2012-05-07 22:39:36

by Julien Ducourthial

[permalink] [raw]
Subject: [PATCH] r8169: fix problem with TSO (TX_BUFFS_AVAIL negative value)

The r8169 may get stuck or show bad behaviour after activating TSO :
the net_device is not stopped when it has no more TX descriptors.
This problem comes from TX_BUFS_AVAIL which may reach -1 when all
transmit descriptors are in use. The patch simply tries to keep positive
values.

Tested with 8111d(onboard) on a D510MO, and with 8111e(onboard) on a
Zotac 890GXITX.

Signed-off-by: Julien Ducourthial <[email protected]>
---
drivers/net/ethernet/realtek/r8169.c | 16 ++++++++++------
1 files changed, 10 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/realtek/r8169.c
b/drivers/net/ethernet/realtek/r8169.c
index f545093..d1e3c51 100644
--- a/drivers/net/ethernet/realtek/r8169.c
+++ b/drivers/net/ethernet/realtek/r8169.c
@@ -61,8 +61,12 @@
#define R8169_MSG_DEFAULT \
(NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | NETIF_MSG_IFDOWN)

-#define TX_BUFFS_AVAIL(tp) \
- (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx - 1)
+#define TX_SLOTS_AVAIL(tp) \
+ (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx)
+
+/* A skbuff with nr_frags needs nr_frags+1 entries in the tx queue */
+#define TX_FRAGS_READY_FOR(tp,nr_frags) \
+ (TX_SLOTS_AVAIL(tp) >= (nr_frags+1))

/* Maximum number of multicast addresses to filter (vs.
Rx-all-multicast).
The RTL chips use a 64 element hash table based on the Ethernet CRC.
*/
@@ -5115,7 +5119,7 @@ static netdev_tx_t rtl8169_start_xmit(struct
sk_buff *skb,
u32 opts[2];
int frags;

- if (unlikely(TX_BUFFS_AVAIL(tp) < skb_shinfo(skb)->nr_frags)) {
+ if (unlikely(!TX_FRAGS_READY_FOR(tp, skb_shinfo(skb)->nr_frags))) {
netif_err(tp, drv, dev, "BUG! Tx Ring full when queue awake!\n");
goto err_stop_0;
}
@@ -5169,7 +5173,7 @@ static netdev_tx_t rtl8169_start_xmit(struct
sk_buff *skb,

mmiowb();

- if (TX_BUFFS_AVAIL(tp) < MAX_SKB_FRAGS) {
+ if (!TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS)) {
/* Avoid wrongly optimistic queue wake-up: rtl_tx thread must
* not miss a ring update when it notices a stopped queue.
*/
@@ -5183,7 +5187,7 @@ static netdev_tx_t rtl8169_start_xmit(struct
sk_buff *skb,
* can't.
*/
smp_mb();
- if (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)
+ if (TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS))
netif_wake_queue(dev);
}

@@ -5306,7 +5310,7 @@ static void rtl_tx(struct net_device *dev, struct
rtl8169_private *tp)
*/
smp_mb();
if (netif_queue_stopped(dev) &&
- (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)) {
+ TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS)) {
netif_wake_queue(dev);
}
/*
--
1.7.7.6



2012-05-07 23:48:57

by Francois Romieu

[permalink] [raw]
Subject: Re: [PATCH] r8169: fix problem with TSO (TX_BUFFS_AVAIL negative value)

Julien Ducourthial <[email protected]> :
> The r8169 may get stuck or show bad behaviour after activating TSO :
> the net_device is not stopped when it has no more TX descriptors.
> This problem comes from TX_BUFS_AVAIL which may reach -1 when all
> transmit descriptors are in use. The patch simply tries to keep positive
> values.

It seems more than good.

Alex, Thomas, can you check if Julien's patch below fixes your broken
kernels as well ?

diff --git a/drivers/net/ethernet/realtek/r8169.c
b/drivers/net/ethernet/realtek/r8169.c
index f545093..d1e3c51 100644
--- a/drivers/net/ethernet/realtek/r8169.c
+++ b/drivers/net/ethernet/realtek/r8169.c
@@ -61,8 +61,12 @@
#define R8169_MSG_DEFAULT \
(NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | NETIF_MSG_IFDOWN)

-#define TX_BUFFS_AVAIL(tp) \
- (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx - 1)
+#define TX_SLOTS_AVAIL(tp) \
+ (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx)
+
+/* A skbuff with nr_frags needs nr_frags+1 entries in the tx queue */
+#define TX_FRAGS_READY_FOR(tp,nr_frags) \
+ (TX_SLOTS_AVAIL(tp) >= (nr_frags + 1))

/* Maximum number of multicast addresses to filter (vs.
Rx-all-multicast).
The RTL chips use a 64 element hash table based on the Ethernet CRC.
*/
@@ -5115,7 +5119,7 @@ static netdev_tx_t rtl8169_start_xmit(struct
sk_buff *skb,
u32 opts[2];
int frags;

- if (unlikely(TX_BUFFS_AVAIL(tp) < skb_shinfo(skb)->nr_frags)) {
+ if (unlikely(!TX_FRAGS_READY_FOR(tp, skb_shinfo(skb)->nr_frags))) {
netif_err(tp, drv, dev, "BUG! Tx Ring full when queue awake!\n");
goto err_stop_0;
}
@@ -5169,7 +5173,7 @@ static netdev_tx_t rtl8169_start_xmit(struct
sk_buff *skb,

mmiowb();

- if (TX_BUFFS_AVAIL(tp) < MAX_SKB_FRAGS) {
+ if (!TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS)) {
/* Avoid wrongly optimistic queue wake-up: rtl_tx thread must
* not miss a ring update when it notices a stopped queue.
*/
@@ -5183,7 +5187,7 @@ static netdev_tx_t rtl8169_start_xmit(struct
sk_buff *skb,
* can't.
*/
smp_mb();
- if (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)
+ if (TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS))
netif_wake_queue(dev);
}

@@ -5306,7 +5310,7 @@ static void rtl_tx(struct net_device *dev, struct
rtl8169_private *tp)
*/
smp_mb();
if (netif_queue_stopped(dev) &&
- (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)) {
+ TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS)) {
netif_wake_queue(dev);
}
/*
--
1.7.7.6

2012-05-08 10:13:27

by Francois Romieu

[permalink] [raw]
Subject: Re: [PATCH] r8169: fix problem with TSO (TX_BUFFS_AVAIL negative value)

Francois Romieu <[email protected]> :
[...]
> Alex, Thomas, can you check if Julien's patch below fixes your broken
> kernels as well ?

You will have better luck with a patch whose lines are correctly formed.

Patch below applies on top of 3.4-rc6.

(Julien, please check your mail user agent)

diff --git a/drivers/net/ethernet/realtek/r8169.c b/drivers/net/ethernet/realtek/r8169.c
index f545093..d1e3c51 100644
--- a/drivers/net/ethernet/realtek/r8169.c
+++ b/drivers/net/ethernet/realtek/r8169.c
@@ -61,8 +61,12 @@
#define R8169_MSG_DEFAULT \
(NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | NETIF_MSG_IFDOWN)

-#define TX_BUFFS_AVAIL(tp) \
- (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx - 1)
+#define TX_SLOTS_AVAIL(tp) \
+ (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx)
+
+/* A skbuff with nr_frags needs nr_frags+1 entries in the tx queue */
+#define TX_FRAGS_READY_FOR(tp,nr_frags) \
+ (TX_SLOTS_AVAIL(tp) >= (nr_frags + 1))

/* Maximum number of multicast addresses to filter (vs. Rx-all-multicast).
The RTL chips use a 64 element hash table based on the Ethernet CRC. */
@@ -5115,7 +5119,7 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
u32 opts[2];
int frags;

- if (unlikely(TX_BUFFS_AVAIL(tp) < skb_shinfo(skb)->nr_frags)) {
+ if (unlikely(!TX_FRAGS_READY_FOR(tp, skb_shinfo(skb)->nr_frags))) {
netif_err(tp, drv, dev, "BUG! Tx Ring full when queue awake!\n");
goto err_stop_0;
}
@@ -5169,7 +5173,7 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,

mmiowb();

- if (TX_BUFFS_AVAIL(tp) < MAX_SKB_FRAGS) {
+ if (!TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS)) {
/* Avoid wrongly optimistic queue wake-up: rtl_tx thread must
* not miss a ring update when it notices a stopped queue.
*/
@@ -5183,7 +5187,7 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
* can't.
*/
smp_mb();
- if (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)
+ if (TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS))
netif_wake_queue(dev);
}

@@ -5306,7 +5310,7 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp)
*/
smp_mb();
if (netif_queue_stopped(dev) &&
- (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)) {
+ TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS)) {
netif_wake_queue(dev);
}
/*
--
Ueimor

2012-05-08 10:21:52

by Thomas Pilarski

[permalink] [raw]
Subject: Re: [PATCH] r8169: fix problem with TSO (TX_BUFFS_AVAIL negative value)

I have already tested the patch with the kernel 3.3.4 and the module
from bugzilla (https://bugzilla.kernel.org/attachment.cgi?id=72682). The
iommu error still occurs.

Am 08.05.2012 12:06, schrieb Francois Romieu:
> Francois Romieu <[email protected]> :
> [...]
>> Alex, Thomas, can you check if Julien's patch below fixes your broken
>> kernels as well ?
> You will have better luck with a patch whose lines are correctly formed.
>
> Patch below applies on top of 3.4-rc6.
>
> (Julien, please check your mail user agent)
>
> diff --git a/drivers/net/ethernet/realtek/r8169.c b/drivers/net/ethernet/realtek/r8169.c
> index f545093..d1e3c51 100644
> --- a/drivers/net/ethernet/realtek/r8169.c
> +++ b/drivers/net/ethernet/realtek/r8169.c
> @@ -61,8 +61,12 @@
> #define R8169_MSG_DEFAULT \
> (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | NETIF_MSG_IFDOWN)
>
> -#define TX_BUFFS_AVAIL(tp) \
> - (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx - 1)
> +#define TX_SLOTS_AVAIL(tp) \
> + (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx)
> +
> +/* A skbuff with nr_frags needs nr_frags+1 entries in the tx queue */
> +#define TX_FRAGS_READY_FOR(tp,nr_frags) \
> + (TX_SLOTS_AVAIL(tp) >= (nr_frags + 1))
>
> /* Maximum number of multicast addresses to filter (vs. Rx-all-multicast).
> The RTL chips use a 64 element hash table based on the Ethernet CRC. */
> @@ -5115,7 +5119,7 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
> u32 opts[2];
> int frags;
>
> - if (unlikely(TX_BUFFS_AVAIL(tp) < skb_shinfo(skb)->nr_frags)) {
> + if (unlikely(!TX_FRAGS_READY_FOR(tp, skb_shinfo(skb)->nr_frags))) {
> netif_err(tp, drv, dev, "BUG! Tx Ring full when queue awake!\n");
> goto err_stop_0;
> }
> @@ -5169,7 +5173,7 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
>
> mmiowb();
>
> - if (TX_BUFFS_AVAIL(tp) < MAX_SKB_FRAGS) {
> + if (!TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS)) {
> /* Avoid wrongly optimistic queue wake-up: rtl_tx thread must
> * not miss a ring update when it notices a stopped queue.
> */
> @@ -5183,7 +5187,7 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
> * can't.
> */
> smp_mb();
> - if (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)
> + if (TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS))
> netif_wake_queue(dev);
> }
>
> @@ -5306,7 +5310,7 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp)
> */
> smp_mb();
> if (netif_queue_stopped(dev) &&
> - (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)) {
> + TX_FRAGS_READY_FOR(tp, MAX_SKB_FRAGS)) {
> netif_wake_queue(dev);
> }
> /*

2012-05-09 04:37:46

by Alex Villacís Lasso

[permalink] [raw]
Subject: Re: [PATCH] r8169: fix problem with TSO (TX_BUFFS_AVAIL negative value)

El 07/05/12 18:42, Francois Romieu escribi?:
> Julien Ducourthial<[email protected]> :
>> The r8169 may get stuck or show bad behaviour after activating TSO :
>> the net_device is not stopped when it has no more TX descriptors.
>> This problem comes from TX_BUFS_AVAIL which may reach -1 when all
>> transmit descriptors are in use. The patch simply tries to keep positive
>> values.
> It seems more than good.
>
> Alex, Thomas, can you check if Julien's patch below fixes your broken
> kernels as well ?
>
No luck. The backtrace still appears after using the patched driver.