wl1251: use ELP wakeup interrupt instead of polling
The chip can generate an interrupt on ELP wakeup, make use of it instead
of polling the status register. This reduces wakeup time to ~7ms from
~8-20ms (the wait loop was sleeping, so could take up to 20ms depending
on how system timer is configured). Note that DEBUG_PSM does not report
this accurately as it only measures jiffies.
This patch also moves atomic portion of interrupt processing to main.c
to avoid code duplication.
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.
While working on PS I've noticed elp_work is kicking rather often, and
sometimes the chip is put to sleep before 5ms delay expires. This
seems to happen because by the time wl1251_ps_elp_wakeup is called
elp_work might still be pending, and if the processing takes longer
(for example interrupts may take some time to process), elp_work might
get started and hit the mutex. After main thread finishes work it
calls wl1251_ps_elp_sleep, which can no longer cancel old work (as
it's already started) and schedules yet another work needlessly. The
elp_work then gets the mutex and puts the chip to sleep too early.
Fix this by cancelling work in wl1251_ps_elp_wakeup instead.
wl1251_ps_set_elp() only does acx_sleep_auth call and takes the chip
from/to ELP, however all callers of wl1251_ps_set_mode() have already
taken the chip out of ELP and puts it back to ELP when they finish.
This makes ELP calls (and register writes they result in) superfluous.
So remove wl1251_ps_set_elp function and call acx_sleep_auth directly.
Finish up filtering code and enable it. mac80211 no longer requires
to be atomic in .configure_filter callback, so remove filter_work.
The new filtering code makes sure it doesn't enable BSSID filtering
while BSSID is not set so that association works properly. This
allows removing hack in wl1251_cmd_join.
Yuri Ershov [Wed, 27 Oct 2010 13:34:39 +0000 (17:34 +0400)]
wl1251: move all transmission to happen in tx_work
Currently TX queue is consumed both in wl->tx_work and in wl->irq_work.
This was problematic with the PS-Poll implementation, and caused stalls
because of synchronisation issues between the workqueues. Fix that by
consuming TX queue only in TX work.
Signed-off-by: Kalle Valo <kalle.valo@nokia.com> Reviewed-by: Juuso Oikarinen <juuso.oikarinen@nokia.com> Signed-off-by: Juuso Oikarinen <juuso.oikarinen@nokia.com> Signed-off-by: Yuri Ershov <ext-yuri.ershov@nokia.com>
9df86e2e702c6d5547aced7f241addd2d698bb11 ported TX queue stopping/waking
improvement from wl1271, but it has some issues:
- it forgot to initialize wl_lock.
- it placed ieee80211_wake_queues in an awkward place between result
handling loops, it's better to do it after all processing is done.
- it left queue wake call in wl1251_tx_packet_cb, which made the patch
ineffective because by the time we reached low watermark check, queues
would already be woken.
- the wl1271 watermark values are too large for wl1251, as it has
shorter hardware tx queue.
Yuri Ershov [Wed, 27 Oct 2010 13:24:55 +0000 (17:24 +0400)]
wl1251: Add vif checking to rssi and beacon loss events
The chipset is initialized and the vif for the interface are created
separately - there is a small window, when the interface is configured,
that the firmware sends rssi or beacon loss events while the vif is not yet
existing, causing an oops in the mac80211.
Add checking for the vif to the handlers, ignoring the events if there is not
yet a vif.
Signed-off-by: Juuso Oikarinen <juuso.oikarinen@nokia.com> Reviewed-by: Kalle Valo <kalle.valo@nokia.com> Signed-off-by: Luciano Coelho <luciano.coelho@nokia.com> Signed-off-by: Yuri Ershov <ext-yuri.ershov@nokia.com>
Yuri Ershov [Wed, 27 Oct 2010 13:21:09 +0000 (17:21 +0400)]
wl1251: Handle failing PS entry
It is possible, that the nullfunc message used to indicate to the
associated-to AP that we are going to power save fails to reach the
intended recipient. In that case, the STA and AP can become out of sync,
resulting in connection loss.
This patch adds code to attempt to recover from such failure by
resending the nullfunc.
Yuri Ershov [Wed, 27 Oct 2010 10:48:55 +0000 (14:48 +0400)]
wl1251: decrease bt coex before beacon time
With the current setup bluetooth coexistance consumes too much power
because firmware wakes up for beacon really early. As a workaround reduce
the wakeup time to 500 us to save power. This patch will be reverted when
a proper solution for bluetooth coexistance high power consumption is
implemented.
This decreases wlan idle power consumption from ~22 mA to ~14 mA.
Signed-off-by: Kalle Valo <kalle.valo@nokia.com> Reviewed-by: Vidhya Govindan <vidhya.govindan@nokia.com> Signed-off-by: Luciano Coelho <luciano.coelho@nokia.com> Signed-off-by: Yuri Ershov <ext-yuri.ershov@nokia.com>
Yuri Ershov [Wed, 27 Oct 2010 10:47:25 +0000 (14:47 +0400)]
wl1251: fix deadlock with ieee80211 beacon filter calls
Lockdep shows that there is a deadlock when calling ieee80211_rssi_changed()
under wl->mutex. Fix it by releasing mutex before calling mac80211 functions.
This patch is based on Work-In-Progress patch sent by Kalle Valo.
here's the original message:
The queue stopping/waking functionality was broken in a way that could
cause huge latencies in TX transfers and even cause the TX to stall in the
right circumstances. Correct these problems.
Signed-off-by: Denis 'GNUtoo' Carikli <GNUtoo@no-log.org> Acked-by: Kalle Valo <kvalo@adurom.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
Poll for join command completion instead of waiting blindly for 10
msecs. There is a timeout of 100 msecs, if the command doesn't complete
by then, we return an error code.
Cc: stable@kernel.org Signed-off-by: Yuri Ershov <ext-yuri.ershov@nokia.com> Signed-off-by: Yuri Kululin <ext-yuri.kululin@nokia.com> Acked-by: Kalle Valo <kvalo@adurom.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
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>
compat: fix uevent_suppress on 2.6.29 or older kernels
Missing uevent_suppress is causing two uevents instead of one, which is
confusing udev and sometimes causing firmware load to fail due to race
condition, so let's add it.
wl1251 has WLAN_IRQ pin for generating interrupts to host processor,
which is mandatory in SPI mode and optional in SDIO mode (which can
use SDIO interrupts instead). However TI recommends using deditated
IRQ line for SDIO too.
Add support for using dedicated interrupt line with SDIO, but also leave
ability to switch to SDIO interrupts in case it's needed.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com> Reviewed-by: Bob Copeland <me@bobcopeland.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
wl1251: register platform_device to pass board data
wl1251 is embedded chip that can be connected using SDIO bus, and is not
an actual SDIO card. For this reason there is a need to pass some board
specific data, like 'EEPROM is attached' flag or power control callback.
However currently there is no way to pass this data through SDIO subsystem,
so this patch registers dummy platform_device to allow that, until we
have a better solution to this.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com> Acked-by: Kalle Valo <kvalo@adurom.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
In case debugfs does not init for some reason (or is disabled
on older kernels) driver does not allocate stats.fw_stats
structure, but tries to clear it later and trips on a NULL
pointer:
Unable to handle kernel NULL pointer dereference at virtual address 00000000
PC is at __memzero+0x24/0x80
Backtrace:
[<bf0ddb88>] (wl1251_debugfs_reset+0x0/0x30 [wl1251])
[<bf0d6a2c>] (wl1251_op_stop+0x0/0x12c [wl1251])
[<bf0bc228>] (ieee80211_stop_device+0x0/0x74 [mac80211])
[<bf0b0d10>] (ieee80211_stop+0x0/0x4ac [mac80211])
[<c02deeac>] (dev_close+0x0/0xb4)
[<c02deac0>] (dev_change_flags+0x0/0x184)
[<c031f478>] (devinet_ioctl+0x0/0x704)
[<c0320720>] (inet_ioctl+0x0/0x100)
Add a NULL pointer check to fix this.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com> Acked-by: Kalle Valo <kalle.valo@iki.fi> Cc: stable@kernel.org Signed-off-by: John W. Linville <linville@tuxdriver.com>
4sec wait is way too pessimistic, TI driver uses 40ms here,
and testing shows that is ebough, so let's also use that.
While at it, add useful sounding comment from the TI driver.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com> Acked-by: Kalle Valo <kalle.valo@iki.fi> Signed-off-by: John W. Linville <linville@tuxdriver.com>
To fix this, add special IO functions for ELP_CTRL access that are
using sdio_readb/sdio_writeb. Similar handling is done in TI
reference driver from Android code drop.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com> Cc: Bob Copeland <me@bobcopeland.com> Acked-by: Kalle Valo <kalle.valo@iki.fi> Signed-off-by: John W. Linville <linville@tuxdriver.com>
Make local functions and data static, also constify
some structures. While at it, clean up unneeded includes.
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com> Cc: Bob Copeland <me@bobcopeland.com> Acked-by: Kalle Valo <kalle.valo@iki.fi> Signed-off-by: John W. Linville <linville@tuxdriver.com>
projects / pandora-wifi.git / log