[1403] in linux-scsi channel archive
Re: ncr53c8xx problem with a CLARiiON disk array.
daemon@ATHENA.MIT.EDU (Richard Waltham)
Fri Feb 7 18:12:49 1997
From: Richard Waltham <dormouse@farsrobt.demon.co.uk>
To: groudier@club-internet.fr (Gerard Roudier)
Date: Fri, 7 Feb 1997 23:07:50 +0000 (GMT)
Cc: linux-scsi@vger.rutgers.edu
In-Reply-To: <Pine.LNX.3.91.970207201728.93A-100000@localhost> from "Gerard Roudier" at Feb 7, 97 09:42:24 pm
Hi Gerard,
I think my previous post which I sent as this was received covers most
of what is here. I must think before typing;)
What are the problems with using the initial value of stest2 as set up by
the bios? Symbios bios on motherboard finds Tekram card and sets it up
incorrectly? That is one problem probably.
As you suggest, using a boot setup command is probably best for now. There
may be a problem even here though if there is more than one scsi adaptor.
The order they are detected seems to depend on the chip type and then the
motherboard PCI connectors. On one of my mother boards with three cards of
the same type the cards are detected in the order of connector 2 first, then
connector 1 then connector 3. If the cards are different then the card with
highest device ID is detected first.
Tekram LED does not use GPIO pin. It is driven from SCSI BSY through some
logic.
I have access to several different types of card so if there are any tests
that you would like me to make let me know. I have:
2 x 53c810a (no name from Taiwan)
3 x 53c815 (another no name from Taiwan)
1 x 53c825 differential (Symbios)
1 x 53c825a (Tekram)
3 x 53c875 (Symbios)
Don't suggest using too many cards as I will not have enough peripherals;)
Now for something different you may find interesting. Two weeka ago I bought
a new motherboard - Giga-byte 586HX (v1.53) - remember the recent problem and
discussion with Giga-byte tech support. The board has NCR 3.06 bios but with
my 53c875 with 4.03 bios the 4.03 bios is loaded, not the bios on the
motherboard:) The main bios is Award 4.51PG. But I have a different
motherboard with similar bios (Award 4.51/NCR 3.06) and on that board the
mother board NCR bios is loaded not the later one on the SCSI Adaptor. Wierd
isn't it? Motherboard manufacturers really need to allow the SCSI bios on
the motherboard to be disabled.
Finally I've been using v1.17 since you released it and it seems very
stable, for me anyway, with the 53c875. Runs many copies of Bonnie at the
same time without problems and also things like copying a 100Mbyte file to 8
other files concurently many times with no problems at all.
I did have some serious problems with a 53c810a card which hung the system
with NO error messages at all, which worried me, but that was a bad cable. I
have a new cable now and a weeks holiday so I will do some more testing with
the 810a just to make sure.
Richard
>
>
> On Fri, 7 Feb 1997, Richard Waltham wrote:
>
> > How about the DIF bit in the STEST2 register. (bit 5) This needs setting to
> > interface to differential drivers correctly.
> >
> > It would be useful to sense this automatically during booting and SYMBIOS
> > cards use GPIO3 to sense single or differential cards. GPIO3 is low
> > for differential cards. Unfortunately at least one other manufacturer
> > (Tekram) uses the same pin for some other function so this rules out
> > automatic sensing of differential mode.
>
> We can do something that seems safe enough to ensure SYMBIOS-like GPIO3
> wiring, for example:
>
> - save initial value of STEST2 register
> - save initial value of GPCNTL
>
> If BIOS setting was bit5(STEST2) set and bit3(GPREG) not writeable,
> then if bit3(GPREG) not set, assume differential and preserve bit5
> of STEST2.
> But Tekram initial GPCNTL is 0x9 and it is in fact not safe enough.
>
> Or, we can base decision only on initial value of STEST2, but we can
> get side effects with incorrect initial settings ...
>
> In my opinion,
> all vendor-specific stuff must be disabled by default, unless
> the driver is able to probe it safely. Unfortunately, that seems not
> possible (at least for the moment).
> A compile option is not safe too, being given that it could make the
> driver broken for not enough compatible boards.
> I would prefer a boot setup command option.
>
> The latest revision (1.17) of the driver supports controller LED
> attached to GPIO0.
> It can be enabled by a CONFIG_EXPERIMENTAL option or enabled/disabled by
> the boot command line for all attached boards.
> The driver checks that initial setting of GPCNTL allows write to GPIO0.
> That works only when BIOS is present and fortunately detects Tekram
> difference (Initial value of GPCNTL is 0x9 for Tekram boards).
> If somebody knows which GPIO pin is used by Tekram boards for LED, I am
> interested in this info.
>
> I am not satisfied about the current check of GPIO0_FETCHN for LED
> support, but I donnot find better.
>
> > May be another compile time option? There may be problems though if there
> > is more than one 53c8xx host adaptor and they are a mixture of single ended
> > and differential.
>
> For the moment, the boot setup command allows to set driver options, so
> for all installed NCR boards.
> A boot setup per controller will help. I will add this in a future
> version of the driver.
>
> Send ideas.
>
> Gerard.
>
