I have a Sun SPARCstation LX (50 MHz microSPARC CPU, sun4m architecture),
with 96M of RAM, and three Fast SCSI-2 controllers (ESP200's: one onboard,
two are 501-2015 X1053A FSBE/S cards), running NetBSD-current as of June
18, 1997 (yesterday). I am attempting to use ccd(4) to mirror two HP C3725S
2GB disks, which are on two different SCSI controllers (the two Sbus
cards). Attempting to newfs the single partition yields:

pmap_extract: invalid pte of type 0
panic: dvma_mapin: null page frame

I was hoping that the recent changes to the pmap module would fix this
problem (it also happened before those changes were commited). The stack
backtrace printed after the panic (subsequently interpreted by gdb) yields:

0xf80ff430 <cpu_reboot+196>:     0x40000093
0xf802c80c <panic+80>:   0x40034ad8
0xf810a190 <dvma_mapin+260>:     0x7ffc898b
0xf8109f8c <kdvma_mapin+20>:     0x40000040
0xf80f2e54 <dma_setup+488>:      0x40005c49
0xf80f3d48 <esp_dma_setup+24>:   0x9fc34000
0xf800e434 <ncr53c9x_intr+4544>: 0x9fc34000
0xf80069ac <sparc_interrupt_common+368>: 0x9fc24000
0xf810584c <pmap_enk4m+428>:     0x7ffff187

The full details can be found in PR# kern/3765 (minor plug: a Macintosh
running MacTerminal makes a dandy console - happily capturing all the
output, all cut&paste-able). This slightly redolent of a missing interrupt
exclusion in the sparc pmap code or in the SCSI controller driver, but I
have been successfully using ccd on a SPARCclassic, interleaved, with both
component disks on the same controller, with a kernel of the same vintage
(very current), for a week or so under light use. No ka-boom, before or
after pmap changes.

The question I have for port-sparc is how many of you are using ccd (on any
platform) with multiple SCSI controller configurations for either mirroring
or striping (interleaving) disks? (in other words, am I the first one fool
enough to try this?)

	thanks,

	Erik Fair <fair@clock.org>