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>
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>
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>
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>
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>
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>
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>
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>
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>
Use the inline function instead of directly indexing the array.
This allows some architectures with hardware instructions
for bit reversals to eliminate the array.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
It is possible to configure driver using mtu_max module parameter
by setting it to value in range of 68..7920 inclusive.
This is sub-optimal performance-wise in case packet is larger than 1 page.
mtu_max default value is 2228.
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
In case of FW error, make it clear (in dmesg) what branch is taken
in the error recovery code.
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Disconnect flow may be invoked either from upper layer request,
or from event reported by the firmware.
In case of firmware event, driver need to release resources for the station but
not send another disconnect WMI command.
In case of upper layer request, WMI_DISCONNECT_STA_CMDID command need to
be issued for the firmware to perform disconnect on the MAC layer. Eventually,
event is expected to confirm MAC disconnect, but it is better to not wait for
firmware event and release station resources immediately. FW may fail to
report disconnect for various reasons, so one could not rely on event always reported.
Introduce parameter to distinguish 2 cases above to prevent double WMI command
issuing.
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
As communicated with the firmware & hardware teams
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
When interface is down, recovery flow should not be attempted.
Next ndo_open() will trigger target reset, that is FW recovery.
Doing recovery while interface is down cause internal "up", leaving
internal driver state in conflict with network stack. Then, when network
stack will call ndo_open(), kernel oops will be triggered.
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Reduce pulse_rssi threshold to 15 in order to improve radar pattern detection
probability on ext channel
Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi83@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The right shift operation has higher precedence than the mask so we
left shift by "(i * 3)" and then immediately right shift by "(i * 3)"
then we mask. It should be left shift, mask, and then right shift.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Some devices have multiple bands enables in the EEPROM data, even though
they are only calibrated for one. Allow platform data to disable
unsupported bands.
Signed-off-by: Gabor Juhos <juhosg@openwrt.org>
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
On some devices (especially little-endian ones), the flash EEPROM data
has a different endian, which needs to be detected.
Add a flag to the platform data to allow overriding that behavior
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This makes the initial NF calibration less likely to fail.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
It can cause inconsistent calibration results or in some cases turn the
radio deaf.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
When NF calibration fails, the radio often becomes deaf. The usual
hardware hang checks do not detect this, so it's better to issue a reset
when that happens.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The "goto chip_reset" is a bit misleading, because it does not actually
issue a chip reset. Instead it is bypassing processing of other
interrupts and assumes that the tasklet will issue a chip reset.
In the case of RXORN this does not happen, so bypassing processing of
other interrupts will simply allow them to fire again. Even if RXORN
was triggering a reset, it is not critical enough to need the bypass
here.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
UB124 is a USB based reference design not supported by ath9k or
ath9k_htc.
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The initvals use up quite a bit of space, and PC-OEM support is
typically not needed on embedded systems
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
* Fix a 11b/EVM issue by adjusting
FIR filter coefficients.
* Fix a problem with receiving probe request
frames sent at 11b rate.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This fixes RX sensitivity issues with AR9580.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
mac80211 currently has a race which can be hit
with this sequence:
* Start a scan operation.
* TX BA is initiated by ieee80211_start_tx_ba_session().
* Driver sets up internal state and calls
ieee80211_start_tx_ba_cb_irqsafe().
* mac80211 adds a packet to sdata->skb_queue with
type IEEE80211_SDATA_QUEUE_AGG_START.
* ieee80211_iface_work() doesn't process the
packet because scan is in progress.
* ADDBA response timer expires and the sta/tid is
torn down.
* Driver receives BA stop notification and calls
ieee80211_stop_tx_ba_cb_irqsafe().
* This is also added to the queue by mac80211.
* Now, scan finishes.
At this point, the queued up packets might be processed
if some other operation schedules the sdata work. Since
the tids have been cleaned up already, warnings are hit.
If this doesn't happen, the packets are left in the queue
until the interface is torn down.
Since initiating a BA session when scan is in progress
leads to flaky connections, especially in MCC mode, we
can drop the TX BA request. This improves connectivity
with legacy clients in MCC mode.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The flush timeout in MCC mode is very small, since
we are constrained by the time slice for each
channel context, but since only the HW queues are
flushed when switching contexts, it is acceptable.
Since the SW queues are also emptied in the mac80211 flush()
callback, a larger duration is needed. Add an override
argument to __ath9k_flush() and set it when flush()
is called in MCC mode. This allows the driver to
drain both the SW and HW queues.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Instead of using ATH_CHANCTX_EVENT_ASSIGN to abort
a HW scan when a new interface becomes active, use the
mgd_prepare_tx() callback. This allows us to make
sure that the GO's channel becomes operational by
using flush_work().
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch makes sure that a GO interface
sends out a new NoA schedule with 200ms duration
when mgd_prepare_tx() is called.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
mac80211 has to be notified when a RoC period
expires in the driver. In MCC mode, since the
offchannel/RoC timer is set with the requested
duration, ieee80211_remain_on_channel_expired() needs
to be called when the timer expires.
But, currently it is done after we move back to
the operating channel. This is incorrect - fix this
by calling ieee80211_remain_on_channel_expired() when
the RoC timer expires and in ath_roc_complete() when
the RoC request is aborted.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
If a GO interface is active when we receive a
mgd_prepare_tx() call, then we need to send
out a new NoA before switching to a new context.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Since both the arguments need to satisfy
the alignment requirements of ether_addr_copy(),
use memcpy() in cases where there will be no
big performance benefit and make sure that
ether_addr_copy() calls use properly aligned
arguments.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
ATH_CHANCTX_EVENT_AUTHORIZED is required to trigger
the MCC scheduler when a station interface becomes
authorized. But, since the driver gets station state
notifications when the current operating mode is AP
too, make sure that we send ATH_CHANCTX_EVENT_AUTHORIZED
only when the interface is in station mode.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Pending frames in the driver can be present
either in the HW queues or SW. ath9k_has_pending_frames()
currently checks for the HW queues first and then
checks if any ACs are queued in the driver.
In MCC mode, we need to check the HW queues alone, since
the SW queues are just marked as 'stopped' - they will
be processed in the next context switch. But since we
don't differentiate this now, mention whether we want
to check if there are frames in the SW queues.
* The flush() callback checks both HW and SW queues.
* The tx_frames_pending() callback does the same.
* The call to __ath9k_flush() in MCC mode checks HW queues alone.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
An offchannel operation also needs to have
a flush timeout that doesn't exceed the NoA
absence duration of a GO context, so use
channel_switch_time. The first offchannel
operation is set a flush timeout of 10ms since
channel_switch_time will be zero.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
In MCC mode, the duration for a channel context
is half the beacon interval and having a large
flush timeout will adversely affect GO operation,
since the default value of 200ms will overshoot
the advertised NoA absence duration.
The scheduler initiates a channel context switch
only when the slot duration for the current
context expires, so there is no possibility of
having a fixed timeout for flush.
Since the channel_switch_time is added to the
absence duration when the GO sets up the NoA
attribute, this is the maximum time that we
have to flush the TX queues. The duration is very
small, but we don't have a choice in MCC mode.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The timeout value for flushing the TX queues
is hardcoded at 200ms right now. Use a channel
context-specific value instead to allow adjustments
to the timeout in case MCC is enabled.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
When an active context transitions to inactive
state, the NoA schedule needs to be removed
for the context that has beaconing enabled.
Not doing this will affect p2p clients.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
When a chip reset is done, all running timers,
tasklets etc. are stopped but the beacon tasklet
is left running. Fix this.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
When a HW reset is done, the interrupt tasklet is
disabled before ISRs are disabled in the HW. This
allows a small window where the HW can still generate
interrupts. Since the tasklet is disabled and not killed,
it is not scheduled but deferred for execution at a later
time.
This happens because ATH_OP_HW_RESET is not set when ath_reset()
is called. When the hw_reset_work workqueue is used, this
problem doesn't arise because ATH_OP_HW_RESET is set
and the ISR bails out.
Set ATH_OP_HW_RESET properly in ath_reset() to avoid
this race - all the ath_reset_internal() callers have
been converted to use ath_reset() in the previous patch.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Instead of having ath_reset_internal() and ath_reset()
as two separate calls to perform a HW reset, have
one function. This makes sure that the behavior will
be the same at all callsites.
Signed-off-by: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
