* [PATCH 1/2] ath10k: reduce rx_lock contention for htt rx indication
@ 2016-02-12 6:10 Rajkumar Manoharan
2016-02-12 6:10 ` [PATCH 2/2] ath10k: process htt rx indication as batch mode Rajkumar Manoharan
2016-03-04 8:42 ` [PATCH 1/2] ath10k: reduce rx_lock contention for htt rx indication Valo, Kalle
0 siblings, 2 replies; 3+ messages in thread
From: Rajkumar Manoharan @ 2016-02-12 6:10 UTC (permalink / raw)
To: ath10k; +Cc: linux-wireless, Rajkumar Manoharan
Received frame indications are queued into a skb list and latest
processed by txrx tasklet. This skb queue is protected by htt rx lock.
Since the entire rx processing till delivering frame to mac80211 and
replenish tasks are processed under rx_lock protection, there might be
some delay in queuing newly received rx frame into that list on
multicore systems. Optimize this by using skb list lock while accessing
rx completion queue instead of htt rx lock.
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
---
drivers/net/wireless/ath/ath10k/htt_rx.c | 18 ++++++++----------
1 file changed, 8 insertions(+), 10 deletions(-)
diff --git a/drivers/net/wireless/ath/ath10k/htt_rx.c b/drivers/net/wireless/ath/ath10k/htt_rx.c
index cc957a6..bedd8c3 100644
--- a/drivers/net/wireless/ath/ath10k/htt_rx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_rx.c
@@ -2011,9 +2011,7 @@ void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
break;
}
case HTT_T2H_MSG_TYPE_RX_IND:
- spin_lock_bh(&htt->rx_ring.lock);
- __skb_queue_tail(&htt->rx_compl_q, skb);
- spin_unlock_bh(&htt->rx_ring.lock);
+ skb_queue_tail(&htt->rx_compl_q, skb);
tasklet_schedule(&htt->txrx_compl_task);
return;
case HTT_T2H_MSG_TYPE_PEER_MAP: {
@@ -2111,9 +2109,7 @@ void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
break;
}
case HTT_T2H_MSG_TYPE_RX_IN_ORD_PADDR_IND: {
- spin_lock_bh(&htt->rx_ring.lock);
- __skb_queue_tail(&htt->rx_in_ord_compl_q, skb);
- spin_unlock_bh(&htt->rx_ring.lock);
+ skb_queue_tail(&htt->rx_in_ord_compl_q, skb);
tasklet_schedule(&htt->txrx_compl_task);
return;
}
@@ -2174,16 +2170,18 @@ static void ath10k_htt_txrx_compl_task(unsigned long ptr)
dev_kfree_skb_any(skb);
}
- spin_lock_bh(&htt->rx_ring.lock);
- while ((skb = __skb_dequeue(&htt->rx_compl_q))) {
+ while ((skb = skb_dequeue(&htt->rx_compl_q))) {
resp = (struct htt_resp *)skb->data;
+ spin_lock_bh(&htt->rx_ring.lock);
ath10k_htt_rx_handler(htt, &resp->rx_ind);
+ spin_unlock_bh(&htt->rx_ring.lock);
dev_kfree_skb_any(skb);
}
- while ((skb = __skb_dequeue(&htt->rx_in_ord_compl_q))) {
+ while ((skb = skb_dequeue(&htt->rx_in_ord_compl_q))) {
+ spin_lock_bh(&htt->rx_ring.lock);
ath10k_htt_rx_in_ord_ind(ar, skb);
+ spin_unlock_bh(&htt->rx_ring.lock);
dev_kfree_skb_any(skb);
}
- spin_unlock_bh(&htt->rx_ring.lock);
}
--
2.7.0
_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k
^ permalink raw reply related [flat|nested] 3+ messages in thread* [PATCH 2/2] ath10k: process htt rx indication as batch mode
2016-02-12 6:10 [PATCH 1/2] ath10k: reduce rx_lock contention for htt rx indication Rajkumar Manoharan
@ 2016-02-12 6:10 ` Rajkumar Manoharan
2016-03-04 8:42 ` [PATCH 1/2] ath10k: reduce rx_lock contention for htt rx indication Valo, Kalle
1 sibling, 0 replies; 3+ messages in thread
From: Rajkumar Manoharan @ 2016-02-12 6:10 UTC (permalink / raw)
To: ath10k; +Cc: linux-wireless, Rajkumar Manoharan
On multicore systems, it is possible that txrx tasket can run
in parallel with pci tasklet (i.e smp affinity of ath10k irq is
assigned to multiple CPUs). Feeding and consuming from the same
rx completion list leads to txrx tasklet runs for longer period.
Prevent this by processing a snapshot of rx queue by moving list
into temporary list. Consecutive received frames will be processed
in next batch.
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
---
drivers/net/wireless/ath/ath10k/htt_rx.c | 16 ++++++++++++++--
1 file changed, 14 insertions(+), 2 deletions(-)
diff --git a/drivers/net/wireless/ath/ath10k/htt_rx.c b/drivers/net/wireless/ath/ath10k/htt_rx.c
index bedd8c3..61d9507 100644
--- a/drivers/net/wireless/ath/ath10k/htt_rx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_rx.c
@@ -2155,22 +2155,34 @@ static void ath10k_htt_txrx_compl_task(unsigned long ptr)
struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
struct ath10k *ar = htt->ar;
struct sk_buff_head tx_q;
+ struct sk_buff_head rx_q;
+ struct sk_buff_head rx_ind_q;
struct htt_resp *resp;
struct sk_buff *skb;
unsigned long flags;
__skb_queue_head_init(&tx_q);
+ __skb_queue_head_init(&rx_q);
+ __skb_queue_head_init(&rx_ind_q);
spin_lock_irqsave(&htt->tx_compl_q.lock, flags);
skb_queue_splice_init(&htt->tx_compl_q, &tx_q);
spin_unlock_irqrestore(&htt->tx_compl_q.lock, flags);
+ spin_lock_irqsave(&htt->rx_compl_q.lock, flags);
+ skb_queue_splice_init(&htt->rx_compl_q, &rx_q);
+ spin_unlock_irqrestore(&htt->rx_compl_q.lock, flags);
+
+ spin_lock_irqsave(&htt->rx_in_ord_compl_q.lock, flags);
+ skb_queue_splice_init(&htt->rx_in_ord_compl_q, &rx_ind_q);
+ spin_unlock_irqrestore(&htt->rx_in_ord_compl_q.lock, flags);
+
while ((skb = __skb_dequeue(&tx_q))) {
ath10k_htt_rx_frm_tx_compl(htt->ar, skb);
dev_kfree_skb_any(skb);
}
- while ((skb = skb_dequeue(&htt->rx_compl_q))) {
+ while ((skb = __skb_dequeue(&rx_q))) {
resp = (struct htt_resp *)skb->data;
spin_lock_bh(&htt->rx_ring.lock);
ath10k_htt_rx_handler(htt, &resp->rx_ind);
@@ -2178,7 +2190,7 @@ static void ath10k_htt_txrx_compl_task(unsigned long ptr)
dev_kfree_skb_any(skb);
}
- while ((skb = skb_dequeue(&htt->rx_in_ord_compl_q))) {
+ while ((skb = __skb_dequeue(&rx_ind_q))) {
spin_lock_bh(&htt->rx_ring.lock);
ath10k_htt_rx_in_ord_ind(ar, skb);
spin_unlock_bh(&htt->rx_ring.lock);
--
2.7.0
_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k
^ permalink raw reply related [flat|nested] 3+ messages in thread* Re: [PATCH 1/2] ath10k: reduce rx_lock contention for htt rx indication
2016-02-12 6:10 [PATCH 1/2] ath10k: reduce rx_lock contention for htt rx indication Rajkumar Manoharan
2016-02-12 6:10 ` [PATCH 2/2] ath10k: process htt rx indication as batch mode Rajkumar Manoharan
@ 2016-03-04 8:42 ` Valo, Kalle
1 sibling, 0 replies; 3+ messages in thread
From: Valo, Kalle @ 2016-03-04 8:42 UTC (permalink / raw)
To: Manoharan, Rajkumar
Cc: linux-wireless@vger.kernel.org, ath10k@lists.infradead.org
Rajkumar Manoharan <rmanohar@qti.qualcomm.com> writes:
> Received frame indications are queued into a skb list and latest
> processed by txrx tasklet. This skb queue is protected by htt rx lock.
> Since the entire rx processing till delivering frame to mac80211 and
> replenish tasks are processed under rx_lock protection, there might be
> some delay in queuing newly received rx frame into that list on
> multicore systems. Optimize this by using skb list lock while accessing
> rx completion queue instead of htt rx lock.
>
> Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Both applied, thanks.
--
Kalle Valo
_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2016-03-04 8:43 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2016-02-12 6:10 [PATCH 1/2] ath10k: reduce rx_lock contention for htt rx indication Rajkumar Manoharan
2016-02-12 6:10 ` [PATCH 2/2] ath10k: process htt rx indication as batch mode Rajkumar Manoharan
2016-03-04 8:42 ` [PATCH 1/2] ath10k: reduce rx_lock contention for htt rx indication Valo, Kalle
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox