NFS in 2.4.8/9ac

NFS in 2.4.8/9ac

[Thomas Dodd]

I've suddenly started getting host not responding messages
on NFS mounts. The mounts are from Solaris8 and HPUX-10.20.
Other Solaris8 machines don't have this problem,
and the machines serving the mounts are unloaded, and responsive.

2.4.7-ac10 worked fine.2.4.8-ac8 through 2.4.9-ac3
have trouble. I have not tested 2.4.8-ac[1..7] yet.

Linux Tulip driver version 0.9.15-pre6 (July 2, 2001)
tulip0:  MII transceiver #1 config 1000 status 782d advertising 01e1.
eth0: Lite-On 82c168 PNIC rev 33 at 0xe400, 00:A0:CC:50:35:C2, IRQ 10
(Tulip FA310TX card)
NETDEV WATCHDOG: eth0: transmit timed out
nfs: server nd not responding, still trying


gcc version 2.96 20000731 (Red Hat Linux 7.1 2.96-85)
binutils 2.10.91.0.2 (from RedHat -3 i386 package)
mount-2.11b
autofs-3.1.7-14 (using autofs.o NOT autofs4.o)
from /proc/mounts:
nd:/export/nd00/ted /home/ted nfs
     rw,v3,rsize=32768,wsize=32768,hard,udp,lock,addr=nd 0 0
hp3:/users/hp02 /users/hp02 nfs
     rw,v3,rsize=16384,wsize=16384,hard,udp,lock,addr=hp3 0 0

	-Thomas

Re: NFS in 2.4.8/9ac

[Alan Cox]

> 2.4.7-ac10 worked fine.2.4.8-ac8 through 2.4.9-ac3
> have trouble. I have not tested 2.4.8-ac[1..7] yet.

Ok can you transplant drivers/net/tulip from a working -ac into the
current tree and see if that fixes it - that tells us if its a network
driver bug or a kernel core bug.

Alan

Re: NFS in 2.4.8/9ac

[Thomas Dodd]

Alan Cox wrote:
>>2.4.7-ac10 worked fine.2.4.8-ac8 through 2.4.9-ac3
>>have trouble. I have not tested 2.4.8-ac[1..7] yet.
>>
> 
> Ok can you transplant drivers/net/tulip from a working -ac into the
> current tree and see if that fixes it - that tells us if its a network
> driver bug or a kernel core bug.
this is the only diff I find in the tulip directory.
(2.4.8 and 2.4.9 have no diff)

--- linux-2.4.7-ac10/drivers/net/tulip/tulip_core.c
+++ linux-2.4.9-ac3/drivers/net/tulip/tulip_core.c
@@ -1409,6 +1409,10 @@
         if (chip_idx == LC82C168)
                 csr0 &= ~0xfff10000; /* zero reserved bits 31:20, 16 */

+       /* DM9102A has troubles with MRM, clear bit 24 too. */
+       if (pdev->vendor == 0x1282 && pdev->device == 0x9102)
+               csr0 &= ~0x01200000;
+
         /*
          *      And back to business
          */
@@ -1751,9 +1755,9 @@
                 kfree (tp->mtable);
  #ifndef USE_IO_OPS
         iounmap((void *)ioaddr);
-#endif

  err_out_free_res:
+#endif
         pci_release_regions (pdev);

  err_out_free_netdev:

---------
the second hunk could be a problem.
I'll try moving the endif back tomorrow.

	-Thomas

Re: NFS in 2.4.8/9ac

[Thomas Dodd]

Thomas Dodd wrote:
> I've suddenly started getting host not responding messages
> on NFS mounts. The mounts are from Solaris8 and HPUX-10.20.
> Other Solaris8 machines don't have this problem,
> and the machines serving the mounts are unloaded, and responsive.

> Linux Tulip driver version 0.9.15-pre6 (July 2, 2001)
> tulip0:  MII transceiver #1 config 1000 status 782d advertising 01e1.
> eth0: Lite-On 82c168 PNIC rev 33 at 0xe400, 00:A0:CC:50:35:C2, IRQ 10
> (Tulip FA310TX card)
> NETDEV WATCHDOG: eth0: transmit timed out
> nfs: server nd not responding, still trying

Just a note: This is still a problem.
eventually I'll get a message about task slots.

It's still there through 2.4.9-ac10.

On a new Athlon 1.4 box, with a LNE100TX NIC
it's there with the default kernel from Red Hat Linux 7.1
(2.4.2-2.i686.rpm), and on a P4-1.8GHz with the same NIC.

The servers are Solaris8 and Solaris-2.6, Sparc machines, and the config
hasn't changed in over a year on the 2.6 boxes.

Anorthe system with RHL-6.2 (2.2.14-5 kernel and eepro100 NIC)
has no trouble talking to the servers.

What/how do I set debugging code on the 2.4 kernels to trak
down this problem? It makes NFS mounted home dirs unusable.

	-Thomas

Re: NFS in 2.4.8/9ac

[Alan Cox]

> > eth0: Lite-On 82c168 PNIC rev 33 at 0xe400, 00:A0:CC:50:35:C2, IRQ 10
> > (Tulip FA310TX card)

Same as mine

> Just a note: This is still a problem.
> eventually I'll get a message about task slots.

Thats the NFS backlog filling. That itself does not indicate a problem of
any kind just congestion - which if the card is getting tx timeouts and
watchdog errors would be reasonable

> On a new Athlon 1.4 box, with a LNE100TX NIC
> it's there with the default kernel from Red Hat Linux 7.1
> (2.4.2-2.i686.rpm), and on a P4-1.8GHz with the same NIC.

Well 2.4.2 is in my prehistory pile. I dont know if this has been fixed
or is known - Jeff Garzik may do.

Alan