[Optimize] Preload user pgd in load_pgtbl hypercall

[zhuangel]

Currently, PVM relies on PTI is enabled inside VM. This means each process inside guest has 2 cr3, and there are 2 PGD in SPT of PVM.

PVM implements a hypercall (PVM_HC_LOAD_PGTBL) to change pagetable of SPT. And PVM is maintaining a pair of process usr/sys CR3s and switch flags, so that PVM could do direct switch to speed up guest process usr/sys switch without return to hypervisor.

We can speed up sys to usr switch by preloading the user pgd in PVM_HC_LOAD_PGTBL.

Here are some lmbench context switch case test results. Using preload user pgd feature, context switch latency could reduce 10%+, if we apply the enlarge the SPT LRU cache patch set, we can speed up context switching in multiprocess environment.

process num       2        3
w/o preload     3.952    5.344
w/  preload     3.555    4.517