iwlwifi: pcie: fix TX while flushing

When flushing TX queues no new TX should go into the system.
However, in the following scenario we get TX:
1. Queues are stopped and there are packets in overflow queue
2. Station is removed and flush begins
3. Flush empties space, and reclaim path TXes SKB from overflow
   queue.

Note that the fact the queues are stopped during the process
doesn't matter - the packet will be TXed since the TX path
doesn't care if TX queues are stopped or not, just if there is
space in the queue, which there is, since we just freed a
packet.

A fix here is rather complicated, since the flow is very racy.

Change code not to warn if we are TXing from overflow TX.
In case there is TX from both overflow TX and TX path we will
miss a warning we optimally had, but we can live with that.

Make sure we don't return before overflow queue is empty, otherwise
we will think queues are empty, but they will be refilled, resulting
with assert.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Fixes: 3955525d5d17 ("iwlwifi: pcie: buffer packets to avoid overflowing Tx queues")
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

1 files changed, 10 insertions, 0 deletions

diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/tx.c b/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
index d8773e0a6062..9fbd37d23e85 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/tx.c
@@ -1182,6 +1182,15 @@ void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
 		skb_queue_splice_init(&txq->overflow_q, &overflow_skbs);
 
 		/*
+		 * We are going to transmit from the overflow queue.
+		 * Remember this state so that wait_for_txq_empty will know we
+		 * are adding more packets to the TFD queue. It cannot rely on
+		 * the state of &txq->overflow_q, as we just emptied it, but
+		 * haven't TXed the content yet.
+		 */
+		txq->overflow_tx = true;
+
+		/*
 		 * This is tricky: we are in reclaim path which is non
 		 * re-entrant, so noone will try to take the access the
 		 * txq data from that path. We stopped tx, so we can't
@@ -1209,6 +1218,7 @@ void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
 			iwl_wake_queue(trans, txq);
 
 		spin_lock_bh(&txq->lock);
+		txq->overflow_tx = false;
 	}
 
 	if (txq->read_ptr == txq->write_ptr) {