Add support for non-contiguous physical page mapping by sangho2 · Pull Request #666 · microsoft/litebox

sangho2 · 2026-02-18T00:55:54Z

This PR introduces a vmap subsystem to the LVBS platform to map non-contiguous physical page frames into virtually contiguous addresses.

Changes

Add map_non_contiguous_phys_frames which is similar to existing map_phys_frame_range except that this new function is for non-contiguous physical page frames.
Reserve a 1 TB VA window for vmap and vunmap
Maintain separate hash tables for tracking PA<->VA for VMAPs (mm/vmap.rs)
Support cross-core TLB flush

wdcui

Over the code looks good to me. I left some comments below.

wdcui · 2026-02-19T20:26:04Z

litebox_platform_lvbs/src/arch/x86/mm/paging.rs

+                            &mut inner,
+                            Page::range_inclusive(
+                                start_page,
+                                start_page + (mapped_count as u64 - 1),


why mapped_count - 1?

rollback_mapped_pages internally calls x86_64 crate's clean_up_addr_range which expects an inclusive range. we can use an exclusive range for rollback_mapped_pages if that is intuitive.

wdcui · 2026-02-19T20:28:44Z

litebox_platform_lvbs/src/arch/x86/mm/paging.rs

+                && FLUSH_TLB
+            {
+                fl.flush();
+            }


Should we log the error at least?

You mean whether we rollbacked mappings? We can log the error, but rollback_mapped_pages is already within an error path.

wdcui · 2026-02-19T20:35:30Z

litebox_platform_lvbs/src/mm/vmap.rs

+    ///
+    /// Returns `Some(VirtAddr)` with the starting virtual address on success,
+    /// or `None` if insufficient virtual address space is available.
+    fn allocate_va_range(&mut self, num_pages: usize) -> Option<VirtAddr> {


Do we need to support address alignment?

Functions in this module are called through vmap which has type-based strict alignment requirement.

wdcui · 2026-02-19T20:38:09Z

litebox_platform_lvbs/src/mm/vmap.rs

+
+        // Try to find a suitable range in the free set (first-fit)
+        for range in self.free_set.iter() {
+            if range.end - range.start >= size {


Here the range's size is measured by bytes, and other places a va range is measured by pages. It's a little confusing.

make sense. let me see whether I can change it.

wdcui · 2026-02-19T20:45:13Z

litebox_platform_lvbs/src/lib.rs

+            }
+        }
+
+        if ALIGN != PAGE_SIZE {


Nit: check alignment before the more expensive loop?

wdcui · 2026-02-19T20:48:18Z

litebox_platform_lvbs/src/mm/vmap.rs

+/// Start of the vmap virtual address region.
+/// This address is chosen to be within the 4-level paging canonical address space
+/// and not conflict with VTL1's direct-mapped physical memory.
+const VMAP_START: u64 = 0x6000_0000_0000;


How do we prevent the page table code from using this va range?

By default, we do use a global offset when we map physical page frames. Unless our target machine has an extreme amount of memory, there are no overlap. We can add extra check to the normal map function if needed.

litebox_platform_lvbs/src/arch/x86/mm/paging.rs

github-actions · 2026-02-21T17:31:58Z

🤖 SemverChecks 🤖 ⚠️ Potential breaking API changes detected ⚠️

Click for details

--- failure enum_variant_missing: pub enum variant removed or renamed ---

Description:
A publicly-visible enum has at least one variant that is no longer available under its prior name. It may have been renamed or removed entirely.
        ref: https://doc.rust-lang.org/cargo/reference/semver.html#item-remove
       impl: https://github.com/obi1kenobi/cargo-semver-checks/tree/v0.46.0/src/lints/enum_variant_missing.ron

Failed in:
  variant PhysPointerError::NonContiguousPages, previously in file /home/runner/work/litebox/litebox/target/semver-checks/git-main/48693f9106fd64135d69f97c954a376d5bd51c97/litebox_common_linux/src/vmap.rs:170

sangho2 marked this pull request as ready for review February 18, 2026 00:58

sangho2 force-pushed the sanghle/lvbs/vmap2 branch 4 times, most recently from debb8fe to 2ed4daf Compare February 19, 2026 04:03

wdcui reviewed Feb 19, 2026

View reviewed changes

sangho2 force-pushed the sanghle/lvbs/vmap2 branch from ad76dac to 4d06a28 Compare February 19, 2026 22:23

sangho2 commented Feb 19, 2026

View reviewed changes

litebox_platform_lvbs/src/arch/x86/mm/paging.rs Outdated Show resolved Hide resolved

sangho2 force-pushed the sanghle/lvbs/vmap2 branch 2 times, most recently from 2e1da17 to 5ccdbae Compare February 21, 2026 00:07

Sangho Lee added 6 commits February 21, 2026 17:05

improve vmap to map non-contiguous physical pages

1ba5280

refactoring

8d975f0

refactoring

4ab6d64

addressed some feedbacks

7fd5d3e

addressed few other comments

13c933a

cross-core TLB flush within VTL1

4e70ee8

sangho2 force-pushed the sanghle/lvbs/vmap2 branch from 5ccdbae to 5407404 Compare February 21, 2026 17:05

support more than 64 VPs

c4e288d

sangho2 force-pushed the sanghle/lvbs/vmap2 branch from 5407404 to c4e288d Compare February 21, 2026 17:24

Comments

Conversation

sangho2 commented Feb 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

wdcui left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions bot commented Feb 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

sangho2 commented Feb 18, 2026 •

edited

Loading