diff mbox

[1/2] rt2x00: pause almost full queue early

Message ID 1513683236-13402-1-git-send-email-sgruszka@redhat.com (mailing list archive)
State Accepted
Commit 3d8f162cb8ecd8cb5f77d92ed27bdc7d8aeb2bae
Delegated to: Kalle Valo
Headers show

Commit Message

Stanislaw Gruszka Dec. 19, 2017, 11:33 a.m. UTC
Do not drop &queue->tx_lock and acquire it again to pause queue when
available entries are below the threshold.

Patch should mitigate problem of frequently printed errors:
"Error - Dropping frame due to full tx queue"

Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
---
 drivers/net/wireless/ralink/rt2x00/rt2x00mac.c   | 10 ----------
 drivers/net/wireless/ralink/rt2x00/rt2x00queue.c |  8 ++++++++
 2 files changed, 8 insertions(+), 10 deletions(-)

Comments

Enrico Mioso Dec. 26, 2017, 11:01 a.m. UTC | #1
On a WL-330N3G device, these patches seems to make the situation much better.
In other words, I wasn't able to reproduce the stall on this chipset.
On MT7620 a stall was never seen directly, but tests with these patches revealed no stall.
So I think it's good to have these merged.

Tested-by: Enrico Mioso@gmail.com


On Tue, 19 Dec 2017, Stanislaw Gruszka wrote:

> Date: Tue, 19 Dec 2017 12:33:55
> From: Stanislaw Gruszka <sgruszka@redhat.com>
> To: linux-wireless@vger.kernel.org
> Cc: Helmut Schaa <helmut.schaa@googlemail.com>,
>     Enrico Mioso <mrkiko.rs@gmail.com>, Daniel Golle <daniel@makrotopia.org>,
>     Felix Fietkau <nbd@nbd.name>, Stanislaw Gruszka <sgruszka@redhat.com>
> Subject: [PATCH 1/2] rt2x00: pause almost full queue early
> 
> Do not drop &queue->tx_lock and acquire it again to pause queue when
> available entries are below the threshold.
>
> Patch should mitigate problem of frequently printed errors:
> "Error - Dropping frame due to full tx queue"
>
> Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
> ---
> drivers/net/wireless/ralink/rt2x00/rt2x00mac.c   | 10 ----------
> drivers/net/wireless/ralink/rt2x00/rt2x00queue.c |  8 ++++++++
> 2 files changed, 8 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c b/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
> index ecc96312a370..c8a6f163102f 100644
> --- a/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
> +++ b/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
> @@ -152,16 +152,6 @@ void rt2x00mac_tx(struct ieee80211_hw *hw,
> 	if (unlikely(rt2x00queue_write_tx_frame(queue, skb, control->sta, false)))
> 		goto exit_fail;
>
> -	/*
> -	 * Pausing queue has to be serialized with rt2x00lib_txdone(). Note
> -	 * we should not use spin_lock_bh variant as bottom halve was already
> -	 * disabled before ieee80211_xmit() call.
> -	 */
> -	spin_lock(&queue->tx_lock);
> -	if (rt2x00queue_threshold(queue))
> -		rt2x00queue_pause_queue(queue);
> -	spin_unlock(&queue->tx_lock);
> -
> 	return;
>
>  exit_fail:
> diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c b/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
> index a2c1ca5c76d1..1ad51e56bc59 100644
> --- a/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
> +++ b/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
> @@ -715,6 +715,14 @@ int rt2x00queue_write_tx_frame(struct data_queue *queue, struct sk_buff *skb,
> 	rt2x00queue_kick_tx_queue(queue, &txdesc);
>
> out:
> +	/*
> +	 * Pausing queue has to be serialized with rt2x00lib_txdone(), so we
> +	 * do this under queue->tx_lock. Bottom halve was already disabled
> +	 * before ieee80211_xmit() call.
> +	 */
> +	if (rt2x00queue_threshold(queue))
> +		rt2x00queue_pause_queue(queue);
> +
> 	spin_unlock(&queue->tx_lock);
> 	return ret;
> }
> -- 
> 1.9.3
>
>
Enrico Mioso Dec. 26, 2017, 11:02 a.m. UTC | #2
Oops, sorry, wrong tag

Tested-by: Enrico Mioso <mrkiko.rs@gmail.com>


On Tue, 19 Dec 2017, Stanislaw Gruszka wrote:

> Date: Tue, 19 Dec 2017 12:33:55
> From: Stanislaw Gruszka <sgruszka@redhat.com>
> To: linux-wireless@vger.kernel.org
> Cc: Helmut Schaa <helmut.schaa@googlemail.com>,
>     Enrico Mioso <mrkiko.rs@gmail.com>, Daniel Golle <daniel@makrotopia.org>,
>     Felix Fietkau <nbd@nbd.name>, Stanislaw Gruszka <sgruszka@redhat.com>
> Subject: [PATCH 1/2] rt2x00: pause almost full queue early
> 
> Do not drop &queue->tx_lock and acquire it again to pause queue when
> available entries are below the threshold.
>
> Patch should mitigate problem of frequently printed errors:
> "Error - Dropping frame due to full tx queue"
>
> Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
> ---
> drivers/net/wireless/ralink/rt2x00/rt2x00mac.c   | 10 ----------
> drivers/net/wireless/ralink/rt2x00/rt2x00queue.c |  8 ++++++++
> 2 files changed, 8 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c b/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
> index ecc96312a370..c8a6f163102f 100644
> --- a/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
> +++ b/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
> @@ -152,16 +152,6 @@ void rt2x00mac_tx(struct ieee80211_hw *hw,
> 	if (unlikely(rt2x00queue_write_tx_frame(queue, skb, control->sta, false)))
> 		goto exit_fail;
>
> -	/*
> -	 * Pausing queue has to be serialized with rt2x00lib_txdone(). Note
> -	 * we should not use spin_lock_bh variant as bottom halve was already
> -	 * disabled before ieee80211_xmit() call.
> -	 */
> -	spin_lock(&queue->tx_lock);
> -	if (rt2x00queue_threshold(queue))
> -		rt2x00queue_pause_queue(queue);
> -	spin_unlock(&queue->tx_lock);
> -
> 	return;
>
>  exit_fail:
> diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c b/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
> index a2c1ca5c76d1..1ad51e56bc59 100644
> --- a/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
> +++ b/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
> @@ -715,6 +715,14 @@ int rt2x00queue_write_tx_frame(struct data_queue *queue, struct sk_buff *skb,
> 	rt2x00queue_kick_tx_queue(queue, &txdesc);
>
> out:
> +	/*
> +	 * Pausing queue has to be serialized with rt2x00lib_txdone(), so we
> +	 * do this under queue->tx_lock. Bottom halve was already disabled
> +	 * before ieee80211_xmit() call.
> +	 */
> +	if (rt2x00queue_threshold(queue))
> +		rt2x00queue_pause_queue(queue);
> +
> 	spin_unlock(&queue->tx_lock);
> 	return ret;
> }
> -- 
> 1.9.3
>
>
Kalle Valo Jan. 8, 2018, 5:40 p.m. UTC | #3
Stanislaw Gruszka <sgruszka@redhat.com> wrote:

> Do not drop &queue->tx_lock and acquire it again to pause queue when
> available entries are below the threshold.
> 
> Patch should mitigate problem of frequently printed errors:
> "Error - Dropping frame due to full tx queue"
> 
> Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
> Tested-by: Enrico Mioso@gmail.com
> Tested-by: Enrico Mioso <mrkiko.rs@gmail.com>

2 patches applied to wireless-drivers-next.git, thanks.

3d8f162cb8ec rt2x00: pause almost full queue early
6dd80efd75ce rt2x00: do not pause queue unconditionally on error path
diff mbox

Patch

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c b/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
index ecc96312a370..c8a6f163102f 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00mac.c
@@ -152,16 +152,6 @@  void rt2x00mac_tx(struct ieee80211_hw *hw,
 	if (unlikely(rt2x00queue_write_tx_frame(queue, skb, control->sta, false)))
 		goto exit_fail;
 
-	/*
-	 * Pausing queue has to be serialized with rt2x00lib_txdone(). Note
-	 * we should not use spin_lock_bh variant as bottom halve was already
-	 * disabled before ieee80211_xmit() call.
-	 */
-	spin_lock(&queue->tx_lock);
-	if (rt2x00queue_threshold(queue))
-		rt2x00queue_pause_queue(queue);
-	spin_unlock(&queue->tx_lock);
-
 	return;
 
  exit_fail:
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c b/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
index a2c1ca5c76d1..1ad51e56bc59 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00queue.c
@@ -715,6 +715,14 @@  int rt2x00queue_write_tx_frame(struct data_queue *queue, struct sk_buff *skb,
 	rt2x00queue_kick_tx_queue(queue, &txdesc);
 
 out:
+	/*
+	 * Pausing queue has to be serialized with rt2x00lib_txdone(), so we
+	 * do this under queue->tx_lock. Bottom halve was already disabled
+	 * before ieee80211_xmit() call.
+	 */
+	if (rt2x00queue_threshold(queue))
+		rt2x00queue_pause_queue(queue);
+
 	spin_unlock(&queue->tx_lock);
 	return ret;
 }