Some background:
We are now debugging the Qemu map cache issue for HVM
PAE guest on PAE host.
We found that a hardcode limit for shadow2 which we
can only support about 7G memory for HVM PAE guest.
With the attached workaround and the Qemu map cache
patch, we can now support almost 16G memory PAE guest.
But it does of course decrease the virtual address
space which Xen can map.
So it is a question now: Is 7G memory good enough for
current PAE HVM guest?
And by the way, if a guest requests too much memory,
Xen should check it first before creation.
Thanks
Xiaohui/Dexuan