From: Karthik M <[email protected]>
During stress test with maximum VAPs and peer connected, below warning
is seen:
[ 1079.110967] ath12k_pci 0004:01:00.0: mgmt tx queue is full
[ 1079.117708] ath12k_pci 0004:01:00.0: failed to queue management frame -28
[ 1079.123191] ath12k_pci 0004:01:00.0: mgmt tx queue is full
[ 1079.129960] ath12k_pci 0004:01:00.0: failed to queue management frame -28
[ 1079.135641] ath12k_pci 0004:01:00.0: mgmt tx queue is full
This is caused by potential race condition while accessing skb_queue_len().
When ath12k_mgmt_over_wmi_tx_work() and ath12k_mac_mgmt_tx() is called concurrently,
then skb_queue_len() might fetch list length which is modified by skb_queue_tail()
or skb_dequeue().
Replace skb_queue_len() with skb_queue_len_lockless() which will
prevent concurrent modified access using READ_ONCE(). And also use '>=',
in case we queue a few SKBs simultaneously.
Tested-on: QCN9274 hw2.0 PCI WLAN.WBE.1.0.1-00029-QCAHKSWPL_SILICONZ-1
Signed-off-by: Karthik M <[email protected]>
Signed-off-by: Ramya Gnanasekar <[email protected]>
diff --git a/drivers/net/wireless/ath/ath12k/mac.c b/drivers/net/wireless/ath/ath12k/mac.c
index 0e5cc477ba56..cc458fed5e78 100644
--- a/drivers/net/wireless/ath/ath12k/mac.c
+++ b/drivers/net/wireless/ath/ath12k/mac.c
@@ -4489,7 +4489,7 @@ static int ath12k_mac_mgmt_tx(struct ath12k *ar, struct sk_buff *skb,
return -ENOSPC;
}
- if (skb_queue_len(q) == ATH12K_TX_MGMT_NUM_PENDING_MAX) {
+ if (skb_queue_len_lockless(q) >= ATH12K_TX_MGMT_NUM_PENDING_MAX) {
ath12k_warn(ar->ab, "mgmt tx queue is full\n");
return -ENOSPC;
}
--
2.17.1