[OpenIndiana-discuss] Antwort: Re: Kernel panic trying to install OI

[Thorsten Heit]

Hi Albert,

> If one of you can reproduce the panic, can you copy the stack trace in
> panic message? I guess it wouldn't have written a crash dump, so you
> may have to boot with the '-k' flag appended to the 'kernel' line in
> GRUB, and the 'splashimage' line as well as 'console=graphics' removed
> (needs both, or the console will still not be visible). That way the
> system won't reboot immediately.
> 
> This should help us identify the bug.

Sorry for the late reply, but in the meantime I haven't tried again with 
OI.

What I did instead was trying to install plain Solaris 10 on my machine. 
Unfortunately only S10u8 could be installed on an empty disk; the recently 
released S10u9 had the same problems as OI, i.e. immediate reboot when 
creating a new zpool on a disc. After some discussions with people on a 
German user group list one of the guys told me why that could not work on 
my machine:


It's a problem with version of the inquiry data the hard disc returns (# 
format -e c4t5d0 -> scsi -> inquiry), and how ZFS works with them. Key is 
the ANSI version and the support for READ_CAPACITY_16 respectively what 
ZFS does with it. ANSI versions < 5 don't support READ_CAPACITY_16; v5 
supports it, but consider byte 0x13 in the response as 'reserved'. ANSI 
versions >=6 support READ_CAPACITY_16 and process byte 0x13.

The ZFS code in S10u9 obviously assumes that the storage only uses ANSI 
version >=6 and uses V6 data for calculating a disc geometry that - at 
least in my case - doesn't fit to the real one.

The guy also told me of a bug Id 6967658 "sd_send_scsi_READ_CAPACITY_16() 
needs to handle SBC-2 and SBC-3 response formats" that has the following 
explanation:

"This CR# is fast turning out to be S10U9 show stopper. Either the 
customers 
will experience panics while creating new zpools on S10U9 based systems or 

(wrose yet) will not be able to import existing zpools with S10U9."

*whoops* :-)


HTH

Thorsten