[PATCH 1/2] ath9k: More xmit queue debugfs information.

[PATCH 1/2] ath9k: More xmit queue debugfs information.

[greearb]

From: Ben Greear <greearb@candelatech.com>

To try to figure out why xmit logic hangs.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
:100644 100644 650f00f... 9e009cc... M	drivers/net/wireless/ath/ath9k/debug.c
 drivers/net/wireless/ath/ath9k/debug.c |   26 ++++++++++++++++++++++++++
 1 files changed, 26 insertions(+), 0 deletions(-)

diff --git a/drivers/net/wireless/ath/ath9k/debug.c b/drivers/net/wireless/ath/ath9k/debug.c
index 650f00f..9e009cc 100644
--- a/drivers/net/wireless/ath/ath9k/debug.c
+++ b/drivers/net/wireless/ath/ath9k/debug.c
@@ -679,6 +679,32 @@ static ssize_t read_file_xmit(struct file *file, char __user *user_buf,
 		PRQLE(tmp, txq_fifo[i]);
 	}
 
+	/* Print out more detailed queue-info */
+	for (i = 0; i <= WME_AC_BK; i++) {
+		struct ath_txq *txq = &(sc->tx.txq[i]);
+		struct ath_atx_ac *ac;
+		struct ath_atx_tid *tid;
+		if (len >= size)
+			goto done;
+		spin_lock_bh(&txq->axq_lock);
+		if (!list_empty(&txq->axq_acq)) {
+			ac = list_first_entry(&txq->axq_acq, struct ath_atx_ac,
+					      list);
+			len += snprintf(buf + len, size - len,
+					"txq[%i] first-ac: %p sched: %i\n",
+					i, ac, ac->sched);
+			if (list_empty(&ac->tid_q) || (len >= size))
+				goto done_for;
+			tid = list_first_entry(&ac->tid_q, struct ath_atx_tid,
+					       list);
+			len += snprintf(buf + len, size - len,
+					" first-tid: %p sched: %i paused: %i\n",
+					tid, tid->sched, tid->paused);
+		}
+	done_for:
+		spin_unlock_bh(&txq->axq_lock);
+	}
+
 done:
 	if (len > size)
 		len = size;
-- 
1.7.2.3

[PATCH 2/2] ath9k: Fix incorrect tx-hang detection logic.

[greearb]

From: Ben Greear <greearb@candelatech.com>

It is not guaranteed that the ath_tx_complete_poll_work runs
after some fixed duration because the channel-reset logic
removes the work and then re-adds it to run immediately.
Two channel-changes 1ms apart, with a transmit was being
attempted, could thus incorrectly trigger a reset by
the ath_tx_complete_poll_work method.

Add a jiffies timestamp to ensure that at least 1 second
has elapsed before triggering the reset.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
:100644 100644 3f5c513... 93209d6... M	drivers/net/wireless/ath/ath9k/ath9k.h
:100644 100644 3aae523... e63de71... M	drivers/net/wireless/ath/ath9k/xmit.c
 drivers/net/wireless/ath/ath9k/ath9k.h |    1 +
 drivers/net/wireless/ath/ath9k/xmit.c  |   15 ++++++++++-----
 2 files changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h b/drivers/net/wireless/ath/ath9k/ath9k.h
index 3f5c513..93209d6 100644
--- a/drivers/net/wireless/ath/ath9k/ath9k.h
+++ b/drivers/net/wireless/ath/ath9k/ath9k.h
@@ -190,6 +190,7 @@ struct ath_txq {
 	spinlock_t axq_lock;
 	u32 axq_depth;
 	u32 axq_ampdu_depth;
+	unsigned long start_tx_timer; /* jiffies */
 	bool stopped;
 	bool axq_tx_inprogress;
 	struct list_head axq_acq;
diff --git a/drivers/net/wireless/ath/ath9k/xmit.c b/drivers/net/wireless/ath/ath9k/xmit.c
index 3aae523..e63de71 100644
--- a/drivers/net/wireless/ath/ath9k/xmit.c
+++ b/drivers/net/wireless/ath/ath9k/xmit.c
@@ -2097,6 +2097,7 @@ static void ath_tx_complete_poll_work(struct work_struct *work)
 	struct ath_txq *txq;
 	int i;
 	bool needreset = false;
+	unsigned long timeout = msecs_to_jiffies(ATH_TX_COMPLETE_POLL_INT);
 
 	for (i = 0; i < ATH9K_NUM_TX_QUEUES; i++)
 		if (ATH_TXQ_SETUP(sc, i)) {
@@ -2104,11 +2105,16 @@ static void ath_tx_complete_poll_work(struct work_struct *work)
 			spin_lock_bh(&txq->axq_lock);
 			if (txq->axq_depth) {
 				if (txq->axq_tx_inprogress) {
-					needreset = true;
-					spin_unlock_bh(&txq->axq_lock);
-					break;
+					if (time_after_eq(jiffies,
+							  txq->start_tx_timer +
+							  timeout)) {
+						needreset = true;
+						spin_unlock_bh(&txq->axq_lock);
+						break;
+					}
 				} else {
 					txq->axq_tx_inprogress = true;
+					txq->start_tx_timer = jiffies;
 				}
 			}
 			spin_unlock_bh(&txq->axq_lock);
@@ -2122,8 +2128,7 @@ static void ath_tx_complete_poll_work(struct work_struct *work)
 		ath9k_ps_restore(sc);
 	}
 
-	ieee80211_queue_delayed_work(sc->hw, &sc->tx_complete_work,
-			msecs_to_jiffies(ATH_TX_COMPLETE_POLL_INT));
+	ieee80211_queue_delayed_work(sc->hw, &sc->tx_complete_work, timeout);
 }
 
 
-- 
1.7.2.3

Re: [PATCH 2/2] ath9k: Fix incorrect tx-hang detection logic.

[Vasanthakumar Thiagarajan]

On Fri, Jan 07, 2011 at 11:30:59PM +0530, greearb@candelatech.com wrote:
> From: Ben Greear <greearb@candelatech.com>
> 
> It is not guaranteed that the ath_tx_complete_poll_work runs
> after some fixed duration because the channel-reset logic
> removes the work and then re-adds it to run immediately.
> Two channel-changes 1ms apart, with a transmit was being
> attempted, could thus incorrectly trigger a reset by
> the ath_tx_complete_poll_work method.

I don't think so. axq_tx_inprogress is reset in ath_draintxq().

Vasanth

Re: [PATCH 2/2] ath9k: Fix incorrect tx-hang detection logic.

[Ben Greear]

On 01/09/2011 09:38 PM, Vasanthakumar Thiagarajan wrote:
> On Fri, Jan 07, 2011 at 11:30:59PM +0530, greearb@candelatech.com wrote:
>> From: Ben Greear<greearb@candelatech.com>
>>
>> It is not guaranteed that the ath_tx_complete_poll_work runs
>> after some fixed duration because the channel-reset logic
>> removes the work and then re-adds it to run immediately.
>> Two channel-changes 1ms apart, with a transmit was being
>> attempted, could thus incorrectly trigger a reset by
>> the ath_tx_complete_poll_work method.
>
> I don't think so. axq_tx_inprogress is reset in ath_draintxq().

Ahhh, I see now.

I'll remove this patch from my queue and make sure it still runs
as well.

Thanks,
Ben

>
> Vasanth


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com