System freezes with SMP support enabled

[Aleksandr Melentiev]

Here's an exact behavior when SMP support is enabled:

Hence, one network card is onboard which is an fxp0.

5.0-RELEASE:
Often spontaneous and constant 'watchdog timeouts' with different network
cards that I have tried including SMC 1255 and 3Com 3C905-TX. An Intel
Pro/100+ showed 'device timeout'. In all cases, system would stop
responding. In most cases network connection would dissapear. Sometimes it
would happen again on the first second after reboot. Onboard network card
didnt show any errors! Only the PCI cards did.

4.8-RELEASE:
Same behavior as above, only no error messages whatsoever, network
connection goes down and system locks up. It is not spontaneous, happens
only when I try to transfer >1MB files locally via ftp (tried several
different ftpd and clients too), maybe other ways of high-speed transfering
are affected too. However, if I throttle my ftp client's upload speed to
15KBytes/sec, it transfers without a problem and systems stays stable.

None of the above happens when SMP support is disabled.
I am attaching dmesg where SMP support is enabled. Might it be because of
the PCI bridge?

Regards,
Alex



----- Original Message -----
From: "John Baldwin" <jhb at FreeBSD.org>
To: "Terry Lambert" <tlambert2 at mindspring.com>
Cc: <freebsd-smp at freebsd.org>; "Aleksandr Melentiev" <tzap at pacbell.net>
Sent: Wednesday, April 09, 2003 9:00 AM
Subject: Re: System freezes with SMP support enabled


>
> On 09-Apr-2003 Terry Lambert wrote:
> > What are both network cards?  Are they fxp?
> >
> > There are a couple of possibilities to consider...
> >
> > The first is that if both cards are identical (same vendor, etc.),
> > you might want to use a network card from a different vendor, to
> > make sure it's not the network card driver.
>
> This sounds like a sensible possibility and a worthy test.
>
> > Second, it seems to me that there's a possibility for a deadlock
> > if an interrupt comes in on one CPU, and an ithread to handle it
> > is scheduled to run on a different CPU.  You may want to try using
> > SCHED_4BSD to see if that changes anything.
>
> Huh?  Where in the code do you see this happening exactly?  All the
> bits you should need to look at for this are in ithread_schedule()
> and ithread_loop() in sys/kern/kern_intr.c.  Not only that, but 4.8
> doesn't have ithreads so I doubt seriously that this is causing the
> lockups on 4.x.
>
> --
>
> John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
> "Power Users Use the Power to Serve!"  -  http://www.FreeBSD.org/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: dmesg
Type: application/octet-stream
Size: 7358 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-smp/attachments/20030409/f2e00381/dmesg.obj