I saw articles (link below) by Ingo Molnar, a famous Linux Kernel enthusiast where he discusses 4GB/4GB split.

In the following paragraphs he states that ..

"using the 4G/4G patch, the kernel could be put together in 4G/4G mode, by which situation there is a full, separate 4GB VM for that kernel, and you will find separate full (and per-process) 4GB VMs for user-space."

My problem begins here ..

1) My assumption is at a monolithic kernel like as with Linux the Kernel and User a part of process shares same PROCESS ADDRESS SPACE. So having a linux kernel without this 4g/4g patch, we've 3/1 split for user virtual space and kernel virtual space correspondingly. Actually kernel's 1gb virtual space is planned in each and every processes' virtual address space. Place it simple, PROCESS ADDRESS SPACE is 4gb, out which 1gb is alloted to kernel. Could it be true ? Please comment.

2) Going through the content under consideration - and when my assumption as stated in 1) is appropriate - does he imply that the kernel address space isn't any more planned into PROCESS ADDRESS SPACE and there exists a different virtual address space for kernel and there's no discussing of address space. That's PROCESS ADDRESS SPACE is 4gb and KERNEL ADDRESS SPACE is yet another 4gb as opposed to the proven fact that PROCESS ADDRESS SPACE has enlarged being 8gb out which 4gb is shared for kernel.


I understand I'm sounding bit stupid here :).

Basically comprehend the article properly, the kernel and also the userspace don't share exactly the same address space following the patch. Nevertheless this costs switching page tables throughout each userspace/system switch.