2016-10-11 09:28:41

by Felix Fietkau

[permalink] [raw]
Subject: [PATCH] mac80211: fix sequence number allocation regression

The recent commit that moved around TX handlers dropped the sequence
number allocation at the end of ieee80211_tx_dequeue and calls
ieee80211_tx_h_sequence instead (for the non-fast-xmit case).
However, it did not change the fast-xmit sequence allocation condition
in ieee80211_xmit_fast_finish, which skipped seqno alloc if intermediate
tx queues are being used.

Drop the now obsolete condition.

Fixes: bb42f2d13ffc ("mac80211: Move reorder-sensitive TX handlers to after TXQ dequeue")
Signed-off-by: Felix Fietkau <[email protected]>
---
net/mac80211/tx.c | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c
index dc60784..8ff81fd 100644
--- a/net/mac80211/tx.c
+++ b/net/mac80211/tx.c
@@ -3212,7 +3212,6 @@ static void ieee80211_xmit_fast_finish(struct ieee80211_sub_if_data *sdata,
struct sk_buff *skb)
{
struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
- struct ieee80211_local *local = sdata->local;
struct ieee80211_hdr *hdr = (void *)skb->data;
u8 tid = IEEE80211_NUM_TIDS;

@@ -3224,8 +3223,7 @@ static void ieee80211_xmit_fast_finish(struct ieee80211_sub_if_data *sdata,
if (hdr->frame_control & cpu_to_le16(IEEE80211_STYPE_QOS_DATA)) {
tid = skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK;
*ieee80211_get_qos_ctl(hdr) = tid;
- if (!ieee80211_get_txq(local, &sdata->vif, &sta->sta, skb))
- hdr->seq_ctrl = ieee80211_tx_next_seq(sta, tid);
+ hdr->seq_ctrl = ieee80211_tx_next_seq(sta, tid);
} else {
info->flags |= IEEE80211_TX_CTL_ASSIGN_SEQ;
hdr->seq_ctrl = cpu_to_le16(sdata->sequence_number);
--
2.10.1


2016-10-12 06:49:51

by Johannes Berg

[permalink] [raw]
Subject: Re: [PATCH] mac80211: fix sequence number allocation regression

On Tue, 2016-10-11 at 11:28 +0200, Felix Fietkau wrote:
> The recent commit that moved around TX handlers dropped the sequence
> number allocation at the end of ieee80211_tx_dequeue and calls
> ieee80211_tx_h_sequence instead (for the non-fast-xmit case).
> However, it did not change the fast-xmit sequence allocation
> condition
> in ieee80211_xmit_fast_finish, which skipped seqno alloc if
> intermediate
> tx queues are being used.
>
> Drop the now obsolete condition.

Hm. I don't know what tree you're looking at, but it looks like I *did*
in fact resolve this correctly; the (now) erroneous condition doesn't
exist in mac80211 nor in mac80211-next (nor net, net-next, linux).

It does seem to exist in wireless-testing, but that's not something I
can control.

johannes

2016-10-12 12:41:26

by Bob Copeland

[permalink] [raw]
Subject: Re: [PATCH] mac80211: fix sequence number allocation regression

On Wed, Oct 12, 2016 at 08:49:45AM +0200, Johannes Berg wrote:
> On Tue, 2016-10-11 at 11:28 +0200, Felix Fietkau wrote:
> > The recent commit that moved around TX handlers dropped the sequence
> > number allocation at the end of ieee80211_tx_dequeue and calls
> > ieee80211_tx_h_sequence instead (for the non-fast-xmit case).
> > However, it did not change the fast-xmit sequence allocation
> > condition
> > in ieee80211_xmit_fast_finish, which skipped seqno alloc if
> > intermediate
> > tx queues are being used.
> >
> > Drop the now obsolete condition.
>
> Hm. I don't know what tree you're looking at, but it looks like I *did*
> in fact resolve this correctly; the (now) erroneous condition doesn't
> exist in mac80211 nor in mac80211-next (nor net, net-next, linux).
>
> It does seem to exist in wireless-testing, but that's not something I
> can control.

Sorry about that, I should've rechecked this after Johannes did the merge.
I applied this patch (to wireless-testing) and pushed tag wt-2016-10-12.
Let me know of any further issues.

As a check I also diffed against mac80211-next/master and found additional
merge damage in the diff which I fixed up. Once the merge window closes
in a few days, I'll resync to the masters of all the trees.

--
Bob Copeland %% http://bobcopeland.com/

2016-10-11 09:41:45

by Toke Høiland-Jørgensen

[permalink] [raw]
Subject: Re: [PATCH] mac80211: fix sequence number allocation regression

Felix Fietkau <[email protected]> writes:

> The recent commit that moved around TX handlers dropped the sequence
> number allocation at the end of ieee80211_tx_dequeue and calls
> ieee80211_tx_h_sequence instead (for the non-fast-xmit case).
> However, it did not change the fast-xmit sequence allocation condition
> in ieee80211_xmit_fast_finish, which skipped seqno alloc if intermediate
> tx queues are being used.
>
> Drop the now obsolete condition.
>
> Fixes: bb42f2d13ffc ("mac80211: Move reorder-sensitive TX handlers to aft=
er TXQ dequeue")
> Signed-off-by: Felix Fietkau <[email protected]>

Acked-by: Toke H=C3=B8iland-J=C3=B8rgensen <[email protected]>

(Think this was introduced in the merging of mac80211-next and net-next?)

-Toke