More than 20000 PPDU id jumping causing status ring and destination
ring processing not sync. The status ring is processed and the
destination ring is not processed. Since destination is not reaped for
so long, backpressure occurs at the destination ring.
To address this issue update the PPDU id with the latest PPDU, this
will allow the destination ring to be reaped and will prevent the
rings from getting out of sync.
Signed-off-by: Karthikeyan Kathirvel <[email protected]>
---
v2:
Corrected the signed off details
drivers/net/wireless/ath/ath11k/dp.h | 3 +++
drivers/net/wireless/ath/ath11k/dp_rx.c | 42 ++++++++++++++++++++++++++++-----
2 files changed, 39 insertions(+), 6 deletions(-)
diff --git a/drivers/net/wireless/ath/ath11k/dp.h b/drivers/net/wireless/ath/ath11k/dp.h
index b0d0e7b..cc36b57 100644
--- a/drivers/net/wireless/ath/ath11k/dp.h
+++ b/drivers/net/wireless/ath/ath11k/dp.h
@@ -101,6 +101,8 @@ struct ath11k_pdev_mon_stats {
u32 dest_mpdu_drop;
u32 dup_mon_linkdesc_cnt;
u32 dup_mon_buf_cnt;
+ u32 dest_mon_stuck;
+ u32 dest_mon_not_reaped;
};
struct dp_link_desc_bank {
@@ -143,6 +145,7 @@ struct ath11k_mon_data {
struct ath11k_pdev_dp {
u32 mac_id;
+ u32 mon_dest_ring_stuck_cnt;
atomic_t num_tx_pending;
wait_queue_head_t tx_empty_waitq;
struct dp_rxdma_ring rx_refill_buf_ring;
diff --git a/drivers/net/wireless/ath/ath11k/dp_rx.c b/drivers/net/wireless/ath/ath11k/dp_rx.c
index 3a9c362..d0b8829 100644
--- a/drivers/net/wireless/ath/ath11k/dp_rx.c
+++ b/drivers/net/wireless/ath/ath11k/dp_rx.c
@@ -4939,6 +4939,12 @@ static int ath11k_dp_rx_mon_deliver(struct ath11k *ar, u32 mac_id,
return -EINVAL;
}
+/* The destination ring processing is stuck if the destination is not
+ * moving while status ring moves 16 PPDU. The destination ring processing
+ * skips this destination ring PPDU as a workaround.
+ */
+#define MON_DEST_RING_STUCK_MAX_CNT 16
+
static void ath11k_dp_rx_mon_dest_process(struct ath11k *ar, int mac_id,
u32 quota, struct napi_struct *napi)
{
@@ -4951,6 +4957,7 @@ static void ath11k_dp_rx_mon_dest_process(struct ath11k *ar, int mac_id,
u32 ring_id;
struct ath11k_pdev_mon_stats *rx_mon_stats;
u32 npackets = 0;
+ u32 mpdu_rx_bufs_used;
if (ar->ab->hw_params.rxdma1_enable)
ring_id = dp->rxdma_mon_dst_ring.ring_id;
@@ -4980,16 +4987,39 @@ static void ath11k_dp_rx_mon_dest_process(struct ath11k *ar, int mac_id,
head_msdu = NULL;
tail_msdu = NULL;
- rx_bufs_used += ath11k_dp_rx_mon_mpdu_pop(ar, mac_id, ring_entry,
- &head_msdu,
- &tail_msdu,
- &npackets, &ppdu_id);
+ mpdu_rx_bufs_used = ath11k_dp_rx_mon_mpdu_pop(ar, mac_id, ring_entry,
+ &head_msdu,
+ &tail_msdu,
+ &npackets, &ppdu_id);
+
+ rx_bufs_used += mpdu_rx_bufs_used;
+
+ if (mpdu_rx_bufs_used) {
+ dp->mon_dest_ring_stuck_cnt = 0;
+ } else {
+ dp->mon_dest_ring_stuck_cnt++;
+ rx_mon_stats->dest_mon_not_reaped++;
+ }
+
+ if (dp->mon_dest_ring_stuck_cnt > MON_DEST_RING_STUCK_MAX_CNT) {
+ rx_mon_stats->dest_mon_stuck++;
+ ath11k_dbg(ar->ab, ATH11K_DBG_DATA,
+ "status ring ppdu_id=%d dest ring ppdu_id=%d mon_dest_ring_stuck_cnt=%d dest_mon_not_reaped=%u dest_mon_stuck=%u\n",
+ pmon->mon_ppdu_info.ppdu_id, ppdu_id,
+ dp->mon_dest_ring_stuck_cnt,
+ rx_mon_stats->dest_mon_not_reaped,
+ rx_mon_stats->dest_mon_stuck);
+ pmon->mon_ppdu_info.ppdu_id = ppdu_id;
+ continue;
+ }
if (ppdu_id != pmon->mon_ppdu_info.ppdu_id) {
pmon->mon_ppdu_status = DP_PPDU_STATUS_START;
ath11k_dbg(ar->ab, ATH11K_DBG_DATA,
- "dest_rx: new ppdu_id %x != status ppdu_id %x",
- ppdu_id, pmon->mon_ppdu_info.ppdu_id);
+ "dest_rx: new ppdu_id %x != status ppdu_id %x dest_mon_not_reaped = %u dest_mon_stuck = %u\n",
+ ppdu_id, pmon->mon_ppdu_info.ppdu_id,
+ rx_mon_stats->dest_mon_not_reaped,
+ rx_mon_stats->dest_mon_stuck);
break;
}
if (head_msdu && tail_msdu) {
--
2.7.4
Karthikeyan Kathirvel <[email protected]> wrote:
> More than 20000 PPDU id jumping causing status ring and destination
> ring processing not sync. The status ring is processed and the
> destination ring is not processed. Since destination is not reaped for
> so long, backpressure occurs at the destination ring.
>
> To address this issue update the PPDU id with the latest PPDU, this
> will allow the destination ring to be reaped and will prevent the
> rings from getting out of sync.
>
> Signed-off-by: Karthikeyan Kathirvel <[email protected]>
> Signed-off-by: Kalle Valo <[email protected]>
Tested-on tag missing, please reply to this message and I can add it to the commit log.
--
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
Karthikeyan Kathirvel <[email protected]> wrote:
> More than 20000 PPDU id jumping causing status ring and destination
> ring processing not sync. The status ring is processed and the
> destination ring is not processed. Since destination is not reaped for
> so long, backpressure occurs at the destination ring.
>
> To address this issue update the PPDU id with the latest PPDU, this
> will allow the destination ring to be reaped and will prevent the
> rings from getting out of sync.
>
> Signed-off-by: Karthikeyan Kathirvel <[email protected]>
> Signed-off-by: Kalle Valo <[email protected]>
No Tested-on tag provided.
Patch set to Changes Requested.
--
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches