Michal Such=C3=A1nek [off-list ref] writes:

On Tue, 24 Apr 2018 14:15:55 +1000
Michael Ellerman [off-list ref] wrote:

quoted

From: Michal Suchanek <redacted>
=20
Based on the RFI patching. This is required to be able to disable the
speculation barrier.

why do you not patch the nospec barrier which is included as part of
the RFI flush code?

We didn't want to put it in the asm like you had, because not all RFI
flush types need the spec barrier.

So it's more complicated than just patching in the spec barrier, we'd
need to only patch it in if the configured RFI flush also needed it.

I think when debugging the code it would make more sense if RFI is
patched by RFI patcher and nospec by nospec patcher.

True. I think what I'm saying is that the spec barrier in the RFI
flush is not a nospec barrier, it's part of that RFI flush type.

A separate question is if the RFI flush would break without the nospec
barrier.

The ORI flush requires it, but the others don't.

cheers