Tracing is limited to 100 characters and this message passes
the limit when there are a few buffered frames. Shorten it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When RX aggregation starts, transmitter may continue send frames
with SN smaller than SSN until the AddBA response is received.
However, the reorder buffer is already initialized at this point,
which will cause the drop of such frames as duplicates since the
head SN of the reorder buffer is set to the SSN, which is bigger.
Cc: stable@vger.kernel.org
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When NSSN is behind the reorder buffer due to timeout
the reorder timer isn't getting re-armed until NSSN
catches up. Fix it.
Fixes: 0690405fef ("iwlwifi: mvm: add reorder timeout per frame")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
In DQA mode, first_agg_queue is initialized to
IWL_MVM_DQA_MIN_DATA_QUEUE. This causes two bugs in the tx response
flow:
1. When TX fails, we set IEEE80211_TX_STAT_AMPDU_NO_BACK regardless
if we actually have aggregation open on the queue. This causes
mac80211 to send a BAR frame even though there is no aggregation
open.
Fix that by simply checking the AMPDU flag that is set on by
mac80211 for AMPDU packets.
2. When reclaiming frames in aggregation mode, we reclaim based on
scheduler ssn and not the SN.
The reason is that scheduler ssn may be ahead of SN due to a hole
in the BA window that was filled.
However, if we have aggregations open on IWL_MVM_DQA_BSS_CLIENT_QUEUE
the reclaim flow will still go to the code of non-aggregation
instead of the aggregation code since IWL_MVM_DQA_BSS_CLIENT_QUEUE
is smaller than IWL_MVM_DQA_MIN_DATA_QUEUE, although it is a valid
aggregation queue.
Fix that by always using the aggregation reclaim code by default in
DQA mode (currently it is implicitly used by default for all queues
except the reserved BSS queue).
Fixes: cf961e1662 ("iwlwifi: mvm: support dqa-mode agg on non-shared queue")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Support new version of beacon template command which deprecates
the use of the tx command inside.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Support differentiating between two phys for a000 devices
in order to load the correct firmware.
Eventually when moving completely to the new phy we will be
able to remove this.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
In DQA mode the check whether to decrement the pending frames
counter relies on the tid status and not on the txq id.
This may result in an inconsistent state of the pending frames
counter in case frame is queued on a non aggregation queue but
with this TID, and will be followed by a failure to remove the
station and later on SYSASSERT 0x3421 when trying to remove the
MAC.
Such frames are for example bar and qos NDPs.
Fix it by aligning the condition of incrementing the counter
with the condition of decrementing it - rely on TID state for
DQA mode.
Also, avoid internal error like this affecting station removal
for DQA mode - since we can know for sure it is an internal
error.
Fixes: cf961e1662 ("iwlwifi: mvm: support dqa-mode agg on non-shared queue")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Our 9000 device supports 64 bit DMA address for RX only, and
not for TX.
Setting DMA mask to 64 for the whole device is erroneous - we
can do it only for a000 devices where device is capable of
both RX & TX DMA with 64 bit address space.
Fixes: 96a6497bc3 ("iwlwifi: pcie: add 9000 series multi queue rx DMA support")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
When driver needs to access the contents of a streaming DMA buffer
without unmapping it it should call dma_sync_single_for_cpu().
Once the call has been made, the CPU "owns" the DMA buffer and can
work with it as needed.
Before the device accesses the buffer, however, ownership should be
transferred back to it with dma_sync_single_for_device().
Both calls weren't performed by the driver, resulting with odd paging
errors on some platforms. Fix it.
Fixes: a6c4fb4441 ("iwlwifi: mvm: Add FW paging mechanism for the UMAC on PCI")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Support getting alive from two LMACs and dumping debug
data from both.
Deprecate older alive notifications no one is using.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
If iwl_mvm_find_free_queue() doesn't find a free queue, it will
return an inactive one.
However, not all the call paths free this queue before reassigning
it, which is a bug.
Check it in other paths and act accordingly.
Fixes: 9794c64f30 ("iwlwifi: mvm: support dqa queue inactivation upon timeout")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For CDB devices we will want to configure scan parameters
per band.
Support the new scan API for now. Logic per band will be
added in the future.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For CDB arch there is another auxiliary mac.
Support statistics APIs that were changed to reflect that.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Currently we have up to 3 phy contexts - defined by NUM_PHY_CTX.
However - some code paths validate the ID by using MAX_PHYS define
which is set to 4.
While there is no harm it is incorrect - since the maximum is 3.
Remove the define and use the correct one.
Cleanup the code a bit while at it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For a000 devices the INIT and RT images are unified to one
image.
The changes in the flow are the following:
* Driver load only RT firmware - meaning that the nvm access
command will be done in the RT image load flow.
* A new command (NVM_ACCESS_COMPLETE) now signals to the FW that
the driver is done accessing the NVM and FW can proceed with phy
calibrations.
* Phy DB is no longer sent from INIT FW to be restored by driver
for the RT FW - all the phy DB is now internal to the FW.
INIT complete will now follow the NVM access command, without
phy DB calls before.
* Paging command is sent earlier in the flow before NVM access
to enable a complete load of FW.
* caution must be care when restart is called since we may have
not completed init flow even though we are in RT firmware.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For old firmwares the memory wasn't allocated, resulting in panic.
Make it dynamically allocated as well. Allow any order of functions
call.
Fixes: eef187a7b8 ("iwlwifi: enlarge number of ucode sections")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Tested-by: Kalle Valo <kvalo@codeaurora.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
When removing inactive queue - use the central disable queue
function. This is needed due to a000 changes to come, but is
a proper cleanup anyway.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For 9000 devices withg upload, FW cannot send the entire phy
flags, and will send only the band - which is what we really
care about anyway. Change is backward compatible.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
In preparation for the new TX cmd - move the rate calculation
to a diffrent function, and make it independent of the TX
command.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Currently the function changes the TX cmd itself.
Make it more generic by returning a value, as preperation
to the new TX cmd.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
The maximum number of firmware sections is now 32 instead of 16 for
a000 devices. Set the appropriate define. Avoid out of bounds access
in case there are more sections than the maximum set by driver.
Make the driver extensible to FW size changes by allocating the
section memory dynamically.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
The logic in the paging calculations is a bit hard to
follow. Replace it with a kernel define.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Some of the code there is duplicate while the only change is
the block size. Unifying it shortens the code and make the
difference clearer.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Currently mac80211 determines whether HW does fragmentation
by checking whether the set_frag_threshold callback is set
or not.
However, some drivers may want to set the HW fragmentation
capability depending on HW generation.
Allow this by checking a HW flag instead of checking the
callback.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
[added the flag to ath10k and wlcore]
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Currently when configuring the cab queue the scheduler is configured
without station id - which results in station id 0.
In DQA mode this causes firmware to assert later on when the actual
station 0 is added with an empty tfd_queue_mask.
Fix that by configuring the queue to the broadcast station.
This is a bit trickier since the queue should not be included in the
tfd_queue_mask of the ADD_STA since it is a multicast queue, and the
tfd_queue_mask is only unicast queue. As a result the queue should be
enabled only after the broadcast station is added.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
When we sync the RX queues the driver waits to receive echo
notification on all the RX queues.
The wait queue is set with timeout until all queues have received
the notification.
However, iwl_mvm_rx_queue_notif() never woke up the wait queue,
with the result of the counter value being checked only when the
timeout expired.
This may cause a latency of up to 1 second.
Fixes: 0636b93821 ("iwlwifi: mvm: implement driver RX queues sync command")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
The firmware has a new smart linker, and this table can now be
in ICCM or in SMEM. It is not hardcoded, but depends on code
size. Allow the full range.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Currently it is logged as UNKNOWN.
Also, 0x6c seems to be the permanent ID for this command, remove
incorrect comment and uncomment the command from the commands
list.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For automatic testing packet injection can be useful.
Support injection through debugfs.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
On default queue we will not receive frame release notification,
but the BAR itself.
Upon receiving the BAR driver should look at the NSSN and adjust
window accordingly.
Fixes: b915c10174 ("iwlwifi: mvm: add reorder buffer per queue")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
There was recently a Full AP mode feature added where hostap adds
the station at auth stage, and not assoc.
However, when running with legacy hostapd, we get the auth response
before station was added, and tx_skb_non_sta fails to allocate
a queue, resulting in a complete failure of association.
Take care of this situation as well.
Add a warning when no valid queue is returned at all and make mvm
drop the packet instead of passing it on.
Refactor the function a bit while at it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Use skb_queue_empty() and not skb_peek_tail() to check for
empty list.
Avoid a redundant check as well - loop will take care of it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Function is very indented. Go to msi section if needed to avoid
it and by that make the code more readable.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Support new format. TX response will not be sent anymore,
so all needed data is in the BA response.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
The original intent was to have the general iwl_queue shared
between RX and TX queues, but it is not the actual status.
Since it is not shared with any struct but iwl_txq, it adds
unnecessary complexity. Merge those structs.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Since TFD was enlarged to 256 bytes, the fetch of the TFD
itself is very expensive.
To make DRAM to SRAM more efficient, bits 12-13 will indicate
the number of 64 byte chunks that should be transferred to
SRAM.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Previous patch introduced the new formats. This patch
allocates the new structures and adjusts code accordingly.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
In future HW the byte count table address will be configured
by ucode per queue. Add API to expose the byte count table to
the opmode
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
In a000 devices we have 15 fifos, so in the shared memory
config the number of tx fifos in the array was changed
accordingly.
As it is in the middle of the struct, the parsing code needs
to be duplicated.
To minimize the duplication, do not save variables we never
actually use.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Firmware may lock those registers for access. This results
in 9000 devices with a bus stall and an endless loop of 0x5a5a5a.
Don't dump those registers.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
New hardware supports bigger TFDs and TBs.
Introduce the new formats and adjust defines and code
relying on old format.
Changing the actual TFD allocation is trickier and
deferred to the next patch.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Since the SCD_QUEUE_CFG command was introduced the driver
calls iwl_trans_txq_enable_cfg() with a NULL for scd_cfg
parameter.
This makes the transport avoid writing to the SCD pointers,
since it can cause races with firmware, which is also accessing
the registers.
The transport only updates the write pointer in that case.
Fix a wrong call to iwl_trans_txq_enable() which caused a
scd_cfg parameter to be sent to transport, resulting with an
access to SCD registers.
Fixes: 58f2cc57dc ("iwlwifi: mvm: support dqa-mode scd queue redirection")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
If device family is 8000 then iwl_pcie_load_cpu_sections()
won't be called at all (iwl_pcie_load_cpu_sections_8000() is
called in that case) so this piece of code never gets called.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Turns out we should access TFH relative addresses.
Also, the FH_UCODE_LOAD_STATUS was replaced by
UREG_UCODE_LOAD_STATUS.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Up till now we accessed SCD configuration only for initial
configuration and for enabling command queue.
For a000 generation the command queue is open by default
and firmware configures the rest. No driver SCD accesses
are expected. Make sure this is the case.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For a000 devices there is a support of 64 bit DMA addressing.
The paging command was changed accordingly - support it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
The 9000 hardware will de-aggregate AMSDUs. In the process
it will copy the mac header "as is" to the new MPDUs.
This means driver should allow the same PN for MPDUs originated
from the same AMSDU.
Do that by incrementing the PN only for the last MPDU in the
sequence.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
All transports has this structure. By moving it to be
shared, we can get rid of casting to the specific transport
in probe and remove.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Centralize the logging of SCD status. The motivation is
that for a000 devices we will have new SCD HW, but this
code was duplicate anyway, so it is a proper cleanup.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For a000 device the FH was replaced by the TFH.
This is the first patch in a series introducing the
changes stemming from this change.
This patch initializes the TFQ queue table with the new
64 bit register and the relevant TFH configuration
registers.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Move the write_prph_64 of pcie to be transport agnostic.
Add direct write as well, as it is needed for a000 HW.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Currently the scratch buffer is set to 16 bytes and indicates
the size of the bi-directional DMA.
However, next HW generation will perform additional offloading,
and will write the result in the key location of the TX command,
so the size of the bi-directional consistent memory should grow
accordingly - increase it to 40.
Generalize the code to get rid of now irrelevant scratch references.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
In MQ environment and new architecture in early stages
we may encounter DMA issues. Track RXB status and bail
out in case we receive index to an RXB that was not
mapped and handed over to HW.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Restart flow zeroes the rx_ba_sessions counter. Mac80211 asks
driver to tear down of the session only afterwards, and as a
result driver didn't free the data. Fix it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Fixes: 10b2b2019d ("iwlwifi: mvm: add infrastructure for tracking BA session in driver")
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
In cases of hardware or DMA error, the vid read from
a zeroed location will be 0, and we will access the rxb
at index 0 in the global table, while it may be NULL or
owned by hardware.
Invalidate vid 0 in order to detect the situation and
bail out.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
This fixes a pretty ancient bug that hasn't manifested itself
until now.
The scratchbuf for command queue is allocated only for 32 slots
but is accessed with the queue write pointer - which can be
up to 256.
Since the scratch buf size was 16 and there are up to 256 TFDs
we never passed a page boundary when accessing the scratch buffer,
but when attempting to increase the size of the scratch buffer a
panic was quick to follow when trying to access the address resulted
in a page boundary.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Fixes: 38c0f334b3 ("iwlwifi: use coherent DMA memory for command header")
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Our HW does not support checksum of fragmented packets.
Fix code accordingly to checksum those packets in the driver.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Fixes: 5e6a98dc48 ("iwlwifi: mvm: enable TCP/UDP checksum support for 9000 family")
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For 9000 family we will get extended statistics notification
with averaged data for RSSI, TCM and rogue AP detection.
Support it. Future patches will added the required algorithms.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Somehow we ended up stopping RX using legacy RX registers
even for devices that support RFH. Fix it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Theoretically we may get only one IRQ from OS, in which
case we will have only 1 queue even in MSIx mode.
This will cause division by zero in the indirection table
calculation.
We do not need send the command in that case, as there is
only one queue so all RX traffic will be directed to it
anyway. Bail out early if there is only one queue.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Currently code calls restock for mq devices during the init
function, unlike sq where restock is called after init.
This causes an harmless but alarming deadlock warning from
lockdep, to fix this - unify the init code.
Rename the restock functions while at it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
We now have 9000 devices that support multiple frames in
a single RB. Enable it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
For 9000 devices we can have PCIe bus for discrete
devices and IOSF bus for integrated devices.
PCIe supports maximum transfer size of 128B while IOSF
bus supports maximum transfer size of 64B.
Configure RB size accordingly.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Integrated 9000 devices have a bug with shadow registers
value retention.
If driver writes RBD registers while MAC is asleep the
values are stored in shadow registers to be copied whenever
MAC wakes up.
However, in 9000 devices a MAC wakeup is not triggered
and when the bus powers down due to inactivity the shadow
values and dirty bits are lost.
Turn on the chicken-bits that cause MAC wakeup for RX-related
values as well when the device is in D0.
When the device is in low power mode turn the RX wakeup chicken
bits off since driver is idle and this W/A is not needed.
Remove previous W/A which was ineffective.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
rx_phy notification is no longer sent in devices with
multiple rx queues.
All the needed data is now set in the metadata - update
code accordingly to reflect all the features as in the
previous RX path.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
In multiple RX queues architecture, the RX_PHY notification
is no longer useful as it is received in the default queue
even for packets that are received on RSS queue, and cannot
be accessed without locking.
All the needed data is in the new RX packet metadata and
firmware will no longer send this notification for 9000
devices. Remove support of it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
We cannot trust NSSN for AMSDU sub-frames that are not the
last.
The reason is that NSSN advances on the first sub-frame,
and may cause the reorder buffer to advance before all the
sub-frames arrive.
Example:
Reorder buffer contains SN 0 & 2.
We receive AMSDU with SN 1 and NSSN for first sub frame 3.
The result us that driver releases SN 0,1, 2.
When sub-frame 1 arrives - reorder buffer is already ahead and
it will be dropped.
If the last sub-frame is not on this queue - we will get frame
release notification with up to date NSSN.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Some hardware (iwlwifi an example) de-aggregate AMSDUs and copy the IV
as is to the generated MPDUs, so the same PN appears in multiple
packets without being a replay attack. Allow driver to explicitly
indicate that a frame is allowed to have the same PN as the previous
frame.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When initializing RX we grab NIC access for every read and
write. This is redundant - we can just grab access once.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
The RX queues have a shadow register for the write pointer
that enables updates without grabbing NIC access. Use them
instead of the periphery registers because accessing those
is much more expensive.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Getting the mvm station out of station id requires dereferencing
the station id to get ieee80211_sta, then checking for pointer
validity and only then extract mvm station out.
Given that there are helpers to do it - use them instead of
duplicating the code whenever we need only mvm station.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
iwl_phy_db_set_section() is get called only from atomic
context, the alloc_ctx parameter is not needed. Remove it.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Driver is agnostic to the number of the phy_db entries and
only serves the firmware as a pipe to move the data from init
image to RT image.
As the size of the arrays may change (as it does in 9000 device)
allocate it dynamically. Firmware sends the largest index first
so we can use this to know how much we should allocate.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Up till now, the reorder buffer uses standard spec based comparison
when comparing the buffer status to NSSN. This indeed works for the
regular case, since we shouldn't cross the 2048 boundary without
getting a frame release notification.
However, this is problematic due to packet filtering that may be
performed by the FW while we are in d0i3. Theoretically we may
filter over 2048 packets, and then the check of the NSSN will get
incorrect.
Change the comparison to always trust nssn unless it is 64 or less
frames behind the head - which might happen due to a timeout.
This new comparison is to be used only when comparing reorder buffer
head with nssn, and not when comparing the packet SN to nssn or
reorder buffer head.
Put this in a separate commit as the logic is a bit tricky and
stands for its own commit message.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Currently code allows mvm reference to become negative and
only warns in case mvm reference is released while reference
counting is 0.
However, we better prevent this from happening at all since
iwl_mvm_unref() may race against iwl_mvm_unref_all_except()
which is called on restart.
As a result we might get the same reference unreferenced twice
ending with a negative value:
An example for an easily reproduced log:
[ 2689.909166] iwl_mvm_ref Take mvm reference - type 8
[ 2690.732716] iwl_mvm_unref_all_except Cleanup: remove mvm ref type 8 (1)
[ 2690.849708] iwl_mvm_unref Leave mvm reference - type 8
[ 2690.849721] WARNING: ... iwl_mvm_unref+0xb0/0xc0 [iwlmvm]()
If there will be yet another another restart iwl_mvm_unref_all_except
will run from 0 up to ref count, and since it is unsigned, we will throw
the transport ref count completely out of balance:
iwl_mvm_unref_all_except[I] -- Cleanup: remove mvm ref type 8 (255)
iwl_trans_slv_unref[I] -- rpm counter: 0
iwl_trans_slv_unref[I] -- rpm counter: -1
iwl_trans_slv_unref[I] -- rpm counter: -2
...
iwl_trans_slv_unref[I] -- rpm counter: -253
iwl_trans_slv_unref[I] -- rpm counter: -254
As there is no valid scenario where we can get to a negative
reference count - prevent it from happening.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Add a constant to allow disabling checksum. This will enable easier
debugging in early phases.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
The firmware will send frame release notification in order
to release "stuck" frames on a queue where no more frames
arrive on.
Upon receiving the message the driver shall indicate the frames
up to the NSSN.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Add a timer in order to release expired frames from the
reorder buffer.
This is needed since some APs do not retransmit frames
to fill in the reorder holes and in TCP it results with
a complete stall of traffic.
This has a few side effects on the general design:
The nssn may not reflect the the head of the reorder buffer.
This situation is valid, and packets with SN lower than the
reorder buffer head will be dropped.
Another side effect is that since the reorder timer might expire
we need to lock the reorder buffer.
This however is fine since the locking is only inside a
single reorder buffer between RX path and reorder timeout and
there is no outside contention.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Next hardware will direct packets to core based on the TCP/UDP
streams.
This logic can create holes in reorder buffer since packets that
belong to other stream were directed to a different core.
However, those are valid holes and the packets can be indicated
in L3 order.
The hardware will utilize a mechanism of informing the driver of
the normalized ssn and the driver shall release all packets that
SN is lower than the nssn.
This enables managing the reorder across the queues without sharing
any data between them.
The reorder buffer is allocated and released directly in the RX path
in order to avoid various races between control path and rx path.
The code utilizes the internal messaging to notify rx queues of when
to delete the reorder buffer.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
According to the spec when a BA session is started there
is a timeout set for the session in the ADDBA request.
If there is not activity on the TA/TID then the session
expires and a DELBA is sent.
In order to check for the timeout, data must be shared
among the rx queues.
Add a timer that runs as long as BA session is active
for the station and stops aggregation session if needed.
This patch also lays the infrastructure for the reordering
buffer which will be enabled in the next patches.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Currently the sync notification is a type of notification. However, it
is better fitted as an attribute of a notification, since there might
be another message in the payload (delba for instance) that should be
sent while control path is waiting for all queues to process.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
mac80211 will call the driver whenever there is a race between
RSS queues and control path that requires a processing of all
pending frames in RSS queues.
Implement that by utilizing the internal notification mechanism:
queue a message to all queues. When the message is received on
a queue it decrements the atomic counter. This guarantees that
all pending frames in the RX queue were processed since the message
is in order inside the queue.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
The firwmare name for 8000 is iwlwifi-8000C. The C is
appended based on a value read from a register. This
allows to load different firwmare versions based on
the hardware step during development. Now that the
hardware development is completed, we can hard code
the 'C' and along the way, fix the input to
MODULE_FIRMWARE.
This fixes:
https://bugzilla.kernel.org/show_bug.cgi?id=116041
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Currently the debugfs entry for starting aggregation session
starts it with timeout of 5 seconds. Allow opening a session
with a custom timeout (according to spec 0 is no timeout).
while at it, refactor the function and remove the magic numbers.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
NETIF_F_RXCSUM is not in the white list, though some
drivers may want to set it in order to enable seeing the
actual RX checksum status in ethtool.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Sara Sharon