On Sun, Apr 27, 2014 at 01:45:24PM -0400, Kevin O'Connor wrote: > On Sun, Apr 27, 2014 at 08:10:39AM +0200, Paolo Bonzini wrote: > > Il 26/04/2014 21:36, Kevin O'Connor ha scritto: > > >Yes, I was thinking of something like that as well. If QEMU > > >internally observes the formula above, then something like the patch > > >below should work instead of my original patch. > > > > > >However, I'm not an expert on QEMU TCG and the patch below would > > >require much more testing. > > > > Yeah, the patch is obviously more complex. On the other hand as you > > point out the code to set hflags was already relying on correct > > eflags as a precondition. > > Looking at this a little closer, I don't see any code generation that > depends on the cpl in hflags. So, maybe another way to fix the root > problem is to just remove cpl from hflags and change all the code that > needs the cpl to call a function which calculates it..
Ignore my mail above - the cpl is used in the translated code - I just didn't originally see how it got there. Sorry for the noise, -Kevin
