Re: irqs and last_ipl[2] already set ...

[Ken T. <ke...@we...>]

On Tue, 8 Oct 2002, Rene Brothuhn wrote:

> It seems that the SCSI-deadlocks has nothing todo with Buster/Zorro. So 

I read hay4091.txt as saying that rev 9 Buster can cause lockups and I
thought I saw that there was one unfixed bug in rev 11 regarding bus
arbitrtion but may have misread the document.

> the only thing I can imagine what the cause of this problem is, is the 
> CS-PPC itself.

That's what I've put it down to uptil now, some flaw in the logic that
makes the PPC look like the 68000 to the motheboard.

> What are the facts? The 53c770 driver causes on some machines the same 
> deadlocks as the 53c710. On slower drives (like ZIP/CDROM) the drivers 
> seems to work. On faster drives you get deadlocks. It is also curious that 
> the deadlocks seems only appear on SCSI.

Adding to that, the sim710 driver which I've just got running for the
A4091 locksup in the same way.

The 53c7xx driver for the A4091 appears to work fine for the fixed address
A4000T 53c710 machines. Likewise the 53c770 on the PPC card is at a fixed
address but doesn't work reliably for me.  Not sure what the hardware
differences are between 'real' zorro plug in cards and zorro like devices
on the PPC card or motherboard.

> Maybe the problem has something todo with DMA and/or SCSI synchronous 
> mode. Because slower drives (like ZIP/CDROM) running in asynchronous mode 
> and don't use much DMA (because they are slow).
> The 53c770 uses DMA, but as far as I look through the 53c710 drivers, they 
> don't use DMA. But I'm not sure.

The ZIP and CDROM drives work for longer but do eventually fail in async
mode, last time I mounted a ZIP drive I worked it fairly heavily for about
three hours without a problem but at other times it has locked up on the
initial read when mounting. Same applies to CDROMS.

I've found a couple of things in the 53c7xx driver, one of them is that
the bootline async negotiation flag does not get used so the testing of
I've done in that mode as been futile. Another one is a bug in the abort
command interrupt processing. I've talked to the author about these.

The 53c7xx driver does use DMA.

Over the years I've tried many things to try to get the 4091 to work,
bounce buffers, much playing cache modes, making the driver less aggresive
but obviously no success.

Things I have to do:

Verify I have a rev 11 buster - does 53c770 being on card use buster for
bus arbitration ?

Keep working on the sim710 driver, calming it down and adding some DMA
address debug output in the hope I can see something of use in fixing the
53c770 driver.

Get out my old GVP Zorro 2 SCSI card and give that a run (Western Digital
chip, not ncr), I recall it being a DMA device but not sure. This might
uncover something in the APUS SCSI code or prove DMA works or not.