From: Johannes Berg <[email protected]>
It turns out that this wasn't a good idea, I hit a test failure in
hwsim due to this. That particular failure was easily worked around,
but it raised questions: if an AP needs to, for example, send action
frames to each connected station, the current limit is nowhere near
enough (especially if those stations are sleeping and the frames are
queued for a while.)
Shuffle around some bits to make more room for ack_frame_id to allow
up to 8192 queued up frames, that's enough for queueing 4 frames to
each connected station, even at the maximum of 2007 stations on a
single AP.
We take the bits from band (which currently only 2 but I leave 3 in
case we add another band) and from the hw_queue, which can only need
4 since it has a limit of 16 queues.
Fixes: 6912daed05e1 ("mac80211: Shrink the size of ack_frame_id to make room for tx_time_est")
Signed-off-by: Johannes Berg <[email protected]>
---
include/net/mac80211.h | 11 +++++------
net/mac80211/cfg.c | 2 +-
net/mac80211/tx.c | 2 +-
3 files changed, 7 insertions(+), 8 deletions(-)
diff --git a/include/net/mac80211.h b/include/net/mac80211.h
index 6d4ea71523d2..64a249445c86 100644
--- a/include/net/mac80211.h
+++ b/include/net/mac80211.h
@@ -1007,12 +1007,11 @@ ieee80211_rate_get_vht_nss(const struct ieee80211_tx_rate *rate)
struct ieee80211_tx_info {
/* common information */
u32 flags;
- u8 band;
-
- u8 hw_queue;
-
- u16 ack_frame_id:6;
- u16 tx_time_est:10;
+ u32 band:3,
+ ack_frame_id:13,
+ hw_queue:4,
+ tx_time_est:10;
+ /* 2 free bits */
union {
struct {
diff --git a/net/mac80211/cfg.c b/net/mac80211/cfg.c
index a11bd1669c13..5715c4606cff 100644
--- a/net/mac80211/cfg.c
+++ b/net/mac80211/cfg.c
@@ -3431,7 +3431,7 @@ int ieee80211_attach_ack_skb(struct ieee80211_local *local, struct sk_buff *skb,
spin_lock_irqsave(&local->ack_status_lock, spin_flags);
id = idr_alloc(&local->ack_status_frames, ack_skb,
- 1, 0x40, GFP_ATOMIC);
+ 1, 0x2000, GFP_ATOMIC);
spin_unlock_irqrestore(&local->ack_status_lock, spin_flags);
if (id < 0) {
diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c
index 4296d9d71311..4520cda6f523 100644
--- a/net/mac80211/tx.c
+++ b/net/mac80211/tx.c
@@ -2443,7 +2443,7 @@ static int ieee80211_store_ack_skb(struct ieee80211_local *local,
spin_lock_irqsave(&local->ack_status_lock, flags);
id = idr_alloc(&local->ack_status_frames, ack_skb,
- 1, 0x40, GFP_ATOMIC);
+ 1, 0x2000, GFP_ATOMIC);
spin_unlock_irqrestore(&local->ack_status_lock, flags);
if (id >= 0) {
--
2.24.1
Johannes Berg <[email protected]> writes:
> From: Johannes Berg <[email protected]>
>
> It turns out that this wasn't a good idea, I hit a test failure in
> hwsim due to this. That particular failure was easily worked around,
> but it raised questions: if an AP needs to, for example, send action
> frames to each connected station, the current limit is nowhere near
> enough (especially if those stations are sleeping and the frames are
> queued for a while.)
>
> Shuffle around some bits to make more room for ack_frame_id to allow
> up to 8192 queued up frames, that's enough for queueing 4 frames to
> each connected station, even at the maximum of 2007 stations on a
> single AP.
>
> We take the bits from band (which currently only 2 but I leave 3 in
> case we add another band) and from the hw_queue, which can only need
> 4 since it has a limit of 16 queues.
>
> Fixes: 6912daed05e1 ("mac80211: Shrink the size of ack_frame_id to make room for tx_time_est")
> Signed-off-by: Johannes Berg <[email protected]>
Fair enough :)
Acked-by: Toke Høiland-Jørgensen <[email protected]>