Junyang Zhang

#713 enables faster VmIO for VmSpace for long buffers. We can bypass VmMappings in this case. And the later designs for process vm management may remove VmMappings for most cases....

I suppose that KVM PCI support is not enabled by kraft. I made it work using Makefile. See [this tutorial](https://github.com/unikraft/docs/blob/cd6589cb6214c5fb95e881e271b7c7c63b2df3d9/content/en/community/hackathons/sessions/behind-scenes/content/Work-Items_06.-Tutorial---Reminder%3A-Adding-Filesystems-to-an-Application.md?plain=1#LL87C34-L87C36).

> Traditionally, global page table frames are kernel page table frames that are shared by all page tables; non-global page table frames are user page table frames that are never...

> @junyang-zh Have the current implementation set the global bit for all kernel-space page table entries? yes

> > 2024/04/18 The scope of this PR is expanded. > > For what purpose? Doesn't the page table refactoring alone already result in a pretty complex PR with thousands...

Ok, this PR is ready now. There's room for further improvement: - use frame metadata to get 2x + space savings; - implement the real copy-on-write page table (currently it...

@tatetian Thanks for the valuable feed back! I'll start fixing these as soon as possible. BTW, the latest commit implements: - the fine grained page table lock protocol; - lighter...

Ok, fixed the first batch. The refactor of page property affects `aster-nix` as well. CI is deemed to fail before #807 gets merged.

> @junyang-zh Does this PR fix #681? Short answer: no. Long answer: we should let framework users use the cursor to access the page table, just like the usage case...

`UserContextAbi` may be another good alternative. Since "which reg is used for arguments" is defined by the ABI.