[2.6.7-bk] NFS-related kernel panic

[2.6.7-bk] NFS-related kernel panic

[walt]

Starting just today with the latest bk changesets I get a kernel
panic when starting the rpc.statd daemon (NFS):

Kernel panic:  Aiee, killing interrupt handler
In interrupt handler - not syncing

Anyone else seeing problems with NFS?

Re: [2.6.7-bk] NFS-related kernel panic

[Grzegorz Kulewski]

On Wed, 23 Jun 2004, walt wrote:

> Starting just today with the latest bk changesets I get a kernel
> panic when starting the rpc.statd daemon (NFS):
> 
> Kernel panic:  Aiee, killing interrupt handler
> In interrupt handler - not syncing
> 
> Anyone else seeing problems with NFS?

This is probably not NFS problem since I have no NFS and I get the same. 
Look at my report. I sent it minute ago to this list.


Thanks,

Grzegorz Kulewski

Re: [2.6.7-bk] NFS-related kernel panic

[walt]

Chris Wright wrote:

> The lockless loopback transmission patch mucks up the preempt count.
> Can you give this patch a try?

I saw an update to loopback.c from linus which I assume was yours --
anyway my panic is fixed.  Thanks.

Re: [2.6.7-bk] NFS-related kernel panic

[Chris Wright]

* walt (wa1ter@hotmail.com) wrote:
> Chris Wright wrote:
> 
> > The lockless loopback transmission patch mucks up the preempt count.
> > Can you give this patch a try?
> 
> I saw an update to loopback.c from linus which I assume was yours --
> anyway my panic is fixed.  Thanks.

Actually that's from Andrew, and Arthur sent in a minor tweak which has
yet to hit mainline.  Glad it's working.

thanks,
-chris
-- 
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net

Re: [2.6.7-bk] NFS-related kernel panic

[Bob Gill]

Ok.  I get (a very similar) error with 2.6.7-bk4.
The error message I get is:
Starting portmapper:                            [OK]
Starting NFS statd: Kernel panic: Aiee, killing interrupt handler!
bad: scheduling while atomic!
 [<c02a2c37>] schedule+0x483/0x488
 [<c0133298>] __get_free_pages+0x33/0x3f
 [<c026974c>] tcp_poll+0x34/0x15a
 [<c02a30b5>] schedule_timeout+0xb5/0xb7
 [<c0248fb3>] sock_poll+0x29/0x31
 [<c015da2e>] do_pollfd+0x4f/0x90
 [<c015db10>] do_poll+0xa1/0xc0
 [<c016dc7e>] sys_poll+0x14f/0x211
 [<c015d09d>] __pollwait+0x0/0xc6
 [<c014afd2>] sys_close+0x63/0x96
 [<c0103eb1>] sysenter_past_esp+0x52/0x71
In interrupt handler - not syncing

And that's where it dies.
I don't have NFS as part of the build either. My build script shows the
following:
# CONFIG_NFS_FS is not set
# CONFIG_NFSD is not set

Something is borken.

Bob

Re: [2.6.7-bk] NFS-related kernel panic

[Chris Wright]

* Bob Gill (gillb4@telusplanet.net) wrote:
> Ok.  I get (a very similar) error with 2.6.7-bk4.
> The error message I get is:
> Starting portmapper:                            [OK]
> Starting NFS statd: Kernel panic: Aiee, killing interrupt handler!
> bad: scheduling while atomic!
>  [<c02a2c37>] schedule+0x483/0x488
>  [<c0133298>] __get_free_pages+0x33/0x3f
>  [<c026974c>] tcp_poll+0x34/0x15a
>  [<c02a30b5>] schedule_timeout+0xb5/0xb7
>  [<c0248fb3>] sock_poll+0x29/0x31
>  [<c015da2e>] do_pollfd+0x4f/0x90
>  [<c015db10>] do_poll+0xa1/0xc0
>  [<c016dc7e>] sys_poll+0x14f/0x211
>  [<c015d09d>] __pollwait+0x0/0xc6
>  [<c014afd2>] sys_close+0x63/0x96
>  [<c0103eb1>] sysenter_past_esp+0x52/0x71
> In interrupt handler - not syncing

The lockless loopback transmission patch mucks up the preempt count.
Can you give this patch a try?

thanks,
-chris
-- 
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net

===== loopback.c 1.15 vs edited =====
--- 1.15/drivers/net/loopback.c	2004-06-20 17:35:52 -07:00
+++ edited/loopback.c	2004-06-21 20:23:06 -07:00
@@ -143,10 +143,11 @@
 
 	dev->last_rx = jiffies;
 	if (likely(loopback_stats)) {
-		get_cpu_ptr(loopback_stats)->rx_bytes += skb->len;
-		get_cpu_ptr(loopback_stats)->tx_bytes += skb->len;
-		get_cpu_ptr(loopback_stats)->rx_packets++;
-		get_cpu_ptr(loopback_stats)->tx_packets++;
+		struct net_device_stats *stats = get_cpu_ptr(loopback_stats);
+		stats->rx_bytes += skb->len;
+		stats->tx_bytes += skb->len;
+		stats->rx_packets++;
+		stats->tx_packets++;
 		put_cpu_ptr(loopback_stats);
 	}
 

Re: [2.6.7-bk] NFS-related kernel panic

[Stewart Smith]

[-- Attachment #1: Type: text/plain, Size: 314 bytes --]

On Tue, 2004-06-22 at 13:35, Chris Wright wrote:
> > bad: scheduling while atomic!
> The lockless loopback transmission patch mucks up the preempt count.
> Can you give this patch a try?

This seems to fix the problem, thanks.

-- 
Stewart Smith (stewart@flamingspork.com)
http://www.flamingspork.com/


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [2.6.7-bk] NFS-related kernel panic

[Grzegorz Kulewski]

Unfortunatelly this does not fix my (similar?) problem described in 
"Network related(?) kernel panic (2.6.7-bk4)".


Thanks,

Grzegorz Kulewski


On Tue, 22 Jun 2004, Stewart Smith wrote:

> On Tue, 2004-06-22 at 13:35, Chris Wright wrote:
> > > bad: scheduling while atomic!
> > The lockless loopback transmission patch mucks up the preempt count.
> > Can you give this patch a try?
> 
> This seems to fix the problem, thanks.
> 
> -- 
> Stewart Smith (stewart@flamingspork.com)
> http://www.flamingspork.com/
> 
> 

Re: [2.6.7-bk] NFS-related kernel panic

[Arthur Kepner]

On Mon, 21 Jun 2004, Chris Wright wrote:

>
> The lockless loopback transmission patch mucks up the preempt count.
> Can you give this patch a try?
>
> thanks,
> -chris
> --
> Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net
>

Yes, there's a problem with preempt count. And this patch is
the minimal fix for it.

However, Andrew Morton noted this problem already and posted
a patch to be tested yesterday. I'd like to suggest that we go
with his patch (very slightly modified) which I'll post in a few
minutes to netdev. (I will cc Chris and Bob on this too.)

--

Arthur


> ===== loopback.c 1.15 vs edited =====
> --- 1.15/drivers/net/loopback.c	2004-06-20 17:35:52 -07:00
> +++ edited/loopback.c	2004-06-21 20:23:06 -07:00
> @@ -143,10 +143,11 @@
>
>  	dev->last_rx = jiffies;
>  	if (likely(loopback_stats)) {
> -		get_cpu_ptr(loopback_stats)->rx_bytes += skb->len;
> -		get_cpu_ptr(loopback_stats)->tx_bytes += skb->len;
> -		get_cpu_ptr(loopback_stats)->rx_packets++;
> -		get_cpu_ptr(loopback_stats)->tx_packets++;
> +		struct net_device_stats *stats = get_cpu_ptr(loopback_stats);
> +		stats->rx_bytes += skb->len;
> +		stats->tx_bytes += skb->len;
> +		stats->rx_packets++;
> +		stats->tx_packets++;
>  		put_cpu_ptr(loopback_stats);
>  	}
>
>