[PATCH net-next] net: ethernet: ti: davinci_cpdma: use idled submit

[PATCH net-next] net: ethernet: ti: davinci_cpdma: use idled submit

[Ivan Khoronzhuk]

While data pass suspend, reuse of rx descriptors can be disabled using
channel state & lock from cpdma layer. For this, submit to a channel
has to be disabled using state != "not active" under lock, what is done
with this patch. The same submit is used to fill rx channel while
ndo_open, when channel is idled, so add idled submit routine that
allows to prepare descs for the channel. All this simplifies code and
helps to avoid dormant mode usage and send packets only to active
channels, avoiding potential race in later on changes. Also add missed
sync barrier analogically like in other places after stopping tx
queues.

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
---

Based on net-next/master

 drivers/net/ethernet/ti/cpsw.c          | 15 ++---
 drivers/net/ethernet/ti/cpsw_ethtool.c  | 12 +---
 drivers/net/ethernet/ti/davinci_cpdma.c | 85 +++++++++++++++++++------
 drivers/net/ethernet/ti/davinci_cpdma.h |  2 +
 drivers/net/ethernet/ti/davinci_emac.c  |  4 +-
 5 files changed, 77 insertions(+), 41 deletions(-)

diff --git a/drivers/net/ethernet/ti/cpsw.c b/drivers/net/ethernet/ti/cpsw.c
index 3430503e1053..7bdd287074fc 100644
--- a/drivers/net/ethernet/ti/cpsw.c
+++ b/drivers/net/ethernet/ti/cpsw.c
@@ -457,16 +457,13 @@ static void cpsw_rx_handler(void *token, int len, int status)
 	}
 
 requeue:
-	if (netif_dormant(ndev)) {
-		dev_kfree_skb_any(new_skb);
-		return;
-	}
-
 	ch = cpsw->rxv[skb_get_queue_mapping(new_skb)].ch;
 	ret = cpdma_chan_submit(ch, new_skb, new_skb->data,
 				skb_tailroom(new_skb), 0);
-	if (WARN_ON(ret < 0))
+	if (ret < 0) {
+		WARN_ON(ret == -ENOMEM);
 		dev_kfree_skb_any(new_skb);
+	}
 }
 
 void cpsw_split_res(struct cpsw_common *cpsw)
@@ -1051,9 +1048,9 @@ int cpsw_fill_rx_channels(struct cpsw_priv *priv)
 			}
 
 			skb_set_queue_mapping(skb, ch);
-			ret = cpdma_chan_submit(cpsw->rxv[ch].ch, skb,
-						skb->data, skb_tailroom(skb),
-						0);
+			ret = cpdma_chan_idle_submit(cpsw->rxv[ch].ch, skb,
+						     skb->data,
+						     skb_tailroom(skb), 0);
 			if (ret < 0) {
 				cpsw_err(priv, ifup,
 					 "cannot submit skb to channel %d rx, error %d\n",
diff --git a/drivers/net/ethernet/ti/cpsw_ethtool.c b/drivers/net/ethernet/ti/cpsw_ethtool.c
index 86697b32194d..f60dc1dfc443 100644
--- a/drivers/net/ethernet/ti/cpsw_ethtool.c
+++ b/drivers/net/ethernet/ti/cpsw_ethtool.c
@@ -464,7 +464,6 @@ static void cpsw_suspend_data_pass(struct net_device *ndev)
 	cpsw_intr_disable(cpsw);
 
 	/* Stop all transmit queues for every network device.
-	 * Disable re-using rx descriptors with dormant_on.
 	 */
 	for (i = 0; i < cpsw->data.slaves; i++) {
 		ndev = cpsw->slaves[i].ndev;
@@ -472,7 +471,9 @@ static void cpsw_suspend_data_pass(struct net_device *ndev)
 			continue;
 
 		netif_tx_stop_all_queues(ndev);
-		netif_dormant_on(ndev);
+
+		/* Barrier, so that stop_queue visible to other cpus */
+		smp_mb__after_atomic();
 	}
 
 	/* Handle rest of tx packets and stop cpdma channels */
@@ -485,13 +486,6 @@ static int cpsw_resume_data_pass(struct net_device *ndev)
 	struct cpsw_common *cpsw = priv->cpsw;
 	int i, ret;
 
-	/* Allow rx packets handling */
-	for (i = 0; i < cpsw->data.slaves; i++) {
-		ndev = cpsw->slaves[i].ndev;
-		if (ndev && netif_running(ndev))
-			netif_dormant_off(ndev);
-	}
-
 	/* After this receive is started */
 	if (cpsw->usage_count) {
 		ret = cpsw_fill_rx_channels(priv);
diff --git a/drivers/net/ethernet/ti/davinci_cpdma.c b/drivers/net/ethernet/ti/davinci_cpdma.c
index 35bf14d8e7af..5cf1758d425b 100644
--- a/drivers/net/ethernet/ti/davinci_cpdma.c
+++ b/drivers/net/ethernet/ti/davinci_cpdma.c
@@ -134,6 +134,14 @@ struct cpdma_control_info {
 #define ACCESS_RW	(ACCESS_RO | ACCESS_WO)
 };
 
+struct submit_info {
+	struct cpdma_chan *chan;
+	int directed;
+	void *token;
+	void *data;
+	int len;
+};
+
 static struct cpdma_control_info controls[] = {
 	[CPDMA_TX_RLIM]		  = {CPDMA_DMACONTROL,	8,  0xffff, ACCESS_RW},
 	[CPDMA_CMD_IDLE]	  = {CPDMA_DMACONTROL,	3,  1,      ACCESS_WO},
@@ -1002,34 +1010,25 @@ static void __cpdma_chan_submit(struct cpdma_chan *chan,
 	}
 }
 
-int cpdma_chan_submit(struct cpdma_chan *chan, void *token, void *data,
-		      int len, int directed)
+static int cpdma_chan_submit_si(struct submit_info *si)
 {
+	struct cpdma_chan		*chan = si->chan;
 	struct cpdma_ctlr		*ctlr = chan->ctlr;
+	int				len = si->len;
 	struct cpdma_desc __iomem	*desc;
 	dma_addr_t			buffer;
-	unsigned long			flags;
 	u32				mode;
-	int				ret = 0;
-
-	spin_lock_irqsave(&chan->lock, flags);
-
-	if (chan->state == CPDMA_STATE_TEARDOWN) {
-		ret = -EINVAL;
-		goto unlock_ret;
-	}
+	int				ret;
 
 	if (chan->count >= chan->desc_num)	{
 		chan->stats.desc_alloc_fail++;
-		ret = -ENOMEM;
-		goto unlock_ret;
+		return -ENOMEM;
 	}
 
 	desc = cpdma_desc_alloc(ctlr->pool);
 	if (!desc) {
 		chan->stats.desc_alloc_fail++;
-		ret = -ENOMEM;
-		goto unlock_ret;
+		return -ENOMEM;
 	}
 
 	if (len < ctlr->params.min_packet_size) {
@@ -1037,16 +1036,15 @@ int cpdma_chan_submit(struct cpdma_chan *chan, void *token, void *data,
 		chan->stats.runt_transmit_buff++;
 	}
 
-	buffer = dma_map_single(ctlr->dev, data, len, chan->dir);
+	buffer = dma_map_single(ctlr->dev, si->data, len, chan->dir);
 	ret = dma_mapping_error(ctlr->dev, buffer);
 	if (ret) {
 		cpdma_desc_free(ctlr->pool, desc, 1);
-		ret = -EINVAL;
-		goto unlock_ret;
+		return -EINVAL;
 	}
 
 	mode = CPDMA_DESC_OWNER | CPDMA_DESC_SOP | CPDMA_DESC_EOP;
-	cpdma_desc_to_port(chan, mode, directed);
+	cpdma_desc_to_port(chan, mode, si->directed);
 
 	/* Relaxed IO accessors can be used here as there is read barrier
 	 * at the end of write sequence.
@@ -1055,7 +1053,7 @@ int cpdma_chan_submit(struct cpdma_chan *chan, void *token, void *data,
 	writel_relaxed(buffer, &desc->hw_buffer);
 	writel_relaxed(len, &desc->hw_len);
 	writel_relaxed(mode | len, &desc->hw_mode);
-	writel_relaxed((uintptr_t)token, &desc->sw_token);
+	writel_relaxed((uintptr_t)si->token, &desc->sw_token);
 	writel_relaxed(buffer, &desc->sw_buffer);
 	writel_relaxed(len, &desc->sw_len);
 	desc_read(desc, sw_len);
@@ -1066,8 +1064,53 @@ int cpdma_chan_submit(struct cpdma_chan *chan, void *token, void *data,
 		chan_write(chan, rxfree, 1);
 
 	chan->count++;
+	return 0;
+}
 
-unlock_ret:
+int cpdma_chan_idle_submit(struct cpdma_chan *chan, void *token, void *data,
+			   int len, int directed)
+{
+	struct submit_info si;
+	unsigned long flags;
+	int ret;
+
+	si.chan = chan;
+	si.token = token;
+	si.data = data;
+	si.len = len;
+	si.directed = directed;
+
+	spin_lock_irqsave(&chan->lock, flags);
+	if (chan->state == CPDMA_STATE_TEARDOWN) {
+		spin_unlock_irqrestore(&chan->lock, flags);
+		return -EINVAL;
+	}
+
+	ret = cpdma_chan_submit_si(&si);
+	spin_unlock_irqrestore(&chan->lock, flags);
+	return ret;
+}
+
+int cpdma_chan_submit(struct cpdma_chan *chan, void *token, void *data,
+		      int len, int directed)
+{
+	struct submit_info si;
+	unsigned long flags;
+	int ret;
+
+	si.chan = chan;
+	si.token = token;
+	si.data = data;
+	si.len = len;
+	si.directed = directed;
+
+	spin_lock_irqsave(&chan->lock, flags);
+	if (chan->state != CPDMA_STATE_ACTIVE) {
+		spin_unlock_irqrestore(&chan->lock, flags);
+		return -EINVAL;
+	}
+
+	ret = cpdma_chan_submit_si(&si);
 	spin_unlock_irqrestore(&chan->lock, flags);
 	return ret;
 }
diff --git a/drivers/net/ethernet/ti/davinci_cpdma.h b/drivers/net/ethernet/ti/davinci_cpdma.h
index 10376062dafa..9343c8c73c1b 100644
--- a/drivers/net/ethernet/ti/davinci_cpdma.h
+++ b/drivers/net/ethernet/ti/davinci_cpdma.h
@@ -79,6 +79,8 @@ int cpdma_chan_get_stats(struct cpdma_chan *chan,
 			 struct cpdma_chan_stats *stats);
 int cpdma_chan_submit(struct cpdma_chan *chan, void *token, void *data,
 		      int len, int directed);
+int cpdma_chan_idle_submit(struct cpdma_chan *chan, void *token, void *data,
+			   int len, int directed);
 int cpdma_chan_process(struct cpdma_chan *chan, int quota);
 
 int cpdma_ctlr_int_ctrl(struct cpdma_ctlr *ctlr, bool enable);
diff --git a/drivers/net/ethernet/ti/davinci_emac.c b/drivers/net/ethernet/ti/davinci_emac.c
index 4bf65cab79e6..5f4ece0d5a73 100644
--- a/drivers/net/ethernet/ti/davinci_emac.c
+++ b/drivers/net/ethernet/ti/davinci_emac.c
@@ -1428,8 +1428,8 @@ static int emac_dev_open(struct net_device *ndev)
 		if (!skb)
 			break;
 
-		ret = cpdma_chan_submit(priv->rxchan, skb, skb->data,
-					skb_tailroom(skb), 0);
+		ret = cpdma_chan_idle_submit(priv->rxchan, skb, skb->data,
+					     skb_tailroom(skb), 0);
 		if (WARN_ON(ret < 0))
 			break;
 	}
-- 
2.17.1

Re: [PATCH net-next] net: ethernet: ti: davinci_cpdma: use idled submit

[David Miller]

From: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Date: Sat, 15 Jun 2019 14:01:32 +0300

> While data pass suspend, reuse of rx descriptors can be disabled using
> channel state & lock from cpdma layer. For this, submit to a channel
> has to be disabled using state != "not active" under lock, what is done
> with this patch. The same submit is used to fill rx channel while
> ndo_open, when channel is idled, so add idled submit routine that
> allows to prepare descs for the channel. All this simplifies code and
> helps to avoid dormant mode usage and send packets only to active
> channels, avoiding potential race in later on changes. Also add missed
> sync barrier analogically like in other places after stopping tx
> queues.
> 
> Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>

Applied.