Re: [Qemu-devel] [PATCH] mips: Correctly save/restore the FP flush-to-zero state

[Maciej W. Rozycki]

On Fri, 5 Dec 2014, Leon Alrae wrote:

> >  I gave it a thought before making this change and concluded it would be 
> > the lesser evil (plus loudly manifesting and easily correctable) if 
> > someone accidentally makes QEMU refuse to load older images where in 
> > fact no compatibility issue exists, than if the reverse is the case, 
> > that is older incompatible images are accepted where they should not 
> > (causing a silent misinterpretation of data), simply because someone 
> > missed the need to change the condition in addition to bumping up 
> > CPU_SAVE_VERSION.  WDYT?
> 
> Finally I got round to reviewing v2 of this patch. Above sounds
> reasonable for me and I haven't seen any objections.
> Hopefully we will convert it into VMState during 2.3 development (Peter,
> thanks for pointing at the commit, it will certainly help).
> 
> Maciej, the only question I have is why you removed only some of "if
> (version_id >= X)" lines in machine.c?

 Argh, because I missed them.  I regenerated the original change and 
didn't think of checking if additional conditionals haven't been added 
upstream since the inception of the original change.  Sorry about that.  
I'll review all the relevant patches, see what has to be updated and 
resend.  Thanks for pointing this out.

 I'll give a priority to the 2008-NaN change though whose final updated 
version has just passed regression-testing (13+hrs per run).  It 
actually updates this save/restore logic here again and introduces the 
CPU_SAVE_VERSION_OLDEST_SUPPORTED macro I proposed here, to support 
older legacy-NaN images that are compatible with the 2008-NaN update.

  Maciej