On Tue, Jul 11, 2006 at 11:04:49AM -0700, Stephen Hahn wrote: > I am still stuck on whether "uname -i" is the correct way to publish > "I'm virtualized"--won't any program that calls uname(2) get an > outcome that should be irrelevant?
Any program using uname -i is asking for the platform. It seems perverse to pretend we're on something we're not. The uname -m case has a stronger argument in that uname -p isn't on other platforms; "uname -m" is widely used in the real world, and defining Xen as part of the "machine hardware" is certainly arguable. In particular things like mdb change depending on the platform, and it's the right information level for this IMO. > (That is, shouldn't you allow them to hit the missing or reduced-capability > feature, rather than using a coarse indicator? It's the "uname -r" versus > feature argument, just along a different axis.) This still wouldn't help the static nature of profiles, anyway. We already have a similar check in our service's methods, and if that's the most reasonable thing to do, then fine. And it certainly seems that the profile mechanism still needs to use a uname -p based method. regards john
