On Tue, 2006-07-25 at 18:38 -0700, Jeremy Fitzhardinge wrote: > Rusty Russell wrote: > > There are several issues to answer here. First, that "set_fixaddr_top" > > above would need to be in an #ifdef or otherwise wrapped for the > > non-paravirt case. > > > > Well, that's simple and self-contained. Besides, by the time we're done > there won't be a need for !PARAVIRT ;) We still have CONFIG_X86_GENERIC and !CONFIG_SMP, so I think that there'll be a !CONFIG_PARAVIRT for a while. Of course, we'll sneak a patch in to turn it on by default at some stage 8) > > Secondly, *if* the only thing that will actually alter the fixaddr_top > > is paravirt, it makes sense to keep it a constant except in the paravirt > > case, ie: > > > > Gerd's original patch also supported a kernel parameter to set it. Was > that for anything in particular? I thought it was merely for testing, and since it will fight badly with paravirt_ops anyway, is not likely to survive long term. > But it still seems to me it could be a variable either way. Are there > any performance-critical users of it? Even so, changing a compile time constant to an unchanging variable is unlikely to be measurable. I guess I feel we're being a little dishonest changing it generically to a variable if it's just for our use. But I can be persuaded if people feel strongly about it... Cheers, Rusty. -- Help! Save Australia from the worst of the DMCA: http://linux.org.au/law
