NetBSD usability on DS20L, part three

[John Klos <john%ziaspace.com@localhost>]

So I haven't been able to nail down exactly when disk corruption will 
occur, but at least we know how to make sure it doesn't occur. Here's what 
I do know:

Use motherboard fxp and disk on either aceride0 or esiop0 and corruption 
hapens within hours.

Use PCI wm and disk on on either aceride0 or esiop0 and there'll be 
eventual corruption, but it takes noticeably longer (days instead of 
hours). Seems more pronounced with SCSI than with IDE.

I've tried SCSI and motherboard fxp with Tobias' sprinkled memory
barriers with no problems even after lots of cvs'ing and compiling:
https://mail-index.netbsd.org/port-alpha/2017/07/29/msg000857.html


Any issues with UDMA with IDE might be complicated by the fact that the 
IDE chip in the DS20L is the same kind as in my SunFire V100:

aceride0 at pci0 dev 16 function 0: Acer Labs M5229 UDMA IDE Controller (rev. 0xc2)
aceride0: bus-master DMA support present, but unused (forced off by config file)
aceride0: primary channel configured to compatibility mode
aceride0: primary channel interrupting at isa irq 14

This IDE chip reportedly doesn't support UDMA with revisions below 0xc4 
(or is it UDMA with LBA48 commands?), so I tested a kernel with flags 
0x0002 to force PIO mode, and I've been running a system with a PCI wm 
card and the primary disk on IDE in PIO for more than a week with heavy 
disk usage and lots of compiling with no issues at all.

I'll test whether corruption occurs with UDMA and disks larger than 
128 gigs with this revision 0xc2 controller. I believe that the system 
can't be booted off of an IDE disk where the first partition is greater 
than 128 gigs, but that's another issue.


Also, it seems that the firmware doesn't like 2 TB SCSI disks. While 
booting with a 2 TB SCSI disk, I get:

*** Hard Error - Error #8 -
Diagnostic Name        ID             Device  Pass  Test  Hard/Soft   1-JAN-2000
exer_kid         00000239       dka0.0.0.3.0     0     0     1    0     12:00:01
Buffer counts differ - buf1:0, buf2:512, location:2a00

*** End of Error ***

Has anyone else run in to issues with large SCSI disks? I'm planning to 
test with firmware version 6.6 (this machine has version 6.0), and I'll 
also try with a 1 TB SCSI disk.

John