[PATCH/RFC] PCI Error Recovery

[Andi Kleen]

> I don't want to expose it that way neither, but the fact is we can't
> just have "generic" states that apply to every architecture. I haven't
> yet looked at Linas latest patch though, but at one point, we need to
> define a few states that may or may not apply to a given architecture
> and give enough info to drivers to deal with them as much as they can.
> I'm afraid we can't completely avoid some of the complexity here.

Perhaps, but Linas' version seems to be far too PPC64 centric to me.

It's really not in your interest either to have a too ppc64 specific
solution because it means much additional work to fix drivers for
ppc64 that have been developed on other architectures. 

What's wrong with just simply telling the driver.

an error occurred. all your recent transactions may be broken.
To use the device again call "foo" first to fix the device.
foo then returns if it fixed the device or not.

I don't get why the driver even needs to know about isolation
or not. It's not fundamentally different from an bus abort
on other systems, just that it lasts longer. 

-Andi