Re: Hardware not detected (9000/801/D200)
On Fri, 2003-02-07 at 09:35, Matthew Wilcox wrote:
> On Fri, Feb 07, 2003 at 08:51:11AM -0600, James Bottomley wrote:
> > I do actually have plans to add it in in the future. I talked to
> > Richard Hirst about getting sim710 working again (migration to the
> > device model, DMA model and cli/sti removal). The upshot is that we're
> > going to gut it and make it use the 53c700 chip driver.
> That makes a certain amount of sense. It's a little bit of a shame not
> to use the features of the 710 chip, but maintaining fewer drivers is
> clearly a Good Thing.
Actually, the 710 is useless in it's true 710 mode (i.e. continuously
operating scripts engine)---and we got the OS data to prove it back at
NCR. The problem is that it has no on-chip caching of script
instructions. Thus, if it's in continuous operation, it keeps accessing
memory to fetch the scripts instructions. Since these chips are usually
behind bridges (like EISA or MCA) for which memory access is slow, it
can end up bogging the machine (NCR actually withdrew the dual 710 card
in favour of a dual 700-66 for this reason). When you drive it in 700
mode, the chip is quiescent until it has to do something.
The script caching problem was the whole reason why the 53c875 and
higher have on-chip scripts memory (although LSI won't admit this).
> > In theory it
> > should be just a matter of adding the EISA probe ID and parsing the
> > registry entries to get it to work. Do you happen to have any docs for
> > this (really, probably all I need to know is the SCSI clock speed
> > feeding the 710 chip).
> I don't have any docs for this card specifically. According to
> "The two types of EISA-SCSI cards use the NCR 53C710 SCSI controller
> at 50 MHz to support either Single-Ended (SE) or Fast-Differential (FD)
> SCSI devices. The Series 700 Configuration Guide (November 1992 Edition)
> states that the older EISA Card (model 25525A) for FD SCSI devices is
> not qualified on the 715/33-50-75 models, and it recommends the 25525B
> EISA card for best results on the other workstations."
Great, thanks, I'll try it.
> > Of course, for this card, it's also predicated on getting EISA to work
> > again on pa-risc after Marc Zyngier's sysfs changes...
> Well, it was only tentatively working anyway... I shall be in the
> presence of the necessary documentation to get DMA working towards the
> end of the month and I'll type it all in then.
I'll look forward to that. The EISA device model needs beefing up, and
PA-RISC might be the place to do it (it doesn't really have a clearly
defined separation between the OS dependent and independent pieces yet).