Kernel Panic New Adaptec - Old Adaptec Works

Kernel Panic New Adaptec - Old Adaptec Works

[David Koretzky]

Hello,

I am running a dual PIII 933 VIA based motherboard (Gigabyte 6VXDC7). 
I have two Seagate Cheetah HDs, and two SCSI cdroms.  There are no IDE 
devices anywhere in the system, and no support is compiled in the kernel or 
as a module for IDE.

Kernel 2.4.7 works perfectly.  I cannot get 2.4.10 to work with the new aha 
driver compiled into the kernel.  Below is what I was able to capture of the 
output 

Disconnected List inconsistency. SCB index == 255, yet numscbs=4.scsi0

Dumping card state while ide, at SEQADDR 0x1a5

ACCNUM=0x0
SINDEX=0x0
DINDEX=0x0
ARG_2=0x7f
HCNT=0x0
SCSI SEQ=0x12 SBLKCTL=0xa
DFNCTRL=0x0 DFSTATUS=0x89
LAST PHASE=0x1 SCSISIGI=0x0 SXFRCTL0=0x88

SSTAT0=0x0
SSTAT1=0x0
STACK=0x0, 0x0, 0x0, 0x0
SCB_COUNT=4
KERNEL NEXTQSCB=3
CARD NEXTQSCB=3

QINFIFO Entries: Waiting Queue Entries: Disconnected Queue Entries:
0:255...31:255

QOUTFIFO entries:
Sequencer Free SCB List:
0...31

Pending List:
Kernel Free SCB List: 2 1 0
Kernel Panic : For Safety

I am currently running 2.4.10 with the old Doug Ledford driver, and it works 
perfectly.

P.S. Could you please CC me any help?

Thank you.

-- 

David Koretzky
david@spanware.com

Re: Kernel Panic New Adaptec - Old Adaptec Works

[Justin T. Gibbs]

>Hello,
>
>I am running a dual PIII 933 VIA based motherboard (Gigabyte 6VXDC7). 
>I have two Seagate Cheetah HDs, and two SCSI cdroms.  There are no IDE 
>devices anywhere in the system, and no support is compiled in the kernel or 
>as a module for IDE.
>
>Kernel 2.4.7 works perfectly.  I cannot get 2.4.10 to work with the new aha 
>driver compiled into the kernel.  Below is what I was able to capture of the 
>output 

You need to disable all of the "speed up my PCI bus even though it violates
the spec) options in your BIOS.  The old driver issues a PCI read after
every device write to overcome such silliness in broken BIOSes/chipsets.
The new driver only issues a read (to flush writes) when synchronization
is required.  The workaround used in the old driver has too high of a 
performance penalty which is why the new driver does not do it.

--
Justin