Subject: port-sparc/3188: 4/110, the 'sw' driver and disk activity panics kernel
To: None <gnats-bugs@gnats.netbsd.org>
From: Brad Spencer <brad@anduin.eldar.org>
List: netbsd-bugs
Date: 02/04/1997 19:52:18
>Number:         3188
>Category:       port-sparc
>Synopsis:       4/110, the 'sw' driver and disk activity panics kernel
>Confidential:   no
>Severity:       critical
>Priority:       high
>Responsible:    gnats-admin (GNATS administrator)
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Feb  4 17:05:01 1997
>Last-Modified:
>Originator:     Brad Spencer
>Organization:
	Sitting at home
>Release:        -current as of 2/2/97
>Environment:
	
    @(#)NetBSD 1.2C (ELROND) #0: Mon Feb  3 19:02:17 EST 1997
    brad@elrond.eldar.org:/usr/src/sys/arch/sparc/compile/ELROND

>Description:

Given a 4/110 with 8MB of core.

Some sort of change between Novemeber 21 [1.2B] and -current has
causes the 'sw' driver in the sparc port to print the following
message:

si_dma_alloc: excessive xlen = 0x10000

followed shortly be a complaint about not being able to transfer more
data and then a kernel panic.

>How-To-Repeat:

Nearly any moderate amount of disk activity will trigger this.  The
simplest method is, perhaps, to do a compile of a reasonably sized C
file.  Moving into multi-user actually seems to not be enough to
trigger this.

>Fix:

No idea.  It is unclear to me if the 'sw' driver can be made to do
polled I/O only.  If so, it might be a possible work around, at a
performace hit.
>Audit-Trail:
>Unformatted: