Re: [PATCH 0/14] Start coping gsis < 16 that are not isa irqs.

[ebiederm@xmission.com (Eric W. Biederman)]

Len Brown <lenb@kernel.org> writes:

>> It appears there are a few systems in the wild that use acpi
>> interrupt source overrides to report a gsi > 16 is an isa irq.
>> 
>> This breaks all kinds of assumptions I figure any BIOS doing that
>> probably should be shot as that is very much not a conservative position.
>
> You might run into trouble here on the ES7000 -- though I don't know
> if anybody is booting a modern kernel on one of those these days.
>
> IIR, ES7000 treated the bottom 16 as a special case.
> When there was an irq shortage, I think they used overrides
> to map higher PCI irqs into the empty spots below 16,
> but I think to make room for them they may have mapped
> some of the ISA irqs to high numbers.  Fuzzy memory
> on this at the moment...

A couple of things.

The ES7000 change is trivially safe because despite differences in how
the numbers are computed I have made that transform that the es7000
does always apply.

I also talked to Natalie about this.  I can't be certain about this
but from your description, from Natalies memories and from looking
at the code.  I believe the problem on the ES7000 is exactly what
I am fixing in the code.  Natalie has threatened to test this on
an ES7000.

The case of interest is this:
Weird but valid platforms where
GSI 0-15 are not ISA irqs.

Some high numbered GSIs are ISA irqs and use interrupt source
overrides to describe them.

That is the case on the IBM platform that regressed a bit ago
and Iranna D Ankad bisected a failure on. On that IBM platform
GSI 0-15 (because the are not ISA irqs) are unusable today.

Everything points to this problem I an fixing being the problem with
early ES7000 machines, including the memory of their boot firmware
developers complaint that they were in spec.

Eric