I see that Keir fixed all of these. Thanks!

The balloon driver in xenlinux was also fixed, but I think this is a syntactic but not a semantic fix. In balloon_init, an assumption is made that all pages below xen_start_info->nr_pages are valid RAM. What if all pages of valid RAM are above nr_pages? Then (I think) the balloon driver will eat up all the domain's RAM. (Unlikely, but you get the drift.)

Or perhaps this assumption is always true on xenlinux/x86 (including PAE and x86_64 and later with VIRTUAL_MEM_MAP)?

I'm still working out what this should be on ia64, but the current code definitely won't work. Is it just trying to allocate N pages where N = max_reservation - initial_reservation? If so, why not just call alloc_page N times? If this works, it is probably shareable with ia64, else perhaps we will need an arch_balloon_init() (and an asm/balloon.h) otherwise...

On a related note, I'm not clear on the relationship between the "memory" variable in xmdefconfig and the two reservation variables needed for ballooning (not sure what they are named, but call them max_reservation and initial_reservation). Which of these latter correspond to "memory" in xmdefconfig and how does the other get determined?

Thanks, Dan