Hello Leo,

Time has gone ..
and I 've made many tests to the two NetBSD-Kernels, namely BOOTX.pol 
and BOOTX.int in the last three days and I've got many expierences,
some of them bad, some very surprised ...


The tests:  (Part 2)
==========

2)  loading  Kernel BOOTX.pol  to Config A:  2 SCSI-disks + 1 ZIP

The Configuration:
	sd0	<IMPRIMIS> 1012 MB  "2nd-disk"
		sd0d on /tos2	MSDOS 	(250 MB)
		sd0f on /tos3	MSDOS 	(250 MB)
	sd1:	<CDC> 528 MB  "System-disk"
		sd1a on /  	ffs	( 85 MB)
		sd1b on swap
		sd1d on /tos1	MSDOS	( 80 MB)
		sd1e on /usr	ffs	( 85 MB)
		sd1f on /home	ffs	(310 MB)
	sd2:	<IOMEGA-ZIP> 96 MB
		sd2c on /zip1	ffs	( 96 MB)
The expierences:
1. Boot ok,  	tar -clf /tos3/test21.tar /  ==> kernel hung
2. Boot ok,  	repeat above	==> OK ; 
		tar -tvf /tos3/test21.tar ==> OK
		tar -tvf /zip1/netbsd-usr.tar 
		==> panic: buffer larger than expected
3. Boot ok,	repeat above (tar from /zip1/..) ==> OK
		newfs -m 1 /dev/sd0f (250 MB) ==> panic: MMU fault
4. Boot ok,	repeat newfs  ==> kernel hung

So it seems too much crashes and hungs with this configuration
(more than with one SCSI-disk + ZIP). 
Therefore next day ...

3)  loading Kernel  BOOTX.int  to  Config B: 1 SCSI-disk + 1 ZIP

The configuration:
	sd0:	<CDC> 528 MB  "System-disk"
		sd0a on /  	ffs	( 85 MB)
		sd0b on swap
		sd0d on /tos1	MSDOS	( 80 MB)
		sd0e on /usr	ffs	( 85 MB)
		sd0f on /home	ffs	(310 MB)
	sd1:	<IOMEGA-ZIP> 96 MB
		sd1c on /zip1	ffs	( 96 MB)
(grr, I have to re-edit /etc/fstab in singleuser-mode, I need
soon a new kernel with sd*-directive = fixed SCSI<-->sd? ;-)
The expierences:
2. Boot, after binpatch _st_pool_size to 999999
	 after line "starting network daemons: printer inetd"
	 panic: UIOmove mode	  
3. Boot, ok	tar -tvf /zip1/netbsd-usr.tar ==> kernel hung
4. Boot, ok	repeat above  ==> OK
5. Boot, ok	tar -cvf /zip1/netbsd-home.tar /home 
		==> kernel hung (very lately, after about 80 % written)
6. Boot, ok	repeat above ==> kernel hung (very early)

So the BOOTX.intr kernel looks very less stable as the BOOTX.poll
kernel tested on the same Configuration (as told in my last mail).
But to realize this I tested this again:

4)  loading kernel  BOOTX.pol  to Config B: 1 SCSI-disk + 1 ZIP 

The expierences:
1. Boot: ok	tar -cvf /zip1/netbsd-home.tar ==> kernel hung
2. Boot: ok	repeat above  ==> OK  
		the same tar-cmd but with MSDOS-FS on ZIP
		==> kernel hung (same position as after 1.Boot)
3. Boot: ok	repeat above  ==> OK
		Hardtest:  tar -tvf /zip1/home43.tar (on ttye1)
			+  tar -cvf /zip1/home44.tar /home (on ttye2)
			at the same time
		==> kernel hung
4. Boot ok	repeat Hardtest ==> OK 
		(why that, for heavens sake ??)

Here you can see a strange rule:
Obvoisly the kernel BOOTX.poll changes his behavior on large SCSI
data transfers from or to the ZIP regulary: 
>>>>	first it hung, second it works OK  	<<<<I think this is too regulary as that is arbitrary !


On the next day (today) I made up my mind that I will do something
else than test large data transfer with  Kernel BOOTX.pol / Config B:
So I tried to make my fovourite shell the  t c s h  (6.06.00):

1. Boot ok,	make tcsh  ==(2 hours)==>> OK, tcsh works fine
		(but I forgot just ONE parameter, so)
		repeat  make tcsh 
		==> Illegal instruction on various commands, because
		sd0(ncrscsi0:1:0) Byte count on read not a multitple
				of 512 (nnnn)
		where (nnnn) mostly = 8176
		then NO COMMAND WORKS, all brought "Illegal instruction"
		even 'reboot' or 'halt'; only 'sync' did.
So I ran into trouble and need, especially then as 
REBOOT HUNG IN LOADING KERNEL !! :-((
e.g.  	D:\LOADBSD.TTP -a D:\NETBSD  hung at once !!
I got this effect one month ago (with NetBSD1.1) and like then 
switching off my Falcon HAS NO EFFECT !!
I couldn't do anything other than wait. . . .
After about two hours ... loading works again (what is this ??)
I realized this strange effect, because the loading of kernel
didn't work some more times. And everytime after switching off AND WAITING
AT LEAST ONE MINUTE loading works again, but after switching off and on 
immediately loading does not work. I wondered what thing can have effect
on a switched off Falcon and is not permanently (like errors on disk) ???

Then I got that annoying message "Byte count not a multiple of 512" 
again. So I checked my filesystem. No effort ! 
Then I changed kernel to BOOTX.int 
Boot ok, 	make tcsh (from scratch) ==(2 hours)==> OK
		(one message: "DMA interrupt discarded")
		kernel works for about 3 hours fine  (astonishing !)

and the last news
changing Kernel back to BOOTX.pol 
1. Boot ok	kernel seems to work fine with "some small cmds"
		(no messages yet)    

WHO CAN EXPLAIN THAT !??!  I'm pry ...

Good night
Helmar 
-- 
          ___   __   Helmar Goettsch (Admin HEG: Lehrstuhl III, Informatik)
  /   /  /     /  '  Orleansstr. 34, D-81667 Muenchen, Phone: 089/48095-190  
 /---/--/-----/--+   email: goettsch@informatik.tu-muenchen.de
/   /  /___  /__/    WWW:   http://www3.informatik.tu-muenchen.de/~goettsch