Skip to contentSkip to conversation

net: stmmac: use correct DMA buffer size in the RX descriptor

PATCH In review Aviraj CJ 4 participants 6 messages Dec 10, 2019 – Dec 11, 2019

Subsystems (4)
lore
Share & tools
lore

Aviraj CJ

author
#0 lore raw

PATCH 1/2 net: stmmac: use correct DMA buffer size in the RX descriptor

We always program the maximum DMA buffer size into the receive descriptor, although the allocated size may be less. E.g. with the default MTU size we allocate only 1536 bytes. If somebody sends us a bigger frame, then memory may get corrupted.

Program DMA using exact buffer sizes.

[Adopted based on upstream commit c13a936f46e3321ad2426443296571fab2feda44 ("net: stmmac: use correct DMA buffer size in the RX descriptor") by Aaro Koskinen <[email protected]> ]

---
 drivers/net/ethernet/stmicro/stmmac/common.h      |  2 +-
 drivers/net/ethernet/stmicro/stmmac/descs_com.h   | 14 ++++++++++----
 drivers/net/ethernet/stmicro/stmmac/enh_desc.c    | 10 +++++++---
 drivers/net/ethernet/stmicro/stmmac/norm_desc.c   | 10 +++++++---
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c |  4 ++--
 5 files changed, 27 insertions(+), 13 deletions(-)
drivers/net/ethernet/stmicro/stmmac/common.h +1 -1
index 623c6ed8764a..803df6a32ba9 100644
@@ -301,7 +301,7 @@ struct dma_features {
 struct stmmac_desc_ops {
 	/* DMA RX descriptor ring initialization */
 	void (*init_rx_desc) (struct dma_desc *p, int disable_rx_ic, int mode,
-			      int end);
+			      int end, int bfsize);
 	/* DMA TX descriptor ring initialization */
 	void (*init_tx_desc) (struct dma_desc *p, int mode, int end);
 
drivers/net/ethernet/stmicro/stmmac/descs_com.h +10 -4
index 6f2cc78c5cf5..6b83fc8e6fbe 100644
@@ -33,9 +33,10 @@
 /* Specific functions used for Ring mode */
 
 /* Enhanced descriptors */
-static inline void ehn_desc_rx_set_on_ring(struct dma_desc *p, int end)
+static inline void ehn_desc_rx_set_on_ring(struct dma_desc *p, int end, int bfsize)
 {
-	p->des01.erx.buffer2_size = BUF_SIZE_8KiB - 1;
+	if (bfsize == BUF_SIZE_16KiB)
+		p->des01.erx.buffer2_size = BUF_SIZE_8KiB - 1;
 	if (end)
 		p->des01.erx.end_ring = 1;
 }
@@ -61,9 +62,14 @@ static inline void enh_set_tx_desc_len_on_ring(struct dma_desc *p, int len)
 }
 
 /* Normal descriptors */
-static inline void ndesc_rx_set_on_ring(struct dma_desc *p, int end)
+static inline void ndesc_rx_set_on_ring(struct dma_desc *p, int end, int bfsize)
 {
-	p->des01.rx.buffer2_size = BUF_SIZE_2KiB - 1;
+	int size;
+
+	if (bfsize >= BUF_SIZE_2KiB) {
+		size = min(bfsize - BUF_SIZE_2KiB + 1, BUF_SIZE_2KiB - 1);
+		p->des01.rx.buffer2_size = size;
+	}
 	if (end)
 		p->des01.rx.end_ring = 1;
 }
drivers/net/ethernet/stmicro/stmmac/enh_desc.c +7 -3
index 7d944449f5ef..9ecb3a948f86 100644
@@ -238,16 +238,20 @@ static int enh_desc_get_rx_status(void *data, struct stmmac_extra_stats *x,
 }
 
 static void enh_desc_init_rx_desc(struct dma_desc *p, int disable_rx_ic,
-				  int mode, int end)
+				  int mode, int end, int bfsize)
 {
+	int bfsize1;
+
 	p->des01.all_flags = 0;
 	p->des01.erx.own = 1;
-	p->des01.erx.buffer1_size = BUF_SIZE_8KiB - 1;
+
+	bfsize1 = min(bfsize, BUF_SIZE_8KiB - 1);
+	p->des01.erx.buffer1_size = bfsize1;
 
 	if (mode == STMMAC_CHAIN_MODE)
 		ehn_desc_rx_set_on_chain(p, end);
 	else
-		ehn_desc_rx_set_on_ring(p, end);
+		ehn_desc_rx_set_on_ring(p, end, bfsize);
 
 	if (disable_rx_ic)
 		p->des01.erx.disable_ic = 1;
drivers/net/ethernet/stmicro/stmmac/norm_desc.c +7 -3
index 48c3456445b2..07e0c03cfb10 100644
@@ -121,16 +121,20 @@ static int ndesc_get_rx_status(void *data, struct stmmac_extra_stats *x,
 }
 
 static void ndesc_init_rx_desc(struct dma_desc *p, int disable_rx_ic, int mode,
-			       int end)
+			       int end, int bfsize)
 {
+	int bfsize1;
+
 	p->des01.all_flags = 0;
 	p->des01.rx.own = 1;
-	p->des01.rx.buffer1_size = BUF_SIZE_2KiB - 1;
+
+	bfsize1 = min(bfsize, (BUF_SIZE_2KiB - 1));
+	p->des01.rx.buffer1_size = bfsize1;
 
 	if (mode == STMMAC_CHAIN_MODE)
 		ndesc_rx_set_on_chain(p, end);
 	else
-		ndesc_rx_set_on_ring(p, end);
+		ndesc_rx_set_on_ring(p, end, bfsize);
 
 	if (disable_rx_ic)
 		p->des01.rx.disable_ic = 1;
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c +2 -2
index f4d6512f066c..e9d41e03121c 100644
@@ -964,11 +964,11 @@ static void stmmac_clear_descriptors(struct stmmac_priv *priv)
 		if (priv->extend_desc)
 			priv->hw->desc->init_rx_desc(&priv->dma_erx[i].basic,
 						     priv->use_riwt, priv->mode,
-						     (i == rxsize - 1));
+						     (i == rxsize - 1), priv->dma_buf_sz);
 		else
 			priv->hw->desc->init_rx_desc(&priv->dma_rx[i],
 						     priv->use_riwt, priv->mode,
-						     (i == rxsize - 1));
+						     (i == rxsize - 1), priv->dma_buf_sz);
 	for (i = 0; i < txsize; i++)
 		if (priv->extend_desc)
 			priv->hw->desc->init_tx_desc(&priv->dma_etx[i].basic,
Signature · 1 line
2.19.1
author
#1 lore raw

PATCH 2/2 net: stmmac: don't stop NAPI processing when dropping a packet

Currently, if we drop a packet, we exit from NAPI loop before the budget is consumed. In some situations this will make the RX processing stall e.g. when flood pinging the system with oversized packets, as the errorneous packets are not dropped efficiently.

If we drop a packet, we should just continue to the next one as long as the budget allows.

[ Adopted based on upstream commit 2170bbf19f6ef6b2740f186ee107827f31395218 ("net: stmmac: don't stop NAPI processing when dropping a packet") by Aaro Koskinen <[email protected]> ]

---
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c +6 -5
index e9d41e03121c..3b514ba570b8 100644
@@ -2176,8 +2176,7 @@ static inline void stmmac_rx_refill(struct stmmac_priv *priv)
 static int stmmac_rx(struct stmmac_priv *priv, int limit)
 {
 	unsigned int rxsize = priv->dma_rx_size;
-	unsigned int entry = priv->cur_rx % rxsize;
-	unsigned int next_entry;
+	unsigned int next_entry = priv->cur_rx % rxsize;
 	unsigned int count = 0;
 	int coe = priv->hw->rx_csum;
 
@@ -2191,6 +2190,9 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit)
 	while (count < limit) {
 		int status;
 		struct dma_desc *p;
+		unsigned int entry;
+
+		entry = next_entry;
 
 		if (priv->extend_desc)
 			p = (struct dma_desc *)(priv->dma_erx + entry);
@@ -2239,7 +2241,7 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit)
 			/*  check if frame_len fits the preallocated memory */
 			if (frame_len > priv->dma_buf_sz) {
 				priv->dev->stats.rx_length_errors++;
-				break;
+				continue;
 			}
 
 			/* ACS is set; GMAC core strips PAD/FCS for IEEE 802.3
@@ -2260,7 +2262,7 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit)
 				pr_err("%s: Inconsistent Rx descriptor chain\n",
 				       priv->dev->name);
 				priv->dev->stats.rx_dropped++;
-				break;
+				continue;
 			}
 			prefetch(skb->data - NET_IP_ALIGN);
 			priv->rx_skbuff[entry] = NULL;
@@ -2291,7 +2293,6 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit)
 			priv->dev->stats.rx_packets++;
 			priv->dev->stats.rx_bytes += frame_len;
 		}
-		entry = next_entry;
 	}
 
 	stmmac_rx_refill(priv);
Signature · 1 line
2.19.1
#2 lore raw
In reply to Aviraj CJ
Quoted 10 lines · Aviraj CJ

We always program the maximum DMA buffer size into the receive descriptor, although the allocated size may be less. E.g. with the default MTU size we allocate only 1536 bytes. If somebody sends us a bigger frame, then memory may get corrupted.

Program DMA using exact buffer sizes.

[Adopted based on upstream commit c13a936f46e3321ad2426443296571fab2feda44 ("net: stmmac: use correct DMA buffer size in the RX descriptor") by Aaro Koskinen <[email protected]> ]

Adopted to what?

What is this patch for, it looks just like the commit you reference here.

totally confused,

greg k-h

#3 lore raw
Quoted 12 lines · Greg KH
Quoted 10 lines · Aviraj CJ

We always program the maximum DMA buffer size into the receive descriptor, although the allocated size may be less. E.g. with the default MTU size we allocate only 1536 bytes. If somebody sends us a bigger frame, then memory may get corrupted.

Program DMA using exact buffer sizes.

[Adopted based on upstream commit c13a936f46e3321ad2426443296571fab2feda44 ("net: stmmac: use correct DMA buffer size in the RX descriptor") by Aaro Koskinen <[email protected]> ]

Quoted 6 lines

Adopted to what?

What is this patch for, it looks just like the commit you reference here.

totally confused,

We're using the patches on the v4.4 -stable branch. It doesn't have these patches and the backport had rejects.

Daniel

maintainer
#4 lore raw

From: "Daniel Walker (danielwa)" <[email protected]>
Date: Tue, 10 Dec 2019 21:40:17 +0000

Quoted 15 lines · Greg KH
Quoted 14 lines · Aviraj CJ
Quoted 10 lines

We always program the maximum DMA buffer size into the receive descriptor, although the allocated size may be less. E.g. with the default MTU size we allocate only 1536 bytes. If somebody sends us a bigger frame, then memory may get corrupted.

Program DMA using exact buffer sizes.

[Adopted based on upstream commit c13a936f46e3321ad2426443296571fab2feda44 ("net: stmmac: use correct DMA buffer size in the RX descriptor") by Aaro Koskinen <[email protected]> ]

Adopted to what?

Quoted 8 lines
Quoted 4 lines

What is this patch for, it looks just like the commit you reference here.

totally confused,

We're using the patches on the v4.4 -stable branch. It doesn't have these patches and the backport had rejects.

As submitted the patch looks like something destined for mainline.

Not specifying the context into which the change is meant to be placed wastes a lot of precious developer time.

#5 lore raw
In reply to Daniel Walker (danielwa)
Quoted 15 lines · Daniel Walker (danielwa)
Quoted 14 lines · Greg KH
Quoted 10 lines · Aviraj CJ

We always program the maximum DMA buffer size into the receive descriptor, although the allocated size may be less. E.g. with the default MTU size we allocate only 1536 bytes. If somebody sends us a bigger frame, then memory may get corrupted.

Program DMA using exact buffer sizes.

[Adopted based on upstream commit c13a936f46e3321ad2426443296571fab2feda44 ("net: stmmac: use correct DMA buffer size in the RX descriptor") by Aaro Koskinen <[email protected]> ]

Adopted to what?

Quoted 8 lines
Quoted 4 lines

What is this patch for, it looks just like the commit you reference here.

totally confused,

We're using the patches on the v4.4 -stable branch. It doesn't have these patches and the backport had rejects.

Ok, but commit "c13a936f46e3321ad2426443296571fab2feda44" is not in Linus's tree, and so I think you really mean 583e63614149 ("net: stmmac:
use correct DMA buffer size in the RX descriptor") which is only included in 4.19 and newer kernels.

So why would this need to go to 4.4.y?

And if so, it needs to be explicitly stated as such, you all have read the stable kernel rules file, right?

Please fix up and resend properly, as well as providing a version for newer kernels also if you really want this in a 4.4.y release.

As David said, to not do so just causes a total waste of developer time trying to figure out what you all are wanting to do here...

thanks,

greg k-h