We don't really need to use different mac colors when adding mac
contexts, because they're not used anywhere.  In fact, the firmware
doesn't accept 255 as a valid color, so we get into a SYSASSERT 0x3401
when we reach that.

Remove the color increment to use always zero and avoid reaching 255.

CC: <stable@vger.kernel.org> [3.10+]
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

This is now implemented by mac80211 (commit below).
mac80211 will flush/drop the frames on the queues before
suspending / disconnecting.
It will then send the deauth and wait until the queues are
empty.

commit 3b24f4c6
Author: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Date:   Wed Jan 7 15:42:39 2015 +0200

    mac80211: let flush() drop packets when possible

This reverts commit 4e6c48e0.

In order to change the usage of U-APSD on the fly later,
move the enabling condition into a new function that is
called when authenticated.

This allows the module parameter to become writable, it
won't take effect immediately but at least on the next
association the new value will be used.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

When the FW is in error status - try to read the RXF and
TXF (all of them) and add them to the dump data.

This shouldn't happen in non-error statuses, as we don't
want to stop the RXF/TXF while they are running.
Signed-off-by: NLiad Kaufman <liad.kaufman@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

Some HW modules have two SRAMs. In such cases add the secondary SRAM to
the list of dumped segments.
Signed-off-by: NIdo Yariv <idox.yariv@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

This change has already been implemented in iwldvm:

commit a260e7b3f0307878b99d57ed1406cf2d497923b8
Author: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Date:   Sun Oct 5 09:11:14 2014 +0300

    iwlwifi: dvm: drop non VO frames when flushing

Since I added the flush() callback implementation in mvm,
we got reports that the queues are stuck while roaming
or suspending.
This commit above helped much for iwldvm, implement the
same behavior for iwlmvm.

CC: <stable@vger.kernel.org> [3.16+]
Fixes: c5b0e7c0 ("iwlwifi: mvm: implement mac80211's flush callback")
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

A number of places (still) use a direct operation, use
iwl_mvm_sta_from_mac80211() consistently. In one place
also move it into the variable initializer.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

Some implementations (i.e. mini_rpm) assume the references
are managed only while the device is started.

Move the stale reference cleanup before stopping the
device in order to make them happy.
Signed-off-by: NEliad Peller <eliadx.peller@intel.com>
Signed-off-by: NGregory Greenman <gregory.greenman@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

The driver and the firmware now support 2 different channels
at the same time. Advertise this capability to the stack.
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

On hw restart, make sure to wait for d0i3 exit
(by checking the IN_D0I3 status bit).

This is needed in order to avoid the stale
d0i3_exit_work from doing harm (e.g. unref
cleared reference).
Signed-off-by: NEliad Peller <eliadx.peller@intel.com>
Reviewed-by: NGregory Greenman <gregory.greenman@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

Enter d0i3 on suspend, and exit d0i3. Wait for the
command responses in both cases.

Use this mode in case of pcie trans.
Signed-off-by: NEliad Peller <eliadx.peller@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

d3 and d0i3 shouldn't be mutually exclusive. Set supported
wowlan triggers by looking for each of them, and check
on suspend/resume which flow should be used ("any" trigger
is supported by d0i3, and all the others by d3)
Signed-off-by: NEliad Peller <eliadx.peller@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

This allows to add the offset. The type of the generic
memory dump will let the parser know that this is SRAM.
Reviewed-by: NLiad Kaufman <liad.kaufman@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

Instead of adding a dump type for each type of memory, change
the SMEM type to be a general purpose memory dump. Add the
type of the memory and its offset in the device in the dump
itself. This will allow an external parser to know where
this memory came from.

Note that since this type isn't really in use yet, this is
not a real problem.
Reviewed-by: NLiad Kaufman <liad.kaufman@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

In NICs that have SMEM - add its content to the dump data
for later debug.
Signed-off-by: NLiad Kaufman <liad.kaufman@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

nvm_file in family 8000 B step and A step differ. This means
that the driver should support 2 file name as default.
Signed-off-by: NEran Harary <eran.harary@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

The HW step member was left out of the core dump information. Fix this.
Signed-off-by: NIdo Yariv <idox.yariv@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

The firmware is able to compensate the rssi when we hear
the frame on a different channel.
This is true for an offset up to 3 channels.
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

The allocation of the DCCM between the data and the stack
can theoretically change without notice to the driver, but
the total size is HW-fixed. Since the stack CCM (runtime
stack) has also data important to the FW - this patch allows
pulling the whole DCCM in one piece and adds it to the dump
data.

If the size isn't known - just use the data part of the
DCCM as it appears in the FW TLVs.
Signed-off-by: NLiad Kaufman <liad.kaufman@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

In order to let drivers have more dynamic U-APSD support,
move the enablement flag to the virtual interface driver
flags. This lets drivers not only set it up differently
for different interfaces, but also enable/disable on the
fly if needed.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>

On stop(), we already cleared our internal state,
and the restart_complete() callback won't be
called, so simply clear the IN_HW_RESTART flag.

Keeping the flag might result in invalid state
on the next start(), preventing the driver starting
properly.

Additionally, don't take IWL_MVM_REF_UCODE_DOWN on stop()
if hw restart was requested, as the ref was already
taken in this case.
Signed-off-by: NEliad Peller <eliadx.peller@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

All the supported firmwares have this TLV flag set.
Reviewed-by: NLuciano Coelho <luciano.coelho@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

This firmware is not supported anymore.
Stop loading this firmware - and remove the code that
handled older versions.
Reviewed-by: NLuciano Coelho <luciano.coelho@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

In case of channel switch, we need to teardown the TDLS peers.
Signed-off-by: NArik Nemtsov <arikx.nemtsov@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

The driver knows whether an rx frame was beamformed and marks
it in the radiotap VHT flags. However it should also declare
that it knows to extract this info otherwise this gets discarded
by sniffers like Wireshark.
Signed-off-by: NEyal Shapira <eyalx.shapira@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

Add the net detect WoWLAN flag to indicate support and use the
nd_config from the WoWLAN configuration to start net detect, if it is
set.  The WoWLAN configuration takes precedence over the debugfs
configuration.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>
Reviewed-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NEmmanuel Grumbach <emmanuel.grumbach@intel.com>

When another vif is also running during a channel switch, we need to
use a session protection when we move to the new channel, so that we
don't miss the beacons.  Without this, sometimes the other vif
repeatedly gets time exactly when we should be hearing the beacons,
preventing channel switch from completing.  Adding a session
protection that lasts from the moment the channel changes until 2
TBTTs later, ensures that we will hear the beacons on the destination
channel.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

After a channel switch, transmission on the new channel is only
started once we see a beacon on it.  Thus, beacon filtering needs to
be disabled during channel switch so that mac80211 receives this
beacon and finishes the process.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

We need to call ieee80211_chswitch_done() ourselves just when the
absence TE started, so we perform the actual context switch early
enough.  To do so, add a dummy channel_switch op, which will cause
mac80211 to skip the countdown timer and allow us to call
ieee80211_chswitch_done() to complete the operation.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

Now that mac80211 waits for the first beacon on the new channel before
calling post_channel_switch, the reconfiguration of the firmware
should be done in the post_channel_switch operation instead of when
assigning the vif to the new context.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

Instead of using a hardcoded number of TUs before beacon 0 as the time
to start the absence and actual channel switch, calculate it in
relation to the beacon interval.  We use 10 TUs + beacon interval
before beacon 0 to target a bit before beacon 1.  This gives us enough
time to switch to the new channel before the AP/GO switches.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

Add an absent time event when pre_channel_switch is called and use the
time event started indication to set the disable_tx bit instead of
doing it in unassign_vif().  This is done so that the firmware queues
are stopped before the actual switch takes place to avoid losing
packets while the AP/GO is performing its actual switch.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

Now that all CSA flows are using the switch_vif_chanctx op, we can
rely on the switching_chanctx boolean that is passed to the
__iwl_mvm_assign_vif_chanctx() and __iwl_mvm_unassign_vif_chanctx()
functions to decide whether the context switch flows need to be
executed.  In this way we make the chanctx switch flow more generic,
without having to rely on the csa_active flag being set.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

Disable PS when pre_channel_switch is called and add the
post_channel_switch operation to re-enable PS when the channel switch
is completed.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

The spec requires to decrement the CSA counters based on TBTT,
regardless if the beacon was actually transmitted. Previously, the fw
would send beacon notifications only for successfully transmitted
beacons. This behavior resulted in inaccurate CSA countdown. In order
to address this issue, the fw was changed to send beacon
notifications also for not transmitted beacons. Such notifications
have TX_STATUS_INTERNAL_ABORT (0x92).
Don't start the CSA countdown before first successfully transmitted
beacon, in order to guarantee that the CSA is announced for a
required period.
Signed-off-by: NAndrei Otcheretianski <andrei.otcheretianski@intel.com>

We have code to recover and go back to the original channel context if
something fails in the middle of switch_vif_chanctx, but we return the
error code of the recover calls instead of the original code, so if
the recovery succeeds, we will return 0 (success).  Fix this by not
assigning the return value of the recovery calls to ret.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

Add support to reassign vif in switch_vif_chanctx.  This is similar to
the existing CHANCTX_SWMODE_SWAP_CONTEXTS mode, but doesn't delete the
old context nor creates a new one, doing to switch between two
existing contexts.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

Currently we only support the CHANCTX_SWMODE_SWAP_CONTEXTS mode, but
we need to support other modes as well.  Spin a new function off in
order to make it easier to support other modes.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

We only need the csa_vif in AP/GO modes, and assigning for other
interfaces may cause problems, because csa_vif is never cleared.  To
prevent this, only assign the value if the iftype is
NL80211_IFTYPE_AP.  Use a switch to do this, even though, for now,
only the AP interface type is handled, because soon other interface
types will be added as well.

Additionally, convert the WARN() in the error case when a
channel-switch is already running to WARN_ONCE().
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>

A new callback has been added to prepare the device for a channel
switch.  Use the new callback instead of the old channel_switch_beacon
operation.

This makes it possible to remove the channel_switch_beacon operation
from mac80211.
Signed-off-by: NLuciano Coelho <luciano.coelho@intel.com>