If the allocation of ivp fails the error handling attempts to
free an uninitialized dma_buf; this data structure just contains
garbage on the stack, so the freeing will cause issues when the
urb, buf and dma fields are free'd. Fix this by not free'ing the
dma_buf if the ivp allocation fails.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Julian Calaby <julian.calaby@gmail.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

Currently mac80211 does not inform the driver of the session
block ack timeout when starting a rx aggregation session.
Drivers that manage the reorder buffer need to know this
parameter.
Seeing that there are now too many arguments for the
drv_ampdu_action() function, wrap them inside a structure.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>

Part of reorganising wireless drivers directory and Kconfig.

Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

Advertise the capability to send A-MSDU within A-MPDU
in the AddBA request sent by mac80211. Let the driver
know about the peer's capabilities.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>

mac80211 requires that rx path does not run concurrently with
tx status reporting.  Add a spinlock which will ensure that.

Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mac80211 requires that rx path does not run concurrently with
tx status reporting.  Since rx path is run in driver tasklet,
tx status cannot be reported directly from interrupt context
(there would be no way to lock it out).

Add tasklet for tx and move all possible code from irq handler
there.

Note: tx tasklet is needed because workqueue is queued very
      rarely and that kills TCP performance.

Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

Rx is run inside a tasklet so ieee80211_rx() should be used
instead of ieee80211_rx_ni().

Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

DMA to variables located on the stack is a bad idea.
For simplicity and to avoid frequent allocations create
a buffer inside the device structure.  Protect this
buffer with vendor_req_mutex.  Don't protect vendor
requests which don't use this buffer.

Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

Kalle Valo says:

====================
Major changes:

mwifiex:

* enhancements for AP mode: support verbose information in station
  dump command and also information about AP link.
* enable power save by default

brcmfmac:

* fix module reload issue for PCIe
* improving msgbuf protocol for PCIe devices
* rework .get_station() cfg80211 callback operation
* determine interface combinations upon device feature support

ath9k:

* ath9k_htc: add support of channel switch

wil6210:

* add modparam for bcast ring size
* support hidden SSID
* add per-MCS Rx stats
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

We expect EEPROM per-rate power table to be filled with
s6 values and warn user if values are invalid.  However,
there appear to be devices which don't have this section
of EEPROM initialized.  In such case we should ignore
the values and leave the driver power tables set to zero.

Note that vendor driver doesn't care about this case but
mt76x2 skips 0xff per value.  We take mt76x2's approach.

Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>