commit e2288b66fe7ff0288382b2af671b4da558b44472 upstream. Since we clear QUEUE_STARTED in rt2x00queue_stop_queue(), following call to rt2x00queue_pause_queue() reduce to noop, i.e we do not stop queue in mac80211. To fix that introduce rt2x00queue_pause_queue_nocheck() function, which will stop queue in mac80211 directly. Note that rt2x00_start_queue() explicitly set QUEUE_PAUSED bit. Note also that reordering operations i.e. first call to rt2x00queue_pause_queue() and then clear QUEUE_STARTED bit, will race with rt2x00queue_unpause_queue(), so calling ieee80211_stop_queue() directly is the only available solution to fix the problem without major rework. Signed-off-by: Stanislaw Gruszka <stf_xl@wp.pl> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3780d038fdf4b5ef26ead10b0604ab1f46dd9510 upstream. Is possible that we stop queue and then do not wake up it again, especially when packets are transmitted fast. That can be easily reproduced with modified tx queue entry_num to some small value e.g. 16. If mac80211 already hold local->queue_stop_reason_lock, then we can wait on that lock in both rt2x00queue_pause_queue() and rt2x00queue_unpause_queue(). After drooping ->queue_stop_reason_lock is possible that __ieee80211_wake_queue() will be performed before __ieee80211_stop_queue(), hence we stop queue and newer wake up it again. Another race condition is possible when between rt2x00queue_threshold() check and rt2x00queue_pause_queue() we will process all pending tx buffers on different cpu. This might happen if for example interrupt will be triggered on cpu performing rt2x00mac_tx(). To prevent race conditions serialize pause/unpause by queue->tx_lock. Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com> Acked-by: Gertjan van Wingerde <gwingerde@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 77a861c405da75d81e9e6e32c50eb7f9777777e8 upstream. The rt2x00 driver gets frequent occurrences of the following error message when operating under load: phy0 -> rt2x00queue_write_tx_frame: Error - Arrived at non-free entry in the non-full queue 2. This is caused by simultaneous attempts from mac80211 to send a frame via rt2x00, which are not properly serialized inside rt2x00queue_write_tx_frame, causing the second frame to fail sending with the above mentioned error message. Fix this by introducing a per-queue spinlock to serialize the TX operations on that queue. Reported-by: Andreas Hartmann <andihartmann@01019freenet.de> Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com> Acked-by: Helmut Schaa <helmut.schaa@googlemail.com> Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Cc: Tim Gardner <tim.gardner@canonical.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Conflicts: drivers/net/wireless/iwlwifi/iwl-core.c drivers/net/wireless/rt2x00/rt2x00queue.c drivers/net/wireless/rt2x00/rt2x00queue.h
The two functions that are in rt2x00ht.c can be much better placed closer to the places where the call-sites of these functions are (one in rt2x00config.c and one in rt2x00queue.c) allowing us to make these functions static. Also, conditional compilations doesn't seem to be necessary anymore as 802.11n support is quite common nowadays. This makes the code a bit easier readable and searchable. Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com> Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
When powersaving is enabled, assocaition times are very high (for WPA2 networks, the time can easily be around the 3 seconds). This is caused, because the flushing of the queues takes too much time. Without the flushing callback mac80211 assumes a timeout of 100ms while scanning. Limit all flush waiting loops to the same maximum. We can apply this maximum by passing the drop status to the driver, which makes sure the driver performs extra actions during the waiting for the queue to become empty. After these changes, association times fall within the healthy range of ~0.6 seconds with powersaving enabled. The difference between association time between powersaving enabled and disabled is now only ~0.1 second (which can also be due to the measuring method). Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com> Acked-by: Gertjan van Wingerde <gwingerde@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
Add a timestamp to each queue entry which is updated whenever the status of the entry changes, and remove the per-queue timestamps. The previous check was incorrect and caused both false positives and false negatives. With the corrected check it comes apparent that the TX status usually times out on rt2800usb unless there is sufficient traffic (i.e. the next TX will complete the previous TX status). Signed-off-by: Johannes Stezenbach <js@sig21.net> Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
Allow passing a void pointer to rt2x00_queue_entry_for_each which in turn in provided to the callback function. Furthermore, allow the callback function to stop processing by returning true. And also notify the caller of rt2x00_queue_entry_for_each if the loop was canceled by the callback. No functional changes, just preparation for an upcoming patch. Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com> Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com> Acked-by: Gertjan van Wingerde <gwingerde@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
The number of flags defined for the rt2x00dev->flags field, has been growing over the years. Currently we are approaching the maximum number of bits which are available in the field. A secondary problem, is that one part of the field are initialized only during boot, because the driver requirements are initialized or device requirements are loaded from the EEPROM. In both cases, the flags are fixed and will not change during device operation. The other flags are the device state, and will change frequently. So far this resulted in the fact that for some flags, the atomic bit accessors are used, while for the others the non-atomic variants are used. By splitting the flags up into a "flags" and "cap_flags" we can put all flags which are fixed inside "cap_flags". This field can then be read non-atomically. In the "flags" field we keep the device state, which is going to be read atomically. This adds more room for more flags in the future, and sanitizes the field access methods. Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com> Acked-by: Helmut Schaa <helmut.schaa@googlemail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>
Conflicts: drivers/net/wireless/ath/ar9170/main.c drivers/net/wireless/ath/ar9170/phy.c drivers/net/wireless/zd1211rw/zd_rf_rf2959.c