Re: [PATCH v4 net] ravb: unmap descriptors when freeing rings

On Thu, Jan 26, 2017 at 03:58:45PM +0300, Sergei Shtylyov wrote:

On 01/26/2017 02:04 PM, Simon Horman wrote:

quoted

From: Kazuya Mizuguchi <redacted>

"swiotlb buffer is full" errors occur after repeated initialisation of a
device - f.e. suspend/resume or ip link set up/down. This is because memory
mapped using dma_map_single() in ravb_ring_format() and ravb_start_xmit()
is not released.  Resolve this problem by unmapping descriptors when
freeing rings.

Fixes: a780893c89f6 ("ravb: unmap descriptors when freeing rings")

   No such SHA1 hash known and the summary is from this patch.

Sorry, it took the HEAD commit rather than a hash I should have provided
as an argument to git fixes:


[alias]
	fixes = log -1 --pretty=fixes
[core]
	abbrev = 12
[pretty]
	fixes = Fixes: %h (\"%s\")

What I wanted to say was:

Fixes: c156633f1353 ("Renesas Ethernet AVB driver proper")

quoted

Signed-off-by: Kazuya Mizuguchi <redacted>
[simon: reworked]
Signed-off-by: Simon Horman <redacted>
--

On inspection it appears that a similar problem is also present in
the SH-Ethernet driver.

v4 [Simon Horman]
* As suggested by Sergei Shtylyov
 - Use bool as type for new parameter to ravb_tx_free()
 - Clean up comment style
 - Only increment tx_bytes for transmitted skbs

v3 [Simon Horman]
* As suggested by Sergei Shtylyov
 - consistently use le32_to_cpu(desc->dptr)
 - Do not clear desc->ds_cc as it is not used
* Paramatise ravb_tx_free() to allow it to free non-transmitted buffers

v2 [Simon Horman]
* As suggested by Sergei Shtylyov
 - Use dma_mapping_error() and rx_desc->ds_cc when unmapping RX descriptors;
   this is consistent with the way that they are mapped
 - Use ravb_tx_free() to clear TX descriptors
* Reduce scope of new local variable

v1 [Kazuya Mizuguchi]
---
drivers/net/ethernet/renesas/ravb_main.c | 111 ++++++++++++++++++-------------
1 file changed, 63 insertions(+), 48 deletions(-)

diff --git a/drivers/net/ethernet/renesas/ravb_main.c b/drivers/net/ethernet/renesas/ravb_main.c
index 89ac1e3f6175..5d66ac721335 100644
--- a/drivers/net/ethernet/renesas/ravb_main.c
+++ b/drivers/net/ethernet/renesas/ravb_main.c

@@ -179,6 +179,48 @@ static struct mdiobb_ops bb_ops = {

	.get_mdio_data = ravb_get_mdio_data,
};

+/* Free TX skb function for AVB-IP */
+static int ravb_tx_free(struct net_device *ndev, int q, bool free_txed_only)
+{
+	struct ravb_private *priv = netdev_priv(ndev);
+	struct net_device_stats *stats = &priv->stats[q];
+	struct ravb_tx_desc *desc;
+	int free_num = 0;
+	int entry;
+	u32 size;
+
+	for (; priv->cur_tx[q] - priv->dirty_tx[q] > 0; priv->dirty_tx[q]++) {
+		bool txed;
+
+		entry = priv->dirty_tx[q] % (priv->num_tx_ring[q] *
+					     NUM_TX_DESC);
+		desc = &priv->tx_ring[q][entry];
+		txed = desc->die_dt == DT_FEMPTY;
+		if (free_txed_only && !txed)
+			break;
+		/* Descriptor type must be checked before all other reads */
+		dma_rmb();
+		size = le16_to_cpu(desc->ds_tagl) & TX_DS;
+		/* Free the original skb. */
+		if (priv->tx_skb[q][entry / NUM_TX_DESC]) {
+			dma_unmap_single(ndev->dev.parent, le32_to_cpu(desc->dptr),
+					 size, DMA_TO_DEVICE);
+			/* Last packet descriptor? */
+			if (entry % NUM_TX_DESC == NUM_TX_DESC - 1) {
+				entry /= NUM_TX_DESC;
+				dev_kfree_skb_any(priv->tx_skb[q][entry]);
+				priv->tx_skb[q][entry] = NULL;
+				stats->tx_packets++;

   We're still incrementing the packet # even when the packet hasn't been sent...

Thanks, sorry for missing that.

quoted

+			}
+			free_num++;
+		}
+		if (!txed)

   Your logic is reversed! :-(

Oops.

quoted

+			stats->tx_bytes += size;
+		desc->die_dt = DT_EEMPTY;
+	}
+	return free_num;
+}
+
/* Free skb's and DMA buffers for Ethernet AVB */
static void ravb_ring_free(struct net_device *ndev, int q)
{

@@ -194,19 +236,21 @@ static void ravb_ring_free(struct net_device *ndev, int q)

[...]

quoted

	if (priv->rx_ring[q]) {
+		for (i = 0; i < priv->num_rx_ring[q]; i++) {
+			struct ravb_ex_rx_desc *desc = &priv->rx_ring[q][i];
+
+			if (!dma_mapping_error(ndev->dev.parent,
+					       le32_to_cpu(desc->dptr)))

   I'd check for zero buffer length (that marks unmapped RX buffers) but
this should be good too... it's just that the length check is shorter. :-)

I'm pretty sure you suggested the logic above at some point.
Lets keep it :)

quoted

+				dma_unmap_single(ndev->dev.parent,
+						 le32_to_cpu(desc->dptr),
+						 PKT_BUF_SZ,
+						 DMA_FROM_DEVICE);
+		}
		ring_size = sizeof(struct ravb_ex_rx_desc) *
			    (priv->num_rx_ring[q] + 1);
		dma_free_coherent(ndev->dev.parent, ring_size, priv->rx_ring[q],

[...]

MBR, Sergei