Re: [RFC PATCH] x86, fpu: Use eagerfpu by default on all CPUs

From: Andy Lutomirski
Date: Mon Feb 23 2015 - 20:00:12 EST

On Mon, Feb 23, 2015 at 4:56 PM, Maciej W. Rozycki <macro@xxxxxxxxxxxxxx> wrote:
> On Mon, 23 Feb 2015, Linus Torvalds wrote:
>> We have one traditional special case, which actually did something
>> like Maciej's nightmare scenario: the completely broken "FPU errors
>> over irq13" IBM PC/AT FPU linkage.
>> But since we don't actually support old i386 machines any more, we
>> don't really need to care. The only way you can get into that
>> situation is with an external i387. I don't think we need to worry
>> about it.
>> But with the old horrid irq13 error handling, you literally could get
>> into a situation that you got an error "exception" (irq) from the
>> previous state, *after* you had already switched to the new one. We
>> had some code to mitigate the problem, but as mentioned, I don't think
>> it's an issue any more.
> Correct, the horrid hack is gone, it was so horrible (though I understand
> why IBM had to do it with their PC/AT) that back in mid 1990s, some 10
> years after the inception of the problem, Intel felt so compelled to make
> people get the handling of it right as to release a dedicated application
> note: "AP-578 Software and Hardware Considerations for FPU Exception
> Handlers for Intel Architecture Processors", Order Number 243291-002.
> Anyway, my point through this consideration has been about the
> performance benefit from continuing the execution of an x87 instruction in
> parallel, perhaps until after a context has been fully switched. Which
> benefit is lost if a FSAVE/FXSAVE executed eagerly at the context switch
> stalls waiting for the instruction to complete instead.

And the save is indeed executed eagerly. Conceivably we could get rid
of that on UP, but that seems to be a lot of complexity for extremely
little gain.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at