Problem w/ SMP and aic7xxx

[Constantine Gavrilov]

Joseph T. Trudeau wrote:
> 
> Problem:
> When running a kernel which supports SMP, I receive errors of scsi time-outs
> and resets under a load (it doesn't take much ... copying files or compiling
> while do the trick).  I enabled as much verbose flags as possible within
> aic7xxx and it seems as though scsi (some of which are completed) commands are
> dropped (see included messages below).
> 
> If I boot with a non-SMP kernel, I CANNOT reproduce the errors (maybe I can't
> generate enough traffic on the hard drives as with 2 CPU's compared to just
> one).  Hence, I suspect SMP, IO-APIC, and/or the aic7xxx driver.
> 
> I have tried multiple combinations of kernels from 2.0.36, 2.2.1, 2.2.3, and
> 2.2.5, with multiple compile options (i.e. PCI Bridging, MTRR, and anything
> that I found possibly related to the problem at hand).
> 
> Hardware:
>   HP Netserver LH Pro
>   128 Meg RAM (2 - 64 Meg DIMM)
>   2 - Pentium Pro 200's
>   2 - aic7880 on-board (PCI):  They share interrupt 11 and cannot be changed
>             to have unique interrupts for each (The EISA config utility
> promptly
>             configures both adapters to the same interrupt when either is
> changed).
> 
> NOTE:  I noticed that the 1st CPU has 512K cache while the 2nd CPU only has
> 256K cache.
> 

The driver has been fixed long time ago -- somewhere around 2.1.131. You must
suspect
hardware. Different CPUs in SMP config (all the more with different amount of
cache)
looks very suspicious. Try using equivalent CPUS. I am using Supermicro P6DGU
with 
two CPUs at home (single channel ULTRA II) with no problems. I have tested it
for 48 hours under load 8, compiling stuff in a loop and there were no errors.
At work, we use Supermicro P6DBS (dual channel ultra-wide) with no problems. It
is being a primary NFS server for 30+ clients and works as a charm (as far as
scsi is concerned). A backup server is Tyan S1836 DLUAN (dual channel
ultra-wide). This one was given to us as an evaluation computer and no problems
were seen as well. 



-- 
----------------------------------------
Constantine Gavrilov
Unix System Administrator and Programmer
Orbotech
Yavne 81102, Israel
Phone: (972-8)-942-3645
Fax:   (972-8)-942-3800
----------------------------------------


To Unsubscribe: send mail to majordomo at FreeBSD.org
with "unsubscribe aic7xxx" in the body of the message