What are the design internals behind zap_xxx_range() APIs?

Rik van Riel to Wu, kernelnewbies

On Sun, 30 Mar 2008 17:55:26 +0800 "Wu Yu" < vestige.lug@gmail.com > wrote:

> Hi all,

> AFAIK, for IA-32 arch, two-level paging is sufficient. But the

> positions of PUD and PMD are kept.

This is done so all architectures can use the same iterative functions, like unmap_page_range -> zap_pud_range -> zap_pmd_range -> zap_pte_range.

> I don't quite understand the word "mapping". Does that mean there > is one entry in PGD for both PUD and PMD?

It means that the kernel pretends that the pgd entry points to a pud (with one entry), even though it really points to a page table.

By pretenting it points to itself unmap_page_range can call the next function down, zap_pud_range.

We do the same thing in zap_pud_range - it calls zap_pmd_range on the same pgd entry, except now it thinks it points to a pmd.

Finally, zap_pmd_range interprets the entry and sees that it points to a page table. Zap_pmd_range invokes zap_pte_range and everything works as normal.

Because, on i386, we pretend that the PUD and PMD page tables only have 1 entry, the loop does not loop, but simply invokes the next function down once.