diff mbox series

[1/2] wifi: mac80211: export ieee80211_purge_tx_queue() for drivers

Message ID 20240819065855.25678-1-pkshih@realtek.com
State New
Headers show
Series [1/2] wifi: mac80211: export ieee80211_purge_tx_queue() for drivers | expand

Commit Message

Ping-Ke Shih Aug. 19, 2024, 6:58 a.m. UTC
Drivers need to purge TX SKB when stopping. Use skb_queue_purge() can't
report TX status to mac80211, causing ieee80211_free_ack_frame() warns
"Have pending ack frames!".

Signed-off-by: Ping-Ke Shih <pkshih@realtek.com>
---
Hi Johannes,

If this patch is accepted, please merge it ahead and I will merge the
second patch via rtw tree after pulling wireless-next tree. I think this
can reduce possible conflicts.

---
 include/net/mac80211.h | 11 +++++++++++
 net/mac80211/status.c  |  1 +
 2 files changed, 12 insertions(+)

Comments

Ping-Ke Shih Aug. 20, 2024, 1:33 a.m. UTC | #1
Johannes Berg <johannes@sipsolutions.net> wrote:
> > +/**
> > + * ieee80211_purge_tx_queue - purge TX skb queue
> > + * @hw: the hardware
> > + * @skbs: the skbs
> > + *
> > + * Free a set of transmit skbs. Use this function when device is going to stop
> > + * but some transmit skbs without TX status are still queued.
> > + */
> > +void ieee80211_purge_tx_queue(struct ieee80211_hw *hw,
> > +                           struct sk_buff_head *skbs);
> 
> Unlike skb_queue_purge()/skb_queue_purge_reason(), this doesn't take the
> lock, that seems important to note here?

Thanks. I wasn't aware of lock, and rtw88 needs a lock version because
its TX work isn't stopped yet while calling ieee80211_purge_tx_queue().

I have three candidate options:

1. add spin_lock by driver before calling ieee80211_purge_tx_queue().

2. move original ieee80211_purge_tx_queue() to __ieee80211_purge_tx_queue(),
   and add an new lock version of ieee80211_purge_tx_queue().

3. change rtw88 deinit flow to ensure stopping TX work before calling
   ieee80211_purge_tx_queue()

I prefer option 3. But want your opinion if option 2 is worth to do?

Ping-Ke
Ping-Ke Shih Aug. 22, 2024, 1:46 a.m. UTC | #2
Ping-Ke Shih <pkshih@realtek.com> wrote:
> Johannes Berg <johannes@sipsolutions.net> wrote:
> > > +/**
> > > + * ieee80211_purge_tx_queue - purge TX skb queue
> > > + * @hw: the hardware
> > > + * @skbs: the skbs
> > > + *
> > > + * Free a set of transmit skbs. Use this function when device is going to stop
> > > + * but some transmit skbs without TX status are still queued.
> > > + */
> > > +void ieee80211_purge_tx_queue(struct ieee80211_hw *hw,
> > > +                           struct sk_buff_head *skbs);
> >
> > Unlike skb_queue_purge()/skb_queue_purge_reason(), this doesn't take the
> > lock, that seems important to note here?
> 
> Thanks. I wasn't aware of lock, and rtw88 needs a lock version because
> its TX work isn't stopped yet while calling ieee80211_purge_tx_queue().
> 
> I have three candidate options:
> 
> 1. add spin_lock by driver before calling ieee80211_purge_tx_queue().
> 
> 2. move original ieee80211_purge_tx_queue() to __ieee80211_purge_tx_queue(),
>    and add an new lock version of ieee80211_purge_tx_queue().
> 
> 3. change rtw88 deinit flow to ensure stopping TX work before calling
>    ieee80211_purge_tx_queue()
> 
> I prefer option 3. But want your opinion if option 2 is worth to do?
> 

I have chosen option 3, so just export ieee80211_purge_tx_queue() and no
need to touch ieee80211_purge_tx_queue() further. Sent v2 along with
suggested comments.
diff mbox series

Patch

diff --git a/include/net/mac80211.h b/include/net/mac80211.h
index 0a04eaf5343c..f11844f0c80f 100644
--- a/include/net/mac80211.h
+++ b/include/net/mac80211.h
@@ -3180,6 +3180,17 @@  ieee80211_get_alt_retry_rate(const struct ieee80211_hw *hw,
  */
 void ieee80211_free_txskb(struct ieee80211_hw *hw, struct sk_buff *skb);
 
+/**
+ * ieee80211_purge_tx_queue - purge TX skb queue
+ * @hw: the hardware
+ * @skbs: the skbs
+ *
+ * Free a set of transmit skbs. Use this function when device is going to stop
+ * but some transmit skbs without TX status are still queued.
+ */
+void ieee80211_purge_tx_queue(struct ieee80211_hw *hw,
+			      struct sk_buff_head *skbs);
+
 /**
  * DOC: Hardware crypto acceleration
  *
diff --git a/net/mac80211/status.c b/net/mac80211/status.c
index dd8f857a1fbc..d1cf987de13b 100644
--- a/net/mac80211/status.c
+++ b/net/mac80211/status.c
@@ -1301,3 +1301,4 @@  void ieee80211_purge_tx_queue(struct ieee80211_hw *hw,
 	while ((skb = __skb_dequeue(skbs)))
 		ieee80211_free_txskb(hw, skb);
 }
+EXPORT_SYMBOL(ieee80211_purge_tx_queue);