On 20.06.2016 08:17, Cédric Le Goater wrote: > On 06/20/2016 04:19 AM, David Gibson wrote: >> On Mon, Jun 20, 2016 at 07:12:38AM +1000, Benjamin Herrenschmidt wrote: >>> On Sun, 2016-06-19 at 19:23 +0200, Cédric Le Goater wrote: >>>>> You can run a 32-bit OS or firmware on ppc64, but it needs to know that >>>>> it's running on a 64-bit chip and do a few things differently. >>>> >>>> yes sure but qemu would still allow rfi under 64bit CPUs, that is what >>>> I was concerned about. Is that ok ? >>> >>> Why ? A real CPU won't allow it, why should we ? >> >> We shouldn't. However, I'm inclined to in for now, until we have an >> OpenBIOS fix actually committed. I'd prefer to keep the existing >> setup sorta-working when the current situation is unlikely to break >> working code, even though it's definitely wrong. >> >> BenH or Cédric, if you want to resend the hrfi fix patch with the >> 64-bit rfi support left in for no, that would be good. > > The current patch does not need fixing. I will send the OpenBIOS patch > shortly after I have looked at the FPU exception. > > Linux ppc behaves the same on a 970. So we will need to fix the 'rfi's > there also.
Really? Wow, that surprises me. That OpenBIOS code likely never ran on a real 970 hardware, so that's not too much surprising that the "rfi" sneaked in there, but the Linux kernel certainly ran on a real 970 once, so I wonder why it's not properly using rfid for the 970 yet? Thomas