rtl8169_get_stats seems to be adding random crap to rx_missed_errors count

rtl8169_get_stats seems to be adding random crap to rx_missed_errors count

[Alex Bennee]

Hi,

I've been seeing odd behaviour in the ifconfig output for my rtl8169
based network device. The ifconfig output had a very high number of
dropped packets in the display which changed massively every time I
checked. I tracked it down to /proc/net/dev and the a few test patches
it seems the rc_missed field is the culprit. I did a test patch of the
form:

+++ b/drivers/net/r8169.c
@@ -3062,10 +3062,16 @@ static struct net_device_stats
*rtl8169_get_stats(struct net_device *dev)
       unsigned long flags;

       if (netif_running(dev)) {
+               u32 missed;
               spin_lock_irqsave(&tp->lock, flags);
-               dev->stats.rx_missed_errors += RTL_R32(RxMissed);
+               missed = RTL_R32(RxMissed);
               RTL_W32(RxMissed, 0);
               spin_unlock_irqrestore(&tp->lock, flags);
+
+               dev->stats.rx_missed_errors += missed;
+               if (missed)
+                       printk(KERN_INFO "rtl8169_get_stats added %d
to missed count\n", missed);
+
       }

       return &dev->stats;

And low and behold I see this behaviour:

[  946.160188] rtl8169_get_stats added 250834943 to missed count
[  946.160354] rtl8169_get_stats added 20768 to missed count
[  946.160411] rtl8169_get_stats added 7032 to missed count
[  946.160425] rtl8169_get_stats added 1677 to missed count
[  946.160439] rtl8169_get_stats added 1577 to missed count
[  946.160452] rtl8169_get_stats added 1543 to missed count
[  946.160465] rtl8169_get_stats added 1517 to missed count
[  946.160479] rtl8169_get_stats added 1572 to missed count
[  946.160492] rtl8169_get_stats added 1586 to missed count
[  946.160504] rtl8169_get_stats added 1284 to missed count
[  948.164206] rtl8169_get_stats added 250864594 to missed count
[  948.164374] rtl8169_get_stats added 21004 to missed count
[  948.164434] rtl8169_get_stats added 7385 to missed count
[  948.164447] rtl8169_get_stats added 1631 to missed count
[  948.164461] rtl8169_get_stats added 1574 to missed count
[  948.164475] rtl8169_get_stats added 1645 to missed count
[  948.164489] rtl8169_get_stats added 1583 to missed count
[  948.164503] rtl8169_get_stats added 1561 to missed count
[  948.164523] rtl8169_get_stats added 2314 to missed count
[  948.164534] rtl8169_get_stats added 1300 to missed count
[  950.168182] rtl8169_get_stats added 250857857 to missed count
[  950.168345] rtl8169_get_stats added 20465 to missed count
[  950.168401] rtl8169_get_stats added 6903 to missed count
[  950.168416] rtl8169_get_stats added 1711 to missed count
[  950.168430] rtl8169_get_stats added 1585 to missed count
[  950.168443] rtl8169_get_stats added 1532 to missed count
[  950.168456] rtl8169_get_stats added 1520 to missed count
[  950.168469] rtl8169_get_stats added 1539 to missed count
[  950.168483] rtl8169_get_stats added 1588 to missed count
[  950.168494] rtl8169_get_stats added 1282 to missed count

It looks very much like the number being read is a decreasing counter
rather than anything else.

Has anyone got a copy of the data sheet and can confirm that:

       RxMissed        = 0x4c,

is the correct offset for the register and it doesn't need any special
handling to be read?




--
Alex, homepage: http://www.bennee.com/~alex/



-- 
Alex, homepage: http://www.bennee.com/~alex/

Re: rtl8169_get_stats seems to be adding random crap to rx_missed_errors count

[Bruno Prémont]

Hi Alex,

This issue has already been reported in
  http://lkml.org/lkml/2008/10/9/235
and a patch exists.

Could you verify that patch #0009 at URL below fixes the issue for you:
  http://userweb.kernel.org/~romieu/r8169/2.6.27-rc9/20081007/


Bruno


On Fri, 10 October 2008 "Alex Bennee" <kernel-hacker@bennee.com> wrote:
> Hi,
> 
> I've been seeing odd behaviour in the ifconfig output for my rtl8169
> based network device. The ifconfig output had a very high number of
> dropped packets in the display which changed massively every time I
> checked. I tracked it down to /proc/net/dev and the a few test patches
> it seems the rc_missed field is the culprit. I did a test patch of the
> form:
> 
> +++ b/drivers/net/r8169.c
> @@ -3062,10 +3062,16 @@ static struct net_device_stats
> *rtl8169_get_stats(struct net_device *dev)
>        unsigned long flags;
> 
>        if (netif_running(dev)) {
> +               u32 missed;
>                spin_lock_irqsave(&tp->lock, flags);
> -               dev->stats.rx_missed_errors += RTL_R32(RxMissed);
> +               missed = RTL_R32(RxMissed);
>                RTL_W32(RxMissed, 0);
>                spin_unlock_irqrestore(&tp->lock, flags);
> +
> +               dev->stats.rx_missed_errors += missed;
> +               if (missed)
> +                       printk(KERN_INFO "rtl8169_get_stats added %d
> to missed count\n", missed);
> +
>        }
> 
>        return &dev->stats;
> 
> And low and behold I see this behaviour:
> 
> [  946.160188] rtl8169_get_stats added 250834943 to missed count
> [  946.160354] rtl8169_get_stats added 20768 to missed count
> [  946.160411] rtl8169_get_stats added 7032 to missed count
> [  946.160425] rtl8169_get_stats added 1677 to missed count
> [  946.160439] rtl8169_get_stats added 1577 to missed count
> [  946.160452] rtl8169_get_stats added 1543 to missed count
> [  946.160465] rtl8169_get_stats added 1517 to missed count
> [  946.160479] rtl8169_get_stats added 1572 to missed count
> [  946.160492] rtl8169_get_stats added 1586 to missed count
> [  946.160504] rtl8169_get_stats added 1284 to missed count
> [  948.164206] rtl8169_get_stats added 250864594 to missed count
> [  948.164374] rtl8169_get_stats added 21004 to missed count
> [  948.164434] rtl8169_get_stats added 7385 to missed count
> [  948.164447] rtl8169_get_stats added 1631 to missed count
> [  948.164461] rtl8169_get_stats added 1574 to missed count
> [  948.164475] rtl8169_get_stats added 1645 to missed count
> [  948.164489] rtl8169_get_stats added 1583 to missed count
> [  948.164503] rtl8169_get_stats added 1561 to missed count
> [  948.164523] rtl8169_get_stats added 2314 to missed count
> [  948.164534] rtl8169_get_stats added 1300 to missed count
> [  950.168182] rtl8169_get_stats added 250857857 to missed count
> [  950.168345] rtl8169_get_stats added 20465 to missed count
> [  950.168401] rtl8169_get_stats added 6903 to missed count
> [  950.168416] rtl8169_get_stats added 1711 to missed count
> [  950.168430] rtl8169_get_stats added 1585 to missed count
> [  950.168443] rtl8169_get_stats added 1532 to missed count
> [  950.168456] rtl8169_get_stats added 1520 to missed count
> [  950.168469] rtl8169_get_stats added 1539 to missed count
> [  950.168483] rtl8169_get_stats added 1588 to missed count
> [  950.168494] rtl8169_get_stats added 1282 to missed count
> 
> It looks very much like the number being read is a decreasing counter
> rather than anything else.
> 
> Has anyone got a copy of the data sheet and can confirm that:
> 
>        RxMissed        = 0x4c,
> 
> is the correct offset for the register and it doesn't need any special
> handling to be read?
> 
> 
> 
> 
> --
> Alex, homepage: http://www.bennee.com/~alex/

Re: rtl8169_get_stats seems to be adding random crap to rx_missed_errors count

[Alex Bennee]

On Fri, Oct 10, 2008 at 10:27 PM, Bruno Prémont
<bonbons@linux-vserver.org> wrote:
> Hi Alex,
>
> This issue has already been reported in
>  http://lkml.org/lkml/2008/10/9/235
> and a patch exists.
>
> Could you verify that patch #0009 at URL below fixes the issue for you:
>  http://userweb.kernel.org/~romieu/r8169/2.6.27-rc9/20081007/

I can confirm that patch works for me. I have further reviewed the
code and it looks OK to me from here. Should I just forward the patch
and tag it to the mailing list or is it now in a tree that gets picked
up?

-- 
Alex, homepage: http://www.bennee.com/~alex/

Re: rtl8169_get_stats seems to be adding random crap to rx_missed_errors count

[Alex Bennee]

On Sat, Oct 11, 2008 at 7:42 AM, Alex Bennee <kernel-hacker@bennee.com> wrote:
> On Fri, Oct 10, 2008 at 10:27 PM, Bruno Prémont
> <bonbons@linux-vserver.org> wrote:
>> Hi Alex,
>>
>> This issue has already been reported in
>>  http://lkml.org/lkml/2008/10/9/235
>> and a patch exists.
>>
>> Could you verify that patch #0009 at URL below fixes the issue for you:
>>  http://userweb.kernel.org/~romieu/r8169/2.6.27-rc9/20081007/
>
> I can confirm that patch works for me. I have further reviewed the
> code and it looks OK to me from here. Should I just forward the patch
> and tag it to the mailing list or is it now in a tree that gets picked
> up?


>From 96b79ad3d9e8c15a0118810e91f0e28ce5575493 Mon Sep 17 00:00:00 2001
From: Francois Romieu <romieu@fr.zoreil.com>
Date: Wed, 10 Sep 2008 22:28:56 +0200
Subject: [PATCH] r8169: fix RxMissed register access

- the register is defined for the 8169 chipset only and there is
  no 8169 beyond RTL_GIGA_MAC_VER_06.
- only the lower 3 bytes of the register are valid

Fixes:
1. http://bugzilla.kernel.org/show_bug.cgi?id=10180
2. http://bugzilla.kernel.org/show_bug.cgi?id=11062 (bits of)

Tested by Hermann Gausterer and Adam Huffman.

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Tested-by: Alex Bennee <kernel-hacker@bennee.com>
---
 drivers/net/r8169.c |   25 ++++++++++++++-----------
 1 files changed, 14 insertions(+), 11 deletions(-)

diff --git a/drivers/net/r8169.c b/drivers/net/r8169.c
index 0f6f974..4190ee7 100644
--- a/drivers/net/r8169.c
+++ b/drivers/net/r8169.c
@@ -2099,8 +2099,6 @@ static void rtl_hw_start_8168(struct net_device *dev)

 	RTL_R8(IntrMask);

-	RTL_W32(RxMissed, 0);
-
 	rtl_set_rx_mode(dev);

 	RTL_W8(ChipCmd, CmdTxEnb | CmdRxEnb);
@@ -2143,8 +2141,6 @@ static void rtl_hw_start_8101(struct net_device *dev)

 	RTL_R8(IntrMask);

-	RTL_W32(RxMissed, 0);
-
 	rtl_set_rx_mode(dev);

 	RTL_W8(ChipCmd, CmdTxEnb | CmdRxEnb);
@@ -2922,6 +2918,17 @@ static int rtl8169_poll(struct napi_struct
*napi, int budget)
 	return work_done;
 }

+static void rtl8169_rx_missed(struct net_device *dev, void __iomem *ioaddr)
+{
+	struct rtl8169_private *tp = netdev_priv(dev);
+
+	if (tp->mac_version > RTL_GIGA_MAC_VER_06)
+		return;
+
+	dev->stats.rx_missed_errors += (RTL_R32(RxMissed) & 0xffffff);
+	RTL_W32(RxMissed, 0);
+}
+
 static void rtl8169_down(struct net_device *dev)
 {
 	struct rtl8169_private *tp = netdev_priv(dev);
@@ -2939,9 +2946,7 @@ core_down:

 	rtl8169_asic_down(ioaddr);

-	/* Update the error counts. */
-	dev->stats.rx_missed_errors += RTL_R32(RxMissed);
-	RTL_W32(RxMissed, 0);
+	rtl8169_rx_missed(dev, ioaddr);

 	spin_unlock_irq(&tp->lock);

@@ -3063,8 +3068,7 @@ static struct net_device_stats
*rtl8169_get_stats(struct net_device *dev)

 	if (netif_running(dev)) {
 		spin_lock_irqsave(&tp->lock, flags);
-		dev->stats.rx_missed_errors += RTL_R32(RxMissed);
-		RTL_W32(RxMissed, 0);
+		rtl8169_rx_missed(dev, ioaddr);
 		spin_unlock_irqrestore(&tp->lock, flags);
 	}

@@ -3089,8 +3093,7 @@ static int rtl8169_suspend(struct pci_dev *pdev,
pm_message_t state)

 	rtl8169_asic_down(ioaddr);

-	dev->stats.rx_missed_errors += RTL_R32(RxMissed);
-	RTL_W32(RxMissed, 0);
+	rtl8169_rx_missed(dev, ioaddr);

 	spin_unlock_irq(&tp->lock);

-- 
1.5.6.4

Re: rtl8169_get_stats seems to be adding random crap to rx_missed_errors count

[Francois Romieu]

Alex Bennee <kernel-hacker@bennee.com> :
[...]

The fix is in davem's net-next tree. David has sent a pull request
yesterday so the fix ought to be in 2.6.28.

Thanks for your testing.

-- 
Ueimor