Skip to contentSkip to conversation

gianfar: Fix TX ring processing on SMP machines

PATCH Applied Anton Vorontsov 4 participants 4 messages Mar 3, 2010 – Jun 11, 2010

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 gianfar: Fix TX ring processing on SMP machines

Starting with commit a3bc1f11e9b867a4f49505 ("gianfar: Revive SKB recycling") gianfar driver sooner or later stops transmitting any packets on SMP machines.

start_xmit() prepares new skb for transmitting, generally it does three things:

1. sets up all BDs (marks them ready to send), except the first one.
2. stores skb into tx_queue->tx_skbuff so that clean_tx_ring() would cleanup it later.
3. sets up the first BD, i.e. marks it ready.

Here is what clean_tx_ring() does:

1. reads skbs from tx_queue->tx_skbuff
2. checks if the *last* BD is ready. If it's still ready [to send] then it it isn't transmitted, so clean_tx_ring() returns. Otherwise it actually cleanups BDs. All is OK.

Now, if there is just one BD, code flow:

- start_xmit(): stores skb into tx_skbuff. Note that the first BD (which is also the last one) isn't marked as ready, yet.
- clean_tx_ring(): sees that skb is not null, *and* its lstatus says that it is NOT ready (like if BD was sent), so it cleans it up (bad!)
- start_xmit(): marks BD as ready [to send], but it's too late.

We can fix this simply by reordering lstatus/tx_skbuff writes.

---
 drivers/net/gianfar.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)
drivers/net/gianfar.c +4 -1
index 8bd3c9f..cccb409 100644
@@ -2021,7 +2021,6 @@ static int gfar_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	}
 
 	/* setup the TxBD length and buffer pointer for the first BD */
-	tx_queue->tx_skbuff[tx_queue->skb_curtx] = skb;
 	txbdp_start->bufPtr = dma_map_single(&priv->ofdev->dev, skb->data,
 			skb_headlen(skb), DMA_TO_DEVICE);
 
@@ -2053,6 +2052,10 @@ static int gfar_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
 	txbdp_start->lstatus = lstatus;
 
+	eieio(); /* force lstatus write before tx_skbuff */
+
+	tx_queue->tx_skbuff[tx_queue->skb_curtx] = skb;
+
 	/* Update the current skb pointer to the next entry we will use
 	 * (wrapping if necessary) */
 	tx_queue->skb_curtx = (tx_queue->skb_curtx + 1) &
Signature · 1 line
1.7.0
maintainer Applied
#1 lore raw

From: Anton Vorontsov <[email protected]>
Date: Wed, 3 Mar 2010 21:18:58 +0300

Quoted 28 lines

Starting with commit a3bc1f11e9b867a4f49505 ("gianfar: Revive SKB recycling") gianfar driver sooner or later stops transmitting any packets on SMP machines.

start_xmit() prepares new skb for transmitting, generally it does three things:

1. sets up all BDs (marks them ready to send), except the first one.
2. stores skb into tx_queue->tx_skbuff so that clean_tx_ring() would cleanup it later.
3. sets up the first BD, i.e. marks it ready.

Here is what clean_tx_ring() does:

1. reads skbs from tx_queue->tx_skbuff
2. checks if the *last* BD is ready. If it's still ready [to send] then it it isn't transmitted, so clean_tx_ring() returns. Otherwise it actually cleanups BDs. All is OK.

Now, if there is just one BD, code flow:

- start_xmit(): stores skb into tx_skbuff. Note that the first BD (which is also the last one) isn't marked as ready, yet.
- clean_tx_ring(): sees that skb is not null, *and* its lstatus says that it is NOT ready (like if BD was sent), so it cleans it up (bad!)
- start_xmit(): marks BD as ready [to send], but it's too late.

Quoted 7 lines

We can fix this simply by reordering lstatus/tx_skbuff writes.

Applied.

#2 lore raw
Quoted 33 lines · David Miller

From: Anton Vorontsov <[email protected]>
Date: Wed, 3 Mar 2010 21:18:58 +0300

Quoted 30 lines

Starting with commit a3bc1f11e9b867a4f49505 ("gianfar: Revive SKB recycling") gianfar driver sooner or later stops transmitting any packets on SMP machines.

start_xmit() prepares new skb for transmitting, generally it does three things:

1. sets up all BDs (marks them ready to send), except the first one.
2. stores skb into tx_queue->tx_skbuff so that clean_tx_ring() would cleanup it later.
3. sets up the first BD, i.e. marks it ready.

Here is what clean_tx_ring() does:

1. reads skbs from tx_queue->tx_skbuff
2. checks if the *last* BD is ready. If it's still ready [to send] then it it isn't transmitted, so clean_tx_ring() returns. Otherwise it actually cleanups BDs. All is OK.

Now, if there is just one BD, code flow:

- start_xmit(): stores skb into tx_skbuff. Note that the first BD (which is also the last one) isn't marked as ready, yet.
- clean_tx_ring(): sees that skb is not null, *and* its lstatus says that it is NOT ready (like if BD was sent), so it cleans it up (bad!)
- start_xmit(): marks BD as ready [to send], but it's too late.

We can fix this simply by reordering lstatus/tx_skbuff writes.

Quoted 7 lines
Quoted 5 lines

Applied.

Anton,

Once this makes it into Linus's tree can you make sure we get it added to -stable.

- k

#3 lore raw
In reply to Anton Vorontsov

drivers/net/gianfar.c static int gfar_start_xmit(struct sk_buff *skb, str=

Quoted 46 lines · Anton Vorontsov

Starting with commit a3bc1f11e9b867a4f49505 ("gianfar: Revive SKB recycling") gianfar driver sooner or later stops transmitting any packets on SMP machines.

start_xmit() prepares new skb for transmitting, generally it does three things:

1. sets up all BDs (marks them ready to send), except the first one.
2. stores skb into tx_queue->tx_skbuff so that clean_tx_ring() =A0 would cleanup it later.
3. sets up the first BD, i.e. marks it ready.

Here is what clean_tx_ring() does:

1. reads skbs from tx_queue->tx_skbuff
2. checks if the *last* BD is ready. If it's still ready [to send] =A0 then it it isn't transmitted, so clean_tx_ring() returns. =A0 Otherwise it actually cleanups BDs. All is OK.

Now, if there is just one BD, code flow:

- start_xmit(): stores skb into tx_skbuff. Note that the first BD =A0(which is also the last one) isn't marked as ready, yet.
- clean_tx_ring(): sees that skb is not null, *and* its lstatus =A0says that it is NOT ready (like if BD was sent), so it cleans =A0it up (bad!)
- start_xmit(): marks BD as ready [to send], but it's too late.

We can fix this simply by reordering lstatus/tx_skbuff writes.

---
=A0drivers/net/gianfar.c | =A0 =A05 ++++- =A01 files changed, 4 insertions(+), 1 deletions(-)

drivers/net/gianfar.c +0 -0
index 8bd3c9f..cccb409 100644
@@ -2021,7 +2021,6 @@ static int gfar_start_xmit(struct sk_buff *skb, str=

uct net_device *dev)

Quoted 3 lines
drivers/net/gianfar.c +0 -0
 =A0 =A0 =A0 =A0}
 

=A0 =A0 =A0 =A0/* setup the TxBD length and buffer pointer for the first =

BD */

Quoted 2 lines

- =A0 =A0 =A0 tx_queue->tx_skbuff[tx_queue->skb_curtx] =3D skb; =A0 =A0 =A0 =A0txbdp_start->bufPtr =3D dma_map_single(&priv->ofdev->dev, =

skb->data,

Quoted 1 line

=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0skb_headlen(skb), DMA_TO_D=

EVICE);

drivers/net/gianfar.c static int gfar_start_xmit(struct sk_buff *skb, st=

Quoted 2 lines
drivers/net/gianfar.c +0 -0
@@ -2053,6 +2052,10 @@ static int gfar_start_xmit(struct sk_buff *skb, st=

ruct net_device *dev)

Quoted 8 lines
drivers/net/gianfar.c +4 -0
 =A0 =A0 =A0 =A0txbdp_start->lstatus =3D lstatus;
 
+ =A0 =A0 =A0 eieio(); /* force lstatus write before tx_skbuff */
+
+ =A0 =A0 =A0 tx_queue->tx_skbuff[tx_queue->skb_curtx] =3D skb;
+

=A0 =A0 =A0 =A0/* Update the current skb pointer to the next entry we wil=

l use

Quoted 2 lines

=A0 =A0 =A0 =A0 * (wrapping if necessary) */
=A0 =A0 =A0 =A0tx_queue->skb_curtx =3D (tx_queue->skb_curtx + 1) &

This patch also makes gianfar work stable on mpc8313 with 2.6.33/RT_PREEMPT= .
WIthout it, I see exactly the same problems as reported by Anton on SMP.

/Esben
--=20
Esben Haabendal, Senior Software Consultant Dor=E9Development ApS, Ved Stranden 1, 9560 Hadsund, DK-Denmark
Phone: +45 51 92 53 93, E-mail: [email protected]
WWW: http://www.doredevelopment.dk