Re: ahc scsi crashes

[Theo de Raadt <deraadt@cvs.openbsd.org>]

This is known, and is a mailing list searchable FAQ.  The newer 7899
(F and up I think) chipsets are enot support at the moment. In the
next 20 minutes the code for that is being commited.  There will be
other fallout, and we will react to that.

So everyone with ahc chipsets, start looking at the new stuff as soon
as it goes in.

> I'm in serious need of help here. I'm currently running two OpenBSD 3.1 servers and am getting the same scsi errors on both of them. Here is the error, hand copied from the server:
> 
> ahc1:A:0 no active SCB for reconnecting target - issuing BUS DEVICE RESET
> SAVED_TCL == 0x0, ARG_1 == 0x13, SEQ_FLAGS == 0x40
> ahc1: Bus Device Reset on A:0. 6 SCBs aborted
> sd0(ahc1:0:0): Unexpected busfree in Message-out phase
> SEQADDR == 0x165
> ahc1: target 0 using 16bit transfers
> ahc1: target 0 synchronous at 40.0Mhz, offset = 0x1f
> ahc1: target 0 using 16bit transfers
> ahc1: target 0 synchronous at 40.0Mhz, offset = 0x1f
> sd0(ahc1:0:0): User command with no ioctl
> uvm_fault(0xf37df920, 0x1000, 0, 3) -> d
> kernel: page fault trap, code = 0
> Stopped at      _pool_get+0x268:      movl      %eax,0(%edx)
> 
> 
> The first machine that had this problem was an Intel STL2 using the onboard AIC 7899 scsi controller with two Quantum Atlas 10KIII 36G hard drives mirrored using RAIDctl. The installation and setup of the server went fine, but two days after the server went live it suddenly crashed (kernel panic). The only thing that seemed to prevent the server from crashing was undoing the RAIDframe mirror and running from standard FFS partitions off the first drive.
> 
> I began testing on another machine, using a supermicro board with an Adaptec 19160 controller card and two Quantum Atlas 10KII 18G hard drives. I experienced the same crash after setting up RAIDframe and letting the server sit idle for two days. After about a week of messing around I've determined that there is nothing wrong with the RAIDframe code (as I intially thought), since I am able to reproduce the crash just by mounting a partition from each disk and simultaneously reading and writing to both disks. This causes a failure within about 15-20 seconds. 
> 

[ rest removed]