Skip to contentSkip to conversation

mac80211: temporarily disable reorder release timer

PATCH In review Christian Lamparter 2 participants 5 messages Oct 8, 2010 mac80211

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 mac80211: temporarily disable reorder release timer

Several serve threading problems in the current release reorder timer implementation have been discovered.

A lengthy discussion - which lists some of the pitfalls and possible solutions - can be found at:
http://marc.info/?t=128635927000001

But due to the complicated nature of the subject and the imminent advent of a new -rc cycle, it was decided to disable the feature for the time being.

---
v1 -> v2:
fix compiling error
net/mac80211/rx.c: In function ‘ieee80211_sta_reorder_release’:
net/mac80211/rx.c:665: error: label at end of compound statement
---

net/mac80211/rx.c +24 -0
index b67221d..aa9b08d 100644
@@ -622,6 +622,26 @@ static void ieee80211_sta_reorder_release(struct ieee80211_hw *hw,
 							tid_agg_rx->buf_size;
 	}
 
+	/*
+	 * Disable the reorder release timer for now.
+	 *
+	 * The current implementation lacks a proper locking scheme
+	 * which would protect vital statistic and debug counters
+	 * from being updated by two different but concurrent BHs.
+	 *
+	 * More information about the topic is available from:
+	 * - thread: http://marc.info/?t=128635927000001
+	 *
+	 * What was wrong:
+	 * =>  http://marc.info/?l=linux-wireless&m=128636170811964
+	 * "Basically the thing is that until your patch, the data
+	 *  in the struct didn't actually need locking because it
+	 *  was accessed by the RX path only which is not concurrent."
+	 *
+	 * List of what needs to be fixed:
+	 * => http://marc.info/?l=linux-wireless&m=128656352920957
+	 *
+
 	if (tid_agg_rx->stored_mpdu_num) {
 		j = index = seq_sub(tid_agg_rx->head_seq_num,
 				    tid_agg_rx->ssn) % tid_agg_rx->buf_size;
@@ -640,6 +660,10 @@ static void ieee80211_sta_reorder_release(struct ieee80211_hw *hw,
 	} else {
 		del_timer(&tid_agg_rx->reorder_timer);
 	}
+	*/
+
+set_release_timer:
+	return;
 }
 
 /*
#1 lore raw
Quoted 8 lines · Christian Lamparter

Several serve threading problems in the current release reorder timer implementation have been discovered.

A lengthy discussion - which lists some of the pitfalls and possible solutions - can be found at:
 http://marc.info/?t=128635927000001

Quoted 5 lines

But due to the complicated nature of the subject and the imminent advent of a new -rc cycle, it was decided to disable the feature for the time being.

This patch lacks documentation of the impact of this patch, while it may fix a lock issue, the does not address what happens when the patch is actually applied.

Luis

#2 lore raw
Quoted 14 lines · On Friday 08 October 2010 22:40:39 Luis R. Rodriguez wrote:
Quoted 12 lines · Christian Lamparter

Several serve threading problems in the current release reorder timer implementation have been discovered.

A lengthy discussion - which lists some of the pitfalls and possible solutions - can be found at:
http://marc.info/?t=128635927000001

But due to the complicated nature of the subject and the imminent advent of a new -rc cycle, it was decided to disable the feature for the time being.

Quoted 5 lines
Quoted 1 line

This patch lacks documentation of the impact of this patch, while it may fix a lock issue, the does not address what happens when the patch is actually applied.

? It just disables the timer. The actual reorder code is left untouched, everything is the same as in Linus' 2.6.36-rcX tree.

Or what impact are you talking about?

#3 lore raw
Quoted 17 lines · Christian Lamparter
Quoted 16 lines · On Friday 08 October 2010 22:40:39 Luis R. Rodriguez wrote:
Quoted 13 lines · Christian Lamparter

Several serve threading problems in the current release reorder timer implementation have been discovered.

A lengthy discussion - which lists some of the pitfalls and possible solutions - can be found at:
 http://marc.info/?t=128635927000001

But due to the complicated nature of the subject and the imminent advent of a new -rc cycle, it was decided to disable the feature for the time being.

Quoted 8 lines
Quoted 3 lines

This patch lacks documentation of the impact of this patch, while it may fix a lock issue, the does not address what happens when the patch is actually applied.

? It just disables the timer. The actual reorder code is left untouched, everything is the same as in Linus' 2.6.36-rcX tree.

Or what impact are you talking about?

The timer is what I'm talking about, we don't run then ieee80211_release_reorder_timeout() right so we don't clear stale frames on aggregates. What impact does this have?

Luis

#4 lore raw
Quoted 26 lines · On Friday 08 October 2010 23:22:57 Luis R. Rodriguez wrote:
Quoted 24 lines · Christian Lamparter
Quoted 14 lines · On Friday 08 October 2010 22:40:39 Luis R. Rodriguez wrote:
Quoted 12 lines · Christian Lamparter

Several serve threading problems in the current release reorder timer implementation have been discovered.

A lengthy discussion - which lists some of the pitfalls and possible solutions - can be found at:
http://marc.info/?t=128635927000001

But due to the complicated nature of the subject and the imminent advent of a new -rc cycle, it was decided to disable the feature for the time being.

Quoted 5 lines
Quoted 1 line

This patch lacks documentation of the impact of this patch, while it may fix a lock issue, the does not address what happens when the patch is actually applied.

? It just disables the timer. The actual reorder code is left untouched, everything is the same as in Linus' 2.6.36-rcX tree.

Quoted 5 lines
Quoted 1 line

Or what impact are you talking about?

The timer is what I'm talking about, we don't run then ieee80211_release_reorder_timeout() right so we don't clear stale frames on aggregates.

What? Who told you that? Because that's just misinformation.

This patch doesn't touch/disable the reorder release mechanism (ieee80211_sta_reorder_release) at all.
It only disarms the timer which would clean out frames after approx. 100ms.

And even without the timer, there are still three other ways for cleaning out "stale" frame aggregates.

1. BAR from peer (ieee80211_rx_h_ctrl)

2. BAW window overflow (ieee80211_sta_manage_reorder_buf, part of the aMPDU rx path - executed for every received MPDU)

3. ieee80211_sta_reorder_release (automatically called by ieee80211_sta_manage_reorder_buf when a reorder gaps are opening up.)

(I repeat, the rx path logic & behavior is identical to vanilla 2.6.36-rcX)

Quoted 1 line

What impact does this have?

everything is back to stock, at least for now?

see:
http://marc.info/?l=linux-wireless&m=128656352920957

"I don't like reverting these patches, but maybe we should simply comment out the code that arms the timer, thereby disabling all of it, while we work on this?"
-- Johannes

Best regards,
Christian