Matthew Jacob wrote:

> http://people.freebsd.org/~mjacob/netbsd.gz
> built off of Top of Tree with some diagnostic code.

Ok, I've been running your kernel and in the meantime it has
actually caught something... unfortunately I wasn't able to
produce a kernel corefile in ddb since I didn't have enough
swap so I can only give you the following kernel message
and I wasn't proficient enough with ddb to actually find out
where it faulted (well, I didn't try hard since I thought
I could produce a crashdump, only to find out I was rather
short on swap...);
Incidentally it occured when I was about to login and reboot
the machine -- the "generic HBA error" hasn't occurred then
for 3 days; I was glad that it _did_ happen right on time. :)
Not much info here but it's probably better than nothing.


login: root
isp0: unable to setup dma (22)
sd0(isp0:0:0:0): generic HBA error

CPU 0: fatal user trap:

CPU 0    trap entry = 0x2 (memory management fault)
CPU 0    a0         = 0x1600a5c00
CPU 0    a1         = 0x1
CPU 0    a2         = 0xffffffffffffffff
CPU 0    pc         = 0x1600a5c00
CPU 0    ra         = 0x120003cc4
CPU 0    pv         = 0x1600a5c00
CPU 0    curproc    = 0xfffffc00077d3d10
CPU 0        pid = 186, comm = login

isp0: unable to setup dma (22)
sd0(isp0:0:0:0): generic HBA error
isp0: unable to setup dma (22)
sd0(isp0:0:0:0): generic HBA error
Jun 28 23:23:07 digibus /netbsd: pid 186 (login), uid 0: exited on 
signal 11 (core dumped)


Since the issue seems to occur when >1GB memory is installed and since
it seems to have something to do with DMA, I wondered if it could be
related to Jason Thorpe's SGMAP DMA improvements he made last year,
according to a quick scan of the port-alpha webpage.
This is just a wild speculation and I'm probably wrong with that but
it's an idea.

--mkb