SDIO funcion.

Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Acked-by: Kalle Valo <kvalo@adurom.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agowireless: fix kernel-doc
Johannes Berg [Mon, 7 Jun 2010 09:12:27 +0000 (11:12 +0200)]
wireless: fix kernel-doc

Fix a whole bunch of kernel-doc warnings
and errors that crop up when running it on
mac80211 and cfg80211; the latter isn't
normally done so lots of bit-rot happened.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: Add netif state checking to ieee80211_ifa_changed
Juuso Oikarinen [Mon, 7 Jun 2010 07:52:12 +0000 (10:52 +0300)]
mac80211: Add netif state checking to ieee80211_ifa_changed

There's a window for ieee80211_ifa_changed() to get called whilst the
managed mode mutex has not been initialized when opening and stopping the
interface. Currently this causes a kernel BUG like the following:

[  132.460013] kernel BUG at /home/wifi/iwlwifi-2.6/net/mac80211/main.c:380!
[  132.460013] invalid opcode: 0000 [#1] SMP

The mutex is initialized during open(), hence once netif_running() is true,
the mutex should be valid. Fix by adding a netif_running() check to the
function.

Reported-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: Juuso Oikarinen <juuso.oikarinen@nokia.com>
Tested-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath5k: add debugfs file for queue debugging
Bruno Randolf [Mon, 7 Jun 2010 04:11:40 +0000 (13:11 +0900)]
ath5k: add debugfs file for queue debugging

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath5k: no need to save/restore the default antenna
Bruno Randolf [Mon, 7 Jun 2010 04:11:35 +0000 (13:11 +0900)]
ath5k: no need to save/restore the default antenna

Since ath5k_hw_set_antenna_mode() always writes the default antenna register
and is called at the end of reset, there is no need to separately save and
restore the default antenna.

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath5k: new function for setting the antenna switch table
Bruno Randolf [Mon, 7 Jun 2010 04:11:30 +0000 (13:11 +0900)]
ath5k: new function for setting the antenna switch table

Collect all pieces concering the antenna switch table into one function.
Previously it was split up between ath5k_hw_reset() and
ath5k_hw_commit_eeprom_settings().

Also we need to set the antenna switch table when ath5k_hw_set_antenna_mode()
is called manually (by "iw phy0 antenna set", for example).

I'm not sure if we need to set the switchtable at the same place in
ath5k_hw_reset() as it was before - it is set later thru
ath5k_hw_set_antenna_mode() anyways - but i leave it there to avoid
problems(?).

Plus print switchtable registers in the debugfs file.

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath5k: update AR5K_PHY_RESTART_DIV_GC values to match masks
Bruno Randolf [Mon, 7 Jun 2010 04:11:25 +0000 (13:11 +0900)]
ath5k: update AR5K_PHY_RESTART_DIV_GC values to match masks

#define AR5K_PHY_RESTART_DIV_GC               0x001c0000
is 3 bit wide.

The previous values of 0xc and 0x8 are 4bit wide and bigger than the mask.

Writing 0 and 1 to AR5K_PHY_RESTART_DIV_GC is consistent with the comments and
initvals we have in the HAL.

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agolibertas: Fix ethtool reporting no WOL options supported if WOL is not already active
Sascha Silbe [Sat, 5 Jun 2010 11:30:12 +0000 (13:30 +0200)]
libertas: Fix ethtool reporting no WOL options supported if WOL is not already active

This patch fixes the libertas driver incorrectly reporting that Wake-on-LAN
is not supported if Wake-on-LAN is currently disabled.

Signed-off-by: Sascha Silbe <sascha-pgp@silbe.org>
Acked-by: Dan Williams <dcbw@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath: Fix uninitialized variable warnings
Prarit Bhargava [Thu, 27 May 2010 18:14:54 +0000 (14:14 -0400)]
ath: Fix uninitialized variable warnings

Fixes 'make -j24 CONFIG_DEBUG_SECTION_MISMATCH=y' warning:

drivers/net/wireless/ath/ath9k/eeprom_4k.c: In function â€˜ath9k_hw_get_4k_gain_boundaries_pdadcs.clone.1’:
drivers/net/wireless/ath/ath9k/eeprom_4k.c:311: error: â€˜minPwrT4’ may be used uninitialized in this function
drivers/net/wireless/ath/ath9k/eeprom_9287.c: In function â€˜ath9k_hw_get_AR9287_gain_boundaries_pdadcs’:
drivers/net/wireless/ath/ath9k/eeprom_9287.c:302: error: â€˜minPwrT4’ may be used uninitialized in this function
drivers/net/wireless/ath/ath9k/eeprom_def.c: In function â€˜ath9k_hw_get_def_gain_boundaries_pdadcs.clone.0’:
drivers/net/wireless/ath/ath9k/eeprom_def.c:679: error: â€˜minPwrT4’ may be used uninitialized in this function

Signed-off-by: Prarit Bhargava <prarit@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
John W. Linville [Mon, 7 Jun 2010 19:13:46 +0000 (15:13 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

Conflicts:
drivers/net/wireless/iwlwifi/iwl-core.h

14 years agomac80211: fix lock leak w/ ARP filtering and w/o CONFIG_INET
John W. Linville [Mon, 7 Jun 2010 19:02:17 +0000 (15:02 -0400)]
mac80211: fix lock leak w/ ARP filtering and w/o CONFIG_INET

"mac80211: make ARP filtering depend on CONFIG_INET" introduced this
potential locking leak.

Reported-by: Juuso Oikarinen <juuso.oikarinen@nokia.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoiwlwifi: fix-up botched revert
John W. Linville [Mon, 7 Jun 2010 18:16:11 +0000 (14:16 -0400)]
iwlwifi: fix-up botched revert

In the revert of "iwlwifi: move _agn statistics related structure", I
need to use CONFIG_IWLWIFI_DEBUGFS instead of CONFIG_IWLWIFI_DEBUG in
the private structure definition.  Without this patch, it is possible
to get this:

drivers/net/wireless/iwlwifi/iwl-rx.c: In function 'iwl_accumulative_statistics':
drivers/net/wireless/iwlwifi/iwl-rx.c:304: error: 'struct iwl_priv' has no member named 'accum_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:305: error: 'struct iwl_priv' has no member named 'delta_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:306: error: 'struct iwl_priv' has no member named 'max_delta'
drivers/net/wireless/iwlwifi/iwl-rx.c:321: error: 'struct iwl_priv' has no member named 'accum_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:323: error: 'struct iwl_priv' has no member named 'accum_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:325: error: 'struct iwl_priv' has no member named 'accum_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:327: error: 'struct iwl_priv' has no member named 'accum_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:329: error: 'struct iwl_priv' has no member named 'accum_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:331: error: 'struct iwl_priv' has no member named 'accum_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c: In function 'iwl_reply_statistics':
drivers/net/wireless/iwlwifi/iwl-rx.c:484: error: 'struct iwl_priv' has no member named 'accum_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:486: error: 'struct iwl_priv' has no member named 'delta_statistics'
drivers/net/wireless/iwlwifi/iwl-rx.c:488: error: 'struct iwl_priv' has no member named 'max_delta'

Reported-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agomac80211: fix function pointer check
Holger Schurig [Mon, 7 Jun 2010 14:33:49 +0000 (16:33 +0200)]
mac80211: fix function pointer check

This makes "iw wlan0 dump survey" work again with
mac80211-based drivers that support it, e.g. ath5k.

Signed-off-by: Holger Schurig <holgerschurig@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agowireless: remove my name from the maintainer list
Zhu Yi [Mon, 7 Jun 2010 06:15:43 +0000 (14:15 +0800)]
wireless: remove my name from the maintainer list

Remove my name from the MAINTAINERS file.

Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoath5k: fix NULL pointer in antenna configuration
Bruno Randolf [Mon, 7 Jun 2010 04:11:19 +0000 (13:11 +0900)]
ath5k: fix NULL pointer in antenna configuration

If the channel is not set yet and we configure the antennas just store the
setting. It will be activated during the next reset, when the channel is set.

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agop54usb: Add device ID for Dell WLA3310 USB
Jason Dravet [Sat, 5 Jun 2010 20:08:29 +0000 (15:08 -0500)]
p54usb: Add device ID for Dell WLA3310 USB

Add Dell WLA3310 USB wireless card, which has a Z-Com XG-705A chipset, to the
USB Ids in p54usb.

Signed-off-by: Jason Dravet <dravet@hotmail.com>
Tested-by: Richard Gregory Tillmore <rtillmore@gmail.com>
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Acked-by: Christian Lamparter <chunkeey@googlemail.com>
Cc: <stable@kernel.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agowl1251: fix a memory leak in probe
Grazvydas Ignotas [Fri, 4 Jun 2010 23:25:47 +0000 (02:25 +0300)]
wl1251: fix a memory leak in probe

wl1251_sdio_probe() error path is missing wl1251_free_hw, add it.

Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Acked-by: Kalle Valo <kvalo@adurom.com>
Cc: stable@kernel.org
Signed-off-by: John W. Linville <linville@tuxdriver.com>
14 years agoiwlwifi: fix wrapping when handling aggregated batches
Daniel Halperin [Tue, 25 May 2010 17:22:49 +0000 (10:22 -0700)]
iwlwifi: fix wrapping when handling aggregated batches

Fairly complex code in iwlagn_tx_status_reply_tx handle the status reports for
aggregated packet batches sent by the NIC. This code aims to handle the case
where the NIC retransmits failed packets from a previous batch; the status
information for these packets can sometimes be inserted in the middle of a
batch and are actually not in order by sequence number! (I verified this can
happen with printk's in the function.)

The code in question adaptively identifies the "first" frame of the batch,
taking into account that it may not be the one corresponding to the first agg
status report, and also handles the case when the set of sent packets wraps the
256-character entry buffer. It generates the agg->bitmap field of sent packets
which is later compared to the BlockAck response from the receiver to see which
frames of those sent in this batch were ACKed. A small logic error (wrapping by
0xff==255 instead of 0x100==256) was causing the agg->bitmap to be set
incorrectly.

Fix this wrapping code, and add extensive comments to clarify what is going on.

Signed-off-by: Daniel Halperin <dhalperi@cs.washington.edu>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: parse block ack responses correctly
Daniel Halperin [Tue, 25 May 2010 01:41:30 +0000 (18:41 -0700)]
iwlwifi: parse block ack responses correctly

Compressed BlockAck frames store the ACKs/NACKs in a 64-bit bitmap that starts
at the sequence number of the first frame sent in the aggregated batch. Note
that this is a selective ACKnowledgement following selective retransmission;
e.g., if frames 1,4-5 in a batch are ACKed then the next transmission will
include frames 2-3,6-10 (7 frames). In this latter case, the Compressed
BlockAck will not have all meaningful information in the low order bits -- the
semantically meaningful bits of the BA will be 0x1f3 (where the low-order frame
is seq 2).

The driver code originally just looked at the lower (in this case, 7) bits of
the BlockAck. In this case, the lower 7 bits of 0x1f3 => only 5 packets,
maximum, could ever be ACKed. In reality it should be looking at all of the
bits, filtered by those corresponding to packets that were actually sent. This
flaw meant that the number of correctly ACked packets could be significantly
underreported and might result in asynchronous state between TX and RX sides as
well as driver and uCode.

Fix this and also add a shortcut that doesn't require the code to loop through
all 64 bits of the bitmap but rather stops when no higher packets are ACKed.

In my experiments this fix greatly reduces throughput swing, making throughput
stable and high. It is also likely related to some of the stalls observed in
aggregation mode and maybe some of the buffer underruns observed, e.g.,

http://bugzilla.intellinuxwireless.org/show_bug.cgi?id=1968
http://bugzilla.intellinuxwireless.org/show_bug.cgi?id=2098
http://bugzilla.intellinuxwireless.org/show_bug.cgi?id=2018

Signed-off-by: Daniel Halperin <dhalperi@cs.washington.edu>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: remove unused parameter
Wey-Yi Guy [Sat, 22 May 2010 19:21:12 +0000 (12:21 -0700)]
iwlwifi: remove unused parameter

framecnt_to_us is not used, remove it

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: queue user-initiated scan when doing internal scan
Johannes Berg [Tue, 18 May 2010 09:29:13 +0000 (02:29 -0700)]
iwlwifi: queue user-initiated scan when doing internal scan

The internal scanning created a problem where
when userspace tries to scan, the scan gets
rejected. Instead of doing that, queue up the
user-initiated scan when doing an internal
scan.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: generic scan TX antenna forcing
Johannes Berg [Tue, 18 May 2010 09:48:36 +0000 (02:48 -0700)]
iwlwifi: generic scan TX antenna forcing

In "iwlwifi: make scan antenna forcing more generic"
I introduced generic scan RX antenna forcing, which
here I rename to make it more evident. Also add scan
TX antenna forcing, since I will need that as well.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: do not use huge command buffer for channel switch
Wey-Yi Guy [Tue, 18 May 2010 16:18:06 +0000 (09:18 -0700)]
iwlwifi: do not use huge command buffer for channel switch

Channel switch host command do not need to allocate huge command buffer
since its size is already included in the iwl_device_cmd structure.

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: reduce memory allocation
Johannes Berg [Mon, 17 May 2010 09:37:34 +0000 (02:37 -0700)]
iwlwifi: reduce memory allocation

Currently, the driver allocates up to 19 skb pointers
for each TFD, of which we have 256 per queue. This
means that for each TX queue, we allocate 19k/38k
(an order 4 or 5 allocation on 32/64 bit respectively)
just for each queue's "txb" array, which contains only
the SKB pointers.

However, due to the way we use these pointers only the
first one can ever be assigned. When the driver was
initially written, the idea was that it could be
passed multiple SKBs for each TFD and attach all
those to implement gather DMA. However, due to
constraints in the userspace API and lack of TCP/IP
level checksumming in the device, this is in fact not
possible. And even if it were, the SKBs would be
chained, and we wouldn't need to keep pointers to
each anyway.

Change this to only keep track of one SKB per TFD,
and thereby reduce memory consumption to just one
pointer per TFD, which is an order 0 allocation per
transmit queue.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: kzalloc txb array
Johannes Berg [Mon, 17 May 2010 09:37:33 +0000 (02:37 -0700)]
iwlwifi: kzalloc txb array

When we allocate queues, we currently don't
use kzalloc() right now. When we then free
those queues again without having used all
entries, we may end up trying to free random
pointers found in the txb array since it was
never initialised. This fixes it simply by
using kzalloc().

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlagn: fix bug in txq freeing
Johannes Berg [Mon, 17 May 2010 09:37:32 +0000 (02:37 -0700)]
iwlagn: fix bug in txq freeing

The iwl_hw_txq_free_tfd() function can be
called from contexts with IRQs disabled,
so it must not call dev_kfree_skb() but
rather dev_kfree_skb_any() instead.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwl3945: fix bugs in txq freeing
Johannes Berg [Mon, 17 May 2010 09:37:31 +0000 (02:37 -0700)]
iwl3945: fix bugs in txq freeing

When we free a txq that had no txb array allocated,
we still try to access it. Fix that, and also free
all SKBs that may be in the txb array (although it
can just be a single one).

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: remove inaccurate comment
Wey-Yi Guy [Fri, 14 May 2010 23:21:55 +0000 (16:21 -0700)]
iwlwifi: remove inaccurate comment

REPLY_REMOVE_STA command is used

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: remove priv->mac_addr
Johannes Berg [Fri, 14 May 2010 13:25:58 +0000 (06:25 -0700)]
iwlwifi: remove priv->mac_addr

This variable is now no longer used, so it
can be removed completely.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: beacon internal time unit
Wey-Yi Guy [Wed, 5 May 2010 18:31:38 +0000 (11:31 -0700)]
iwlwifi: beacon internal time unit

use TIME_UNIT define for beacon internal calculation

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: support channel switch offload in driver
Wey-Yi Guy [Thu, 6 May 2010 15:54:11 +0000 (08:54 -0700)]
iwlwifi: support channel switch offload in driver

Support channel switch in driver as a separated mac80211 callback
function instead of part of mac_config callback; by moving to this
approach, uCode can have more control of channel switch timing.

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: beacon format related helper function
Wey-Yi Guy [Thu, 6 May 2010 15:54:10 +0000 (08:54 -0700)]
iwlwifi: beacon format related helper function

Move the ucode beacon formation related helper function from 3945 to
iwlcore, so both _3945 and _agn devices can utilize those functions.

When driver pass the beacon related timing information to uCode in both
spectrum measurement and channel switch commands, the beacon timing
parameter require in uCode beacon format; those helper functions will do
the conversation from uSec to the correct uCode format

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: remove unused parameter in iwl_priv
Wey-Yi Guy [Fri, 7 May 2010 22:31:05 +0000 (15:31 -0700)]
iwlwifi: remove unused parameter in iwl_priv

restrict_refcnt is no longer used, remove it from iwl_priv
structure

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: do not clear data after chain noise calib
Shanyu Zhao [Tue, 11 May 2010 22:25:03 +0000 (15:25 -0700)]
iwlwifi: do not clear data after chain noise calib

Chain noise calibration data are cleared after the calibration is done
in iwlagn_gain_computation() and iwl4965_gain_computation(). This cause
the debugfs entries for those data useless. To provide valid debugging
info, clear those data right before starting the calibration instead.

Signed-off-by: Shanyu Zhao <shanyu.zhao@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: remove debug frame dumping
Johannes Berg [Wed, 12 May 2010 10:33:13 +0000 (03:33 -0700)]
iwlwifi: remove debug frame dumping

This can now be much better achieved using
tracing and post-processing of the trace,
rather than doing the processing in place
in the driver, so remove a lot of code.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: use virtual interface address for scan
Johannes Berg [Wed, 12 May 2010 10:33:12 +0000 (03:33 -0700)]
iwlwifi: use virtual interface address for scan

For probe request frames sent during scan, we
should use the virtual interface's mac address
that the scan was initiated on to avoid issues
when the wrong address is used.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: set MAC address in RXON from interface
Johannes Berg [Wed, 12 May 2010 10:33:11 +0000 (03:33 -0700)]
iwlwifi: set MAC address in RXON from interface

When we do not have an interface, priv->mac_addr
is all zeroes, so the memcpy() is not useful as
the RXON buffer has been cleared previously.
Therefore, use the interface's address that we
are setting up the RXON for, if available.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: remove mac_addr assignment
Johannes Berg [Wed, 12 May 2010 10:33:10 +0000 (03:33 -0700)]
iwlwifi: remove mac_addr assignment

priv->mac_addr is the address of the operating
interface, not the permanent MAC address. They
are usually the same, but the user can override
the operating address, so we shouldn't set the
variable to the permanent one, it is assigned
when an interface is added.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: remove unused wlap_bssid_addr assignment
Johannes Berg [Wed, 12 May 2010 10:33:09 +0000 (03:33 -0700)]
iwlwifi: remove unused wlap_bssid_addr assignment

There's no microcode that actually uses this
variable, and it is reserved for functionality
that the driver doesn't support anyway, so we
shouldn't be setting it.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: remove useless node_addr assignments
Johannes Berg [Wed, 12 May 2010 10:33:08 +0000 (03:33 -0700)]
iwlwifi: remove useless node_addr assignments

iwl_connection_init_rx_config() will already
have set up the entire RXON command, so these
assignments are duplicate.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: move sysfs_create_group to post request firmware
Emmanuel Grumbach [Sun, 23 May 2010 07:14:08 +0000 (00:14 -0700)]
iwlwifi: move sysfs_create_group to post request firmware

Move the sysfs_create_group to iwl_ucode_callback after we
have safely got the firmware.

The motivation to do this comes from a warning from lockdep which detected
that we request priv->mutex while holding s_active during a sysfs request
(show_statistics in the example copy pasted). The reverse order exists upon
request_firmware: request_firmware which is a sysfs operation
that requires s_active is run under priv->mutex.

This ensures that we don't get sysfs request before we finish to request
the firmware, avoiding this deadlock.

=======================================================
[ INFO: possible circular locking dependency detected ]
-------------------------------------------------------
cat/2595 is trying to acquire lock:
 (&priv->mutex){+.+.+.}, at: [<facfa598>] show_statistics+0x48/0x100 [iwlagn]

but task is already holding lock:
 (s_active){++++.+}, at: [<c0580ebd>] sysfs_get_active_two+0x1d/0x50

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (s_active){++++.+}:
       [<c0489b74>] __lock_acquire+0xc44/0x1230
       [<c048a1ed>] lock_acquire+0x8d/0x110
       [<c0581499>] sysfs_addrm_finish+0xe9/0x180
       [<c057f64a>] sysfs_hash_and_remove+0x4a/0x80
       [<c05829d4>] sysfs_remove_group+0x44/0xd0
       [<c0714b75>] dpm_sysfs_remove+0x15/0x20
       [<c070dac8>] device_del+0x38/0x170
       [<c070dc1e>] device_unregister+0x1e/0x60
       [<c071838d>] _request_firmware+0x29d/0x550
       [<c07186c7>] request_firmware+0x17/0x20
       [<fad01bf1>] iwl_mac_start+0xb1/0x1230 [iwlagn]
       [<fa46ba06>] ieee80211_open+0x436/0x6f0 [mac80211]
       [<c0808cd2>] dev_open+0x92/0xf0
       [<c0808b2b>] dev_change_flags+0x7b/0x190
       [<c08148e8>] do_setlink+0x178/0x3b0
       [<c0815169>] rtnl_setlink+0xf9/0x130
       [<c081453b>] rtnetlink_rcv_msg+0x1bb/0x1f0
       [<c0827ce6>] netlink_rcv_skb+0x86/0xa0
       [<c081436c>] rtnetlink_rcv+0x1c/0x30
       [<c08279c3>] netlink_unicast+0x263/0x290
       [<c0828768>] netlink_sendmsg+0x1c8/0x2a0
       [<c07f85fd>] sock_sendmsg+0xcd/0x100
       [<c07f964d>] sys_sendmsg+0x15d/0x290
       [<c07f9e6b>] sys_socketcall+0xeb/0x2a0
       [<c040ad9f>] sysenter_do_call+0x12/0x38

-> #0 (&priv->mutex){+.+.+.}:
       [<c0489f84>] __lock_acquire+0x1054/0x1230
       [<c048a1ed>] lock_acquire+0x8d/0x110
       [<c08bb358>] __mutex_lock_common+0x58/0x470
       [<c08bb84a>] mutex_lock_nested+0x3a/0x50
       [<facfa598>] show_statistics+0x48/0x100 [iwlagn]
       [<c070d219>] dev_attr_show+0x29/0x50
       [<c057fecd>] sysfs_read_file+0xdd/0x190
       [<c052880f>] vfs_read+0x9f/0x190
       [<c0528d22>] sys_read+0x42/0x70
       [<c040ad9f>] sysenter_do_call+0x12/0x38

other info that might help us debug this:

3 locks held by cat/2595:
 #0:  (&buffer->mutex){+.+.+.}, at: [<c057fe25>] sysfs_read_file+0x35/0x190
 #1:  (s_active){++++.+}, at: [<c0580ecd>] sysfs_get_active_two+0x2d/0x50
 #2:  (s_active){++++.+}, at: [<c0580ebd>] sysfs_get_active_two+0x1d/0x50

stack backtrace:
Pid: 2595, comm: cat Not tainted 2.6.33-tp-rc4 #2
Call Trace:
 [<c08b99ab>] ? printk+0x1d/0x22
 [<c0487752>] print_circular_bug+0xc2/0xd0
 [<c0489f84>] __lock_acquire+0x1054/0x1230
 [<c0478d81>] ? sched_clock_cpu+0x121/0x180
 [<c048a1ed>] lock_acquire+0x8d/0x110
 [<facfa598>] ? show_statistics+0x48/0x100 [iwlagn]
 [<c08bb358>] __mutex_lock_common+0x58/0x470
 [<facfa598>] ? show_statistics+0x48/0x100 [iwlagn]
 [<c08bb84a>] mutex_lock_nested+0x3a/0x50
 [<facfa598>] ? show_statistics+0x48/0x100 [iwlagn]
 [<facfa598>] show_statistics+0x48/0x100 [iwlagn]
 [<c0580cf9>] ? sysfs_get_active+0x69/0xb0
 [<facfa550>] ? show_statistics+0x0/0x100 [iwlagn]
 [<c070d219>] dev_attr_show+0x29/0x50
 [<c057fecd>] sysfs_read_file+0xdd/0x190
 [<c05ff314>] ? security_file_permission+0x14/0x20
 [<c0528242>] ? rw_verify_area+0x62/0xd0
 [<c052880f>] vfs_read+0x9f/0x190
 [<c047745b>] ? up_read+0x1b/0x30
 [<c057fdf0>] ? sysfs_read_file+0x0/0x190
 [<c04af3b4>] ? audit_syscall_entry+0x1f4/0x220
 [<c0528d22>] sys_read+0x42/0x70
 [<c040ad9f>] sysenter_do_call+0x12/0x38

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwlwifi: add name to Maintainers list
Wey-Yi Guy [Wed, 2 Jun 2010 22:17:49 +0000 (15:17 -0700)]
iwlwifi: add name to Maintainers list

Add "Wey-Yi Guy" to maintainers list for iwlwifi.

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwl3945: fix internal scan
Abhijeet Kolekar [Thu, 3 Jun 2010 04:15:10 +0000 (21:15 -0700)]
iwl3945: fix internal scan

Port of internal scan to iwl3945 missed introduction
of iwl3945_get_single_channel_for_scan.

Fix the following bug by introducing the iwl3945_get_single_channel_for_scan
http://bugzilla.intellinuxwireless.org/show_bug.cgi?id=2208

Signed-off-by: Abhijeet Kolekar <abhijeet.kolekar@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
14 years agoiwl3945: enable stuck queue detection on 3945
Reinette Chatre [Thu, 20 May 2010 17:54:40 +0000 (10:54 -0700)]
iwl3945: enable stuck queue detection on 3945

We learn from
http://bugzilla.intellinuxwireless.org/show_bug.cgi?id=1834 and
https://bugzilla.redhat.com/show_bug.cgi?id=589777
that 3945 can also suffer from a stuck command queue. Enable stuck queue
detection for iwl3945 to enable recovery in this case.

Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>