summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorSara Sharon <sara.sharon@intel.com>2017-10-17 12:06:52 +0300
committerLuca Coelho <luciano.coelho@intel.com>2017-11-03 11:56:08 +0200
commit14a1f85bdc27283898358af09ee506d7f3344d3c (patch)
treedf543e07d4ea9d2a9587bf68fe1b5feb3c91b81d
parent82d2b9a62699f3d6a61323774c41cebe920e5dc5 (diff)
downloadop-kernel-dev-14a1f85bdc27283898358af09ee506d7f3344d3c.zip
op-kernel-dev-14a1f85bdc27283898358af09ee506d7f3344d3c.tar.gz
iwlwifi: mvm: improve latency when there is a reorder timeout
When there is a reorder timeout, we may get to a situation where we have the timeout latency for all the next 64 frames. This happens since NSSN is behind for a while, and the driver won't release the frames, since it is not allowed by NSSN. As a result the frame is stored in the reorder buffer although there is no hole, and released 100 ms later. Add a direct comparison to the reorder buffer head, and release immediately if possible. For example: Frame 0 is missed. We receive frame 1, and store it in the buffer. After 100 ms, frame 1 is released and reorder buffer head is 2. We then receive frame 2, with NSSN 0, and store it instead of releasing it. Signed-off-by: Sara Sharon <sara.sharon@intel.com> Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
-rw-r--r--drivers/net/wireless/intel/iwlwifi/mvm/rxmq.c16
1 files changed, 16 insertions, 0 deletions
diff --git a/drivers/net/wireless/intel/iwlwifi/mvm/rxmq.c b/drivers/net/wireless/intel/iwlwifi/mvm/rxmq.c
index b84756d..9852a4d 100644
--- a/drivers/net/wireless/intel/iwlwifi/mvm/rxmq.c
+++ b/drivers/net/wireless/intel/iwlwifi/mvm/rxmq.c
@@ -719,6 +719,22 @@ static bool iwl_mvm_reorder(struct iwl_mvm *mvm,
return false;
}
+ /*
+ * release immediately if there are no stored frames, and the sn is
+ * equal to the head.
+ * This can happen due to reorder timer, where NSSN is behind head_sn.
+ * When we released everything, and we got the next frame in the
+ * sequence, according to the NSSN we can't release immediately,
+ * while technically there is no hole and we can move forward.
+ */
+ if (!buffer->num_stored && sn == buffer->head_sn) {
+ if (!amsdu || last_subframe)
+ buffer->head_sn = ieee80211_sn_inc(buffer->head_sn);
+ /* No need to update AMSDU last SN - we are moving the head */
+ spin_unlock_bh(&buffer->lock);
+ return false;
+ }
+
index = sn % buffer->buf_size;
/*
OpenPOWER on IntegriCloud