Re: NCR53C8XX: IRQ blundering

Gerard Roudier (groudier@club-internet.fr)
Sun, 22 Sep 1996 18:43:20 +0000 (GMT)


Alan,

On Sun, 22 Sep 1996, Alan Cox wrote:

> > All the manual says is:
> >
> > Chipset Special Features
> > When disabled, the chipset behaves as if it were the earlier 430FX
> > chipset. This option should be enabled for best performance.
> > The default is enabled
> >
> > Anyone know what I lose by disabling this option?
>
> No but I can tell you the NCR controllers do a lot of PCI bus things that
> catch out magic hacks and PCI timing violations. Because the NCR chip runs its
> code from main memory it generates assorted read/writes to the PCI bus, often
> back to back.
>
> It may be Gerard has something that happens to trip your bus but Drew doesn't.
> It could also just be turning the option off slows the machine down a tiny
> fraction to mask some other software or hardware bug....

You are probably right.

Prior to resetting the chip, Drew saves some IO registers (set by the
BIOS) that affect PCI things, and restore some part of those registers
after chip reset.
The BSD driver, does not trust the BIOS for PCI related features and so
ignore BIOS setting.
These registers are DCNTL, CTEST4 and DMODE.

For example, CTEST4=(1<<7) means BURST DISABLE. If the BIOS has disabled
burst, Drew will keep that unchanged, but BSD driver will not.

I have sent Jon a patch that print a chip register dump, just before the
driver resets the chip. This dump will allow to guess if there is a major
difference of chip setting between the 2 drivers.

> Alan

Gerard.