1 files changed, 108 insertions, 12 deletions

diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c
index 48d8fbb1c220..8d35ce317f67 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/sge.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/sge.c
@@ -406,7 +406,7 @@ static void free_tx_desc(struct adapter *adap, struct sge_txq *q,
  */
 static inline int reclaimable(const struct sge_txq *q)
 {
-	int hw_cidx = ntohs(q->stat->cidx);
+	int hw_cidx = ntohs(ACCESS_ONCE(q->stat->cidx));
 	hw_cidx -= q->cidx;
 	return hw_cidx < 0 ? hw_cidx + q->size : hw_cidx;
 }
@@ -613,6 +613,7 @@ static unsigned int refill_fl(struct adapter *adap, struct sge_fl *q, int n,
 				       PCI_DMA_FROMDEVICE);
 		if (unlikely(dma_mapping_error(adap->pdev_dev, mapping))) {
 			__free_pages(pg, s->fl_pg_order);
+			q->mapping_err++;
 			goto out;   /* do not try small pages for this error */
 		}
 		mapping |= RX_LARGE_PG_BUF;
@@ -642,6 +643,7 @@ alloc_small_pages:
 				       PCI_DMA_FROMDEVICE);
 		if (unlikely(dma_mapping_error(adap->pdev_dev, mapping))) {
 			put_page(pg);
+			q->mapping_err++;
 			goto out;
 		}
 		*d++ = cpu_to_be64(mapping);
@@ -663,6 +665,7 @@ out:	cred = q->avail - cred;
 
 	if (unlikely(fl_starving(adap, q))) {
 		smp_wmb();
+		q->low++;
 		set_bit(q->cntxt_id - adap->sge.egr_start,
 			adap->sge.starving_fl);
 	}
@@ -1029,6 +1032,30 @@ static void inline_tx_skb(const struct sk_buff *skb, const struct sge_txq *q,
 		*p = 0;
 }
 
+static void *inline_tx_skb_header(const struct sk_buff *skb,
+				  const struct sge_txq *q,  void *pos,
+				  int length)
+{
+	u64 *p;
+	int left = (void *)q->stat - pos;
+
+	if (likely(length <= left)) {
+		memcpy(pos, skb->data, length);
+		pos += length;
+	} else {
+		memcpy(pos, skb->data, left);
+		memcpy(q->desc, skb->data + left, length - left);
+		pos = (void *)q->desc + (length - left);
+	}
+	/* 0-pad to multiple of 16 */
+	p = PTR_ALIGN(pos, 8);
+	if ((uintptr_t)p & 8) {
+		*p = 0;
+		return p + 1;
+	}
+	return p;
+}
+
 /*
  * Figure out what HW csum a packet wants and return the appropriate control
  * bits.
@@ -1320,7 +1347,7 @@ out_free:	dev_kfree_skb_any(skb);
  */
 static inline void reclaim_completed_tx_imm(struct sge_txq *q)
 {
-	int hw_cidx = ntohs(q->stat->cidx);
+	int hw_cidx = ntohs(ACCESS_ONCE(q->stat->cidx));
 	int reclaim = hw_cidx - q->cidx;
 
 	if (reclaim < 0)
@@ -1542,24 +1569,50 @@ static void ofldtxq_stop(struct sge_ofld_txq *q, struct sk_buff *skb)
 }
 
 /**
- *	service_ofldq - restart a suspended offload queue
+ *	service_ofldq - service/restart a suspended offload queue
  *	@q: the offload queue
  *
- *	Services an offload Tx queue by moving packets from its packet queue
- *	to the HW Tx ring.  The function starts and ends with the queue locked.
+ *	Services an offload Tx queue by moving packets from its Pending Send
+ *	Queue to the Hardware TX ring.  The function starts and ends with the
+ *	Send Queue locked, but drops the lock while putting the skb at the
+ *	head of the Send Queue onto the Hardware TX Ring.  Dropping the lock
+ *	allows more skbs to be added to the Send Queue by other threads.
+ *	The packet being processed at the head of the Pending Send Queue is
+ *	left on the queue in case we experience DMA Mapping errors, etc.
+ *	and need to give up and restart later.
+ *
+ *	service_ofldq() can be thought of as a task which opportunistically
+ *	uses other threads execution contexts.  We use the Offload Queue
+ *	boolean "service_ofldq_running" to make sure that only one instance
+ *	is ever running at a time ...
  */
 static void service_ofldq(struct sge_ofld_txq *q)
 {
-	u64 *pos;
+	u64 *pos, *before, *end;
 	int credits;
 	struct sk_buff *skb;
+	struct sge_txq *txq;
+	unsigned int left;
 	unsigned int written = 0;
 	unsigned int flits, ndesc;
 
+	/* If another thread is currently in service_ofldq() processing the
+	 * Pending Send Queue then there's nothing to do. Otherwise, flag
+	 * that we're doing the work and continue.  Examining/modifying
+	 * the Offload Queue boolean "service_ofldq_running" must be done
+	 * while holding the Pending Send Queue Lock.
+	 */
+	if (q->service_ofldq_running)
+		return;
+	q->service_ofldq_running = true;
+
 	while ((skb = skb_peek(&q->sendq)) != NULL && !q->full) {
-		/*
-		 * We drop the lock but leave skb on sendq, thus retaining
-		 * exclusive access to the state of the queue.
+		/* We drop the lock while we're working with the skb at the
+		 * head of the Pending Send Queue.  This allows more skbs to
+		 * be added to the Pending Send Queue while we're working on
+		 * this one.  We don't need to lock to guard the TX Ring
+		 * updates because only one thread of execution is ever
+		 * allowed into service_ofldq() at a time.
 		 */
 		spin_unlock(&q->sendq.lock);
 
@@ -1583,9 +1636,32 @@ static void service_ofldq(struct sge_ofld_txq *q)
 		} else {
 			int last_desc, hdr_len = skb_transport_offset(skb);
 
-			memcpy(pos, skb->data, hdr_len);
-			write_sgl(skb, &q->q, (void *)pos + hdr_len,
-				  pos + flits, hdr_len,
+			/* The WR headers  may not fit within one descriptor.
+			 * So we need to deal with wrap-around here.
+			 */
+			before = (u64 *)pos;
+			end = (u64 *)pos + flits;
+			txq = &q->q;
+			pos = (void *)inline_tx_skb_header(skb, &q->q,
+							   (void *)pos,
+							   hdr_len);
+			if (before > (u64 *)pos) {
+				left = (u8 *)end - (u8 *)txq->stat;
+				end = (void *)txq->desc + left;
+			}
+
+			/* If current position is already at the end of the
+			 * ofld queue, reset the current to point to
+			 * start of the queue and update the end ptr as well.
+			 */
+			if (pos == (u64 *)txq->stat) {
+				left = (u8 *)end - (u8 *)txq->stat;
+				end = (void *)txq->desc + left;
+				pos = (void *)txq->desc;
+			}
+
+			write_sgl(skb, &q->q, (void *)pos,
+				  end, hdr_len,
 				  (dma_addr_t *)skb->head);
 #ifdef CONFIG_NEED_DMA_MAP_STATE
 			skb->dev = q->adap->port[0];
@@ -1604,6 +1680,11 @@ static void service_ofldq(struct sge_ofld_txq *q)
 			written = 0;
 		}
 
+		/* Reacquire the Pending Send Queue Lock so we can unlink the
+		 * skb we've just successfully transferred to the TX Ring and
+		 * loop for the next skb which may be at the head of the
+		 * Pending Send Queue.
+		 */
 		spin_lock(&q->sendq.lock);
 		__skb_unlink(skb, &q->sendq);
 		if (is_ofld_imm(skb))
@@ -1611,6 +1692,11 @@ static void service_ofldq(struct sge_ofld_txq *q)
 	}
 	if (likely(written))
 		ring_tx_db(q->adap, &q->q, written);
+
+	/*Indicate that no thread is processing the Pending Send Queue
+	 * currently.
+	 */
+	q->service_ofldq_running = false;
 }
 
 /**
@@ -1624,9 +1710,19 @@ static int ofld_xmit(struct sge_ofld_txq *q, struct sk_buff *skb)
 {
 	skb->priority = calc_tx_flits_ofld(skb);       /* save for restart */
 	spin_lock(&q->sendq.lock);
+
+	/* Queue the new skb onto the Offload Queue's Pending Send Queue.  If
+	 * that results in this new skb being the only one on the queue, start
+	 * servicing it.  If there are other skbs already on the list, then
+	 * either the queue is currently being processed or it's been stopped
+	 * for some reason and it'll be restarted at a later time.  Restart
+	 * paths are triggered by events like experiencing a DMA Mapping Error
+	 * or filling the Hardware TX Ring.
+	 */
 	__skb_queue_tail(&q->sendq, skb);
 	if (q->sendq.qlen == 1)
 		service_ofldq(q);
+
 	spin_unlock(&q->sendq.lock);
 	return NET_XMIT_SUCCESS;
 }