To enable the TKIP/CCMP replay protection, the frames has to be
processed in the right order. However, the device is not able to
re-order the frames during BlockAck sessions.
Mac80211 is able to reorder the frames, but it need the information
about the BlockAck sessions start and stop. Unfortunately, since the
BlockAck is fully handled by the hardware, these frames were not
forwarded to the host. So, if the driver ask to mac80211 to apply the
replay protection, it drop all misordered frames.
So, until now, the driver explicitly asked to mac80211 to not apply
the CCMP/TKIP replay protection.
The situation has changed with the API 3.4 of the device firmware. The
firmware forward the BlockAck information. Mac80211 is now able to
correctly reorder the frames. There is no more reasons to drop
cryptographic data.
This patch also impact the older firmwares. There will be a performance
impact on these firmware (since the misordered frames will dropped).
However, we can't keep the replay protection disabled.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200701150707.222985-12-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It seems that in the old days, the RSSI information could be missing. In
this case, in order to not pollute the RSSI stats, the frame was
dropped (!).
It is far better to mark the frame with the flag RX_FLAG_NO_SIGNAL_VAL.
In add, the problem seems now fixed in the firmware (at least, it has
not been encountered with recent firmwares).
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200701150707.222985-10-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When a station go to sleep, the driver receive the status REQUEUE and
forward this answer to mac80211. So, neither the driver, neither the
hardware buffer the frames. So the call to ieee80211_sta_set_buffered is
useless.
In add, it seems that mac80211 does not expect to receive
ieee80211_sta_set_buffered(false) after the station is asleep(). When
the device send data to a station, the following sequence can be
observed:
- Mac80211 call wfx_sta_notify(awake).
- The driver calls ieee80211_sta_set_buffered(true). Since the
station is awake, its TIM is not set.
- Mac80211 receive a power save notification from the station, so it
calls wfx_sta_notify(asleep).
- Then, since the driver has declared it has buffered some frames,
the TIM of the station should be set. This action is delayed by
mac80211.
- The device also notice the station go to sleep. It replies the
REQUEUE status for the buffered frames. The driver forward this
status to mac80211.
- There is no more frames in queues, so the driver call
ieee80211_sta_set_buffered(false).
- Mac80211 updates the TIM but since there is no frames buffered by
the driver, it set the TIM for the station to 0.
Anyway, correctly use the ieee80211_sta_set_buffered() API solves the
problem.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200701150707.222985-8-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
During the loading of the firmware, the WFX_DCA_GET register provide the
number available bytes in the receiving buffer. It is not necessary to
access to the WFX_DCA_GET after sent of each firmware fragment.
This patch allows to send the firmware:
- in 64ms instead of 130ms using SDIO bus
- in 78ms instead of 115ms using SPI bus
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200701150707.222985-6-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The device is able to send multiple Tx confirmations in the one reply.
In this case, there is only one vif identifier for all the
confirmations.
Unfortunately, to generate this kind of messages the device squashes all
the confirmations whatever their vif ID and use the vif ID of the first
confirmation. So, the driver cannot rely on the vif ID mentioned in the
header. Fortunately, using the packet_id, the driver can retrieve the Tx
request and the associated vif.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200701150707.222985-4-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The device handles 4 queues (one per AC) for each virtual interface (and
maximum 4 virtual interfaces). Until now the driver unified the queue of
all interfaces and handled only 4 queues for whole device.
This architecture did not allow to balance the traffic between the vif. So,
this patch relocate the queues into the vif and change the API accordingly.
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200701150707.222985-2-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This is the movement of the wilc1000 driver out of staging, pulled in
here so that we do not end up doing duplicate work.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
* 'wilc1000-move-out-of-staging' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers-next:
wilc1000: move wilc driver out of staging
All the functions that vchi currently provides are a 1:1 mapping to its
vchiq counterparts. Get rid of vchi altogether and use vchiq's on all
services.
In the process also get rid of the vchi directory, as the only remaining
file was a TODO file, which now lives in the parent directory.
Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
Link: https://lore.kernel.org/r/20200629150945.10720-44-nsaenzjulienne@suse.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
vchi created an opaque structure to be held by services while they
process callback messages. The contents of this opaque structure are the
service handle, which all services already maintain, and a pointer to
vchiq's message structure, struct vchiq_header.
Let's get rid of this opaque structure and directly pass struct
vchiq_header, which is what ultimately vchiq consumes when handling
callback messages.
Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
Link: https://lore.kernel.org/r/20200629150945.10720-33-nsaenzjulienne@suse.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
As vchiq_shim's callback does nothing aside from pushing messages into
the service's queue, let's bypass it and jump directly to the service's
callbacks, letting them choose whether to use the message queue.
It turns out most services don't need to use the message queue, which
makes for simpler code in the end.
Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
Link: https://lore.kernel.org/r/20200629150945.10720-29-nsaenzjulienne@suse.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This has historically been handled by vchi, but there is no reason why
this couldn't be handled directly in vchiq.
The patch tries to avoid altering any behavior, with the exception of
the msg_queue size, which is now fixed to VCHIQ_MAX_SLOTS (it was set to
VCHIQ_MAX_SLOTS / 2). This is done to match vchiq's user_service message
queue, which could be merged with this one in the future.
Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
Link: https://lore.kernel.org/r/20200629150945.10720-24-nsaenzjulienne@suse.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
